Oracle Aktie

Great to see everybody. I personally am very excited about this meeting, as you may know. So let's dive in.

Knowing this is my last meeting with you all. My wife asked me over the weekend. She goes, "What are you going to miss most about being at Oracle?" And I thought about it. I know what it is. So you don't want to know what I'm going to miss most about being at Oracle? Yes. Safe harbor slides. My friend, the safe harbor slide. Those of you who've been here before, you know that I always like to have just a little bit of fun with the safe harbor slide. But kidding aside, I'm showing these slides because you're not going to see them in the decks later on as we go through the day. It is our safe harbor slide. And in there, really what we're trying to tell you is that we're going to be making some forward-looking statements today. And I just want to remind you that these are forward-looking statements. And to be clear, as forward-looking statements, there are risks and uncertainties that go around those statements. Now if you'd like to know more about those, you want to open up any of our 10-Qs, 10-Ks, look in the Risk Factors section and you get a lot more information about that. If we see circumstances change in the future, I just want you to understand that the forward-looking statements that we're making today, we're not obligating ourselves to refresh those at a later point in time, okay?

For Hilary, she will be using a lot of financials. In that presentation, she will be making use of non-GAAP measures. Just want to make sure you're clear that she will be using those and they're non-GAAP. Okay?

And then finally, more from the product people and the services people. Just to understand that today's presentations are just really for informational purposes for you, and they're not any type of commitment to deliver on any material code functionality that we're discussing today.

With that, let's talk about the agenda. We're going to start with Mike who's going to come on stage shortly. He's back here, ready to go. Mahesh will then come up and talk about infrastructure and database. Chris Leone will come up and talk about enterprise applications. Samon will -- Simon will follow with industry applications in NetSuite. Mark Hura, I see, is here, we've got a couple of customers. We'll be having come up on stage as well, talk about customers and how they're actually using Oracle technology in their businesses, and then we'll take a break.

I think that's going to be around 11. So just if you're thinking about when, about then. Coming back from the break, Hilary will come on stage and give you the CFO perspective, give you some sense of how we're expecting things to play out over the next few years. And then Hilary will have Clay and Mike join her on stage for Q&A, okay?

Following the event. So somewhere around 1-ish, the break will be about 11, will finish around 1, and then there would be a box lunches for you in the back, where you had breakfast. And you're welcome to stay in this room if you'd like, up to 2:00. I know that we've got some companies announcing, if I remember right, Microsoft, ServiceNow. And so if you want to work from here from 1 to 2, you're more than welcome to do so. So with that, let's get going. And please, let's roll a video.

[Presentation]

Good morning, everyone. I've had the good privilege of hearing Ken read that safe harbor for 18 years in person, and I'm certainly going to miss it. It's been a timeless classic like the voice of the monorail at Disney World. It's going to be hard to get used to a new safe harbor read. But anyway, thanks for being with us here today. I hope you had some time to enjoy both AI World and Las Vegas.

We find ourselves at a very interesting time. And throughout our history at Oracle, we've never been afraid to make bold moves, as many of you know who have been covering the company for so long. Five decades ago or so, Larry saw that potential for relational databases, to not just change but fundamentally change how businesses organize, manage and leverage information. And along the way, we turn that technology into the foundation for modern enterprise computing. We then expanded into applications, infrastructure cloud. We've made many significant investments. Always try to challenge conventional thinking, always try to be different and differentiated and therefore, built the technologies that some of the world's most important businesses and governments depend on every single day.

So, in so doing, we weren't just responding to at any point in time where technology was at that time. We're building for where we believe the technology needs to go. And in the age of AI, all of that matters even more. First and foremost, it's going to make it easier for customers to move to modern cloud platforms through the automation of go-live services. We think about the time, effort and capital that it takes to move large mission-critical systems to modern cloud systems. Anything that we can do to accelerate that has always been appreciated. We can now dramatically accelerate that. That's going to be a fundamental unlock not only for our growth in the applications business, but also a terrific value return for our customers and allow them to get time to value far more quickly. And second, once our customers get to the modern platforms, they get the AI built right in. You've heard that, and you'll hear that again from Chris and Simon, as we talk about the sheer amount of AI features that we're putting into our application platforms and all of our platforms every single quarter as we release these products. In fact, we believe that AI has the potential to fundamentally change, just like the database, fundamentally changed enterprise computing. What's possible for businesses, for people and for the world by solving problems that have challenged us for generations. You heard in the video, Larry talked about curing cancer, which I think is not at all far-fetched. I hear it from our customers, our hospital systems, our life sciences customers. They tell me things like, I think we have the opportunity to, at a minimum, turn cancer into something that we can live with. It's not a death sentence. But we wouldn't be able to do that without the Oracle technology. So if we can detect cancer years earlier, we can rethink how we produce food, energy and the things that our world depends on, unlocking really extraordinary new levels of productivity and economic growth. Because of our full stack advantage, we have never been in a better position to deliver that to customers. It's -- we find ourselves in a very unique, I think, position to benefit humanity. And when we do that, we want to be careful that it doesn't seem like virtue signaling. These are real deliverable outcomes for customers, food, agriculture, health care, that are going to some problems that previously could not be solved. So think about health care, for example, whereas there's a lot of positive sentiment around health care, AI and the intersection of clinical research and clinical care and why the full stack matters. That requires high-performance, secured infrastructure with very close provisioning to the AI models. It also requires the aggregation and vectorization of hundreds of thousands of health care data sets. Further, it requires purpose-built AI applications that face researchers, patients and providers. I know of no other company besides Oracle Corporation that's providing every single piece of that. And if we think about taking -- Larry spoke a lot yesterday about robotic surgery and think about the complications of taking robotic surgery at scale into the operating room. First and foremost, patient safety is paramount. Secondly, from an opportunity standpoint, it allows you to unlock and potentially perform surgeries that were impossible before. In order to deliver that completely and safely, we need the entire stack. We need scale GPU infrastructure. We need scaled AI infrastructure, the connection to that infrastructure from the models and all the inferencing and all the referencing that's going to make, that's going to be unique to that particular procedure and then that particular patient requires near-zero latency permission -- provisioning to those models. You can't have a gap, can't have a gap. We can't have a time out. We can't have a small gap in latency or a small increase in latency. We're in the middle of a very complicated surgery. We need clinical applications. We need vector services. We need IoT edge devices like the things that we unveiled in [ an ambulance ]. You need all of it. Today, that is happening with little stovepipes, piecemeal, pieces that people are gluing together. It's inefficient. It's very costly. It's not exactly secure, and it doesn't allow us to create a scaled infrastructure that can democratize health care. So maybe the top clinical research organizations in the United States can afford that, community hospitals can't. They can't do it because they can't buy it as a service. They can't take it as a service from a vendor because nobody is supplying that. We have a unique opportunity to do that. Now I can go industry by industry. I can give lots of similar examples, but the point remains. Our full stack provider provides an outsized and differentiated results and there's no other scale provider beyond Oracle who is delivering that. If you think about these examples, if any piece of the stack doesn't exist, we can't deliver that as a turnkey service because somebody else has to come in to be able to do that. Somebody else has to come in to connect it. Somebody else has to implement it. That's a tremendous difference. And when outcomes matter and they will in AI, I think we should be in a really good position to deliver. The way I think about it is that AI then is the catalyst that's really activating the full potential of everything that we build. In fact, it's making every part of our portfolio more valuable. So we don't think about it just as the next chapter of cloud. It's a fundamental expansion of the opportunity, and we really believe that we're built to lead it. And as AI becomes more pervasive, every layer of our stack becomes both an enabler and a beneficiary. Enterprise AI powered by OCI becomes exponentially more useful when it's connected to enterprise data. Oracle Database and AI Data platform provide a foundation for securely managing and accessing business critical, mission-critical, highly sensitive data. Oracle applications already run critical business processes. They understand workflows, roles, permissions, business rules and industry requirements. These aren't separate AI stories. These are the conditions that we believe will allow AI to thrive. So a few recent shifts, as you know, things are evolving very quickly. And there have been a few shifts that I'd like to just briefly discover and how I think our stack is aligned for this moment. So the first shift, and you heard Clay covered this in the keynote yesterday, if you had the chance to see it, is around trust and security. So as AI moves some answering questions to taking actions, the stakes become much higher. Think about an AI agent making a financial decision, executing a transaction or accessing sensitive employee information. Those agents need to operate with the same security, the same permissions, the same business rules that govern the enterprise.

An AI agent going rogue is not an unfortunate side effect in many businesses and many missions, it's simply not acceptable. It's not something that we can accept. Our integrated stack helps secure that experience. It helps keep the agents on the rails. Since we've spent many decades protecting the world's most critical enterprise data, we engineer security into the infrastructure, protect the data at the database lever and enforce the business rules and permissions with our applications. These are the governance rules that are already in place to run mission critical systems. We're embedding AI right into those rules to keep it on the rails, to keep it safe, to keep it secure. So rather than relying on a bunch of separate systems to establish those controls, we have designed them to all work together. And as AI enterprise -- as enterprises give AI more responsibility, I think that becomes increasingly important.

Okay. So another thing that's happening in shifts is the way that end users, customers are interacting with technology. Increasingly, people are starting their work with AI. They're asking questions, they're making decisions, getting things done through intelligent assistance and agents, not necessarily traditional use -- what we would call traditional user interfaces. Those agents need a very powerful foundation. They need to access the right data. They need to understand their business processes. They need to know exactly what actions they're authorized to take and they need the infrastructure to connect all of that very efficiency -- efficiently. Our stack also powers that experience. For example, in NetSuite Next, we announced this week, AI is the UI. Workflows are aggregated by agents, operating right within the system, not outside. So AI is the UI. Fusion Claw makes AI the UI not just for a single app, but across many applications. Be they Oracle apps or be they somebody else's app, doesn't matter. AI is the UI. Fusion Claw is the orchestration layer and AI UI to enable these outcomes. So this means that AI can become the UI for our customers, for business outcomes regardless of the underlying systems. It doesn't matter. And AI becomes the trusted interface and the partner for getting work done faster and more efficiently. That is quite unique to Oracle's full stack offering. In fact, again, if we're not delivering the full stack, these problems get much harder to solve.

All right. And then the third major shift is moving from what I'll loosely call AI experimentation to now real AI outcomes. Customers don't just want to see tokens being consumed. In fact, there's a pushback on token mixing. They want really AI to provide gains and productivity. They want to lower cost. They want faster decisions. They want more efficient operations. They want to achieve the expected ROI from -- it's not -- it's about moving from reactive information to helping businesses proactively take action. And the full stack, we think makes a tremendous difference in delivering on this benefit and producing financial results. So I think about example of a customer in a mining operation where a freight train is hauling tons of thousands of material, and there's break failure. We've got IoT that detect that break failure. And in that moment, those milliseconds matter. Because the immediate safety response needs to happen locally, but through specialized control systems that act in real time on the edge, but then what? What's the rest of the story? How quickly can the train be repaired? Do we have the necessary parts? How do we adjust production? How do we minimize downtime? So that's where an integrated architecture creates tremendous value, infrastructure data, applications and AI, all working together to connect the immediate response with the broader business context. The same principles apply, whether you're managing a complex supply chain or supporting critical health care operations or rebalancing your workplace. It doesn't really matter. It's not just about reducing technical latency. It's about accelerating the path from intelligence to action. Again, we are uniquely positioned to do all of this with optimized performance across our entire platform. And our customer base, is leveraging AI. We're seeing AI adoption accelerate with very tangible benefits for customers. So as an example, I'll give you some recent stats on Fusion AI adoption. In Q1 of this fiscal year, so we're in Q2 now. In Q1 of this fiscal year, customers leveraged our embedded Fusion AI capabilities more than 150 million times with usage growing 42% just from June to August. So 150 million times customers made AI calls in our Fusion applications just in the quarter. In September of this year, AI usage across Fusion consumed nearly 500 billion tokens just in September. Month-over-month, that's almost a 20 -- it's almost a 23% increase, 22.7% increase to be exact, month-over-month increase from August to September. I didn't know that, that was possible, right? I had a double check with Chris, but 500 billion tokens were consumed in Fusion Applications in one month alone. Well, somebody is using AI out there at scale because that's a lot of tokens to consume.

And our products, the number of agents that we have deployed in our products have grown from over 1,000 in May to 3,500 in September. By the way, that's in the same year in the same calendar year. We had 1,000 agents live in May. We have 3,500 live in September in our Fusion Applications stack. Stunning progress and stunning consumption of AI functionality. We see major brands like DHL, Avis, Orange and ADT all expanding their use of Fusion Applications and AI. We expect adoption trends to significantly accelerate with our existing cloud customers, given many of the technology deployments that we've unveiled here this week at AI World. We're also expecting further adoption as our on-premises customers can now move far more easily to the cloud by leveraging the AI tool set that we announced this week that automate migrations and decrease the cost and time to implement. I actually think -- and we spend a lot of time internally discussing this. This is just as important as all of the features, all of the AI features that we're putting into all of our applications. Time to value, time to live, automating migrations of on-premises customers. By the way, they don't have to be Oracle customers. They can be a competitor of customers. What we're unveiling here to automate migrations to the Fusion stack, to the industry stacks and the NetSuite stack, doesn't have to be an Oracle workload. This is for everybody to consume. That has been a big headwind. When you had early adopters that moved to the cloud, there were a lot of folks that got on boarded. And then you get into a little bit of a stall period because the time, effort and energy to make that happen becomes sometimes not palatable for certain organizations. I think that goes away. I expect that our large remaining base of on-premises customers that have been eagerly awaiting these developments are going to start to adopt this technology quite quickly.

All right. And then as a benefit, as a reminder, when we move customers from on-premises to cloud, we see a multiple increase in revenue. It's a good business for us to be in. Cloud is a very good business for us to be in. And it's good for our customers because we put them on a modern platform, we deliver AI out of the box. And obviously, from a cybersecurity standpoint, we put them in a much safer spot. So I also think that in addition to our Applications business, we're poised to capture market share across our entire technology stack from top to bottom. So that's a bit about customer outcomes. I want to also talk about our shareholder -- taking good care of our shareholders as well because that's just as important for us, and it's something that we think about. So we have -- as we think about our investments, we think about differentiated growth. We're different on purpose. We offer a full stack, and we think that creates a tremendous advantage. We're committed to strong profitability and strong free cash flows and, we think, very respectable EPS results as well. We're also committed to disciplined capital allocation, understanding that we're making the most efficient use of our capital to create the most amount of return for both our customers and our shareholders simultaneously. So these are things that we are spending a lot of time internally as we think about investments, and we think about building things. We're making sure that everything that we do internally aligns to these three pillars to create and accelerate value for our shareholders that is extremely important for everybody at Oracle, and we all know that this is a mission.

Hilary is going to share a lot more detail on our targets, and our commitments after you hear a little bit more about some of our business investments that we're making, and I'm sure you won't want to miss that session as you'll see those targets and commitments lining up to these three pillars. But we wanted to share that incredibly important to us. We're very excited about what we're what we're building. We're very excited about what we're delivering for our customers. We're also very excited for our shareholders because we think that's -- what we're building here is going to create fantastic returns for all of our investors. So -- thanks again for being here. Thanks for covering us. Thanks for always pushing to make us better and giving us all of your feedback, and I really hope that you enjoy the rest of the day, and I'll see you back up on stage a little bit later. Thank you.

[Presentation]

Please welcome to the stage, Mahesh Thiagarajan.

All right. Good morning, everyone. Cool. What I'm going to talk about today is it's very clear that the next phase of AI is going to be about who gets faster access to compute, and how can customers generate the most optimized TCO for the outcomes that they actually want to drive. So a lot of my session today is going to be about popping the hood, talking about the engine. If you're some of the people who don't like infrastructure, you're probably going to have a very bad time. But let's dive right in.

So as you know, this is data. This is probably data that's already available. So this is the last 6 quarters of sequential growth that OCI has actually shown. And what I'm going to do today is explain the set of things that we've actually done over the last year and why we believe this is accelerating, why we believe we've scaled this out enormously; third, how fungibility works at the infrastructure layer across multiple customers; and lastly, talk about the durability and the differentiation on what we're building.

So I want to start from the bottom of the stack. Let's start with power, right? When you think about the power and the challenge with it is, first, moving to locations where you can actually get access to energy resources. And the first challenge about moving to a place where you need access means that you need to build your cloud right next to that place. In Oracle, at OCI, we've built a foundational differentiator of building our regions in less than 30 days as soon as the power is available and the data store is available. That gives us an edge, an opportunity, an advantage to go to places where the power exists.

The second, you go to places where there's energy but you need to build the power and the energy plants. And so a lot of what we've done is, first, we were the first hyperscaler at Abilene who decide who use gas turbine-based power generation for our behind-the-meter -- as behind-the-meter solution for a backup power. And this was in 2024, we're talking 2 years ago.

Second, when we picked our next gigawatt scale data center, once again in Texas, where there's a lot of natural gas resources. We selected VoltaGrid, which is actually a reciprocating engine-based solution.

Third, when we went to New Mexico and we looked at feedback from our -- the community, we actually pivoted and evolved to building a gigawatt scale fuel cell solution. And when you think about all of these different solutions, people inherently, the AI workload has different sort of challenges. So it's easy to say, hey, we need to go build all of these solutions and put it together, it doesn't actually work. You have to understand the challenge that comes with it. First, when you think about AI workloads, they operate as one giant AI supercomputer as one computer, which means when something breaks or the workload has stopped, the power drops. And then when it ramps back up, it ramps back up pretty fast. So the challenge of handling this across various different power sources is a unique challenge, but that's where Oracle differentiates itself. We are in the business of accelerating the delivery of our compute. And so we have done tremendous amount of engineering work with our partners and many different solutions around existing technologies that exist like UPS to make sure this power comes online, and it's ready for consumption. It doesn't mean that we are building all of our data centers with fuel cells or other solutions, we do connect with the grid, but we have a very simple principle in terms of how we connect to the grid, where we take -- where we say, the grid -- the way we consume the power from the grid should look identical to their being 1,000 homes. And that requires a tremendous amount of engineering and building technology that ensures that the grid interface is presented in a very smart way. I'm going to take a simple example. So when you think about that load ramped up and down, we have what I call is like the UPS systems that exist to handle sort of the swings in any data center. What we did is we went to those UPS systems, we understood the profile of an AI load, tweaked the software and the firmware that runs on the UPS systems to essentially protect and present an interface to the grid when these load ramps up down, it doesn't actually see. The great part about that technology is it doesn't apply just only to the grid, but it applies to the gas turbines, it applies to the fuel cells. So a lot of this is foundational technology that we innovate even on something as simple as power. So you build all of these solutions, but then how do you know it's actually going to work. We do not take that for granted. So what you're seeing is a real 10-megawatt AI power loading low testing simulator that is right now deployed and testing our replicating engine solution at Shackelford, Texas. And this is a solution that we ensure whenever we deliver that power, it actually comes out and is ready to start running compute immediately. And we can do this with our partners. And so we have a ton of partners both from grid providers and the energy solution providers like VoltaGrid and Bloom Energy, to bring all of this online on time and accelerate it.

Another interesting opportunity or a challenge is working and figuring out how do we explain the responsible data center to the communities, right? And we have a 5-pronged approach in terms of what we do across our data centers, and we explain it to the community. First, we create good-paying jobs. Two, we prove with facts. For example, at Abilene, we've actually taken local talent, converted them -- over 200 people who have not done like we're just from high school or community colleges and employed them at Abilene. That story resonates with the various communities that we're part of. You saw a closed-loop liquid cooling video. We employ innovation and technology to use closed-loop liquid cooling systems, not just for the data center, in many cases, like the fuel cells, even for the fuel cells that ensures we're not using the water, which actually helps in our community strategy. Third, we invest and engage in the community. And lastly, we actually drive accountability. So one of the projects that we announced is about the transparency initiative where we're transparently going to share usage of air, water, heat, light and everything on a public dashboard is very confident about our technology.

So we talked about power and what we're doing to bring power online, stable power, reliable power online for AI workloads. Next, I want to talk about what we do in the data center side of the house. For me, it's a few things. One, when you think about an entire construction project and the set of things that you do to build a house is one task. The second task is going inside the house and actually remodeling it the way you want it. One of the fundamental things that we did in the last 2 years is to not think of them as two projects, but think of them as one project. So we found the hardest path and complete parallelization of doing both of these activities together. Second thing that we did is to ensure all of our data centers are modular, so you can actually repeatably stamp all of those designs across multiple different data centers. So you're not taking a longer time to do supply chain, longer time to finish the designs or get it approved. The third thing that we do is we design those data centers to be flexible, independent of the compute architectures that we're going to apply. As you heard, and I'll talk a little bit about it, we have many different types of hardware platforms that have come online. But we may not know at the time of the decision making, what hardware platform is going to be there. So our data center designs are innovative to be very, very flexible.

And the last thing that I want to remind is that these data centers are faced, right? When you think about a large data center, it's not like magically 1 gigawatt of data center show up. I wish it did. It would be amazing. I don't have a job. But it does move in very small phases over a 1.5-, 2-year period of time. So I just wanted you guys to understand that.

With that, I'm going to show you some of the modular work that we did in one of our data centers. Can we play the video?

[Presentation]

All right. So what you guys just saw was a ton of pre-fabricated modular designs that allows us to go air drop a lot of these things and move them into a position, so it enables us to deliver these data centers really fast. The second part of our -- once you get the power and the data center, the next thing to think about is the rack, the hardware design and a lot of that work that needs to happen. But one of the things that we have done over the last year is significantly scale out our supply chain. When I talk about supply chain, everybody starts thinking about only the few accelerators or the CPUs, but I'm talking about every small little thing that goes into the rack. Just to again to give you an idea, there's about 6,000 components that go into a single GB300 rack. And it is our job to completely scale out every single one of them across multiple diverse partners for every component. And then you go down the stack at the data center supplier level, at the MEP supplier level, we're completely diversified with many channels that ensures our rack delivery will be on time.

The next part of it is manufacturing capacity. I'm going to take one example about driving capacity is one element, but it's about the quality of the racks that come out of it, and they get ready to work in the data center. So more recently, for Vera Rubin racks, Oracle took over the software that runs in their manufacturing floor to give us that precision and scale that is required. So when the rack show up of the data center, it starts working immediately.

The other aspect is in the last year, we've doubled the capacity, but at the same time, we've tripled our GPU rack shipments without any delays.

Another number that I want to talk about is looking back at Q1 FY '27, Clay in the earnings call, talked about how we delivered about 850 megawatts in a single quarter. But if I pop the hood and look inside, that 850 megawatts delivery in a single quarter was about 38 different projects. And it was happening at in about 11 countries globally. And only one of them was our Abilene data center. So just to kind of give you a sense of the scale and the global nature of what we do and how we deliver data centers, that's the magnitude we're talking about.

I want to take another data and an example to talk about a sample customer from customer contract signing to revenue generation. There's a lot of details here, and I'll let you take the data. But the thing that I want to point out is the 21 days. The time is completely optimized in our infrastructure rack delivery and cabling and making sure the customer acceptance goes really, really well. We were able to achieve that in about 21 days from the time that we were able to get into the data center to the time that those racks started generating revenue. And you cannot achieve this without the optimization of the network, a tremendous amount of optimization on the rack manufacturing, making sure the rack show up on time and the quality of those racks, the quality of the power that is going to be on the data center, all of those come together into these 21 days.

And so just to recap on FY '27, Q1 numbers. So we talked about 850 megawatts. It's roughly 8.5 megawatts of power delivered per day, and that is 4x as larger compared to the prior year. And all of this work that I kind of talked about essentially enables us to deliver on an accelerated pace.

And the scale of doing all of this and moving really fast only matters if it works for the customer. And so our gigawatt scale Abilene data center, we have customers -- in our customer OpenAI, accepting the capacity in less than 24 hours, and the availability has been over 99%.

Let's talk about fungibility. So when we think about these AI super clusters, they come in various shapes and forms and Oracle has the ability and the network designs to achieve this. But I want to remind one thing. If you take a single 512 GPU cluster, we were the first in the industry, and we created this for Exadata, by the way. But we're the first in the industry to actually bring security onto our multi-tenant network cluster. So a single cluster can be sliced, a 512 GPU cluster can we sliced across 32 customers with complete security, and all of this can be done in a matter of minutes.

I want to show you fungibility in action. In the last quarter, we had a customer reallocation where a customer wanted to move to a different GPUs, but new capacity became available. And the interesting part about our technology and what we do is that we were able to reallocate that capacity in a matter of hours from customer A to customer B completely derisking our ability to actually have capacity that is on the ground and is not used. And as -- just as a reminder, we have a very large AI infrastructure customer base of over 1,000 customers.

So one of the questions that I've gotten a lot about is like, hey, you're building a lot of these liquid cooled data centers and these accelerated racks are looking different and then the CPU racks are completely different. And we understand the challenge because there are existing data centers that are air cooled, future days and there are liquid cooled. So that you have a mix in a combinatorial matrix that is happening on the floor. And for us, that fungible hardware and data center design is a foundational principle that ensures our assets are durable over a very long period of time. And the only way to get through it is to actually innovate. And so what we did is we did two things. One, we actually made our CPU racks to be liquid cooled or network racks to be liquid cooled. And we work with our hardware partners to drive extreme density into these racks where a single CPU rack can draw the power as much as an accelerated rack. That allows me to fungibly place the of these racks in any of the data centers. That's one.

The second, but what happens to all of these existing air-cooled data centers, Mahesh, like, are you going to throw them away? No. What we did is we innovated and invented this concept of a sidecar called an OCI L2ARC, which can actually power and cool as up to 1 megawatt of a rack, and it sits right next to the GPU racks. In fact, in the data center in the United States, we actually have over 100 megawatts of these ultra rack deployments that are deployed already and serving customers for accelerators in an air-cooled data center.

Let's talk about durability. We've talked at length, and we've been the foundational innovator when it comes to the networking space. We did that for Exadata, as I talked about, in 2018 because we needed to bring RDMA networking directly into our virtual networks. We've been continuing to innovate on that front, either it be intelligent traffic control, be it distributed load balancing, better security on our RDMA networks, the highest performance, so we've been continuing to innovate. But last year, we also announced a concept of a Converged NIC that gives customers a better price economics and security, where we actually embedded our Zero Trust Packet Routing protocols directly onto the network. We also introduced better reliability and resiliency where customers can get most out of their GPUs on OCI and any other cloud by innovating a brand-new network type called multiplanar fabrics, where each one of the GPU nodes can completely be available if one of the links flop. We also find innovative ways to shrink the time that packets take to move across our network by introducing fabric accelerators in the path that reduces the amount of time it traverses multiple switches.

And at AI World, we also brought that RDMA technology to the front end that enables and powers a next generation of AI workloads that accesses storage, that does a bunch of different work around more agent swarm and agentic activities. Second, we also are bringing the technology very close to block storage, so you can get very high performance. And lastly, we also released plug-ins that enables our customers to easily adopt this on our cloud with a new plug-in framework.

And one of the questions that I ask -- that I could constantly asked about is what happens to the CPU business and where it's headed? So I thought it's better to talk about it with numbers and not just tell my story. So what you see here is a recent commitment from one of our customers for about 1 million CPUs, 99% of those CPUs are older generation hardware. When I talk about older generation, those are chips and platforms that we launched in years 2017 to 2021. And just as a reminder, we're in 2026. And these are contracts that we signed which are not pay-as-you-go for 1 hour or 2 hours, I'm talking about 18 to 48 months of 100% committed consumption. And this is our general purpose CPU business.

Let's talk about durability. So one of the things that you all heard a lot about was at announcements across multiple hardware platforms. As I mentioned at the start of my presentation, we think the next phase of AI is going to be around the fastest access to compute and the most optimized TCO. And we're building towards that world, and that's why you see a ton of support across all of our hardware partners in our platform -- in our cloud.

Another innovative way to reduce the capital allocation on our platform is our innovation around bringing your own hardware. And the easiest way to explain this is what Larry asked Clay 8 years ago, Larry's like, how can you bring Exadata in 4 months onto our cloud? The good news is we had innovated on our SmartNIC technology and off-box virtualization that allows us to bring any hardware directly and place it on our network with the security properties. We took the technology. We enforced standard designs and industry standard designs with our customers, and we can virtualize the technology and place those racks our data centers and give them complete access to all OCI services. And you may ask, what happens when you place those services, how does it actually work? The first thing that I want to tell you is that when customers -- AI infrastructure customers use OCI, they don't just use the AI accelerators. On average, 13% of the spend of these AI accelerated customers is actually on non-accelerator spend.

Second, they use, on average, about 40-plus additional OCI cloud services when they're an AI infrastructure customer for OCI.

I'm going to jump into data. The -- if you think about the last before -- even 2 years ago. The amount of data that moves across the Internet and the amount of data that's being generated by machines versus humans, 30% of the data was generated by machines. What we're observing is a giant opportunity where AI agents are going to generate exponentially much higher volume of data, where this is going to go up to 50% to 80%. The volume of data is going to be so high that it actually creates an opportunity both for customers to create new business models and better outcomes and for Oracle to actually drive more data workloads onto our cloud.

The second thing is that as AI agents take over, the next billion users won't be humans, they'll be agents for database. So what that opportunity presents itself is, one, how do we actually offer the greatest and the greatest technologies for these agentic apps? Second, also how do you embed security at each layer, not just at the infrastructure layer, but also the database layer. And so we announced a series of products and capabilities today at AI World, where Oracle's Deep Trust AI architecture allows you to kind of get those capabilities on your platform.

One part of database and data is about storing the information, but that allows you to power applications where you can drive meaningful business outcomes without a data platform. And Oracle AI data platform and what we announced is the concept of an AI discovery agent that allows that goes into your environment, reads the application, reads the application code, reads your semantic models and essentially builds out an ontology of your business. It allows you to generate business context on top, which could be embedded into more meaningful outcomes. That is a very powerful technology. And if you connect the amount of data that is going to get generated, the mission-critical data that sits on Oracle databases and how it's going to evolve and how we can shrink time for our customers to realize that value we have -- we are poised for a bigger opportunity over the next few years. And a lot of this AI database technology is not just available on our cloud. It's live. And as in this quarter, to date, we have about 71 regions and it's one of our fast-growing businesses at over 209%. And all of these greatest -- latest and greatest features will be made available, not just in our cloud, but in all clouds.

I'm going to take one last example to talk about what is actually happening in our multi-cloud space with our customers, where we're seeing in a traditional world, migration, data migration, bringing the database online and bringing it into the cloud was a month-long process -- a month's-long process. But what we're seeing with AI and with the great teams that we actually have across the companies, we're seeing adoption move at a much faster rate. For example, Zema Global was a great customer who actually migrated their database on to Azure, Oracle Database@AWS with no downtime. And we also saw another similar example with Conduent where they were able to accelerate their cloud journey on an Oracle database really, really fast.

And so in summary, what you heard today is the amount of work that we're doing on OCI and innovating at every layer to create a durable, scalable and forever accelerating business. Thank you.

Please welcome to the stage Chris Leone.

All right. That was great. Love it. All right. Welcome, everyone. All right. No builder vest for this session. I promised Hilary, I'd put the suit on, so we're suited up. Let's get started.

I thought I would start with a little bit of overview of how we fit into the entire stack. I'm going to be talking about the application layer. I'm going to be talking about Fusion Applications primarily. But I wanted to kind of drill down a little bit on what kind of Mike talked about at the beginning and Mahesh kind of brought in a little bit, about where we fit in the overall enterprise AI stack. So I'm going to spend the entire session on the application layer, how we built AI in and not bolted it on. But let me talk about the enterprise foundation and how that truly enables what we're able to deliver at the application level.

So first, there's a shared set of AI capabilities that all the application teams benefit from. On the OCI side, huge value. This is where we get all of our models from. So whether they're an open weight model like gpt-oss, that is part of what we deliver in Fusion Applications. Or they're an enterprise model, we get the end points directly from OCI, but it goes beyond that. We get all the guardrails, all the observability, all the explainability, the traceability, all the audit level capabilities, prompt injection in the guardrails, all of this type of work happens at the OCI level that protects the application layer and makes it auditable for what AI is doing in our applications. And then what AI data platform or the enterprise data platform gives us is the ability to bring data in from Oracle and non-Oracle systems, create agents that cut across the enterprise. And then we can use those agents and that data in at the application level to actually execute work. And that's a big part of what we're going to be talking about today, where we're really transitioning the application layer from AI-assisted work, which we've been doing for 18 months to AI actually doing the work and doing the work more and more autonomously. And I thought Larry had a great analogy, simple analogy. The difference from AI-assisted work to AI doing the work and doing the work autonomously is like getting in a car and driving it and getting in a Tesla and having it drive you. It's that big of a leap. That's what AI doing the work and an AI doing the work autonomously. That's where we're going. And actually, that's where we're at today.

So let me jump in and talk about the announcements. Huge announcements, this is probably the largest set of -- certainly the largest set of AI announcements that we've had at any conference. On the excitement level from our customers and our partners has truly been amazing. Let me hit on these, and then we'll go into some details about what they're all about. And we'll geek out a little bit and talk a little bit about truly how they're differentiated and the architectural innovations that we've delivered.

So let me start with the first announcement which is we've been delivering agenetic applications for a while. They're built around an outcome. I'll talk about that in just a second. But we extended those with a whole new technology called Fusion Claw. So Fusion Claw is a technology that allows us to do more specialist type work in the enterprise. And so when I say specialist type work, you can think of what a financial analyst in accounting. I know you guys are industry financial analysts, but financial analyst in accounting, supply chain planner might do, an inventory specialist might do. They're doing analytic type work, and we'll spend a little time on that, but huge value within the enterprise and Fusion Claw technology allows us to do that work.

Second is we announced a new front door, for lack of a better term, entry into Fusion Applications with Fusion Companion. And we have a native component to that, which is Fusion Agentic Home, which is our native experience. And what Fusion Companion does, it is a universal intelligent layer that cuts across the entire surface area of Fusion Applications. So any user doesn't have to understand which page they need to go to, which transaction they need to work against, which report they need to look at, which workflow, and we go beyond that too, they don't even need to worry about what agent, what agentic application or what long-running job they want to run with Fusion Claw, a single universal entry layer. We do all the work behind the scenes. A huge, huge step forward for us from an engagement perspective.

And then a great innovation where now we're delivering Fusion Set up and Configuration Registry. This allows us to deliver a machine-readable set of AI skills to setup and configure Fusion Applications. And we'll talk about this, but it should reduce the mechanical work of implementing and extending our application from 50% to 70%, and we'll talk about what that means in a little bit more detail.

Okay. So one of the transitions that we're making or the extension that we're making is extending the subscription-based business that we have today. So we sell seats, we sell employees with a usage-based metric. And we've rolled this out a few months ago, and it's around a concept called AI units, which has tokens involved and has some compute involved, but it's a calculation of work being done in our system. And what this allows us to do is drive another layer of revenue across our application tier without necessarily adding more subscription seats. And so I'm going to make 4 points about what we're doing, and you'll see this play out throughout the presentation. At the end, I'll summarize those four points, so you don't have to write them all down right now. So the first one is this transition. So we're moving from a -- we're adding to our subscription-based revenue, a whole usage-based revenue. So the more value, the more work that gets done by AI in our system, as Mike gave some statistics earlier on, the more work that gets done, the more revenue we will see because they're consuming AI units as that work is getting done. So that's point one. We don't have to add more seats in order for us to now generate more revenue. We can do more work, and I'll talk to you about the type of work we can do.

The second part of it, with the innovation that we've delivered around Fusion Claw, we can do different types of work that was traditionally done outside of Fusion Applications now inside Fusion Applications, and I'll talk about that type of work that we can now do that again, the more work we do, the more usage we have, the more revenue we will generate.

The third part of this that you'll see throughout this session is we've delivered a machine-readable set of skills that allow implementations to go 50% to 70% faster. So this is going to allow us to move our installed base on-premise customers significantly easier and faster and more consistently because we have best practices built into the cloud. So we're going to get more moving to Fusion, so more subscription revenue that will ultimately also lead to more usage revenue.

And then the third piece is just as important. There's a whole economic side. So the way we're delivering AI architecturally allows us to deliver that scale within the enterprise, and it changes the economics. So our margin on the AU component is actually really good. We're not reliant on token and the cost of a token because part of it is token cost, part of it is compute cost and when that's work that's getting done, and I'll explain how that works as we go through the session. So those are the four points for us to drive more revenue through the usage of AI with great economics and margins.

So let's jump into the transition that we've made in enterprise applications. So we've transitioned from a system of record. We've extended the system of record into what we call a system of outcome. So -- we've been doing this for a while. Everybody is talking about outcomes now. 18 months ago, we reimagined in the world of AI, what our applications should be like, how should we organize applications? And we decided, and we came to the conclusion that, look, it would be great if we organized applications to run an outcome. So an outcome can be pay people accurately on time within policy, or I'm going to show you a demonstration later and it could be get shipments out on time within capacity cost and the right service levels. You can define an outcome and the application is organized around that outcome. You don't have to go from module to module to module to complete work. An agentic application owns the outcome and brings together agents that move work forward 24/7 to help the user get work done and do the work. And so that's what moving from systems of record to systems of outcomes means. And the base layer is our foundation is the authoritative source which is Fusion Applications where all of the controls, the permissions, the policies the actual state of the transactions you need in order to execute work. And that's the change that we've made over the last 18 months, and that's what part of what we've been delivering against. Now let's bring this into a little bit more detail so you understand what this looks like. So as we transition to outcome-based applications, agentic applications. And these agentic applications are designed to move work forward, do more work in the enterprise, the authoritative source is incredibly important because we're going to do more work. We cannot do that work based on an estimation of customer, of supplier, of orders, of inventory. It has to be on the actual state if we're going to commit transactions and move work forward. So the operating model becomes relatively simple. The human, person running the system is accountable for the outcome. Agentic applications own that outcome. They handle the cognitive load of the process. They move work forward. They know what the next immediate action needs to be in the system so they can move that work forward. And then it sits on top of the system of record that actually knows the current state of the business and it can move the transactions forward and commit work via AI and via autonomous AI that we're delivering. So this is how we've transitioned going forward. Now in order to do more work with agentic applications, we needed to go deeper into the type of work we were able to do. And so if we were going to do more work in the enterprise, we had to take on harder work, what I call specialist work. And this is the work that was traditionally done in applications outside planning applications, reconciliation applications, maybe Excel for ad hoc analysis. Now that specialist work done by financial analysts or inventory specialists can now be done inside Fusion. And the way we can do this is a new technology that we've just announced called Fusion Claw. So Fusion Claw allows us to do -- take on long-running work, allows us to do multistep computational intensive work, allows us to do reconciliation against millions of records, allows us to do simulations, allows us to do optimizations, allows us to do sophisticated high-value work within the enterprise. And the second part of this is this is a huge leap from an architecture perspective. And so when I say that, it's a huge leap from an architecture perspective. We have cracked the economic side of handling extremely large workloads. And so what I mean by that is when you're doing enterprise AI within Fusion Applications, you're talking about extremely large operational and transactional data sets that you need to do work against. And so think of a ledger with 1 million or 2 million records that you need to go reconcile or you need to go analyze. You might have 1,000 calculations or 100 scenarios that you need to go through. If we applied an LLM or an agent to that type of workload, it would never be efficient enough, it would never be precise enough, it would never be economical to do that. And if you think about it, just the way an agent works, LLM reasons, it looks at a record, it makes a decision and then it goes back and does the exact same thing and looks at record after record. So just think when we get to record 992,000, and the context window fills up and it starts making the wrong decisions, that's never going to scale from a precision perspective, from an efficiency perspective, I don't know if it will ever finish. And it's never going to scale from an economic. The cost to do that type of work in the enterprise is never going to be cost effective. So what we did was architecturally, we separated reasoning and planning, which AI does fantastically well. And so we leverage it to do all the planning side. And then we use deterministic compute to do the execution side. So what that means is, we use AI to understand the problem, to assemble the tools and capabilities and skills and the execution logic, and then we hand that execution logic over to compute, and then we can run it against a million transactions, a thousand scenarios and do it without touching the token side of the LLM, that's complete compute. So we can handle extremely large workloads, do high-value work, optimization work, reconciliation work at really good economics for enterprise customers, completely different architecture. We're really the first to deliver this in the enterprise, truly the first to deliver in the enterprise at this scale, and partners' customers are extremely excited about it. And so I'll drill into this, just one more thing to bring it home. We didn't have to build 5 different execution run time engines. We built one. And we can solve thousands of use cases across the enterprise, probably even more use cases that we've traditionally solved in the past and whole new use cases that we haven't even thought about or have been able to do in our application layer. So we can do analysis against millions of records against a policy document or a standard operating procedure. We can do reconciliation in the ledger against 1 million or 2 million records and look at discrepancies in the general ledger and subledger balances and come back with a deep analysis. That type of work traditionally would have been taken out of Fusion, done in a data cloud or an Excel and then the result would have been put back into Fusion. Now we can keep it all together and do those large AI workloads in one system.

And then optimizations. I'm going to show you an optimization example in just a second, but we can optimize inventory. We can optimize capacity. We can optimize cost. We can optimize schedules. So all of these high-value use cases we can apply to AI and actually do the work.

So before I run out of time -- and we can execute. So the last key piece is we take action. So this is not just like we're recommending moving 20 million inventory positions, we're actually -- we can do and transact that 20 million movement of inventory around the world to get inventory to the right data centers for Mahesh, and make sure we have the right equipment. They're all that type of work we can do with AI and then we can execute in the system.

So let's go to a demo. Now I'm going to tee this up. But I gave Ken, one last one. I usually like to do the demos myself. I did it in the opening keynote, but he was worried that it might not come off. So we did a video. But I finally found someone, I trust enough to do the demo for me. So let's take it and go to the demonstration.

Cloud database will have very high growth, well in excess of 50% and moderate capital intensity, similar to SaaS cloud database has very high gross margins and relatively high operating expenses as a percentage of sales. Our infrastructure business is a very different business model, with relatively lower gross margins and high capital intensity.

However, it also has relatively lower operating expenses as a percentage of sales, particularly supported by the networking technology differentiation that Oracle has already created and continues to create in its database business. There is a big differential in gross margin between these business models. However, at the level of operating model -- margin, excuse me. This differential closes quite a bit. So all three of our business models contribute strongly to our profitable growth journey. Historically, our non-GAAP operating margin has trended in the lower 40s percent. We show here in fiscal 2026, it was 43%. We do anticipate some near-term margin pressure from MIT as well as the ramp-up of our projects over the next few years.

However, with the continued strong contribution from our apps and database businesses and focused cost management will be back at a 40%-plus margin in fiscal 2030. And at the peak of our ramp-up impact that's in fiscal '28 and fiscal '29 were not significantly below that. To dive a bit deeper into the contributions from apps and database. There are a couple of points I'd like to draw out from these slides. First, on app. Chris covers the evolution of our pricing models with the addition of outcome-based pricing. And we continue the journey, make tester by AI to migrate customers to the cloud. That results in funded. I think it's 3 to 4x, 4x pricing uplift versus on-prem. In database, AI is driving a shift to higher-value database services. We continue to scale our multi-cloud business and AI data platform offers another great opportunity to expand our monetization model as customers focus on using intelligence to drive results. We've received a lot of questions about our infrastructure business model.

So we thought it'd be helpful to provide a 1 gigawatt example for education and modeling purposes. This is roughly reflective of the economics of our next few years of projects. And this is, of course, based on Oracle's business model for data centers. So it assumes releasing the shell, the power, et cetera. A few things to call out here. Each of our five large sites is planned at plus or minus a few hundred megawatts from 1 gigawatt. So this is a good proxy for each of those sites. One of those sites is close to fully executed Abilene. And then we have four sites expected to start first service delivery in calendar 2027.

Of course, our financial guidance takes into account up to a couple of quarters delay on any of those. Mahesh has highlighted those sites are not built as a single project. In fact, they roll out as a series of data halls and buildings over time. As a result, the lease obligations, CapEx commitments and revenues ramp up as the build progresses. This means, in the case of a delay, most of the costs are delayed as well. Plus a lot of the hardware is fungible site by site. We've also given the details on average project-level economics, reflective of our current RPO. CapEx at around $37 billion is a bit less than 50% of total contract value revenues are around $16 billion per gigawatt annualized. So that's once we ramp up.

And while our gross margin is at around 35%, operating margin at the project level is around 34% due to that OpEx right business model I described earlier. This business model generates strong free cash flow fairly quickly as we ramp up. fully ramped a 1-gigawatt project will convert 100% of its nonoperating profit after tax plus depreciation into cash flows annually. And I've spoken about our returns on investment in the high 20s in the past. Here, you can see the details of the calculation itself. We've added back depreciation in the numerator because this business has limited maintenance CapEx. Aside from new IT equipment around them would expect to recontract or later. And please note, effectively all of our large AI infrastructure contracts are cost plus or similar. So revenues may change, but the margin and ROI will remain.

As we demonstrated in our prior slide, this is overall a very attractive business model, albeit capital intent when self-funded. In the model shown on this slide, we pay for all $37 billion of the required CapEx in the first 2 years of the project. We've been looking at some models to reduce that capital intensity, and we show the directional impact of those in the next couple of slides. One key lever we are availing of is vendor financing. We have strong partners on the supply side who are willing to support us. And you can see an illustrative example here of a simple 70% 5-year vendor financing. This reduces the capital needed from Oracle in the first 2 years to $18 billion, a 50% decrease. Bringing on hardware is where our customer provides their preferred chipset to us, reducing our capital requirements to a minimal $6 billion or lower, again, based on this 1 gigawatt example.

Of course, with bring your own hardware, there is some impact to the project economics with lower revenues and higher margins. And Mahesh talked through the value add to our customers in that scenario. This slide summarizes the simple examples I just walked through. In the case of self-funding and vendor financing, the total level of CapEx and project economics remain roughly the same. What changes is the timing of our capital needs with vendor financing, shifting those payments to be better aligned with our revenue ramp. Bring here on hardware changes the capital intensity of the project itself, driving higher margins and higher returns on investment.

I'll close this section with a few further points of guidance. Based on the strength of our 3 business models, including the strong cash conversion from our infrastructure projects, I described a few minutes ago, we expect to drive greater than $30 billion in free cash flow in fiscal '29. I mentioned the 40% plus operating margin we expect in fiscal 2030, and we also anticipate $21-plus in non-debt EPS. Of course, there's a question of recontract and the necessity for the renewal of GPUs at the end of our various contracts. Mash walked through the differentiation we have in terms of technology. and how we'll continue to drive value for our customers over time. We're building a business for today and tomorrow. And we expect to recontract at similar margin rates or better than what we have today.

In the case of renewal of GPUs or other IT equipment, we gave a real life example in our Q1 where we're selling or renewing 4-year GPUs 20% premium to their original contract price. We put that information into the slide here. We think that's a powerful example of the potential upside from the GPUs we have in our portfolio today. We won't count on it, but we do think that these are valuable assets, both for today and into the future. Turning now to accelerating value through disciplined capital allocation. Our balance sheet and capital allocation priorities remain clear. First and foremost is to maintain investment-grade credit rating. That's extremely important to us as a business. Second, to deliver returns to our shareholders in the form of a dividend. And third is to fund the significant business opportunities we continue to have in front of us.

I mentioned we've been looking at some different models to reduce the capital intensity of our current build-out as we bring online an incremental 7 gigawatts over 2 years. I already spoke about the impact of vendor financing and bring your own hardware. Lastly, as we discussed over the past couple of quarters, we continue to be in a very strong demand environment. And within this environment, customers are willing to make prepayments to preserve the right to receive services from Oracle in the future. Why does this all matter to us? Well, we've been working super hard on sizing the level of final capital needs we can commit to for the remainder of this investment cycle. So between now and 2030 or really between now and when we can be self-funding as a business. This slide reflects how we've gotten comfortable with those calculations. The most important driver of our funding need is, of course, the remaining CapEx needed to fund the build-out of our projects.

This slide is illustrative. So I simply use the 1 gigawatt example to project our CapEx needs starting in Q2 2027. And it may not capture all timing impacts, but directionally, it's around correct. After working with our partners and customers, with the capital intensity reduction models, we have in place. We're confident we'll reduce our upfront cash requirements between now and when we'll be sales funding by around 50%. And of course, we have the operating cash flows from the business itself. Based on this, we're able to determine and to commit to today the final capital needed to finish our current investment cycle. $20 billion in incremental capital is what we're committing to today to complete the funding of this investment cycle. We'll split that evenly between $10 billion equity and $10 billion additional gross debt. Given this amount is quite a bit less than what might have been anticipated, we'll look to execute that opportunistically as early as calendar year 2.

I'll mention here again that, that doesn't mean we intend to stop growing our business. With the flexible models presented here today and with our RPO converting into strong free cash flows in the fairly near term, we expect to be able to fund our business for continued differentiated growth in the future. Let's say, a slide on our balance sheet expectations. Based on everything we discussed here today, we expect to improve the strength of our balance sheet in the near future. Following the $10 billion in new debt, we anticipate our gross debt will remain flat over the next couple of years as we refinance in the ordinary course. And we're committing to a less than 3.5x net debt-to-EBITDA ratio by fiscal 2029.

I'll finish with the same slide I presented at the beginning. We've solidified our capital requirements for this cycle. We're confident in our path to differentiated growth to positive free cash flows, strong operating margins and a strong balance sheet. We look forward to showing you with each coming quarter, the stepping stones in that journey. With that, I'll pause, and we can ship to Q&A with myself, Clay and Mike.

We've got Mike, Clay and Hilary upfront. We'll start with John. If you want to ask a question, just raise your hand or shoot me a text message.

Thanks. You guys really left the best for last there, and everybody is waiting for all that. So it was great to see. I mean, I think everybody in this room or most people in this room anyway, and you guys probably do to realize that the stock of Oracle as not necessarily trading on how you're running the business, but on more tactical things that you actually just addressed. I guess I want to hit one of them anyway in a little more detail. I guess you said, Hilary, and there's a lot of little things. So it's a little thing, I think. But you said that the CapEx in fiscal '28 was going to peak and you said, and so I think a lot of people that I speak to anyway believe that, that means that '28 will be about equal to -- but when I think of gross CapEx, at least in my model, when I try to model and I spend too much time doing that, I guess. But it actually goes up a lot because you have so much new business coming on into '28 and '29. So how should we think of that for gross CapEx? You gave a great -- that bar chart of how much is going to come in from early payments and stuff that helps a lot.

Sure. So the way I think -- actually, that's one of the reasons we gave that 1 gigawatt example so that you guys can do a little bit of modeling yourselves a little bit more easily. What I would say is we're ramping up part of those big lots this year and probably even more next year. So maybe that can help you a little bit directionally as you model.

Tal Liani from Bank of America. Just one clarification before my question, a clarification. I think the most positive surprise here was how little money you need to raise for the next 3 years. So I just wanted to make sure that I understand the slide, it's 10 and 10, so $20 billion for 2027 and nothing for the other 2 years after that?

Yes. That's what we're committing to today is we have been able to release flexible mechanisms we explained, vendor financing or sort of vendor financing similar, bring your own hardware prepayment we can really have that confidence to commit to that $20 billion today. And that's a real commitment. We don't expect other forms of structured financing or other things using our Oracle credit out in the market. aside from, of course, we'll continue with the same business model that we have today of leasing data centers.

Got it. So my question is about pricing of capacity and your strategy around pricing. There is a difference between pricing for 5- to 6-year deals and pricing of 1- to 3-year deals and spot pricing, how what's your plan? What's your strategy behind selling this capacity?

Yes. It's a good point. I like -- the way that we think about it is that you have to kind of pick a portfolio mix, right? And it comes down to what's your risk appetite and how you want to kind of balance off your different options. So as an example, if you sell something kind of on very short-term reservations, you can get a much higher price at the beginning of the life cycle of the hardware. But you're taking on more risk about what's going to happen in the future.

There's -- so there is a game to be played around how much cash do you front, how much of a return do you get at what period of time? Which partners you're using, which -- how the customer is allocating their capital towards you, all of that comes into play. What I would say is that ignoring the -- how do we think about how many 1-year deals versus 5-year deals that we're doing. I think the thing that everyone is massively underappreciating is that the data centers that we built that came online this year Five years from now, whether it's we made x or y on how we got there, these are going to be incredibly cheap data centers compared to what people are able to build 5 years from now.

The hardware that will be in them, I think, will be incredibly cheap. And so regardless of when there is a renewal cycle and it's part of the reason why Hilary showed the renewal options that we're seeing. There's been a lot of fear in the generic industry, right, that, hey, people are assuming a much longer life cycle for this hardware, that is very much the opposite of what we're seeing. And so the way I would think about the situations where we have plentiful cash and that we think the risk appetite is there, we're going to invest and take more risk, but we think we can get a higher return.

In situations where we think that that's not appropriate, and we need a lower risk profile. Maybe we need access to different capital or we want to watch something in, we didn't do longer-term contracts. But we're constantly balancing that off. And the thing to remember is that you don't just get one bite at the apple. If you sell a 4-year contract tomorrow, four years will be here sooner than we all like as we age very rapidly. And then when that happens, do you renew it on 2 or 3 years? Do you sell that into the spot market? There's a constant it's kind of like asking what's the airline or hotel room -- hotel chains philosophy to pricing.

Well, it fundamentally has to be extremely dynamic. And that's a huge part of what we do is to actually have extremely dynamic pricing, and you put a portfolio together to optimize your risk along with your overall return.

I would also add, Clay, you mentioned the word partnering, which I think is very important. As we think about these relationships, we also keep in mind that we ultimately become a distribution channel for these customers as well. Even call is a terrific example. There's a lot happening behind the scenes in Fusion call. We're aggregating some of our model customers. And so there's another chance for us to monetize that output as well because we want those customers to also become partners of ours, too. So -- and that's unique Oracle because we're delivering the entire ecosystem, the entire stack. Everybody else is basically in the business of providing computers. We're in the business of providing the computers that also enable outcomes as well. So it certainly factors into what we do.

Stefan Slowinski from BNP Pariba. Hilary, so much has happened in the last year in the industry, your guidance hasn't really moved for 2030. And so just wondering how that all nets out. Obviously, we've had lots of inflation over the last 12 months and lots of debate about the inflation escalators in your contracts. Interest rates have gone higher. So just wondering if that guidance a year ago, there's a lot of cushion in it or how you would characterize the guidance today and how those moving parts has impacted the lack of change in the guidance?

To be fair to Hilary, because she wasn't here. The number went from a number to a number plus.

Indeed. So I hope that we clarified the question around pricing in the large AI infrastructure contracts that we have today. So we do have effectively some form of cost plus pricing or for the rest of like the CPU business, we're anyway contracting much closer to when we bring the compute online. So we feel pretty confident in being able to cover the increase in costs, for example, that we might see during the build-out of projects that we have, whether for CPU or GPU. In terms of the guidance, look, I think adding $158 billion is pretty impressive. Not a 100% sure that the market is price that in yet. So I think the $225 billion plus. We're 1 year into the journey. We feel even more confident. We feel very confident with that $2.25 and that $21 in EPS, we added the plus really exactly for this reason.

Brad Zelnick, Deutsche Bank, over here waving in the same room. Thanks so much for a great presentation all around. Really very helpful. Hilary, you made reference to this cycle as you presented the next few years. How should we think about future cycles beyond the current one and your ongoing commitment to growing free cash flow beyond the $30 billion in fiscal '29? And how also should we think about your appetite to take down more deals and the size of the types of commitments that you'll make from here to customers?

So first, I think the $30 billion plus leaves us with opportunity to continue to grow the business. Like I mentioned, I think we're showing truly differentiated growth than what we're showing to the 225, 225-plus in future, that's in fiscal 2030. This business is a very strong free cash flow positive business driver on all three business models. I think we have plenty of space to grow that business. I think that we'll be in a very different place in fiscal '29 fiscal '30, is we're growing far in excess of $30 billion plus in free cash flow. So I'm quite confident that we'll have all of the opportunity that we need for that business beyond 2030 to keep growing at the rates that we should as the business and keep up with the market we have available to us. I don't know, Clay, if you want to add anything else?

Yes. I mean I think that clearly, we don't want to -- the world is dynamic, and we want to have flexibility there, but it's very clear that I think that Hilary has talked about this in the past. It's a lot about ratios, right? It's very -- going from 1 to 10 gigawatts, right? You're going up a curve is very set. As those things throw off cash, okay, I'd love to tell you that I think that in the next -- from 2030 to 2034 that we could go from 10 to 100 gigawatts, I don't think that's going to happen. And so part of, I think, what you're seeing is that we have gone up a very fast ramp.

And I would -- while we will continue to push as hard as we can and grow as you -- I don't think it's possible to add 90 more gigawatts in 2030 to 2034. And so the ratio of what -- how things get funded in just inherently going to be different in that environment.

Alex Zukin from Wolfe Research, echoing the congrats on the presentation and the clarity and kind of the obviously, the confidence that it conveys and what you guys are doing. Maybe just given, I think, some of the dynamism that's inherent right now in the industry and particularly the pricing dynamics, I think, Clay, that you're alluding to, it feels like you guys have discovered something pretty meaningful in this creativity to drive down that level of financing that you need, whether it's from the price points that the compute you're able to take online, the price points they are able to deliver it from. Can you maybe get a little bit more specific on what changed in the last call it, year that's enabling this decline in the density of financing that you're going to need to complete the cycle.

Well, I don't think that -- I mean I think if you actually look at what we've been saying over the years, it's been consistent, right? Hilary did an amazing job of showing you the different forms we didn't -- it didn't like surprise. We invented bring your own hardware and prepayments and vendor financing in the last 3 weeks. We've been talking about that for many quarters. I think that part of what Hillary has done is he's explained to you the size and the scope of it. I think if you now take a step back and go back and read everything we've told you over the past several quarters, you'd be like, oh, this is entirely consistent with what we have been saying. I think there's just other people didn't -- I only say they didn't calculate it the same way that we were.

And I think I mentioned today that we've now solidified quite a bit of this planning. There's a big difference between when we first set out our vision. And now what we put into place, which is all the mechanics around it. So we're really in a place today where we have enough. These aren't just forecast. We have enough done that we can really make the commitment that we're making today on the lightening of capital intensity mechanisms, so not just some dream like now we actually have confidence and we have those mechanics in place.

Jackson Ader at KeyBanc Capital Markets. The question I had was on the mix, I guess, of vendor financing, bring your own hardware prepayments. What kind of mix is assumed in these gigawatts that's expected to come online in the next couple of years? And is there -- I guess, number one, is there any risk that if that mix changes, it's like, oh, shoot, we got to -- we actually do have to go out and raise outside capital? And are there -- are there certain type of customers that you might have to say no to in order to maintain that mix?

So the 70 gigawatts that I showed, this is for contracted capacity already. And that mix between vendor financing, bring your own hardware and prepayments is primarily solidified. So that's the kind of confidence that -- otherwise, we wouldn't have the confidence to say what we do today on the $20 million. we don't take that lightly as a company. That's a big, big statement to make. So we really feel comfortable with what we have in place with those 7 gigawatts with the customers that we have with the suppliers that we have today. Now will we have some additional growth, primarily probably beyond fiscal '28, I think between now and 2018 is about what we have coming online.

Karl Keirstead, UBS. Clay, I've got two chip related questions for you. One, I think Hilary just answered, which is that if you're relying a lot more on vendor financing, given that you're very heavily weighted to one particular chip supplier, just how solidified is this commitment from that chip supplier? Is it locked in agreement?

And then, Clay, congrats on the big news around hosting Google TPUs. For me, that was actually one of the biggest announcements from Oracle this week. Maybe you could just spend a second on how meaningful that is for Oracle. And does that have any effect on the revenues and backlog and margin outlook? That might be helpful.

Yes. I think on the first question, I think that's presumptuous. I don't think that we said that we have specific financing with any city vendor or in what magnitude or quantities. So I don't think -- I don't think you should walk away here with any of the assumptions. I think there's a potential reason that we're very confident in the different ways in which we're financing the business. Now on to the second question about TPUs. Look, I think TPUs are a really cool accelerator. And if you look at our overall strategy, it's that we want to have all the accelerators that our customers want. So I think that every customer will suddenly want to buy TPUs? No, that's okay. In the same way that we have AMD and Intel CPUs.

We also have NVIDIA arm CPUs, and we have [ MPR ] arm CPUs. We have a lot of different options. And what I would say in general is that our overall strategy has been is to make the expense to us and the expense for our partners to be extremely low to integrate meaning if we can make it simple and easy for us to do the integration between our products, then why would we not want to offer those options? And I think that has served us well, not just because it helps you have accessibility to more customers. But that strategy actually makes us more flexible across the ecosystem today, but it also gives us confidence that we're being flexible across the ecosystem in the future.

So if you wonder why is it that we're confident that we're going to be able to plug in different types of hardware 5 or 6 or 7 or 8 years from now, it's because we're able to plug in a lot of different types of hardware today. That's the only way we can be confident that the stuff that we're designing is actually truly flexible. So look, in summary, TPUs are great. There's many customers of those. We have some of them that are very excited to use them on our cloud. But obviously, A lot of people would like to buy a lot of [indiscernible] as well. Same thing, right? People are excited about [ Cerebrus ]. They're also -- Broadcom Jalapenos is also very, very exciting. So that's our overall strategy. And we're working with all of the different players and all the different customers to try to give them the acceleration that they want. What I would say in aggregate is that demand for all of it is far exceeding the ability to go out and supply it at least for what I see for the next several years.

Derrick Wood at TD Cowen over here. We all read the news, obviously, on some of the challenges with permitting and data centers. You've responded recently on some of that. Can you just talk about the fungibility that you have in case there are delays and the ability to put equipment into other shelves and kind of how much slack you have and like withstanding some of those delays out there?

Yes. Well, first thing I would say is, like I think there's a misunderstanding in that I think a lot of people who are not very familiar with certain industries, it becomes very popular for some reason. And then whether it's kind of normal course of action suddenly becomes front-page news. Sometimes it feels like you wake up and there's like actors showed up to the set today and they had to wait until the lighting was right before they could record their lines, gen. Like that's how movies are made.

The process that we use like building large construction projects of this scope are complicated. And I think anyone that goes into those projects without assuming that they're going to be complicated is sadly mistaken. Now in terms of the actual technical piece, there's what we put inside of the data centers themselves, it's also the data center components, right? I think that people misunderstand that if you look at the total capital that goes into a completed data center, right? And Mahesh, I thought I had a pretty cool video upfront. The building itself does cost money, but it is not as significant. If you take out the racks and you take out the network gear, you take out all the fiber optics and you take out all the cabling because all of that you can put somewhere else.

And then you take out all the chillers because you can move those and you could take out all the UPSs because you can relocate those. And suddenly, what you find is that as a percentage of the total site expense, yes, you have to do a bunch of grading and there is specific if you're looking into the grid, certain substation transformer work and certain things. But in terms of the overall percentage of the project, it's very, very small. And so sometimes it's look like Bob Ross would say you have happy accidents. You may have something that's delayed over here, but then you can use those components to accelerate a different project over here. That's as part of what Mahesh is trying to explain was, hey, in a single quarter delivery, we have on the order of 40 or more things delivering, you have a complex apparatus that you're hyper optimizing all the time.

So look, do I think that it requires a huge amount of effort and focus every day for us to wake up and focus on it and to make them work Yes. But that is our job. And that's why people trust us to deliver capacity for them. And I think if you see the predictions that we've made in terms of revenue and capacity and then the success that we're having about hitting the deliverables that we're talking about, people keep asking, are you confident that you can plug in their stuff? It's like, yes, and you can see it, right? It's -- we show you the videos of the sites. We showed you the revenue it's associated with the customer paying for it, show you the profitability that comes directly in operating income and how that flows through to EPS each quarter.

This is not the future, right? When we said we delivered 850 megawatts last quarter. I mean if you just -- it doesn't take many quarters to get to -- that's a lot. And it's not some abstract thing in the future. This is happening right now every day.

Siti Panigrahi from Mizuho. Thanks for all the clarity today on funding and precast inflection great job today. But I'll switch the topic to application. I think this is one of the underappreciated topic today. AI world. I think Chris talked about the AI units, how you are launching now the new revenue stream, a base revenue stream that doesn't depend on you're adding sites with higher margin. So Hillary, in your guidance for double-digit SaaS growth, how much contribution are you assuming now from the AI unit on [indiscernible]? And when could AI [indiscernible] become material enough to accelerate that a slow beyond where it is today.

Well, double digit, obviously, encompass a fairly large range of outcome. But we're just getting started, I would say, in terms of the outcome-based pricing. So at least in my financial models, we haven't incorporated too much yet, but the opportunity is certainly there. And Mike, I don't know if you want to talk a little bit.

Yes, I would say, look, this is -- it's new, but things gone pretty quickly in the world the catch with outcomes-based business is you want to make sure that you can generate the outcome, right? And since we've been an operator and a custodian for so long, our confidence is quite high. Our customers have shown us what good looks like, and in many cases, what great looks like, if we can replicate those heuristics across all of our customers by industry, we think that becomes a terrific business for us. And so we said, double digits is a wide range. We'll stick with that range right now and certainly provide more clarity as we roll it out. But I think the other thing that's equally important, I said it this morning, you still have in the ERP base, about half the customer is still running on premises, right?

So getting them to Cloud is incredibly important. I think it's also incredibly important in the age of all these advanced models that are discovering as Sam said on stage yesterday, incredible vulnerabilities, the on-prem stuff is going to become even more vulnerable. And it's going to be very, very difficult to patch that stuff at scale. So I think that business is actually really important for us to automating implementations, Chris said, 50% to 70% faster. I mean that is an unlock that customers have been waiting for a long time. So we're really thrilled with both of those announcements.

Mark Moerdler, Bernstein Research. Thank you for all the great detail, impressive and congratulations Ken's been great working with you and enjoy the next stage of life. We'll miss you. Hilary, you mentioned that you built in data center delays. Clay answered this from a technical perspective. But, Hilary, can you give us more color on how you're modeling in and managing all the potentials of regulatory or other delays on the data center? Does these delay create potential for clients to walk away? So does that allow you to manage the price with the clients? Any color so we understand what happens if there's a data center delay from a financial perspective?

Sure. So coming out of infrastructure for some portion of my career, and I think Clay said it a few times on a few of our quarterly calls as well. One would be crazy probably to assume a particular impact set date with yourself and your partners. And that extends into our customer contracts as well. I think the customers understand that we are building not so much something new. Data centers have been built for a long time, but at a scale that's quite new. So I would say I wouldn't arrange any particular concerns from a customer standpoint. We really view it more like a partnership with the large-scale AI sites.

And in our own financial -- another thing I would say is that -- and I mentioned this throughout my presentation, the costs generally ramp up as we build. So if we're going to have a delay, the costs won't ramp up. So we're able to fairly match, and I thought Clay did a good job describing that. Not a lot of the costs that we have in front of us are actually the shell or sort of flattening the land or anything like that. Most of the cost ramp up as we go by data hall, by building. So we're able to see -- we're able to manage, I would say, well financially if we did run into any delays, and we do have fungibility between the sites for the vast majority of what we're spending money on.

I think we've all said that Mahesh, myself, Clay today as well. So in fact, there's a -- from a financial modeling standpoint, there's a forgivability I would say, in terms of the plus/minus of the cost that you actually have. The financial targets that we've put out we've already actually included into those, and I mentioned during my presentation, in my mind, infrastructure is effectively on time, if it's a quarter or a couple of quarters around what you anticipated. Again, given that we have customers that are more like partners. So we're not talking about LDs or anything like that. So our financial modeling takes into account any of those large sites being delayed up to a couple of quarters here and there. And I think I mentioned that in the presentation.

Patrick Colville here in the middle from Scotiabank. I've got a clarification question for Hilary, please and then I guess, play more of a strategic one for you. I guess, Hilary, in the slide about cutting funding by 50% with vendor financing, ringroadship and prepayments. Optically from the slide, it looked like it was 1/3, 1/3, 1/3. Is that mix directionally right?

So the slide is just illustrative. We didn't give you the perfect breakdown between those. I think that's probably a little bit too much information to give out for our partners being customers, partners being suppliers. So we've just given an illustrative mix there. But each one of those three plays a decent part in providing that around 50% capital intensity reduction that we put in the graph.

Okay. Very great. And I guess, the more interesting question, if I may. I mean, your comments earlier were very clear about demand exceeding supply. I think we all know that right now. But looking at the industry pipeline, there's about 40 gig that's due to come on next year and then 60 gigs worldwide in 2028. If I think about Oracle, I think probably uniquely well protected given longer-term contracts. But I guess could you talk about that demand supply balance industry-wide? And whether you expect that to persist and what that means for Oracle?

Sure. Well, I will start maybe from the beginning. First, I think it's really important to separate out the marketing from reality. And so one of the things that I think people get -- you can get -- it's kind of like people will read the press releases, and they seem to ignore this really important term called UPT. UPT is a very important concept. It means irrelevant. So, I could be up to 100 feet tall, not, but I could be up to. So I think if you actually understand our industry, right, there are some built-in self governors. There is only so much memory in the world. There's only so much wafer fab.

I think everybody who spends time thinking about how much power is coming online is kind of missing the point. I think you can actually look at how much memory being produced and how much fabrication capacity is, and that's a much clearer example of how much you could put it because we don't need power without chips, right? So that's a much smaller set of things that you have to analyze. So I don't understand what people get caught up in I've got people e-mailing me that they've got a farm in Iowa, I could put up to 300 gigawatts on. So if you add that to your number, then it's 340 just next year. but it doesn't mean that I could actually get any computers to put it.

And what I can tell you is that any smart provider, and I have to be fair, I think get the marketing, a lot of the people who are making most of the decisions are actually quite intelligent. They're not building data centers. They don't have network fiber optic cables, transceivers, you don't go out and put all this money into mechanical electrical plumbing if you don't have the stuff to go with it, right? Your job is to make [indiscernible]. So I think that now if you're asking is that's kind of the supply side. I also think that people do not understand the immense nature of the demand, right, I think if you look at any of the stuff that we talked about over the past couple of days, AI world, it is changing very, very rapidly. And I don't think people can understand yet the rate at which compute is going to be needed to fund the AI demand.

And so the cool part is that we continue to -- a year ago using AI for software development with an idea, but it was not rocking. It's there now. I think every year, you're going to see significant new industries that are turned on, right? If you look at kind of what's happening with mathematics or science, and then you look -- part of this is it's nice to be in my position because you get to talk to a lot of the people doing the research, and you see what's coming. I don't -- I just don't think people understand the breadth of demand that's out there.

So, I am not worried about a situation. Like I think people are like, what is somehow we're able to build just so many more GPUs than there is demand for. I just don't see that. Not for the not for the time line that I feel like I can confidently have some perspective on over the next 3 to 5 years, I don't see that happening.

This is Bill Fitzsimmons from Piper Sandler. I appreciate the free cash flow targets. We talked about prepayments and the ability costs already, but given everyone is kind of setting the CapEx dollars in their models right now. My question is when you think about CapEx dollars and expect some financing needs. If we look back over the last couple of years, the cost to bring on a megawatt or a capacity increase for a variety of reasons, whether chips, components, turbines or energy. And when we -- can you just frame what's been contemplated in the targets in terms of additional potential cost inflation in future years? And what capital expenditures have potentially been locked in already?

Can I give you -- one second, Hilary. But I feel like you're all asking questions that are kind of missing the point. The next 18 months are the area of the biggest growth for us. That's also the area where we have the most certainty around pricing and clarity. We then, as Hilary said, hit free cash flow positive. So the question around, well, plus or minus this money, great. But the cool part is if we understand how much revenue is going to be at and how much cash that's growing off that gives us a lot of flexibility to absorb things out in '29 or '30.

So when people say, hey, well, how do you confident this is how much capital you need, it's for exactly that reason. And people say, well why didn't you give this out a year ago? Because we need some more of the certainty that we have available now to get from that period over the next 18 months. Thankfully, 18 months is within the time horizon where you actually have issued POs and have locked prices. right? That's what I mean like it's not -- it's -- I don't know what a chip will cost 5 years from now, but I'm also very confident our free cash flow looks very different 5 years from now.

Yes. And that 1 gigawatt example we gave that up to date. I think last year, we talked about $10 billion probably per gigawatt this year, it's 16%. A lot of that is actually the pricing increase that we might have seen in memory, for example. So we've given you updated economics. I mentioned that cash plus mechanism we have in pricing. So therefore, that revenue amount might change, that CapEx amount might change, whereas the gross margin and the return on investment. That's really what we think about in terms of creating value for our business. So this is a business where there's enough demand. We have enough technological differentiation that we have pricing power, right? So that's what we're doing with the business. So in 3 years, will those numbers look different in the 1 gigawatt cycle. I would imagine, yes, in one way or another, aside from the strength of the business model that we have.

Tyler Radkee from Citi here in the front. Thanks for the great presentation. One of the stats that stood out to us was just the diversity of capacity coming from the smaller sites, sites outside of the ones we all track. Can you just talk about how you see that opportunity going forward? Like where is this demand coming from? What's your appetite to really expand this globally? And is that mix the right way to think about the large site or small site contribution going forward?

That was -- I'm not -- there were many questions there. I don't -- is your question like, how do we think about the mix of when to do large versus when to be small? Or it felt like it was a -- Do you want to answer the question?

I think one thing we definitely wanted to demonstrate actually today is exactly what you said, which is which is that all of the growth that we're seeing doesn't come from the few sites that you tend to ask us the most questions about. And I think, Clay, you've been thinking more and more about -- in fact, we like diversification of sites. We like diversification of size and sites. And maybe if you can talk a lite bit about customer...

I think the way to think about it is that large sites are great except for all the downside of large sites. So -- or in other words, I actually think that the gigawatt scale campuses make a lot of sense for training workloads. Because you need to be able to harness all of that capacity in one area. But as you see inferencing get larger and larger and larger, but this not inferencing is not going to be 2-megawatt sites everywhere but we're delivering a lot of 50, 100, 200 megawatts, which, by the way, used to be considered very large sites, right? And those are available all over the world, things to be distributed partly for latency reasons around certain inferencing use cases, and where is it -- where is the power available and where does it make sense. So I think you'll continue to see a mixture because of the customer demand in the next year, right? You have training, but you also have lots of influencing demand. So for that.

And then in terms of the customers, I mean, I think I had said a good job with we have 1,000-plus kind of AI customers on the platform. We see it from all angles. It is the big model providers that you think about, but it's this trading analytics firm who wants to do their own analysis or this startup who is going into space and existing like I think there's a misunderstanding of where AI demand is coming from, as AI becomes more and more ubiquitous. This is just computing. Like I had a slide yesterday in my keynote where I showed that we're selling far more AI accelerated compute and traditional compute, even while positional compute has gone way up.

This is just a new -- this is not a fab like we don't -- when the mini computers came along, people didn't ask, but what about the punch cards. The reality is this is the new normal for computing and the demand for it is immense because we had a fundamentally different capability than we've seen before. And I think we will see massive demand from the big players, but also we see huge demand across the portfolio from new start-ups to establish enterprises. We will continue to build playing a more diverse set of options just from a risk management perspective. But obviously, big sites still a role that -- the thing [indiscernible] all allocations. You can't take a bunch of small data centers and join them together as well as you can to the large campus. So you'll see -- continue to see a mixture of both from us. And I think you'll see that kind of throughout the entire industry.

It's not just Oracle, you'll see a wide diversity of that. And what I would say, though, I think that people focus on the big ones. There's a lot, there's a lot of little stuff, right? If you look at that 10 gigawatts, forget how we got here, Oracle will be definitely less than half of its capacity is in these Gigawatt sites, right? That's 5 out of 10 is 50%. So that's what I mean is it like -- I think the difference is that it's so much harder to write a headline about 17 different data centers right across 4 countries. But that's why we're able to actually go out and predictably deliver on things. It's how you manage risk around regulation, it's how you manage risk around supply chain. That's just -- I think you'll continue to see that from everybody.

Okay. We'll take our final question here on your far right.

Kirk Matern with Evercore. Clay, just following up on that train of thought. One of the things that's differentiated you for a long time is the composable nature of your data centers. And as we think about the diversity of AI demand, including what I think is going to be a lot of sovereign II and countries want to have their own version of AI. Can you just talk about your positioning within that world?

Today we have the option with a lot of our different providers to actually offer Alloy and they're partner operated, and they have complete control of those regions. All of those regions are the same. We deployed GPUs in all of those. What kind of limits the scale and scope of them, it's the appetite of that customer and/or partner provider, right? If we have Alloy partners who show up and they're like, hey, I want to deploy 10,000 GPUs. We go awesome, let's do it. That's -- so I think as you look through like how sovereign AI evolves or just sovereign compute evolves, we're very well set up for that. And you'll continue to see other -- there's things you're continuing to see, I think, more and more evolution from us in different geos around the world, right? We've invested very heavily in that kind of strategy in the Middle East. We have many partners there. We have a lot of partners in Asia.

I think you'll continue to see that evolve. And as different areas of the world become more concerned with sovereignty, we have the product that I think is by far the best poised. It allows them to start small, taking -- we grow with them incrementally. We actually have a proven track record, having done this for 5, 6, 7 years. So yes, I think there will be an immense amount of demand for that. But it's a little bit slower. It's kind of like the smaller data centers, right? It doesn't rise to the level of headline, but it's funny I was -- on Hilary slide, she was sitting there and showed I think by '28, we'll have like 200-plus regions. And I was like, of course, you just have.

That's not normal. I think if you look at many of our competitors, that's not how they do it, and we just kind of take it for granted now, but that will continue to pay dividends for us over the next few years.

Thanks, Clay. Before we let you guys go, we do have one more thing. We'd like to bring Ken to the stage along with Bill. I think most of you know, we mentioned it a little bit throughout the day that Ken is retiring as the Head of Investor Relations from Oracle. So we wanted to make sure that we give them a big, big thanks. I think they play a big role here and thanks very much.

In the spirit of disciplined and capital allocation, we didn't get you anything. As a shareholder, thank you. Well -- okay. Well, first of all, my plan to slip out the back door has not worked out so well. I was saying I want to just [indiscernible]. In all the years that I've been here working with Safra, she talked a ton of things a ton. And one of my favorites is less is more. So I'm going to be brief.

I want to let everybody know how appreciative I am to Safra for having given me the opportunity all those years ago to run the IR program. It fundamentally changed my life [indiscernible]. And for that, Safra, thank you. Truly, thank you. I also was very fortunate to have great mentors and they're here somewhere. They're hiding somewhere. Like that Doug carrying a [indiscernible] were phenomenal mentors to me, they guided me along the way. And I was also lucky enough to have Luis Montano on the team in that Luis quietly and efficiently, get a lot of the hard work in the group. But most importantly, I don't want to get a motion about this one, but I'm most grateful to my wife Barbara and my families who have supported me through all these years.

The reality is Oracle, it's just a passion project and it takes a lot of time away from your home life. And yes, PARP was awesome. [indiscernible]. Now for the good news. The good news is I could not be more pleased to have Bill coming on board. Bill Choi, I've known for almost 20 years. And in all those 20 years, I have been a fan. The guy is absolutely great. And I know as I step out, I know that you all are in great hands with build. So with that, thank you all. God Bless. And Bill, let me turn it to you for closing.

As you said, I've known you for a very long time, and I personally very much appreciate our relationship over the years. Congratulations on your retirement, and thank you for your many years of service to Oracle.

Well, it's great to see all of you. I'm extremely excited to be here at Oracle at this important time. As some of you know, I have been on your seat for much of my career as a sell-side analyst covering technology. And then joining in an in-house role as Head of Investor Relations and Strategic Finance. I see many familiar faces. So I certainly look forward to reconnecting with many of you. And for those who I don't know yet, I look forward to meeting you. Before we close, I also want to make a quick announcement that all of our slides today will be uploaded to the Oracle Investor Relations website and the video replay will also be available shortly.

With that, I want to thank Mike, Clay and Hillary and all of our presenters today. I also want to thank everyone of you here who showed up in person. And for those joining us on the webcast. We thank you for your interest in Oracle, and we look forward to seeing you soon. Have a great afternoon.

Good day, everyone, and welcome to the Oracle Corporation First Quarter Fiscal Year 2027 Earnings Call. Just a reminder that this call is being recorded. [Operator Instructions]

I would now like to hand the conference over to Mr. Ken Bond, Head of Investor Relations. Please go ahead, sir.

Thank you, Mariam. Good afternoon, everyone, and welcome to Oracle's First Quarter Fiscal Year 2027 Earnings Conference Call. On the call today are Chief Executive Officer, Mike Sicilia; Chief Executive Officer, Clay Magouyrk; and Chief Financial Officer, Hilary Maxson. A copy of the press release, including financial results, tables and supplemental financial metrics and guidance is now available on our Investor Relations website. Also available on our website is the slide deck that will be used in this call and a GAAP to non-GAAP reconciliation.

As a reminder, today's discussion will include forward-looking statements, and we will make some important comments around factors relating to our business. These forward-looking statements are also subject to risks and uncertainties that may cause actual results to differ materially from the statements being made today. As a result, we caution you from placing undue reliance on these forward-looking statements, and we encourage you to review our most recent reports, including our 10-K and 10-Q and any applicable amendments. And finally, we are not obligating ourselves to revise our results or these forward-looking statements in light of new information or future events. Before taking questions, we'll begin with prepared remarks.

And with that, I'll turn the call over to Hilary.

Thanks, Ken. Hi, everyone. Great to be here with you all today. And like Ken said, you can follow along with our remarks in the earnings slide deck on our website and via the webcast.

If I need to describe this quarter in one word, I think it would be acceleration, as we're seeing an acceleration in execution across the company translating into our top and bottom line results. Q1 was another record quarter, driven by strength in both our Cloud Infrastructure and Cloud Apps businesses. Total revenue was a record $19.3 billion, up 30% year-over-year in U.S. dollars. And for the first time, Q1 total revenue grew sequentially, an important sign of our continued progress in building scaled infrastructure.

Historically, a record Q4 was followed by a lighter Q1. But as we accelerate across the full technology stack from infrastructure to database to software, that's no longer the case. Cloud Infrastructure revenue for Q1 was $7.4 billion, up 121%, reflecting strong execution as we brought record levels of new megawatt capacity online, supported by a continued strong demand environment for compute and our database services. Cloud Apps were up 10%, with Fusion and our industry apps tracking well of that, and Mike and Clay will give more details on those businesses in just a moment.

Our non-GAAP operating income increased 31% in U.S. dollars to $8.2 billion, driven by strong revenue progression boosted by operating leverage. Our gross margin did decline as expected, driven by impacts from ramping up our data centers and the acceleration of infrastructure revenue. However, this was offset in the quarter by lower operating costs and strong operating leverage tied to simplification and efficiency actions.

Net-net, our operating margin remained around flat for the quarter at 42% on a non-GAAP basis. This all translated to a strong increase in our non-GAAP EPS of plus 30% in U.S. dollars, reaching $1.92 for the quarter.

The last point I'll make on the financial highlights is that our remaining performance obligations, or RPO, increased $26 billion from Q4. There are two things happening here. First, we continued to grow our RPO during the quarter to support future revenues. And the vast majority of those new contracts would be a prepay or bring your own hardware or similar mechanic, so [indiscernible] incremental capital from Oracle. Also, that new RPO won't impact our CapEx or revenues until fiscal '28 or beyond.

Second, we started to see a strong conversion of our RPO into revenues this quarter, driving our Cloud Infrastructure results. We've added a few slides here where you can see that strong inflection point in our RPO converting into revenues and operating profits.

First, in Cloud Infrastructure revenues, I already mentioned the plus 121% growth for this quarter, and that's after a Q4 of plus 93%. And we expect acceleration to continue in the remainder of fiscal '27 as we convert more RPO into revenues. We now expect around half of our RPO to convert into sales over the next 36 months.

Next, not surprisingly, you can see the acceleration in our total company revenues here shown on a trailing 12-month basis, driving growth in Q1 to 5 points higher than our Q4 as Cloud Infrastructure accelerates and becomes a larger and larger contributor. Lastly, our growth in operating income, also on a trailing 12-month basis, shows a similar strong acceleration from 16% to 21% between Q4 and Q1.

Now to our balance sheet and cash flows. We drove record cash flow from operations of $23 billion in Q1, again reflecting our strong execution against a backdrop of strong demand as well as collection of customer prepayments. Our CapEx for the quarter was $28 billion, leading to negative free cash flow of $5 billion. And our net cash CapEx, so net of prepayments, was $18 billion for the quarter.

To note, our CapEx will not be linear throughout the year. We continue to anticipate $90 billion to [ $95 billion ] in CapEx for the full year, with not more than $70 billion in net cash CapEx. Lastly, we're quite pleased to announce that we completed our previously disclosed $20 billion at-the-market equity issuance in entirety during the Q1.

With that, let me turn the call over to Mike and then Clay to get into more deals on our Cloud Apps and Infrastructure businesses.

Thanks, Hilary I'll start with some additional color on the applications business. We continue to see the power of application suites in the minds of our customers. They are investing in trusted complete solutions that now seamlessly blend agents and applications together to run their businesses. The introduction of AI is an accelerator, not a replacement for packaged applications. As such, our decade if experience and expertise running business processes across every industry in every geography for organizations of any size gives us the understanding of how to help them succeed.

Before AI came along, application suites had already proven their effectiveness. Companies have been able to increase their profit margins because end-to-end automation with standardized and efficient business processes proved to be much more effective than one-off custom solutions. But that did require organizations to follow workflows and processes as designed in the system, something that many struggle to achieve consistently across functions, teams and regions.

AI changes this dynamic. Rather than asking every employee to navigate and execute a process exactly as the system expects, AI agents can perform tests using the organization's established workflows and business rules. Employees then shift to overseeing agents, resolving exceptions and applying human judgment where it matters most. By combining applied AI with decades of sophisticated [ rules ], regulatory compliance, security models, data models and customer configurations, we enable customers to continuously realize AI's value while keeping their data secure and their operational guardrails intact. This allows organizations to harness the power of our application suites more easily than ever before. We are incredibly confident in the potential for this new paradigm to deliver much more rapid ROI for our customers.

At AI World in October, we will unveil a new agentic AI [ rater ] poised to redefine how customers deploy Oracle applications faster, simpler and at a dramatically lower cost. Working alongside Oracle and customer teams, AI agents will automate and orchestrate implementation at an unprecedented scale, compressing SaaS deployments from years to months and months to weeks. It's really the power of these things together that reinforce my belief that the growth of our applications business is only going up from here.

In Q1, we had a strong quarter in many of our SaaS offerings, driven by the demand environment that I just described. In total, our SaaS business grew 10%, with Fusion growing at 14%. Our Oracle Health business continued to accelerate. And although we don't specifically call it out, our industry applications grew at greater than 20% in Q1.

As I mentioned last quarter, NetSuite saw some slower decision cycles last fiscal year, and therefore, the growth is a little lower than the rest, but we have an exciting new product generally available that I'll speak about just a bit. Now a few customer callouts from a much longer list than the quarter. Uber Technologies, Stanford University and Mitsubishi UFG Bank in Japan all went live and/or accelerated their usage of Fusion. [ Bivarkifier and Safety ] chose Oracle's complete application suite from industry apps to Fusion, including Fusion agentic applications. [ Johnson Controls ], the Saudi National Bank, [ Guidewell Mutual ] Holding Corporation, a health solutions company serving more than 45 million people, and Petronas, Malaysia's national energy company, each added Fusion agentic applications this quarter to drive better outcomes.

So let me share just a few stats around our embedded AI usage and progress in the quarter. Customers used our embedded AI capabilities more than 150 million times during the quarter, with usage growing 42% sequentially. Our AI agents executed more than 3.5 million times in production during the quarter, nearly doubling quarter-over-quarter. Customers have over 2,300 AI agents in production, and that's up 90% quarter-over-quarter. Overall, AI production usage across Fusion alone consumed 900 billion tokens during the quarter. I think it's fair to say that customers are using our AI built into our Fusion applications and our application stack every day.

Now turning to NetSuite. We're announcing the general availability of our new AI-powered offering called NetSuite Next. This presents an agentic experience that is simpler more powerful, and it's infused with AI across the workflows that customers rely on every day. It's easier to adopt. It's more productive from day 1, and it's more valuable as customers grow.

Additionally, the NetSuite AI Connector service, which lets customers securely connect their NetSuite data to leading AI assistants of their choice, including ChatGPT and Claude, is already 1 of the fastest adopted capabilities in the whole -- the entire history of NetSuite with more than 10,000 customers already using it. Personal care company [ Every Man Jack ] estimates that the service alone will save $350,000 annually and nearly 5,000 hours of work.

This month at our Oracle Health and Life Sciences customer event, we will debut our all-new agentic care management system alongside a world-class lineup of external speakers. More than an EHR, this system connects clinical research and care delivery, reduces the burden of records management for patients and enables providers to practice at the top of their license with AI as a user interface.

There are a few AI missions that matter more, helping deliver better care while freeing providers focus on what matters most: their patients, not computer systems. The proof points around Oracle's AI offerings are definitive, and we remain confident in the opportunity ahead.

And with that, I'll turn it over to Clay.

All right. Thanks, Mike. OCI continues to grow quickly by delivering the capacity our customers need. We delivered 850 megawatts of AI capacity containing more than 300,000 GPUs to customers since the end of Q4. Delivery in Q1 is almost 3x what we delivered in all of Q4 and 73% of the total capacity we delivered last fiscal year. This reflects years of investment in every aspect of infrastructure, from data center design, through supply chain and manufacturing to installation and operations.

Customer demand continues to support this investment. We closed more than $30 billion of additional AI contracts in Q1 without requiring additional capital from Oracle. Our ability to operate a large multi-tenant fleet remains a significant advantage. GPU utilization remains extremely high at 97.9% in Q1. GPU longevity and value continue to impress. Of all the GPUs that came up for renewal in Q1, that capacity was renewed or resold at a 20% premium to prior contracts. The majority of those GPUs are 4 years or older.

We see a long useful life with increasing value for the AI capacity we're deploying. Abilene continues to deliver at an extraordinary pace. We delivered 131,000 GPUs there in Q1, 1.9x the volume delivered in Q4. Six of the 8 campus buildings representing 618 megawatts and 75% of total capacity have now been delivered to the customer. Customer acceptance has compressed to only 24 hours, showing that the systems arrive ready for customer workloads.

The recently released GPT 6 Astra was trained at our site in Abilene. Shackelford is our next gigawatt scale campus and is progressing well. NVIDIA Vera Rubin systems are performing better than expected across hardware quality, manufacturing yield and performance. We will deliver our first Vera Rubin systems to customers in Q2.

Our database cloud business is also growing quickly, Multicloud Database database revenue grew 353% year-over-year, and multicloud customers grew 180% year-over-year. We completed our planned Azure and AWS regional footprint expansion, reaching 70 multi-cloud database regions and 119 availability zones. This gives customers a consistent way to run Oracle AI Database next to their applications and data in the cloud they choose.

We also made Oracle Interconnect for AWS generally available. With this launch, OCI now has private high-speed connections to all hyperscalers with no data transfer charges. That makes it easier for customers to build distributed applications and migrate workloads across clouds.

The pace of AI innovation is increasing across the ecosystem. We expanded our OpenAI relationship to offer OpenAI API access, ChatGPT for Work and [ Codex ] through Oracle Marketplace, including GPT 6 Astra. We are bringing Gemini models to Oracle's enterprise applications, and we released new Grok reasoning, multimodal and text to speech models. We also continue to expand the open source model catalog, including new models from NVIDIA, [ Quinn ], Google, DeepSeek and others.

We announced a multiyear partnership with [ Continuum ] to offer its Helios quantum computer through OCI. Helios will operate in a U.S.-based OCI AI data center, enabling hybrid quantum and AI workloads for enterprise, research and AI customers. Oracle Apex 26.1 brings these infrastructure, database and model capabilities together for application developers. Apex already runs more than 2 million applications, with thousands more added every day.

[ Apex Lang ] is a new technology that represents an Apex application as structured human readable application definitions that can be stored in source control, validated and governed. AI coding agents generate and modify those definitions while the Apex engine continues to provide the security, reliability and operational controls required for enterprise applications. [ Filters ] gain the speed of generative development without the downside of difficult-to-maintain opaque application code.

We are taking the same approach with the Oracle AI Data Platform. AI Data Platform is now integrated with [ Codex ] and Claude Code, allowing developers to work with AI data platform data, knowledge and capabilities from the coding environments they already prefer. We are also adding advanced ML ops capabilities and working with early launch customers on business knowledge models, semantic region engine and deep insight agents. These capabilities ground AI in the meaning of an enterprise with context, [ cement ] and governance. They also bring AI-driven analysis into Fusion data intelligence and Oracle Analytics Cloud, where many customers already manage their most important business data.

Taken together, Q1 shows how the pieces reinforce 1 or other. We are delivering data center and GPU capacity at a pace that would have seemed impossible only a year ago. Customers are signing new contracts, renewing capacity at higher prices and keeping the fleet almost fully utilized. We are placing Oracle AI Database in every major cloud and making more proprietary and open models available on OCI. We are then [ initiate ] models to enterprise data, applications and developer workflows. That combination is why demand continues to grow and while we remain confident in the long-term value of the technology we are building.

And with that, I'll hand it back to Hilary.

Thanks, Clay. I'll finish with our guidance for Q2 2027 and the full year. In Q2, we would expect growth in total revenues of between 30% and 34% in U.S. dollars. Of that, we'd expect growth in cloud revenues in U.S. dollars of between 65% and 71%. In non-GAAP EPS, we expect between $1.85 and $1.93 in, up between 21% and 25% in U.S. dollars. And that excludes the gains from Ampere we booked in Q2 of last year.

For the full year, reflecting the strong execution and acceleration you see in our Q1, we're upping our guidance to at least $90 billion in total revenues. That's a year-over-year increase of plus 34%, and to $8.10 for our non-GAAP EPS. Lastly, a small note from my side to please make sure to mark your calendars for our Investor Day coming up in October.

With that, I'll turn the call back to Ken for the Q&A.

[Operator Instructions] Your first question comes from the line of Brad Zelnick of Deutsche Bank. Please go ahead.

Great. And first off, Ken, knowing that this is your last earnings call, I just wanted to congratulate you on your retirement. For nearly 2 decades, you've been the face of Oracle to the investment community, and you've done so reliably and with such high integrity, and I know you'll very much be missed.

As for my question, I fully expect Oracle will continue to be among a small handful of market leaders for AI infrastructure. And you've told us that fiscal '27 and '28 are peak CapEx years. But at the same time, others in the market are spending hundreds of billions of dollars on capacity. With no end in sight, how should we think about Oracle possibly slowing down spending beyond the next 2 years if others aren't? And is there a scenario where we could see even higher peak beyond fiscal '28? What will guide your investment? And then related to that, given the current state of the backlog as it stands today, when should we expect to see the company return to generating positive free cash flow?

Thanks, Brad. This is Clay. I like -- I think -- I think there's like 4 questions in there. So I'm going to answer the parts that I want to answer, and then I'll make Hilary do the hard work. So look, I think you've heard us talking about for the past several quarters, we're constantly finding [indiscernible] ways to fund the business. And 1 of the mechanisms that we have to fund our business is obviously that we go out and we spend our own capital.

But we've invested very heavily in relationships with different suppliers and vendors, invented new business models, including bring your own hardware, all of which have different ways of spreading out that capital. And so I think that we have to separate out in our minds what Oracle spends as CapEx directly, uncouple that directly from how we think about how the business can grow. Because from our perspective, I think we see ways that -- we see ways that clearly, capital is still required to do this work. But it doesn't all have to flow from Oracle's side, doesn't have to be Oracle CapEx.

So I don't see it as a limitation to the growth of our business. I think it just represents the continued evolution of the business model that we're developing as kind of this AI expansion continues. Hilary, do you want to talk about some of the other answers?

Yes, sure. So the other question I think you asked was about free cash flow. We haven't given a particular time frame on that yet, and we don't expect to give that today. What I would say, though, is that each of these projects that we're doing by nature is a strong free cash flow generating project. So as soon as they ramp up, very shortly thereafter, they're delivering a free cash flow conversion ratio of something like 100% to post-tax EBITDA. So in fact, the business by nature is somewhat "self-funding" at some point in terms of throwing off a lot of free cash flow.

So we haven't given the time frame on that, but you'll see as we continue to ramp over the next quarters, what that could look like. Again, we expect it to be a reasonably time frame past the big ramp-up that we have going on today. The only decision there, and you pointed it out, would be about more growth CapEx, and that's something that we want to continue to deploy at the right levels to grow the business.

Your next question comes from the line of Siti Panigrahi with Mizuho. Please go ahead.

Great. And can I also echo my congratulation, and best wishes on your retirement, you will certainly be missed. And going back to the question, there's been a lot of speculation about potential data center delays in New Mexico and Wisconsin. But you delivered on your Q1 target and now is in raising full year 2027 revenue guide to now at least $90 billion.

So can you give us an update on where both sites actually stand today? And is there any risk to the delivery time line for those data centers that could pose a risk to your '27 revenue expectation? And as you continue to grow RPO, how comfortable do you feel in your ability to secure and bring capacity online to support that growth going forward?

Yes. Thanks, Siti. Okay. So I'll answer the specific pieces of that, but I think it's important to have some context before we go forward. New Mexico and Wisconsin are very important large sites for us. But I think it's important to have some context. We talk about these sites kind of as being around 1 gigawatt a piece. We just delivered 850 megawatts in Q1. And what that means is that, and as you can see, neither Shackelford, nor New Mexico or Wisconsin or Michigan were delivering in Q1. So we have a large, diverse broad set of data center developments going on throughout the U.S. and around the world to deliver capacity to customers.

Now some of these sites like New Mexico and Wisconsin, obviously garner a lot of attention. There's a lot of discussion about them. But I think it's important people realize that all of our eggs are not in a single basket. The next thing I would make sure everybody understands is that when these large sites are built, they don't all come online at once. So let's say that you have a gigawatt site. It doesn't -- and it's supposed to start delivering, let's say, in January of a year. It's not like in January, you get a gigawatt of capacity. It's phased over many quarters.

So it's not as though if you have a delay compared to a plan that you have some massive thing that hits in a single [indiscernible]. So -- and then the other thing I would say is that anyone that's been in the business of doing construction or large-scale infrastructure development, if they -- if their plan relies on 100% achievement of every 1 of their deliverables, we have a term for that. It's called a bad plan. And so we try real hard not to make bad plans.

So we obviously know the difficulty and the complexity of what we're doing. And we don't assume 100% of everything is going to work all the time. And we have backup options for those things as well as we don't count that everything is going to get done exactly on time, all the time.

So now moving specifically to the question around Mexico and Wisconsin. New Mexico is an interesting location. We're making very good progress in terms of construction, data center is definitely on track. We're going through the process of acquiring our air permit. And we've got -- the technology that we'll be deploying there is Bloom fuel cells, which is by far the most environmentally friendly way that we can do on-site power generation, has extremely low water consumption, has extremely low emissions compared to really any other way to kind of do on-site generation.

And so we're very confident that as we continue through this process, we'll work with the local regulators and community citizens in Dona Ana County and with everybody else in New Mexico. But we're just going through the process. And I don't think that's rare for large projects like this one.

In Wisconsin, we're not doing on-site generation. We're really working with our partners across the board to design and deliver that energy capability through the grid. But again, working through the process, these are complex projects. And again, in Wisconsin, data center delivery is actually very much on track and going well, working with the Public Service Commission and ATC and [ We Energies ]. We're constantly evolving aspects of the energy design and delivery plan. But we feel very confident in both of those sites.

To the question about impact on FY '27 revenue, neither of these sites will have any impact into our FY -- our previously stated FY 2027 revenue or earnings guidance. So I would say is -- we have some sites that are ahead of schedule, some sites that are more difficult than others. We work through all of them, and we have a plan in place both in terms of risk management, as well as we take that into account when we do our forecasting for how we go about setting expectations for ourselves and then communicating those expectations for you.

And the last question is like how do we feel about bringing more capacity online. I think we feel very excited about it, both in the -- obviously, the environment continually changes. It used to be that the constraints were GPUs and fabs, then constraints moved to power generation. There's data center constraints.

But the world is a big place. There's a lot of demand for this capacity, and we're pursuing all the different avenues to bring the capacity online. So we remain very excited and very confident in our ability to keep meeting both current RPO as well as the future RPO growth that we are expecting to see.

Your next question comes from the line of Raimo Lenschow of Barclays.

Thank you. First of all, all the best for Ken, and thank you for the team for making Ken's last quarter like such a great quarter. My question is, again, on the data center side. Obviously, there's a lot of talk about component price increases, et cetera.

Clay, can you talk a little bit about like how your pricing is evolving between contract pricing, spot pricing? And you gave us a framework on gross margins last year. Is that holding? How should we think about that in this ongoing dynamic on component pricing?

Sure. But definitely, prices -- in a world where demand exceeds supply, we -- typically, prices don't go down, they do go up. And so I think the net effect is that obviously [ costs ] more, but then we have to charge more money for them so that we get compensated. And we're doing that across all of these different businesses.

No, we don't expect this to have an impact on our gross margins. So I think the previous guidance that we've given you there is still remains true. I will say is that, as I mentioned in my prepared remarks, I think there's -- a lot of people have been very concerned about what is the useful life of the different hardware assets? What is the useful -- what is the demand going to be for data centers and the power capabilities that come with it?

I've been in the infrastructure business for pretty much all of my professional career. I've been doing it now for 12 years at Oracle. 1 [ in Variant ] that has been true so far is that the demand for server side computing and data centers has only gone up. And it turns out so far that everything we see about the AI use case is the same way, but only more so, right? The fact that when we go out and we have capacity up for renewal, that we actually can achieve higher prices to the order of 20%, all of those are very positive signs for -- both the continued demand, the growth and then the profitability of this business.

Your next question comes from the line of Mark Moerdler of Bernstein. Please go ahead.

Ken, we're going to be sorry to see you go, but I'm sure you have an amazing bucket list to add of trips and things to do. So I'd like to focus in on the RPO. Can you give us more details on the drivers of the growth of RPO that is not requiring additional CapEx, the prepaid and bring your own chips? Is this AI labs, semiconductor companies? Or is it that sovereign? In fact, we've not discussed sovereign cloud in a while and we haven't discussed that sovereign AI at all. Can you explain how that side of OCI is going and how that will impact CapEx and margins?

Sure. So let me make sure we clarify 1 thing first, and then I'll dive into the pieces. I didn't say -- and I don't think myself nor Hilary said that it doesn't require additional CapEx. We said it doesn't require additional cash from Oracle, right?

So at the -- I'm not an accountant, but I get to play 1 on TV sometimes. But the intention of what we're trying to say is that while there are -- clearly, our capital expenditures does not require Oracle to go out and find additional cash to do it. Now the question becomes, well, how do you do that? Well, we have a variety of different models. Sometimes, it's working with our suppliers through different financing arrangements that allows us to pay for the capacity as the customers pay us. That's 1 mechanism.

Another mechanism is that a customer says, hi, I'd like to pay for the hardware, but use your operational ability and your cloud infrastructure technology assets in your data center to go out and actually turn that into an AI cluster. It's a different option. The third option is that the customer has been able to raise a start-up, maybe it's an established company and says, hi, I would like to pay you upfront as a prepayment. And in return, that doesn't require you to front the cash to go out and spend those -- your dollars and that capital expenditure. So we have different models for achieving that goal.

In terms of the [ customers ] and where we see that demand, it's really broad based, right? It doesn't matter if it's a startup or the most valuable investment companies. There's an understanding of this model right now that the access to capital and different ways of funding that are a constraint and the industry adapts to allocate that in the most efficient way possible.

Now specific to your sovereign cloud question, look, our alloy business is doing well. We've got a lot of partners in Japan. We've got great partners in the Middle East. We've got partners that are both going after the commercial business in a more sovereign way, as well as kind of more government-focused sovereign cloud. That continues to expand broadly. We're seeing lots of demand.

And also, it's actually tied to AI as well because many of those customers, we offer GPU capabilities for those customers, and they deploy that for those on sovereign workloads. Obviously, we also have a very large and very rapidly growing general-purpose cloud business that we don't talk about quite as much, but that is growing rapidly and has great growth rates, great margins and does require some capital, but not as much as the giant AI clusters.

Your next question comes from the line of John DiFucci of Guggenheim Securities. Please go ahead.

I'm going to make a quick comment here, too. I've covered Oracle for almost 3 decades. I think Ken, you've been there for close to 2. In this business, it's always nice to develop professional friendships, but it's even nicer when they become personal friends, which is what I count you as. And importantly, I've learned a lot from you. So I just want to say thank you.

But as far as what Raimo said, I'm glad, team, you had a good quarter, but not so much for Ken, for me because you're my best idea, and that's -- it's all about me. So I think my question here is for Hilary. Hilary, gross margins came down meaningfully this quarter. But you had so much capacity coming online this quarter. I assume that before that, there were a lot of expenses associated with those deals coming online, but they didn't have much revenue yet. Nevertheless, it was a big drop in gross margin. It is something that we all on this call talk a lot about when we're talking to investors.

But as you pointed out, operating margin is flattish, up a little bit year-over-year. I know operating margin is the North Star because that's closest to the bottom line profit in eventual cash flow. But how should we think about gross margins going forward in addition to operating margins, that whole ecosystem?

Sure. So for me, gross margin, obviously an important thing to follow both internally and externally because particularly -- actually in both of the business models or all 3 of the business models that we have. It's really a health indicator of the business. Are you -- can you price? Are you pricing at the right level? Are you able to manage your input costs? So gross margin, to me, is always an important indicator, probably even more important internally for us to double check. And I think investors obviously want to double check. It's something that will change very quickly, for example, if we don't have the right pricing model.

When we talk about driving value, though, and driving value for the business, for me, operating margin is probably the ultimate point that we want to follow. So gross margin, like you mentioned at the moment, there's a number of things going on. We have both the ramp-up in data centers, plus we have an adjustment across the 2 business models that we have in the business, software being a much higher gross margin business, but with higher R&D and sales costs below gross margin, infrastructure being a lower gross margin business, and we've talked about that, and we gave the numbers. Clay has given the expectations for quite a bit of that business, not database, obviously, but the more AI infrastructure and cloud side.

But that business, by nature, has much lower R&D in sales associated with it, at least a company like Oracle, where we can effectively gain from all of the R&D that's already been done and being done across rest of the company. So really how to watch how we're going to drive value out of the business over time, I think operating margin is really the key metric that we would look at. Again, not forgetting about gross margin, something to look at, but operating margin being very, very important tied to value.

That's for your -- and that's where we'll focus on. But how should we just so -- because most people -- I think the Street typically overestimates which your gross margins are going to be the underestimated what the impact is going to be to operating margins. So it should -- I don't know if you're prepared to say, like should gross margins continue to go down now from here? Or should they be steady around these levels for the rest of this year? And how should we think of that?

Yes. So we've mentioned already, I mentioned in the Q4 that we would expect a step down in gross margins this year. You can see the EPS [ give ] those. So you can see what we might expect in terms of operating margin. Over the next couple of years, as we finish the ramp-up, you can reasonably expect that gross margin would flatten, I would say. But we haven't given any particular guidance today. We'll speak more about some of the stuff on the upcoming Investor Day, though, in October.

Your next question comes from the line of Brent Thill with Jefferies. Please go ahead.

Mike, I'm seeing double-digit growth in SaaS. Maybe if you can walk through what you're seeing over the next couple of years? I think there's been the fear of the SaaS business across the industry taking ahead because of AI, but it seems like that's not your view. Give us your perspective on what you're seeing in terms of keeping sustainable double-digit growth.

Yes, sure. Thanks for the question. Here's how I think about our SaaS business, and I'm going to take the liberty of broadening it a little bit because I think there are a couple of key adjacent things that are very important in SaaS.

First is we have a highly differentiated offering in that we have end-to-end suites that automate complete industries: healthcare, retail, telecommunications, construction. There's a whole list of these industries where we have horizontal applications, vertical applications. And if you look at our 2 very big categories in the quarter, both Fusion and industries, our industry locations is growing very nicely.

The fact that customers can get them as a complete suite and a complete package has been a differentiator for Oracle for years now. And now you layer in AI embedded into those workflows, adding AI as a service as part of the regular application updates that our customers are getting. And we think, Brent, it's a really nice way for customers to get value from AI or ROI from embedded AI very, very quickly without having to bolt anything as a sidecar.

Now the next layer is our agentic, our Fusion Agentic AI studio, which allows customers and/or partners to build their own AI agents right inside the same platform. That's not a different platform. It's not a different control plane. It is the same control plane and the same platform that our applications are running on, which means that Agentic Studio allows customers to build their own agents or partners, gets all of the same quarterly updates, gets all of the same security passing and is available as a complete service to our customers. And you're allowing customers to position AI as a UI on top of a very complex set of business rules, on top of a highly differentiated security model and, of course, data models that have evolved for years and years.

So you put the onsite horizontal applications, the vertical applications, the agentic studio together, and we think that's very compelling. But if that's not enough, the next piece that we're -- as I mentioned, that we're unveiling is AI assistant go lives. These are -- in many cases, we're taking on very hard work in helping our customers solve very mission-critical industry-specific problems.

And sometimes those go-lives can take a long time. They're complex. There's data migrations. Well, AI has given us a tremendous ability to accelerate those go-lives. And we have some proof points for the tooling that we've rolled out to date, and we'll roll out a bunch more of it at AI World. But we've seen like, for example, complicated go-lives in our healthcare applications that were in the high double-digit months coming down to single months for go lives.

In our NetSuite applications, we've seen early customers leveraging these AI tools coming down from double-digit months down to single-digit weeks in order to be able to go live in production. So we think that does 2 things. Number one, it helps us -- it helps customers, rather, get to value from AI more quickly than ever and certainly at a lower cost. So number two, in some of these very complex industries, there's ramps associated with these go-lives and it allows us to unlock the ramp and recognize revenue more quickly. Then we may -- we have in the sort of manual implementation piece.

But I think the other piece that's quite important is, as we mentioned in the press release is, again, part of the same platform running our applications on the AI data platform, which automates the creation of enterprise ontology. So while we are very honored to serve our customers with a very widespread and very large suite of applications, there are other applications as well. And being able to take those applications that our customers are running create an enterprise ontology as an automated service and then layer those AI agentic studios on top of that. When you put all that together, right, I just don't thing that anybody else in the market is delivering all of that as a cloud service, certainly not in highly regulated industries, certainly not at the scale that we are.

And then 1 other thing I'll mention, our SaaS business is that our SaaS business is also a wonderful lead generation business for our IaaS business. Our SaaS customers are also buying OCI. They're also have other workloads, non-Oracle workloads that they're more than happy to leverage OCI. They get a very good test of the performance and scalability of OCI as a SaaS customer because they are inherently an OCI consumer at that point as well.

So we think about it as really the SaaS business being part of our overall solution set that we're delivering to customers. We continue to invest in that solutions that we continue to make it easier to go live and add more and more tooling to allow customers to configure and leverage AI. So for all those reasons, we're quite optimistic on the future of our applications business.

Your final question comes from the line of Kirk Materne with Evercore ISI. Your line is open. Please go ahead.

Ken still has an Analyst Day to get through. So I'll wait and say congrats to him in person in a month or so. But Mike, you mentioned the AI data platform, you guys put it in your press release. So I just wanted to double click on that. It seems like an important way in allowing customers to plan against the proprietary data. Can you just help us understand the business model around that? Does it drive incremental consumption of Oracle Database, OCI? Or do you see that being sort of a stand-alone software revenue opportunity? I was just curious kind of how we should think about that playing into the financials over time.

I think you sort of clicked on the answer really to the business model is all the above. I mean certainly, we can run this in a model where we're consuming 100% of non-Oracle work. This is by no means specific to the Oracle Database on the Oracle applications. The AI data platform is agnostic and able to pull in the ontologies from any data source. In fact, we have hundreds of data sources that we're doing automation for today.

So whether it's just pure consumption of AI data platform, whether it's used in concert with our applications or whether it's used in concert with OCI, we're more focused on allowing the customer to make the best choice or the partner to make the best choice to use them. So I think it really helps all of our business.

I would say also that the Oracle Database still remains the most -- probably the most coveted custodian of mission-critical data in the world. And certainly, as we're able to automate ontologies and have the Oracle [ Space ] as 1 of the primary fees, at least for the mission-critical data that, it also helps us unlock our ever-growing Multicloud Database as well, where certainly, customers would like to take those on-premises versus the Oracle Database and very easily move them into the cloud. As we continue to invest in making that available in every cloud, in every region, we think the AI data platform will help drive that growth as well.

I know you guys had a huge services system, but will you guys be putting forward deployed engineers around that to try to help customers understand some of the opportunities things they can expose? Perhaps they haven't been able to expose before? The database just kind of is there a go-to-market to it, too?

Yes, absolutely. And we are already investing and deploying forward deployed engineers at our customers. That's true for the Data platform. It's also true for our Fusion agentic studio as we see them really as a combination platform running on a single control plane in OCI. So yes, no doubt that service offering, we think is necessary. I actually think we help get the customers live as quickly as we possibly can. And as I said, we're measuring some of the successes now weeks, which in heavily regulated industries, you really would not have dreamed of maybe even a year ago.

Thank you, Mike. For next quarter, we expect our Q2 fiscal year '27 earnings results will be announced on December 14 of '27 -- '26, excuse me. Any change to the date will be publicly announced. A telephonic replay of this conference call will be available for 24 hours on our Investor Relations website.

Thank you for joining us today. And with that, I'll turn the call back to Mariam for closing.

Thank you. This concludes today's conference. We would like to thank you all for your participation today. You may now disconnect.

I think we're live. Good morning, everybody. I'm Brad Zelnick, Head of Software Equity Research here at Deutsche Bank. And on behalf of myself and all my colleagues, extremely delighted to welcome you all to the 20th Annual Deutsche Bank TMT Conference here in sunny Monarch Beach, the Waldorf Astoria in Southern California.

I got to say that this is an exciting time for tech, extremely dynamic, a lot of things happening, but I feel like event-wise, this is something really special that we've carved out. And I know that because I heard from all of you, many of you that were at the welcome event yesterday. I can see over the years, it's now the fourth year that we're here in this location. And I can tell you 4 years ago, I was an exception in bringing my son.

And many of you who have seen him and have been here all 4 years, you said, wow, he's kind of like doubled in height. So now I look up to him. But I see more and more families, and this has become a really special event. So I should only hope that for the next 2 days, we make it really productive, really fun. And in the next year, if you didn't bring your family or if you're listening on the webcast, you're going to come and join us as well.

But as I think about technology and I think about paradigm shifts, which seem to occur every decade or so, there's a lot of change going on. And there's very, very few companies that have not only survived but thrived over the decades. And when I think about Oracle that's been around in a leadership position nearly the last half century, I couldn't think of a more fitting company to kick us off. So with that, welcome, Mike Sicilia. Mike, thank you so much for being here.

Thanks for having me.

So Mike, for everybody in the room, can you maybe just give us a quick overview? Mike is the CEO of Oracle Corporation, along with Clay Magouyrk, who is your co-CEO. Can you just give us a sense of where you spend your time most across apps, infrastructure and the broader Oracle portfolio?

[indiscernible] events like this. And basically running and operating our go-to-market functions, our marketing functions, all the things that we need to do to put our best foot forward with our customers. And Clay spends a lot of time on the infrastructure side, particularly the large infrastructure build-out that we have right now. Although it's not that we exclusively split our time that way.

Clay and I have been working together for a very long period of time here long before we took these roles. So we collaborate on many things, including investment decisions, operating the company and things like that. So it's a partnership that's been well formed over the years, working together with Larry for quite a long period of time, and it's one that works quite well for us.

I sort of came up or grew up, I guess, you would say, at Oracle in the applications business and spent a lot of time there, particularly in our mission-critical industries application. But it's part of what we do at Oracle. It's not all of what we do.

Very good. Maybe getting right into it, AI, I mean, we -- I guess we haven't gone 2 or 3 minutes without getting straight into AI because it's so topical. Just at a high level, it's creating both massive opportunity, yet at the same time, disruption across the industry. Can you just help us understand your dialogue with CEOs, CIOs, corporate boards, how that dialogue around AI has evolved over the past year and how having a full stack offering positions you in those conversations today?

Yes. Well, I think what's most interesting, Brad, is that you've said how has that evolved? That conversation evolved over the last year. I actually think about it and say, how has it evolved over the last month because that's how quickly things continue to change here in the AI. So if I went through the whole last year, we take up the whole 45 minutes that we have here because every month, it changes.

But I'll sort of summarize it in the disruption side of AI. We said this on probably a couple of earnings calls ago. Perhaps the first industry to be really disrupted by AI is actually -- is our own industry. It's the industry that sort of created, if you will, the AI, but the disruption factor, the positive disruption factor that we're seeing is really stunning. I mean I started a long time ago in those days called a programmer, eventually changed it to software engineer and things but same function.

If you would have told me today, if you told me then that the productivity that we could get from a single software engineer was what we're getting and the dramatic output that, that person could have in production grade, I'm talking mission-critical. I'll use the application space here, but it's not exclusive to applications, would be what it is today. I'd say there's just -- that's like Star Wars, futuristic stuff. There's no way, but it's real. I mean it is super real. It is incredibly interesting.

And so I think the software industry will probably be likely disrupted in a positive way. Now what I don't mean by that, let me be clear. What I don't mean by that is that the SaaSpocalypse fears and all these things are -- I mean that we no longer need software engineers and stuff. The fact that we're able to disrupt ourselves and become so productive is because we have decades of experience with building and delivering mission-critical applications, mission-critical infrastructure like Oracle Database.

And we understand the domain, we understand the regulatory constraints. We understand all the things you need to do to be able to deliver that as a turnkey service to customers. And when you have that domain expertise, you can just get a lot more done more quickly with a lot of the tooling. So that's the first piece on disruption. Now come to your question about how are customers thinking about this, what are the conversations like with CEOs of our customers.

Where I think the acceptance is, okay, I mean, yes, there's something to this AI stuff. This is pretty real. How do I get value from it? How do I understand the cost around it? And how much of a disruptive factor do I have to put into my organization? What's the change management? Where is the low-hanging fruit? Where is the midterm return? And what should I be thinking about from a longer perspective? So that's become the nature of these conversations.

And I think what that leads to is that it really opens up what we largely refer to as the inferencing market, which I think is sort of the next big thing, if you will, in AI is that, as we've said, look, the models stunning, right? Codex, I mean, the ability to create source code, the ability to read and write everything on the Internet, understand, have conversation, just stunning technology. Of course, from a contextual standpoint, these models don't know anything about private data, government data and all the enterprises and governments that we serve on a daily basis.

The intersection of those 2 things or the union of those 2 things, the application of that is where I think there's just a tremendous value unlock as we go forward. What's -- where people have stumbled early on is just not understanding like what a token is let alone, how much it's going to cost them and how do you get value from this. Where we have, I think, a compelling differentiated conversation with our customers is like I'll just use one example. we're lots of applications, but our HCM application, right?

So our hiring and screening application, how do you hire candidate -- how do you screen candidates, how you hire candidates? Because we have the heuristics and the patterns that we have for 14,000 customers that are live on our Fusion application today, we can have a conversation with a customer that says, okay, you are today are screening 15 high-value and hiring 15 high-value candidates per month. You'd really like to get -- given the trajectory of your business, you'd really like to get to 40 or 50.

How much of that -- how many tokens do you need to consume? And how do we package that up into a bundle for you so that you've got predictable cost and you've got an ROI. You've got an ROI case built into that. And how do you price to package that all as a service. And if you're not doing all of it, right, if you're not closely provisioned to in this case, the large language model that's going to go through the resume posting and reading and all of these screening things.

If you're not in the infrastructure business of providing the lowest price performance compute on the market, you're not in the applications business and you don't have patterns of 14,000 live ERP customers of all sizes. It's very difficult to get into that predictable ROI discussion. And I think that, that is -- that's become an evolution of the customers that I'm having. I could go on and on and on with a ton of examples, but -- that was 8 minutes on one question. No...

It's great context, Mike. And there's a lot of threads to pull on there. And if I think about Mr. Ellison over all those decades, the one thing that hasn't changed is price performance has always been a message around Oracle. And if I think about software monetization and pricing models. They've evolved the metrics even within Oracle from power units many years ago to processors and database to users -- to named users.

But it had always been across the industry an ROI-based sale. And as we move forward, though, over those decades, it feels like the risk has shifted from the customer onto the vendor as we went from decades ago, big ERP implementations nearly bankrupting some customers. But here in the AI era, I'd love to explore a little bit more AI monetization and how that maybe changes things.

So Oracle today offers a number of options, token bundles, usage-based pricing for agents, Agent Studio, more recently, even outcome-based pricing. Is there any one of those that you see becoming more popular or in your view, becomes more prevalent over time?

Well, I think since we offer a really diverse and yet complete stack that all 3 will enjoy popularity as we go forward. From an -- I just gave the example of outcomes-based for the HCM hiring and screening use case. We've talked about our Fusion AI Agent Studio in which customers or partners can create their own AI agents on top of our Fusion platform, not on a custom platform, not on a different platform, but right on top of that platform, that's going to be in a token consumption perspective where customers are going to need to consume tokens to build their own agent.

So I think the combination of the 3 things that you mentioned will continue to be popular choices for our customers, given that we're in each one of those businesses. We're not in any single piece of that business. But I think the key element is what's the predictability around it. I mean you've seen -- we've all seen -- look, I mean, we had -- I had -- we're leveraging AI internally at Oracle, and I had some salespeople run up some token bills.

And I said, what are you actually doing with Codex in a sales function? And actually, it was fairly compelling what they're doing. That said, we had to tamp down the token throttle a little bit here because you don't want people just run away. That's kind of been the early stumbling block is that you've got a bunch of people out there with their credit card or the equivalent thereof, kind of buying things for a science project.

And I think having that governance over top of it and having the ability to sort of marshal all these different consumption models and outcome models is going to be the key differentiator that enterprise, particularly enterprises and governments as well are looking for. But the monetization angle to your point, right, there's not just a single monetization model here. It's -- there's still some customers that are user by seat.

Outcomes, I think, become increasingly popular as people are looking to get that ROI from AI. And then just token bundling as well, understanding it, give me a certain amount of tokens and then tell me when I'm getting close to that. But by the same token, help measure the ROI for what's been consumed. How do I measure the ROI for what's been consumed for the consumption of those tokens? Well, we can actually easily do that because we're delivering the platform that is the orchestrator of the token being consumed together with a large language model, we can actually see and demonstrate the ROI.

That makes a lot of sense. Maybe as we think about AI-native applications, I think Oracle has had a point of view and emphasize that making AI native to the application and workflow rather than taking a bolt-on approach is the way to be and the way to go. When customers are evaluating Fusion today, what role does embedded AI play in accelerating or even expanding a customer commitment?

Yes. I mean it's part and parcel of every conversation we're having with customers today. We like to say that one of the quickest ways to get value from AI today is actually to move to Fusion applications because there's nothing you have to do except to consume the application, right? It's built into the application. It's not something separate you need to think about.

And frankly, you're probably going to put yourself in a -- well, not probably, but I think going to put yourself in a much better cyber defense perspective rather than taking bolt-ons from third parties and bolting it on to your existing on-prem infrastructure and potentially creating a bunch of Internet-facing stuff that wasn't Internet-facing before accidentally or in some cases, maybe on purpose as a result of the feature. So the fact that we built it in is huge.

The fact that we also say to customers, hey, you can build your own as well is another big differentiator. If you think back to like applications business in general, what's one of the things that -- what's one of the things that's plagued you run away go-lives, cost everyone, a lot of it is customizations, right? -- customizations and it's done 2 things. Number one, they become much more expensive than thought. The go-lives take a lot longer. And the second thing is you've got people that are stranded.

They're stranded on-prem because there is no SaaS equivalent for what they've built. With the AI Agent Studio and the AI agent builder provisioned together with and running on the same database, running on the same infrastructure, running on the same platform services together with Fusion, you really get the best of both worlds. So that you can say, here's a package application, right? But here is a token bundle, here's a studio builder.

By the way, you don't need to be an engineer or a coder to build these agents. And you can do what you'd like to your heart's content, but here's the good news. You can't go off the rails. We won't let you go off the rails. This is that embedded data layer, an API level that we guarantee not to deprecate interface with all the applications, not just Fusion, but vertical applications as well so that you can do both because there is -- look, the industry in general has always said, don't customize. It's a bad idea, right?

But there are always some things that you have to sympathize and empathize with customers and that they're pretty specific to their business. It's really hard to sometimes do that to say, I'm not going to have any customizations at all. I'm just going to run this as is. And we absolutely still want that to be the North Star. But now I think AI really kind of brings that conversation back to the forefront where you've got a configuration, not a customization.

And by the way, all of those agents and all that studio, we upgrade, we patch that just like we do our Fusion applications on a quarterly basis. Every quarter, in some cases, monthly patches if there are security things that we're thinking deserve attention. All that's just built in as a service.

Makes sense. Just following that, should we think about the larger opportunity being direct AI monetization or in using a rich AI-native feature set to drive greater adoption of applications, expand seats and consolidate more workflows onto Oracle Fusion?

Yes. I think they're both equally important. It's hard to say which is the bigger -- I mean, things are changing so quickly. If we look at this over a 5-year horizon, it's hard to say that -- which is the bigger market. I'm happy that we're in both markets. I'm happy that we're I mean there's -- here's what I think is likely to happen, right?

You've got very wide moat applications or particularly when you're in heavily regulated industries like banking, utilities, health care, things like that. It's unlikely to me that there's a vibe coding phenomenon that goes and replaces all of that mission-critical infrastructure that the regulators are going to say, sure, go right ahead and go run that for the American health care system, the banking system, stuff like SWIFT. It's unlikely, I think that's going to happen.

What is likely is that people will want to continue to tailor and continue to make things germane to either their specialty or their subspecialty industry. And the fact that you can do both, you can consume the wide, the sort of heavy moat stuff, as is you can create something in parallel, I think, is certainly compelling. The other thing I think that's interesting is that you think about this from a security standpoint, from a business rule standpoint, everything that we're doing in our application stack is all API-driven.

So it's not even out of the question that you can consider that the applications really run headless, if you will, and the agent becomes the agents, the customer-generated agents, Oracle-generated agents or partner-generated agents become face -- the face of those applications. It's also not out of the question. And we see this in like in our health care applications as some of the recent AI agents that we've -- that we've released and that the user interface itself almost entirely disappears, right? User interface is really almost gone. And it's voice activated or it's just -- when I say user interface, I mean, user interface in the traditional clicking for and things like that.

AI is the new UI...

Right. AI is the new UI. But I think what's the -- why there's a continued and certainly even, I think, continued growth opportunity for applications is that all the stuff in the middle that it needs to -- that the AI interface needs to interact with is it's pretty -- at least the businesses that we're in, in Oracle, right, these are pretty complex mission-critical, heart of the industry, heart of the business type stuff or for governments, a similar vein.

So I think that both of those things, that's why it's hard to say which one is the bigger -- which one is the bigger generator? It might come down to how you classify the revenue. But I think the combination of both -- like if you didn't have both, we wouldn't be in as good of a situation going forward, right? And where the revenue falls on either side of it is probably less interesting. What's more interesting is that we're generating differentiated outcomes for our customers from a turnkey service.

Well said. Very well said. Just as we contemplate this moment elsewhere in the industry, we have seen some moderation in SaaS growth rates, longer deal cycles as customers figure this all out as they face pressure, whether it be in rising component costs and taking care of prioritizing their hardware needs or dealing with runaway token spend for the moment.

That all said, the last few quarters, Oracle's deferred revenue has continued to grow in excess of in-period revenue growth that you've put up. Can you just unpack that for us? What in your view is driving this divergence?

Well, I think if you segment the market a little bit and you look at where there's been slower sales cycles or even maybe a couple of quarters of freezing sales cycles, which is now unlocked, it's really more been at the SMB side of this thing. When you look at the kind of the most acute -- on the SaaS side of things, when you look at the sort of the most acute part -- the most acute peak of the kind of doom and gloom that the SaaS business was gone forever and never coming back, and it was all going to go into a garage somewhere and be reinvented.

That definitely has an impact, particularly probably more outsized on small and midsized businesses than it does on large enterprises, particularly again, particularly those in regulatory environment. So -- and we said that on our last earnings call, if you look at our NetSuite business. If you look at our NetSuite business as one example of that, you had some slowdown in sales cycles.

But now I think we're -- with our new AI-enabled NetSuite applications, we've -- we actually feel good about the fact that AI actually helps the businesses too and actually gets to a faster go-live time. In the enterprise market, both government and large enterprise market, we did not see a slowdown in our growth. In fact, as our deferred position continues to grow more quickly than our in-quarter revenue, which is a sign of booking strength in the future, we actually saw an acceleration because of the built-in AI to our Fusion applications and our industry applications.

So actually quite the opposite. I think the kind of next -- the piece that we've spoken about is that the next unlock on that is the ramp, which is how do you leverage AI with forward deployed engineers in our services business and actually providing toolkits to our customers to get to live even more quickly because the opportunity is still big. And we've said publicly, about half of our installed base has moved from on-prem to cloud in the application infrastructure.

That means there's half to go, right, at a 4 to 5x conversion rate. And there's half to go in big media industries like banking, right, and utilities and health care. So there's still -- the tried and true kind of on-prem to cloud business, we're not through that transition. There's still room to go there, quite a bit of room to go. On the more -- on the CPU side of things, I think the reason we've continued to accelerate is that our cost performance is superior, right?

And we -- as you said, it's always been a bellwether for Larry and for Oracle in general. We've always thought about we want to be the most efficient provider of technology to our customers. And if you can spend x with one hyperscaler and you can get twice as much done with Oracle for the same x, I mean, you're going to make that choice. So look, I don't have any secret sauce for you, except to say it's all about innovation.

It's about innovation at every single layer of the stack. And that has been our differentiator at Oracle that we've been in multiple -- we've been in what we think to be all layers of the stack. We continue to innovate in every layer of the stack. Just as important to our acceleration is what we're doing in database. I mean what we're doing in database with blockchain tables and encryption keys and vector search and all of those things, that helps us -- that helps our application business be stronger.

It's a differentiated situation when you're running that on our OCI infrastructure, whether natively or you're running it in our multi-cloud infrastructure throughout the world. So that's another piece that we've unlocked. So I can go on and on and on.

This is fantastic. Yes. I want to maybe touch on another subject, the importance of having an industry focus, which if I remember, 20-plus years ago, Oracle had the IBUs, the industry business units. And then there was a period where the company made a number of acquisitions to get to the heart of mission-critical business processes in each industry, including Primavera, by the way, which I remember very well, where you had come from.

I guess before taking on your current role as CEO, you had a prominent role leading Oracle Industries. Why is the vertical approach to the market more or less relevant in the age of AI? And how does that inform the ongoing investment you'll make in industry solutions and industry go-to-market?

Well, I think it's absolutely more important than it's ever been for us because for the simple reason that outcomes matter, right? And outcomes matter and outcomes are going to matter to the business. So we've always been -- with our industry approach, we've always been in a situation where we sold to both the IT function of the business, but also to the business itself because we're in the heart of what the business does.

I mean we're providing the core banking systems. We're providing the merchandising supply chain management systems. We're supplying the meter data management and grid sharing systems for utilities, the property management systems, reservation systems for hotels. And these are all businesses that we've been in. You don't see the Oracle logo on them because they're private labeled by the business itself, but we are in the guts of what matters in health care, electronic, which we all been publicized very well, largest custodian of electronic health care records on the planet.

Now when you want to talk about how do you get value from AI, it's both an IT conversation, of course, but also increasingly a business conversation. And it's one that CEOs are increasingly involved in our customer situations. We've always been selling to the CEO in that industry perspective. And we've been able to recruit people in. People matter here in a big way, Brad.

We've been able to recruit people in who are domain experts in these -- both for acquisition, but also organically. If I look at the people that are running our industries today, these are people that we've recruited in from the industry. Unlike me, most of them didn't come from an acquisition, but they actually like the center of gravity that we have in the industry and they recruited into Oracle as a result. So it's always been a competitive differentiator for us, probably more important than ever.

In terms of our go-to-market function, reorganized our sales team to be completely industry focused across the globe. And that's even true in our infrastructure business as well. Even in our infrastructure business, we have people that call on the same industry because it's unlikely that in our IaaS business, unlikely that's the only thing we're selling to our customer is infrastructure. We're also selling database. We're selling ERP. We're selling some operating technology as well.

So we've actually gone through a major restructuring of our go-to-market function so that we have 2 things. We have complete industry alignment. And number two, we have less sellers calling higher on our customers and speaking more at the executive level. And because we need less sellers per customer, it's allowed us to expand our reach and pick up some uncovered territory today that was previously uncovered.

Helpful. Mike, you already touched on the advantage of having the full stack. And I think you've said in the past, customers increasingly think about the applications and database layers together rather than as separate decisions. For those less familiar in the room, can you just remind us what differentiates the Oracle AI database and how the layers come together to create additional value for customers?

So just in general, the Oracle AI database is probably a quite popular custodian of the world's mission-critical data and has been incredibly sticky in that regard. That mission-critical data is the key for inferencing, the key for retrieval augmented generation. I'll say contextual AI, contextual applied AI. So we've built into the database, things like vector search, where we are automatically vectorizing all of that data for customers as a service, just as an upgrade that's built into the database, just one example of innovation.

The other piece is that we're doing that on top of our blockchain tables as well, which for certain vertical industries is incredibly -- having something that is completely immutable is incredibly important for certain vertical industries. Then you put on top of it the way that we encrypt data. So I'll use our health care example, our health care database, for example. It's -- let me put it in a different way.

How did the health care business make our database features stronger? Because we are the largest custodian of electronic health records on the planet, we have to think about encryption keys. We have to think about what's called longitudinal records, which means your entire patient record. So the way the database works for health care is that we have an encryption key per patient. So databases -- most databases in the market today are encrypted at rest and encrypted in transit. We actually have an encryption key for patients.

So if you were to steal the database, you'd have to steal how many patients worth of encryption keys or decode how many patients worth of encryption keys you also had in the database at the same time. Now how do you do all that and also not impact performance because encryption is one of the things that can make databases operate more slowly. Well, that's -- I mean, that's just decades and decades of innovation that we put into the Oracle database.

Then you put all that together and say, how do I make it highly secure? How do I make it immutable for our mission-critical government and enterprises? And how do I make it applicable for AI. So it's not just any -- again, it's not any one of those features that gives us an advantage is the fact that we can do all of them as a service. It's just a service that you built in. You don't have to go buy a third-party application or a third-party pack or have somebody come in and custom build this.

It's a service that's built into the Oracle database. The reason that I think we're able to get to that level of innovation is, again, because we're in the entirety of the business. Our testing cycles start in health care, for example, in a patient exam room, right? They're not just starting with unit testing with the database team. And that's just a different -- it's a different perspective on the market and a different philosophy on the market.

It's clearly distinct and it's clearly driving your success. If I take a lot of what you said and I align that with what were the financial targets that were put out at the last financial analyst meeting, fiscal '27, this year marks an important inflection point for cloud database growth. The environment is clearly supportive, right?

You've got data migrations, accelerating coding agents that are in part reducing the associated friction of that. Multi-cloud reaching global regional availability. But beyond these factors, what else underpins your confidence in hitting these targets over the next few years, specifically in cloud database?

Well, I mean -- so we just announced that we have 22 regions live now with Amazon in multi-cloud database. If you look back at our last earnings, and we had already gone live with Azure. It was our first partnership with Microsoft and with Google and now with Amazon. And in Q4, we announced over 400% growth in our multi-cloud database business.

The good news is that, that's largely concentrated in the earlier partnerships and not in the Amazon partnership because we're now just delivering those regions. And now you have 22 of them live, which is pretty quick growth here as we go across the board. We feel good about our position in that business. I think the other thing that gives us confidence in database and cloud growth around database is that more than ever, data matters.

More than ever, as we think about, again, not to repeat what I said, but get back into this idea of applied AI, differentiated AI, outcome-based AI, without the database, you can't do it, right? You just can't do it. So whether the customer is running AI infrastructure on Amazon, on Google, on Microsoft, we're agnostic. We even running on OCI, we're agnostic. We're model agnostic, right?

We arbitrage open source models. We oversource OpenAI inside of our own applications. But all against the Oracle database, right? So we can take the Oracle database. We've made the Oracle database very portable. We've made our infrastructure very portable. And the same reason that we're able to take the Oracle database and make it available in all of our competitors' infrastructure is the same reason that we, I think, have a unique competitive advantage in the infrastructure business in general, the good old CPU business.

Long ago, I remember that business like last year, the good old infrastructure business is that our form factor is very differentiated. Our form factor is smaller. It would be not as easy. If you were to say, well, we're going to go build Oracle database, Oracle Cloud database inside Amazon for 22 sites. If we had -- if we needed hundreds of racks per site to do that, that wouldn't be such a good value prop for us, right?

But the fact that we can do this with such a small number of racks and then we take that same amount of -- it's that same technology that allows us to create fully sovereign infrastructure with 12 racks. And I don't mean like an edge solution here. I'm talking about full feature OCI, all of Fusion, all of our applications, everything that you want in 12 racks. And you can imagine that, that becomes a price point that's very attractive for countries and enterprises throughout the world.

So again, it's I'd like to -- I'm sorry, it's the same boring story, but the fact that you're in every -- the fact that we're in every one of these businesses gets us to more of a turnkey solution at the best price point for the money. And my other -- why am I also continue to be confident in our cloud database growth? Well, there are a lot of Amazon customers, AWS customers that are also Oracle database customers, right? So -- and early days of that infrastructure build-out, 22 sites live, I feel good about it.

12 racks and can scale horizontally to.

By the way, 12 racks is probably high. We're getting down to...

Even smaller.

Yes, we're getting into 6. I mean -- so you think about where does that -- does it -- well, this becomes a really interesting kind of infrastructure question because look, the big gigascale stuff gets all the headlines today, right? It gets all the headlines and rightfully so, right? There's a lot going on in the industry there.

But think about like how many banks, utilities, even large health care systems, how many of them have their own existing data centers today, lots, right? How many legacy data centers are there out there? And as you see banks modernize from COBOL, there's a lot of space that's going to free up on the floor. Pretty easy to think about getting 12 racks or 6 racks into that from an infrastructure standpoint and having a turnkey cloud solution.

So -- and by the way, that's not a very capital-intensive business. These databases are already -- these data centers are already energized. You might do a little retrofit. You might put some different chillers in there and things like that. But it's not like you need to go build something brand new or go acquire the land or figure out how you're going to get the power for it, right?

So that same form factor that allows us to take the Oracle database and make it completely portable to any of our competitors' database also gives us full-scale advantage to be able to deliver a much smaller form factor cloud for existing database infrastructure. And I think -- the other thing that I like about that business is because asset-intensive industries or heavily regulated industries have been slow to move their mission-critical stuff, if you will, to the cloud, whether it's database or applications, now kind of getting on board in this Mythos moment to say, we actually might want to think about at least ring-fencing our mission-critical stuff in the cloud, if not thinking about a SaaS or database cloud transition, that starts to unlock this value proposition conversation around saying, well, how do we have a best of both worlds conversation for you?

We can bring it to you in private cloud. We'll bring it to you in private cloud in your data center with a number of racks that's very palatable and we'll deliver the racks, we'll deliver OCI, we'll deliver the applications, we'll deliver the database, we'll deliver the analytics. And that, I think, is an interesting.

It's a really interesting point in a world where powered shells is a constraint to all this explosive growth that we're seeing. So thanks for pointing that out. While we're on the topic of infrastructure, and you touched on this a little bit, there's a lot of focus today on OpenAI models and their implications. Oracle has taken more of a Switzerland approach of not competing at the model layer and instead wanting to be a compute provider to everyone. Can we just get your thoughts on the emergence of open models becoming one of the primary ways of serving inference in the enterprise?

Yes. Look, I think there's a place for open models. In fact, some of those Fusion examples that I gave earlier with HCM, we're arbitraging both, in this case, OpenAI and open models, right? And we're using ChatOSS and some of these things on the back end. And if there's a good enough result from the model that results in a high-quality answer for customers, we're by no means opposed to leveraging that and keeping -- again, keeping our costs low as a result.

I think there will be a place for both, just like open source infrastructure in that you tend to see on the mission-critical side that the key mission-critical infrastructure is largely running on enterprise-grade or enterprise provided infrastructure, more of the ancillary pieces on open source. And I think when you think about the way that AI intersects with data, there's going to be lots of questions around safety and security from customers around how that whole thing works.

So we're in a good position to have that conversation with customers because we are arbitraging models. We are not necessarily -- we're not locked into any particular model. I think there will be multiple winners in the space. I think there will be open source winners as well. I do. I think customers are going to make the choice as to which one they'd like to consume.

And I think in certain regulated industries, it's probably going to tend more to the enterprise provided models than it does to the open source. But do I think there's a place for open source? Absolutely. Just like -- I mean, we're in the Linux business, right? So it's not unfamiliar territory for Oracle. I mean -- so we understand and we've supported open source for decades.

For sure. And I remember, I was there at OpenWorld many years ago when Unbreakable was announced, I'm showing my age. But Mike, I've always been interested in looking at adoption patterns of enterprise technology and enterprise software and people talk about classic S-curve adoption. I always -- I often think of it as a value prop just becoming so compelling that from CFO to CFO on the golf course or CEO to CEO, you see that your peers are doing it, you feel that you have to do something.

I'd be curious your perspective, is there a specific customer profile or maybe set of workloads that you see acting with a greater sense of urgency to migrate today and to become AI infused? Or has anything surprised you about which customers are moving first?

Yes. Well, it's a really interesting time for that question because, again, because of the cyber concerns in the world, right? So I think you're seeing a little bit, Brad, of a paradigm shift in that think about the general move to the cloud, right? One of the things that made it sometimes go slower for folks than they would like or made it more expensive is because they try to move and transform at the same time, right?

How do you move this workload and how do I transform it into a modern application stack whether it's SaaS, whether it's PaaS, whatever layer it's at. And how do I modernize it at the same time. That added cycles, that added change management, that added unforeseen circumstances. I think this idea of ring-fencing everything that somebody has, ring-fencing it into at least being able to defend having a perimeter around this because of the impact, the potential downside impact of a cyber -- the cyber trade-off with AI is probably becoming a more popular conversation than the transform at the same time conversation.

So which is obviously going to lend itself to our CPU infrastructure business as well. On the heels of that, where I think there's another big opportunity for AI is the time to go live, is reducing the time to go live, particularly when you have like big ERP transformations, which, as you mentioned before, sometimes have been plagued with cost overruns, time overruns and expectation. expectations not being met.

These models, and we're leveraging with our own services team have become very good at understanding customers' environments, understanding, for example, an E-Business Suite implementation, understanding all the customizations that have been made and figuring out what's the best path for you to get to a more vanilla Fusion application and then what are the agents that you should create to kind of go replace those customizations. And you're seeing compression from years to months.

And I'll give you an example. So -- and we spoke about this. So we're a large provider to the U.S. Veterans Administration for electronic health records. And we saw the time to go live because of the AI tooling that we -- and health care is like probably the most complicated industry from a go-live perspective just because of all the regulatory patient safety and all the things, all the boxes that you need to tick very important boxes you need to tick as you go live. We saw that come from 18 months down to 8 months.

Same set of people, same consultants, same implementation people. But with AI tooling, understanding the before and the after is -- and that one does involve a transformation. That does involve new clinical processes. That does involve new workflows and new configurations. Now if you just wanted to say, how do I get from here to here, get the same functionality without doing any transformation, I think you're going to see that curve continue to bend.

Good news is that we talk about our deferred position being greater than our in-quarter position is one of the things that would be to our advantage is to decrease the ramp, decrease the go-live times because you've got so much sitting -- so much of that revenue sitting in deferred at this point. And the more quickly that you can bring it in, obviously, the better off you are.

Very good point. I think we're almost about out of time, but I'm going to ask you one final question. If I look out a year from now, what do you expect will have changed most meaningfully in how investors view Oracle's growth profile and competitive position?

Well, Here's a long time in the age of AI. But I think, look, as you mentioned, we're approaching the sort of the peak of our scaled-out infrastructure delivery. And in the short term, short term meaning let's use your year as a short-term horizon, we're probably in a position to deliver an industry-leading amount of compute to the market of scale compute to the market. So I think proof points around that delivery are going to be important to the market.

I think innovations. And I think the other thing that will be interesting over the next year is that I feel good about the fact that we're in a very good position to deliver meaningful AI proof points that are going to do 2 things. Number one, they're going to help enterprises and governments get more comfortable with the application of AI. But I also think they're going to be particularly acute in businesses like health care.

And I think those things are going to be good for the world. I think what the world really wants to hear is some very positive proof points around this. And if you look at some of the backlash and some of the sentiment around AI in the market, there's one shining star that is if you look at all the polls, people largely believe that AI can have a dramatic impact on health care, on better outcomes for patients.

And I don't think the world is that far away from actually delivering some of that stuff. And that keeps -- that makes me very excited. It's good for the world. It's good for the country. It's also good for business. So I think that we put all those things together. And I would hope that in the next year that those are the things we talk about -- we talk about next year in terms of accomplishments.

Awesome. Well, Mike, it's always great to see you even better here at the Deutsche Bank.

Yes. Thanks for having me, and thanks, everyone, for your attention and participation.

Thank you.

Well, good day, everyone, and welcome to the Oracle Corporation Fourth Quarter Fiscal Year 2026 Earnings Call. [Operator Instructions]. I would now like to hand the conference over to Mr. Ken Bond. Please go ahead, sir.

Thank you, Lisa, and good afternoon, everyone. Welcome to Oracle's Fourth Quarter and Fiscal Year 2026 Earnings Conference Call. On the call today are our Chief Executive Officer, Mike Sicilia; Chief Executive Officer, Clay Magouyrk, and Chief Financial Officer, Hilary Maxson.

A copy of the press release, including financial results, tables, supplemental financial metrics and guidance are now available from the Investor Relations website. Also is a slide deck being introduced this quarter, which you'll see momentarily, a GAAP to non-GAAP reconciliation, other supplemental financial information and list of many customers who purchased Oracle Cloud services or went live on Oracle Cloud recently. These items will be available after today's call.

As a reminder, today's discussion will include forward-looking statements, and we will make some important comments around factors relating to our business. These forward-looking statements are also subject to risks and uncertainties that may cause actual results to differ materially from statements being made today. As a result, we caution you against placing undue reliance on these forward-looking statements, and we encourage you to review our most recent reports, including our 10-K and 10-Q and any applicable amendments. And finally, we are not obligating ourselves to revise our results or these forward-looking statements in light of new information or future events.

Before taking any questions, we'll begin with a few prepared remarks. And with that, I'll turn the call to Hilary.

Thanks, Ken, everyone. Great to be here with you today. And as a CFO, I thought I'd start with a few thoughts on why I'm so excited to join Oracle at this time. I said my career all around the world by companies that use technology and data to drive transformation, both internally and for customers. And I believe that the most valuable transformational change fits as a just deposition of the physical and virtual worlds across business models, from infrastructure and enterprise software. Also understands that intersection and is now uniquely positioned for one of the most significant technology transitions we've seen in decades. Various 2 companies can help customers across the entire technology stack from the cloud infrastructure that powers AI workloads to the mission-critical applications that run their businesses.

Oracle -- plus, this is a company with deep technical expertise, differentiated technology and a long history of helping customers turn technology innovation into tangible business value. And now I've only been here for 2 months, nothing I've seen has reinforced my confidence in the company's strategy, execution and opportunity ahead. I'm excited to be part of the team and look forward to helping Oracle capitalize on the opportunities in front of us to drive return on investment and shareholder value. And as we pursue these opportunities, we'll remain focused on disciplined capital allocation, maintaining a strong balance sheet and preserving our investment-grade credit rating.

With that, let me turn to our Q4 and fiscal year '26 results. And like Ken said, we've introduced a short presentation to accompany our earnings call, so you can follow along with the numbers and the comments we'll make today. In our of Q4, it was a record quarter, driven by strength in both our cloud infrastructure and cloud apps business.

Revenue was $19.2 billion, up 21% in U.S. stockers. Cloud infrastructure revenue grew 93%, reflecting strong demand for both AM workloads and our database services, and cloud was up double digit at plus 10%. And Mike later will give more detail on these businesses in just a moment.

Our net operating income increased 22% in U.S. dollars to $8.6 billion, driven by our strong revenue progression. Our operating margin increased slightly with our gross margin declining driven by impacts from ramping up our data centers and the acceleration in our infrastructure revenue.

This was more than offset in the quarter by a reduction in operating costs and for -- access the lines in our P&L, starting with sales and marketing due to efficiency actions in our cost structure. Our North EPS reached $2.11, an increase of 24% in U.S. dollars for the quarter -- onetime net gain on investment. Excluding our non-GAAP EPS increased by 20%.

Turning to the full year, we surpassed revenues of $67 million for the first time which translated into strong non-GAAP operating income of $29 billion, up 16% in U.S. dollars for the year. Our non-GAAP EPS was up 27% in net dollars to $7.63, including onetime gains on investment. Excluding these gains, our non-GAAP EPS was $6.83.

For the full year, our gross margin stepped down around 5 points as expected as we start to see the impacts from the build-out of our infrastructure business, the acceleration in its revenues primarily offset by lower operating costs as a percentage of revenue, driven by operating efficiencies.

All of this translated into strong cash flow from operations of $32 million, up 54%. We did continue with our program of capital investment tied to unlocking the strong growth opportunities in front of us. Our net cash outlay for capital expenditures for the full year was $48 million, taking into account equity payments and timing impact of around $8 [ billion ].

You can see the table showing the details of net cash outlay for CapEx in our press release. We think this measure is important to better understand our funding needs. And our remaining performance obligations, or RPO, finished $638 million, up [ 356% ]. This unprecedented level of RPO provides exceptional visibility into our future revenue growth all supported by some long-term contractual customer commitments and reflects the strong customer demand we see across both AI infrastructure and cloud services.

To give a bit more detail on our RPO, we expect 12% to be recognized in the next 12 months and another 34% between 13 and 36 months. And these percentages are both expected to accelerate over the coming quarters based on our current long-term outlook. Mike will now get into a bit more detail on our cloud businesses, and then I'll be back with our outlook for fiscal year '27.

Thank you, Hilary, and welcome to Oracle. So I'm going to cover our cloud apps and cloud database business in a bit more detail both of which performed quite well in Q4. We're on the front of one of the most testing times in the technology business.

Our customers are now focused on how to leverage AI in their own businesses. They want AI to increase productivity, enhance customer service, and create real competitive advantages. So let's go quickly and within their existing budget open.

Oracle is indeed advantage is that we deliver the applications, the data, the infrastructure, the AI tooling and the industry expertise together. That combination invariably puts us at the center of customer conversations, whether they're existing Oracle customers or not. And our customers have moved past the expertise with AI.

They are ready to implement enterprise rate complete agentic solutions to help run their businesses. Over the past year, we have delivered more than 1,000 AI agents across our application suites. These e-based offerings will reason decide and execute work across processes.

So the quickest, most affordable and most productive way, customers can begin to consuming AI is just to continue using Oracle applications. since every 3 months again more and more of the AI features built for them and we're ready to do. This is a major shift in enterprise software, and Oracle is uniquely positioned to be good. And you can see it in our Q4 results. orchard applications generated revenues of $4.1 billion, which is up 10%, and our SaaS deferred revenue was up 16% in the quarter. Across the company, we took thousands of customers like last quarter, over 300 infusion alone.

Acceleron adopted our utilities platform to manage operations. Bright County share software linked with our Public Safety suite. Westfield Insurance, implemented Fusion ERP and Mirage Bank with life of the work banking just to name a few. All of these customers are upgrading to a better and modern applications platform that also comes with AI built right in. In Q4, we also continued our electronic health market deployment at the United States Department of Veterans Affairs. In Q4, we added 4 VA medical centers in Michigan and in early June, added another 4 VA medical centers in Ohio. Oracle now supports 14 VA medical centers serving 29,000 clinicians and 500,000 [ beds ] across the United States.

Okay. While not part of our Q4 bookings, United States -- Office of Personnel Management today announced an agency-wide award to Oracle for Fusion HCM. So this is obviously a strong start for us in our FY '27 in applications business.

In addition to discussions around AI in our applications, I'm also having very interesting conversations with our customers around leveraging their own proprietary data sets with AI. Much of this data already sits in an Oracle database or is generated by work applications. For many enterprises interfering against decades of rich operations data is where the benefits of AI. Compound exponentially.

Oracle's full stack offerings allow customers to get up and running quickly, leveraging AI together with their private data sets. This is like Claro, a major telecommunications provider in Latin America chose OCI, field services applications and our AHA data platform to automate customer service for their 30 million subscribers this quarter. U.K. National Health Services share business services, owners the Brazilian retailer and QXO, the fastest-growing building products distributor in the other states. Combined AI-ready oral infrastructure or database products with Oracle applications to move their businesses forward. Again, just to name a few.

Last quarter, we also released a long list of new AI functionality in the Oracle database. Here are just 2 examples. The Oracle AI agent memory is a library that help developers build agents that can remember reason and act with enterprise context. Oracle Deep data security as data access rules at the database level. This protects against both authorized access, it limits precisely what data a user -- and any AI agent acting on their behalf and see we're act upon. All of these innovations I've just described and many more are available in our cloud our partners' clouds and in our customers' environments.

In Q4, our cloud database business revenue grew by 29%, with multi -- growing much faster. Multi revenue was up 404% year-over-year, and bookings were up 325% year-over-year. One example of an enterprise using a wide range of Oracle technologies in Vodafone who turned on us in Q4 to consolidate and modernize their operations.

Vodafone selected OCI dedicated region in their data centers, our multi-cloud database offering and a partner cloud, and our applications to reduce costs and run their processes faster, in some cases, up to 60% faster.

Finally, we are working with our customers to deliver quick ROI within their AI budgets. To do so, we are simplifying how customers consume and pay for agenda capabilities. Our new agented pricing aligns with customer value.

Now much of our AI innovation in our core applications continues to be included at no extra charge. However, customers can also purchase additional agent capacity and a simple, predictable way by purchasing bundles of tokens that can use proper application suites. We're also introducing outcome-based commercial models that align pricing directly to the value derived.

For example, in a few agents that are priced based on the number of candidates screened more hospitality upsale agents priced on the percentage of end consumer upsale transactions. In Q4, we started a limited rollout of our toping bundles and had 33 customers like Aon Services Corporation and Liberty Energy, repurchased tokens that have access to more advanced reasoning and bottles. All of this helps our customers control their costs and align their setting with the value being generated.

And with that, I will turn it over to Clay.

Thanks, Mike. Okay. You just heard from Mike about our applications and database businesses. Oracle has been in these businesses for decades, and they continue to impress us because of their ability to continually grow aggregate margin dollars, through a combination of durable differentiation and increasing market size. I want to share how we see our infrastructure business in that same category and the evidence that enforces that move.

Okay. differentiation comes in many forms. Technological innovation, supply chain execution, operational ability and -- we created OCI as the most highly secure, highest performance, most flexible, lowest cost infrastructure available anywhere. We deliver that through innovation across all layers from deploying the smallest and the largest clouds to inventing technologies like Acceleron that provide the highest performance and lowest cost networks.

We combine the power of OCI infusion applications to implement an incredibly efficient and flexible supply chain. We architect across data center design, power distribution, data hole layout and networking to deliver the most efficient and the most flexible infrastructure available team wear.

Oracle had a long track record of durable differentiation. This is because we know the real differentiator is the organization, the people, the company itself that can adapt to new requirements, invest solutions and deliver them to customers rent. OCI has been the fastest-growing class provider for years. And now with AI infrastructure, we've shown to everyone the power of the organization we built, the technology we created and the value we're delivering to customers.

OCI is continually releasing new services, hardware, networks and cloud regions to ensure we are always the best place for our customers' infrastructure workloads. Cloud infrastructure has become a very large market because of the ever-growing demand for server sign computing. AI infrastructure makes the existing cloud infrastructure market looks small. Everything we see shows this market size is trillions of dollars per year. Combined with our previously outlined 30% to 40% margin profile, OCI should grow into an extremely large and extremely profitable business.

These beliefs are supported by compelling and multiplying amounted evidence. We signed $67 billion in AI infrastructure contract this quarter, the majority of which was either bring-your-own hardware or prepaid. This increases our combination of bring-your-own hardware or prepaid customer contracts $75 billion. with those contracts having no degradation in margin compared to our other contracts.

Customers are showing they chose OCI to deliver their infrastructure, even when they are bringing the capital themselves. Design, delivery and operation of this large-scale infrastructure is extremely demanding.

Q4 finalized an impressive FY '26 where we delivered more than 1.2 gigawatts of customers. Our pace of delivery continues to accelerate. With our FY '27 Q1 delivery approaching 1 gigawatt, near the same capacity as we've delivered in the previous 4 quarters combat. There will be many winners name, and our strategy is to have them all those customers. We continue to diversify across our largest customers with 4 customers contracting for more than $8 billion this quarter.

Our infrastructure is fundamentally multi-tenant, and we continually allocate capacity between customers. In Q4, 35,000 GBUs from 59 separate customers were up -- 49% of those customers renewed for 92% of those GPUs. That doesn't mean though that 8% of those GPUs arrived. And both of those GPUs themselves were subsequently sold to other customers in the same quarter.

Our global GBU utilization rate is 97.5%. It's also true to -- AI here to stay. AI is delivering value on multiple fronts, but the most clear and obvious is a into coating. This is an area where we have a front row seat as both the provider and the consumer. Existed coating tools has completely changed at Oracle operates, and we see no slowdown in our own demand for such capabilities. The same is true for all the customers and partners we work with. The demand for AI infrastructure in this domain as well is enormous, ignoring the many, many other growth areas.

Okay. Now before the end, let's look at a summary of our 5 largest sites and the significant progress we're seeing across polite. To begin, let's look at Abilene, Texas. Abilene, Texas today has delivered 42% of the total capacity. An additional 35% of capacity will be delivered in the next 90 days, was the remainder delivering in the subsequent quarter.

Moving forward to Shackleford, Texas. We contracted this in August of 2025. Customer delivery begins in the first half of FY '27 -- sorry, first half of calendar year '27. 115 megawatts of power capacity already available online, more than 1 month ahead of schedule. If we take a look at [ Don and ] County, New Mexico, we contracted this in September of 2025. Customer delivery begins in the first half of calendar year '27 as well. Power design is based on gigawatts of clean, energy-efficient bloom fuel cells.

We look at Saline, Michigan, we contracted this in October 2025 and Customer delivery began in the second half of 2027. The network core is ahead of schedule and delivered at the end of this calendar year.

And then to the final slide, I want to touch on [ Port Washington -- Port ] Washington, Wisconsin. This was contracted in September of 2025 and delivery begins in the second half of calendar year '27. I think you can see from all of the pictures the massive progress that we're making across a very large number of sites. It's an incredible time to be a technology and to have the privilege of doing that in a company like Oracle. It's especially fun to have a decided view of the birth of a new business that can join the likes of our applications and database businesses. Hopefully, the lease and the data points give you some insight into why we are so excited about [ OCI ] and where that's going to take work.

And with that, I'm going to hand it back to Hilary.

Thanks, Clay. Before we get to our fiscal year '27 and Q1 guidance, I'd like to share some comments on our funding expectations. We already mentioned throughout the call, the compelling opportunities we see at Oracle based on our portfolio positioning. And our strong Q4 results reflect this well. customer demand and our growing visibility into future revenues is what underpins the long-term financial outlook we shared at our most recent Analyst Day of plus 31% revenue CAGR and plus 28% EPS CAGR through our fiscal year 2030.

In order to unlock this unique growth opportunity, we started a program of capital investments. We'll continue those investments in our fiscal year 2027, with an expected net cash outlay for capital expenditures of around $70 million. This includes customer prepayments and timing impacts expected at around $20 billion to $25 billion, so our reported CapEx will be higher by this amount.

Importantly, these investments are being driven by committed customer demand reflected in our record RPO, giving us confidence in our long-term outlook as well as strong returns on the capital we're deploying. As Clay already mentioned, this demand is allowing us to garner customer prepayments and bring your own hardware at similar or better margins than the rest of our contracts. To support our capital investment program, we expect to raise around $40 million in debt and equity in our fiscal year '27 and that includes our already announced aftermarket equity issuance.

We don't anticipate raising additional debt funding in calendar year 2026 -- our fiscal year 2027 guidance, you can start to see the strong translation of our RPO into revenues with expected growth in our total revenues of plus 34% in constant currency, surpassing the 5-year revenue CAGR included in our long-term outlook. Our fiscal year 2027 gross margin will step down due to timing for the ramp-up of our data center projects into their full revenue contribution, what impact from mix. What these investments are creating pressure on the near term to gross margins in our infrastructure business, we expect margin performance in infrastructure to improve rapidly as we reach full contractual revenue levels at our data centers.

Operating costs, we expect to be slightly negative year-over-year in dollar terms due to efficiency actions driving improved operating leverage. Net-net, we expect our non-GAAP EPS for the year to be $8.05, up 18% in constant currency, excluding the net onetime investment gains we booked in fiscal year '26 from [ AMP ] and Bloom Energy.

I'll finish with guidance for our Q1 2027. In Q1, we expect growth in total revenues of between 27% and 29% in U.S. dollars Of that, we expect growth in cloud revenues of between 58% and 64%. In non-GAAP EPS, we expect between $1.72 and $1.76, up between 17% and 20% in U.S. dollar. And anticipate revenues and earnings will accelerate in the second half of the year as we bring further megawatts online at our data centers to fulfill customer demand.

I look forward to speaking further with all of you over the next 2 weeks and months leading into our Q1 and that our next Oracle Investor Day scheduled for October 28 in Las Vegas.

With that, I'll turn the call back to Ken for the Q&A.

Thank you, Hilary. Lisa, do you please pull the audience for any questions they might have.

[Operator Instructions]. The first question comes from John DiFucci from Guggenheim Securities.

Thank you to -- my question is a question that I've dealt with all this quarter. And Clay, that was a ton of information you gave, which is helpful, really helpful. But there's one little nuance here. You spent a little bit more this quarter on CapEx than we expected, and that's sort of the topic a little bit. It adds into it. We know we all know that component costs have gone up a lot, especially memory. It's grown significantly, right?

And even though you said that most 3Q and 4Q contract or large-scale AI contracts that were prepaid for GPUs, you have a lot of other contracts. This has been an issue for a lot of software companies and large cloud companies. I don't think it's as much of an issue for you given my understanding of how you construct your contracts. But can you explain that to investors like when it comes to these very long-term contracts like between you and the end customer and the suppliers.

Sure. Yes. Good question, John. Good to talk to you again, always Look, I'll answer I think that in 2 parts. In terms of the capital expenditures, at least from what we're seeing in Q4 any increase in CapEx that is not due to component prices from our perspective, that's largely around timing, right? I mean, part of my job is to figure out ways to actually accelerate CapEx it's -- tough life. My job is starting to spend the money a little bit faster, so I can get ramped revenue sometimes. So I don't see that as related to component prices. Now talking about component prices in general.

Look, I think everyone knows that memory prices have definitely gone up at the steep prices, hard drive prices, et cetera. So one of the things that we do, John, it's actually quite simple. When we're selling stuff. In a time period where we have certainty, but there will be certainty because the capacity is already deployed or we have certainty because we have lost price is across the spectrum, whether it be space and power costs, energy costs, people costs, component costs.

When we know those costs, we will then do fixed-price contracts. Time that we don't know those costs because it's out far in the future or we have too much supply chain risk, whether that be due to just the way the world works or a lack of things being locked in. We then do not do fish price contracts with our customers and we have a mechanism whereby those costs end up being flipped.

So I don't like it when costs go up. Our customers don't like able to cost go up. But honestly, I don't think our suppliers do love to -- they'd love to be able to deal with everything we want. But when the costs do go up, we have, I think, a very robust set of mechanisms ensure that Oracle is not sitting there with reduced margins.

That is really helpful. It makes a ton of sense. And if I could, just a quick one, Hilary, you kind of alluded to what I'm going to ask. But this is a second question, I get a ton of questions on. You have long-term targets out there. You're a new CFO, right? And congrats. It's great to have you on the line. But can you just comment on those long-term targets at all? I know you've only been there a couple of months.

Yes. I think -- and that would the intention of putting a slide. I think that we're reconfirming those long-term targets and the in the sense of the CAGRs that we put into the slide today. So we feel comfortable with that. And you can see the RPO building to the level that you can start to have a lot of belief, I think, in those long-term targets exactly. So full reconfirmation from my side on the long-term targets.

The next question comes from Brad Zelnick, Deutsche Bank.

Great. And Hilary, welcome to Oracle. Hilary, as you come to Oracle from a capital-intensive business in another industry, how would you suggest that investors evaluate Oracle's progress and returns during this period of heavy investment.

Yes. So the way I think about it, and as we said in the earnings call, we feel the returns for the infrastructure business. So that CPU and GPU business are quite strong. probably from a back even envelope standpoint, the way I think about return from that business model is in return on invested capital. And what we see is return on invested capital in the high 20s at a steady state once the revenues have ramped for large projects at the project level. And that doesn't take into account upside like who knows if the GPUs don't need to be replaced over the long term and things like that. Just purely in the steady state, where we're at the steady state of the contracts that we have.

And as we're generally able to preserve and improve margins in the case of things like bring your own hardware, the ROIC structures, the ROIC for those will be even higher. And again, that back of envelope, I'm just calculating return on invested capital is after-tax operating margin plus depreciation divided by gross investments, so total gross CapEx at the project level. Maybe that gives you a little bit of an idea. And of course, we're happy to talk more about that over the next couple of quarters.

That's really helpful, Hilary. And congrats to the whole team on the execution this quarter. nice shop to everybody.

Next, we'll take a question from Mark Moerdler from Bernstein.

And also congrats on the quarter and or welcome, and we're really looking forward to working with you. Clay and Hilary, with so many vendors entering the market to deliver AI data centers, including the Neo Cloud, SpaceX, which is now going to build data centers in Spain, et cetera, where does Oracle see itself in the competitive landscape? And how do you see that increase in capacity impacting your ability to, one, retain customers renew contracts to capture new customers; and three, maintain or improve margins?

Yes. Thanks, Mark. Look, I think that First, I think it's very important that we stay focused on customers. So the nice thing is that I think whether you see it from existing RPO or increased contracts that we're getting, yes, there's a lot of things happening in the market. but we have a large, diverse set of customers, both very large and also smaller customers. And what I spend all of my time doing is I wake up every day and I go, how do I make sure those customers are as happy as possible with us.

And that's -- when I shared the numbers, for example, in my -- in the prepared remarks about the extremely high utilization rate, even when things come back for renewal, they're instantly snapped up. Those are all indicators that we have great customer relationships. They're happy with their products, and they're very satisfied with the prices that we're charging for them. But I think there's going to be a lot of people who enter the space. I think there's clearly several years in, there's still a massively higher demand than there is supply.

And so I think that's going to -- there are going to be more and more people trying to figure out how to meet that demand -- but I don't worry about that. I really focus on how do we make sure that we can meet as much of that demand at a reasonable margin profile. And that's what I think you've seen us invest new business models to -- In terms of how does that affect our future renewals I find that largely what affects future renewals is that several years relationship that we're going to have between now and then. And we're fundamentally in service business. If you think that you're just buying something and then you're done with it, it's not the way it works, right? These people are relying on what we do with Oracle to run and maintain these massive clusters every day. Our ability to do that extremely well, creates an extremely positive relationship, then ensures that the renewal goes well.

And then in terms of the margin profile, look, I've been at Oracle now for 12 years, the whole time I've been working on OCI. What I can tell you is it's not easy to build an extremely efficient, highly secure robust cloud. So I think that our customers see and appreciate the value of what we provide, the flexibility that we give them, the comprehensive set of services that we provide. So I think that over time, as the market continues to mature, and we deploy more and more of our research and development dollars and make it make more efficient. I think there's ways that Oracle gets higher and higher margins, but we actually now -- lower and lower prices to our customers. That's ultimately the job that is on our shoulders and what we've been doing over the past decade is why the biggest and most robust customers come our way.

The next question today is Keith Bachman from Bank of Montreal.

Yes. Thank you very much for the question. Mike, I wanted to direct this to you, if I could. You mentioned 2 things as net new. One was moving towards outcome-based commercial pricing models, the other was rolling out some incremental token packages. And I wanted to see if you could flesh out the why. And more specifically on the commercial -- outcome-based commercial pricing models -- how do you think this reduces friction. And what is this related to what modules? In other words, I assume this is the SaaS portfolio, ERP, ACM, what models might this relate to? And then finally, how do you think this might impact Brooks? That's it for me.

Sure, Jeff. Thanks for the question, Keith. So alpha pricing is not entirely new for us. So this is something we've been doing in our construction business. Based upon construction value under management, subcontractor -- a general contractor, a subcontractor cash flow and payments of sellers, as I mentioned in my prepared remarks, with hospitality even in health care, in our new AI-based automated agents where we're automating doctors' notes, we're automating lab orders. We're able to measure and actually price based on patient throughput, which is what the providers. One of the things providers care about is people can get through a health care system, reduce waiting queues -- service to patients.

What is new is that we're now expanding that offer across our entire fleet, as you mentioned, across all of our applications, including our -- fusion piece. Now the sort of difficult thing is that you're not creating the outcome in the first place, that's a tricky thing to present. But since we've made this full stack invested. And since we're able to very easily take the best of the output from the large language models to our customers, pair that with our -- both our horizontal applications and our industry applications, we have a very easy way to measure outcomes for our customers.

And as I mentioned, one of the things we're increasingly hearing from customers is -- how much are we going to spend on AI? And how do I get ROI very quickly. So I think we have a very unique advantage since we're in the infrastructure business, we have large -- lines large vendors training out there. We've got -- we've got all of our applications business, both horizontal and vertical businesses. We are naturally generating these outcomes for customers, and it really gives us the ability to help them understand their own AI budgets as well as align that to the value, again, which is really easy to measure. So I think it's a unique offering. It relies on the full stack investment that we made.

And as I mentioned, early days, early days, but certainly resonating very -- customers. They appreciate the transparency, appreciating about outcomes -- align outcomes to AI spec. I also mentioned the token models, customers want this, again, a lot of we're doing our fusion applications, our industry applications. We continue to add at no additional charge.

The customers want access to advanced reasoning, they want essentially for more tokens at the models, we have prepackaged bundles to allow them to do that. So we're allowing as much liability and as much aligned with the value in our pricing models across our entire application set as we possibly can. And I expect that, that will continue to resonate well with customers as it did in the quarter. And as we roll it out across our entire fleet, certainly should be helpful for our growth story as well.

Your next question is from Raimo Lenschow from Barclays.

Perfect. And welcome to the team as well for me. The question I had was we talked a lot about like AI and the with great momentum you have there, but you still have like the classic Oracle business that we all grew up with. And there's a lot of kind of noise in the market at the moment, especially on the investor side, what's happening to software, et cetera. Can you kind of address a little bit what you're seeing on the database side, there's OCI at Azure, et cetera, and overall database momentum? And then on the application side, the growth rate ticked down a little bit, but then you also mentioned on the call some very nice customer wins. Can you see what you see -- can you talk to what you see in the traffic business?

Yes. Sure, it's Mike. So I'll take a sad here and then ask Clay to jump in as well. So on the applications business, we think double-digit growth on a -- quarter result, inquiry run rate of $4.1 billion is pretty good. And we're certainly happy with our continued double-digit growth.

As I mentioned, our deferred position in the quarter grew by 16%. So we preferred position is growing faster than our in-quarter revenue and it gives us confidence that as far as impact of -- I would say maybe a couple of quarters ago, there were some delayed decision cycles out there as customers saw through that.

But really, particularly in the mission-critical systems space, which is where we play at Oracle, people have quickly moved on to that and realize that's enterprise software, particularly when you have AI built into our SaaS solutions is certainly a very good approach and is necessary to move forward for the modernization and protection of their businesses. So I expect that our applications business will continue to be a healthy contributor to Oracle as it has been.

As far as the anybody, we -- they caught innovation of 29% in the quarter, with multiple -- mentioned growing multi-cloud revenue growing 4x poking through 3.25x in the quarter. And here's the really good news on database. We're in early innings, very early days on multi-cloud database. We continue to unlock new regions and unlock new partnerships in some cases with our competitor cloud, we expect that business to continue to be an outsized growth and engine for going forward.

And I'll say the final piece is that -- in addition to multi-cloud, the innovation and the data I mentioned a couple of the detailed security and agent memory that we put in the database, things like better database surge and features that we've been adding into the database, our part and parcel to company's strategies, the data strategy matters, data architecture matters. And as the -- market starts to take hold, which is, again, also in early days, a lot of that data is just in the Oracle database across the world. And we expect to see continued investment in growth in our database business as a result multi-cloud.

All facets of database, Oracle Cloud, multi-cloud as an underpinning support pillar for all of our applications and all of the spoke workflows in the world that are running work database, prognosis is very good.

And our last question today comes from [ Kirk Materne ] from Evercore ISI.

Yes. I had one, maybe sort of 2-parter around the bring your own hardware and prepaid dynamics. Maybe for Clay and then for Hilary. I guess Clay for you, about 12% of RPO is now related to these type of deals. When you look at the pipeline, where do you think that split could ultimately go? And when you go into these type of deals, especially on the bring-your-own hardware side. What's the value differentiation that maybe those deals have that some of the ones that include -- or I guess how does that differ versus the GPU type deals?

And then Hilary, just to clarify on the CapEx guide. I think you said $70 billion in CapEx, but that was excluding, I think, $25 billion from some of these prepaid deals. So could you just, I guess, talk about this dynamic as it relates to sort of your CapEx outlook?

Yes. So I'll start and then hand it to Hilary. Look, I mean I would like to tell you that yes, we know all of us how things are going to change in the future. I can't say that today. The reality is that what's driving, I think, the mix change is an evolving business model. right? You have a lot of different types of accelerators. You have a lot of different customers. You have a lot of the business in the arrangements.

And so ultimately, one of the things that Oracle can provide to our customers is that we can go out and put up front capital and then depreciate that over a period of time and help finance the customers use to that. But that's not -- the only thing we provide and for a lot of customers is not even the most important thing to provide. What they contract with us for is the ability to go out and get these data centers constructed, design them properly, secure them design networks that go inside of them, install a cloud, give us a complementary set of services around the -- hardware because it turns out that a set of these accelerators on their own is not functioning cloud. You need general-purpose compute, general purpose storage, you need load balancers, you need security function, you need identity. You need all of that to actually make this stuff usable and Oracle provides all of that.

And then anyone that thinks that these things are easy to operate is very confused. So you're not just buying a single rack and putting it into your and your data hall. These are extremely complex clusters that require constant care and feeding cost and maintenance across the network and the hardware itself. And so when you add all that together, I think that what you're seeing is that you've got different entrants in the market around accelerators that are helping customers find ways to procure their accelerators.

And you've got different customers who have different ways of thinking about that. And so I can't tell you exactly the mixture of when we do bring your own hardware versus when we do prepay versus when we bring the capital. But what I can say is that I think you'll continue to see innovation and evolution in this model, given the rapid changes that are happening across this entirely ecosystem.

So let me just start -- we said on the call a couple of times that we also see the margins in those structures, either at or better than the prior contracts that we have. So that's good news in terms of economics. We did introduce this quarter this net cash outlay for capital expenditures, which I think is pretty important to understand our funding requirements.

So indeed, for 2027, fiscal year 2027, we expect around $70 billion in net cash outlay for capital expenditures. That does the exclude $20 billion to $25 billion in prepayments that we will -- or there's some timing difference is in there, but it's just associated with third-party manufacturers, not vendors, not vendor financing, just third-party manufacturers.

So $20 billion to $25 billion. The sum of those is our reported CapEx. But from a funding standpoint, what's happening here is that these structures are enabling us to have a lower cash CapEx requirement when we look at how we plan our business. And also from an economic standpoint, of course, because we're collecting money upfront.

So in normal cases, we would put out the CapEx amount and then later, we would collect money from customers. Here, we collect money from customers upfront and actually -- so that's going to come out of our funding to pay for the CapEx or not 100% of it. Therefore, the return on capital is going to be a bit better as well.

Thank you, Hilary. So for next quarter, this is new for everybody. We expect our Q1 fiscal year '27 earnings results will be announced on September 10. Any changes in the date will be publicly announced.

Also, as a reminder, Hilary brought this up earlier, but our Investor Day will be held on October 28 in Las Vegas. We look to see you all there as part of AI world. A telephone replay of this conference call will be available for 24 hours on our Investor Relations website. And as a reminder, the slides that you saw today will be posted to the website shortly.

Thank you for joining us today. And with that, I'll turn the call back to Lisa for closing.

And once again, ladies and gentlemen, that does conclude today's conference. We would like to thank you all for your participation today. You may now disconnect.

Hello, and thank you for standing by. My name is Regina, and I will be your conference operator today. At this time, I would like to welcome everyone to the Oracle Corporation Third Quarter Fiscal Year 2026 Earnings Conference Call. [Operator Instructions]

I would now like to turn the conference over to Ken Bond, Head of Investor Relations. Please go ahead.

Thank you, Regina, and good afternoon, everyone. Welcome to Oracle's Third Quarter Fiscal Year 2026 Earnings Conference Call. On the call today are Chairman and Chief Technology Officer, Larry Ellison; Chief Executive Officer, Clay Magouyrk; Chief Executive Officer, Mike Sicilia; and Principal Financial Officer, Doug Kehring. A copy of the press release and financial tables, which includes the supplemental financial details on our most recent quarter, guidance for our future results a GAAP to non-GAAP reconciliation and a selected list of customers who purchased Oracle Cloud Services or went live on Oracle Cloud recently will be available from our Investor Relations website. .

As a reminder, today's discussion will include forward-looking statements, and we will discuss some important factors relating to our business. These forward-looking statements are also subject to risks and uncertainties that may cause actual results to differ materially from the statements being made today. As a result, we caution you from placing undue reliance on these forward-looking statements, and we encourage you to review our most recent reports, including our 10-K and 10-Q and any applicable amendments.

Finally, we are not obligating ourselves to revise our results or these forward-looking statements in light of new information or future events. Before we go to the Q&A portion of the call, we'll begin with a few prepared remarks. And with that, I'll turn it over to Doug.

Thanks, Ken. Let me start by highlighting the changes we are making to our earnings press release and this call. In the press release, we have laid out clearly and explicitly the supplemental financial metrics that we otherwise would have provided on the earnings call so that each of you has the information in writing and in advance.

Then as it relates to our approach to the earnings call itself, I will be very brief and then turn it over to Mike and Clay to provide more substantial thoughts on our business. After which, all of us, including Larry, will be available to take questions.

In terms of the results for Q3, we had a tremendous quarter that exceeded expectations across the board. Our momentum continues to accelerate, with Q3 being the first quarter in over 15 years, where both organic total revenue and organic non-GAAP EPS grew at 20% or better in USD, as we highlighted in the press release. I'll quickly mention a couple of things and then hand the call over to our CEOs.

First, in January, TikTok U.S. completed the separation of its U.S. data operations from ByteDance into an independent company in which Oracle now holds a 15% equity stake along with a seat on the board. In terms of impact to our financials, there is no impact to the revenue related to the services we have been providing as their technology vendor [Audio Gap] equity investment, we will be accounting for this under the equity method, and we will recognize our share of the new company's earnings for the period from the close of the investment in late January to March 31, and in our Q4 results as there is a 2-month reporting period time lag.

It will be recorded as nonoperating income or loss on our income statement and is incremental and additive to our financials. Second, in February, we announced our intent to raise up to $50 billion in debt and equity financing along with the statement that we do not expect to issue any additional bonds beyond this amount in calendar year 2026. Within days of the announcement, we raised $30 billion through a combination of investment-grade bonds and mandatory convertible preferred stock with a record order book that was substantially oversubscribed.

As noted in our release, we have not yet initiated the at-the-market equity portion of the financing program. Finally, I'd be remiss not to remind everyone, we are reporting our financial results just 10 days after the last day of the quarter despite the increasing size and complexity of our business. Using Oracle Fusion, we continue to close and file our financial results faster than any other company in the S&P 500, providing us with a significant strategic advantage as well as an opportunity to help our Fusion customers do the same with their businesses.

With that, let me now turn the call over to Mike. .

Thanks, Doug. And as Doug just detailed, we really had an excellent quarter across the board and continue to see strong execution. So let me say a few words about our applications business. Oracle has the fastest growing, most complete suite of cloud applications in the market full stop. Our SaaS solutions are industry complete platforms with highly scalable, trusted, secure and regulatory compliance systems and processes in which our customers trust us to run the systems that run their businesses.

In constant currency, cloud applications revenue was up 11% in the quarter, reaching an annualized run rate of $16.1 billion. Within that, Fusion ERP was up 14% and using SCM, up 15%; Fusion HCM, up 15%, using CX up 6%. NetSuite was up 11%. Industry SaaS solutions for hospitality, construction, retail, banking, restaurants, local governments and telecommunications combined, were up 19%. So certainly, very happy with the applications growth in the quarter.

In the context of that, I'll say a few words about the reported [ SaaS apocalypse ]. You've all heard the thesis or theory that new companies coding quickly using AI will spell the death of SaaS. I don't agree with that at all. I do think that AI tools and their coding capabilities would be a threat if we weren't adopting them, but we are and very rapidly. Oracle is using the best AI coding tools and the best developers not only to accelerate our SaaS business but to deliver solutions that enable entire ecosystems across numerous industries.

The use of AI coding tools inside Oracle is enabling smaller engineering teams to deliver more complete solutions to our customers more quickly. We are building brand-new SaaS products using AI and also embedding AI agents right into our existing applications suites. By embracing AI with small engineering teams, we have just built 3 brand-new CX applications, lead generation and qualification, sales orchestration and automated selling and our new website generator.

In fact, we just used the website generator to build and launch the new oracle.com. We've built these new CX products to help our customers sell not simply to forecast or generate e-mail opens, these are 3 products that Salesforce.com does not have. And of course, salesforce.com also doesn't have OCI, the AI data platform, Fusion ERP and complete industry suites. Complete AI-powered end-to-end ecosystem automation platforms are quite unique to Oracle.

In addition to that, we've already delivered well over 1,000 agents right inside our horizontal back office and industry applications. This doesn't even include the agents that our customers are building themselves or the fleet of agents that we're using internally. These are AI features built right into our applications and existing processes. And a great example, I think, is in health care, where our brand new AI-powered ambulatory EHR, electronic health record system is live in the market and the results are quite clear. We are reducing administrative overhead, we're allowing clinicians to see more patients. We're improving access to care, and we're increasing provider satisfaction.

In another example, in banking, we provide a comprehensive AI-powered SaaS platform, including everything from commercial banking, retail banking, investment banking, anti-money laundering, financial crime and compliance, payments, supply chain financing, CX, ERP and HCM. That banking suite alone contains hundreds of embedded AI agents, all available at no additional cost to our customers.

In retail, our AI-enabled solutions span merchandising, assortment planning, supply chain management, point-of-sale commerce and of course, ERP, CX and HCM. In summary, these are not systems that can be replaced by a small collection of these features cobbled together and bolted on in the name of AI. So yes, some smaller or single focused SaaS players may well be disrupted, but Oracle will not be among them.

Now let me focus on key -- a few key wins in Q3 in the application space. And by all means, this is a very short list, not an exhaustive list. Memorial Hermann Health System selected Fusion ERP, SCM and HCM. This was a win over Workday. University of New South Wales also selected Fusion ERP and HCM, also a win over Workday. Gregg Media selected Fusion EPM and ERP, a win again over Workday and also over SAP. Investec Bank selected Fusion EPM and ERP over SAP. HID Global Corporation also selected Fusion ERP and SCM over SAP.

Ethiopian shipping and logistics services enterprises selected Fusion ERP, SCM, and HCM, again, over SAP. A major Wall Street Bank elected to standardize on Fusion ERP for the entirety of their business and all of their business units replacing SAP full stop. Loudoun County Public Schools selected Fusion ERP, EPM, HCM and SCM. The J.M. Smucker Company selected Fusion ERP and EPM, Westfield Insurance [ spec ] Fusion ERP, EPM, HCM and procurement. Mitsubishi UFG Financial Group is an existing cloud customer and database customer, they are now moving into both our Fusion ERP and industry SaaS applications. SDC Kuwait, an existing major tech customer is moving [ EBS ] to the cloud to support their growth. So just a very small list of major application wins in the quarter.

In the quarter, we had over 2,000 customers go live in Q3, 2,000 customers when you think about our industry applications and our fusion applications put together over 2,000 went to live. And more importantly, we continue to see the median time from live decreased a very small sample of go-lives in the quarter. Herst expanded their ERP with enterprise -- with EPM as well as HCM. J.M. Huber companies across Fusion ERP and Hcm. Emirates Health Services went live with HCM, which enabled a comprehensive HR, payroll and talent suite to elevate their workforce management.

Niagara bottling that went live on SCM moving from on-premise ERP to Fusion. Seadrill is now live across ERP, HCM, SCM and EPM. Again, very -- with 2,000 lives in the quarter, that's just a very, very short list of go-lives, but you can see, hopefully, not only momentum but multi-pillar momentum with these customers. I also have an equally shortlist compared to the overall list of key tech wins in Q3.

Lockheed Martin selected OCI high-performance compute to scale AI across their environment efficiently. Rambus selected OCI compute, networking and storage for AI video and security across all of their workloads. Lucid Motors selected OCI core services for data and connectivity in order to expand into European markets. Infomart in Japan selected OCI for their mission-critical B2B platform. Claro Brazil selected OCI alloy for sovereign AI.

Air France KLM, which is a multi-cloud win featuring a win with the Oracle database at Azure, and that led to a 13x performance improvement at a significantly lower cost for Air France KLM and Activision Blizzard, an existing oracle E-Business Suite was also an Oracle database at Azure win. Oracle's embrace of AI across our strategic applications is leading to broader enterprise conversations with our customers involving our full stack, OCI, AI data platform, Fusion applications, industry suites. These conversations are about ecosystem automation. They're not about single apps. They're about automating the entire ecosystem, and they further been able to by our simplified go-to-market model, which we spoke about in our last earnings call.

This is allowing us to close more multiproduct deals with more customers combining the power of the Oracle database, our OCI platform, our AI tooling and our complete application suites. In constant currency, cloud applications deferred revenue was up 14% versus in-quarter cloud applications revenue growth of 11%, which further supports our acceleration thesis.

With that, Clay, I'll turn it over to you.

Thank you, Mike. Okay. So I'm going to talk about 2 segments of our business: our multi-cloud database and AI infrastructure. Both are growing extremely quickly. Multi-cloud database revenue grew 531% year-over-year. AI infrastructure revenue grew 243% year-over-year. Both also have demand that exceeds supply and a clear execution plan for Oracle that will rapidly turn that demand into profitable recurring revenue.

Oracle database has run on any hardware and operating system for decades. Oracle Database Cloud Services up until recently, were only available in a single cloud, OCI. We created our multi-cloud partnerships with First Microsoft, then Google and finally, Amazon to bring the best database platform to all cloud. Those partnerships unlock an enormous backlog of demand. Our database customers who want to use our database in other clouds. This quarter, we achieved an important milestone.

We have global region coverage in all of our partner cloud. We now have 33 regions live with Microsoft and 14 live with Google. We delivered significant growth with AWS beginning Q3 with 2 AWS regions live, exiting Q3 with 8 AWS regions live, and we will exit Q4 with 22 AWS regions live. AI is also accelerating the adoption of our database cloud services. The improvement in model coding fields and agentic capabilities pushes customers to move their most valuable data into our cloud services. They need access to the latest AI features to support vector embeddings and CP server access and advanced security controls. Customers also need their data to be co-located with the agents themselves, and our multi-cloud database makes that easy. Our multi-cloud architecture brings the best of Oracle Cloud into our partner regions.

This ensures that we will rapidly turn billions of pipelines into highly profitable database service revenue. Demand for AI infrastructure, both GPU and CPU continues to exceed supply. This is directly visible in our $553 billion RPO. I want to share a model for how that RPO turns into profitable recurring revenue as well as some operational metrics that are early indicators of our progress.

AI infrastructure begins with data centers and power generation. Through our partners, we have secured more than 10 gigawatts of power and data center capacity coming online over the next 3 years. Those infrastructure investments also need funding and greater than 90% of that capacity is fully funded through our partners, with the remainder planned to finish this month. Once the data center is secured, several things must come together. The data center and on-site power generation has to be constructed. Compute, networking and storage has to be designed, manufactured, delivered and installed. All the capacity inside the data center also has to be funded.

We continue to innovate across each of these steps. We optimize our data center construction through standardized designs. Our supply chain has improved with more suppliers and deeper relationships. We have tripled our manufacturing sites and increased rack output by 4x all in the last year. We have scaled our installation processes to enable multiple phases of delivery in parallel. Time from rack delivery to revenue has reduced by 60% in the past several months.

We also continue to innovate on our business models. On our last earnings call, I shared multiple ideas for how we can incrementally grow our AI infrastructure without Oracle raising more debt or issuing equity. We have signed more than $29 billion of contracts since then, across multiple customers using that new model. A combination of bring your own hardware and upfront customer payments enables us to continue expanding without any negative cash flow from Oracle.

Of course, this $29 billion is in addition to other deals we signed this quarter. Ultimately, all of this results in capacity delivered to customers and revenue to Oracle. In Q3, we delivered more than 400 megawatts to customers. 90% of that committed capacity was delivered on or ahead of schedule as we've consistently done over several quarters. This is why customers continue to choose Oracle for their infrastructure needs.

Investing in the AI infrastructure is capital-intensive, but our operating model is optimized to ensure profitability. Flexible infrastructure design, high utilization and rapid handover combined with diversified customers create an incredible, Increased scale spreads our fixed costs over a larger base increasing profitability. It's unprecedented to scale a capital-intensive business so quickly while also increasing profitability.

Looking at the AI capacity we delivered in Q3, our gross margin for that remained above our 30% guidance at 32%. Now combine that with our other segments of OCI, which have much higher margins, like our database services, and you can see why Oracle is growing so quickly and profitably. Our numbers speak for themselves. We are overdelivering on FY '26 revenue and earnings, and we are constantly raising our FY '27 forecast. This is made possible by Oracle's transition from a predominantly seasonal license business into a highly predictable recurring revenue class business.

Demand for AI and advanced compute will continue to expand broadly across the economy. There will be many successful models, agentic platforms and businesses that emerge. We support hundreds of the most advanced AI customers today and more continually want to work with us. We build infrastructure that is flexible, fungible and can support the smallest workloads up to the largest. We continue to offer the latest in accelerators from the most recent NVIDIA AMD options to emerging designs from companies like [ Cerebro and Peloton ]. Altogether, we are confident that the investments we make now in data centers, compute capacity and customer relationships will only grow more valuable time. Back to Ken for questions.

Thank you, Clay. Regina, if you could please poll the audience for questions.

[Operator Instructions] Our first question will come from the line of John DiFucci with Guggenheim.

Thank you. Well, a lot going on here. So listen, I'm going to let others ask about the AI infrastructure question. But we've heard Doug talk about a halo effect that the AI infrastructure business is having on the rest of your business. This quarter was strong, and you said that the RPO increase was from large-scale AI contracts. At the same time, we're hearing from the field now that halo effect is actually turning into business.

Outside of AI infrastructure, it sounds like the go-lives are steady, but the business activity and especially the pipeline are up materially from more traditional cloud workloads, including dedicated regions, sovereign clouds, even [ alloy ] deals we've started hearing about. In addition to what Mike started talking about, with the often related apps deals. I realize these types of deals aren't the scale of these AI deals. But can you talk about what seems to be an underlying momentum building in these businesses. Am I right to be thinking of this? And if I could, on a sort of related topic, can you give us any visibility into CapEx for fiscal '27?

Okay. John, this is Mike. I'll take the question. So yes, we absolutely are seeing a halo effect, and let me add a little bit of color on that. As far as the apps business, the fact that we're training so many models on OCI and so closely provisioned to our applications allows us to embed very high-quality AI services right into our applications, as I said, as features.

So not only are we serving these customers, serving the model vendors for training. But we're also embedding a lot of the output, right, into our applications. Of course, we're doing prompt engineering and things like that to make it relevant to the business. But the fact that that we are the custodian in our applications business of so much of the world's mission-critical data. We have very close provisioning, very close proximity to these models, putting those 2 things together allows customers to get value from AI very, very quickly.

And if you've heard any criticism with AI and well, I can't get value quick enough. Well, actually, when you bundle up as a service and expose private data to AI that we are the custodian of the applications that we've seen terrific wins. I mentioned some of the verticals you heard about there, but I think that's true across the board. The other piece that is a very interesting halo effect is leveraging our infrastructure, just OCI, infrastructure as a budget creator for customers.

You've heard us say it before, we're faster and cheaper than everybody else. And when customers are thinking about these large-scale application or large-scale infrastructure transformations, we can also help them get to a position of a budget creation to be able to fund that transformation simply by moving their workloads to OCI because we can run them more quickly and more efficiently and less expensively than our competitors.

And then finally, the other halo effect before I turn it over to Doug, for your question on CapEx is around sovereign AI. Our sovereign story is not new and it's not a knee jerk reaction to the things that are happening in the world, combine other with our Alloy story, we're really seeing increasing pipeline across the world. The fact that our form factor, and we're so differentiated in our form factor, and we can deliver not just the smaller form factor, but complete OCI services on top of that form factor no matter how many racks are involved, whether it's 3 racks or 500 racks, we think that's a huge differentiator in the market. So you put you put apps together, you put OCI AI services together, you put sovereignty together. And yes, it's a pretty big halo effect.

Yes. And John, just let we start by acknowledging the creativity in getting 2 questions at the same time, it's always fascinating to watch. So on CapEx, I think we'll get back to everyone next -- after the end of the fiscal year and talk about next year's CapEx at that point in time. But I will say a couple of things. Obviously, from what Clay has gone through, the most interesting thing that you just start thinking about is the uncoupling of CapEx with capital requirements from Oracle.

Obviously, when we have these additional funding mechanisms, there may be additional CapEx, but it doesn't require out-of-pocket cash from Oracle, which is quite interesting. So underlying that, as we remain committed to what we talked about last quarter, which is maintaining the investment-grade rating at Oracle as well as staying within the financing envelope that we talked about, obviously, of which we've announced that we're doing $50 billion this calendar year of that total. So more to come, John, on the CapEx after next quarter.

Very much appreciate the color on that, Doug. And Mike, your prepared remarks on AI and how Oracle approaches it, everybody should use that because it's a logical approach.

Our next question will come from the line of Mark Murphy with JPMorgan.

Congrats on the acceleration play. As Oracle transitions to higher levels of AI inferencing, what do you view as the right strategy for trying to optimize the location of your data centers. For instance, you have these huge centralized data centers in Texas and Wyoming, they're very close to power, but they're pretty far from the population centers and the fiber routes that are out there on the seaboard.

So it crosses our minds that the users and the devices are a long distance away. So as you make us move more into inferencing, are you seeing any reason to try to pivot those locations a little closer to where the users and the traffic are.

Sure. This is Clay. Great question, Mark. So let me start by, I think, highlighting our perspective on inferencing and how that impacts kind of data center deployment. So first thing, I would say, is I think we are -- for a while, there's a lot of training going on. inferencing is very rapidly growing, everywhere and anywhere. I think it's because of higher and higher utilization of the models themselves and also new use cases. As anyone who's been using Claude or Codex recently in the software space nodes.

These are incredible tools that are changing how we do everything. So inferencing is going to have a huge amount of demand. Now when you talk about data center location, you mentioned latency is the one. Realistically, there are several reasons you might care about the location. It might be the cost, it might be just overall availability, it might be for sovereignty. So there's different reasons to pick a location. But honing in on your point about latency, the thing, I think, to understand is that latency is all proportional, meaning. If what you're trying to do is you're trying to do a very, very low legacy trade on the stock market, waiting for the 100-millisecond round trip from coast to coast is a bad idea. If what you're doing is you're asking a question for your business, it's going to take an AI model several seconds to think about an extra 40 milliseconds of latency from New York to Wyoming is not going to hurt you.

And so when you actually talk to customers about use cases where they need lower latency, the latency problem right now is not actually the location of the hardware. It's the type of hardware that's being deployed. And that's why you're seeing so much innovation going on around these AI accelerators. If you look at what the Groq Rock does or Cerebras or Positron, all of these different types of customers are saying, well, not only how do we reduce the cost of inferencing, but also how can we significantly reduce the latency of it.

I think if you look forward to GTC from NVIDIA next week, you'll see announcements from them. But across the board, I think the way that an industry we're going to consolidate and kind of reduce latency has to first start with a different architecture for that inferencing. And thankfully, the data center location is actually a very tiny part of that. So it makes much more flexible for us to go out and put the data centers where power is abundant, land is plentiful, and we can actually optimize for what's available to meet this ever-increasing demand.

Our next question comes from the line of Siti Panigrahi with Mizuho.

I want to ask about the opportunity with your AI database and the AI data platform. So with recent excitement on AI and on enterprises now adopting tools from Frontier LLMs. So what are you hearing from customers about training their private data and building their private LLMs. And how confident are you in seeing the inflection in your AI database growth that you talked about at the Analyst Day in October. .

Thanks. This is Clay. So look, I think there are 2 parts to that question. One is, how much adoption are we seeing a kind of private LLMs. And then how much are we seeing of using AI with private data.

I think in the early days, people -- a lot of people thought that most customers will be doing very specific training of their own large language model. I think that has largely proven to not be the case. Instead, what I think is incredibly popular growing in popularity is people taking the best models and wanting them to combine that in a private way with their private data. .

And we're seeing a lot of demand for that. If you listen to Mike earlier talk about, right, how we're embedding these AI models into our applications, that's one use case. But obviously, not everything, unfortunately runs inside of an Oracle application and lots of customer applications are written. So we added a lot of functionality to our Oracle AI database, make it easy to connect whether it be through the MCP servers or natural language to SQL that you can use these models to use. But also, we have our AI data platform product, which is really about solving this exact problem. You have a lot of data, maybe application data, it may be custom data and different data lakes and lake houses and may be data in a structured database.

All that together gives you an agentic platform to quickly build applications on as well as access to all of the greatest models from multiple providers. So across the stack, we're seeing a lot of momentum across that. And that's why in my prepared remarks, I talked about the growth that we're seeing with our multi-cloud database. .

What we see is that for customers to take advantage of the latest and greatest AI, they first have to be in the cloud, and there's still a lot of data that's not in the cloud. And so we see acceleration of moving that most important private data to a cloud environment, so they can then take advantage of the latest and greatest AI with that data.

Our next question comes from the line of Mark Moerdler with Sanford Bernstein.

Congratulations on what's a really good quarter, really great work. I'm going to change over a little bit and discuss on the financial side a little bit. Now that you've completed your major debt raise, can you explain, given the blend of the cost of the -- building out the AI data center and the cost of raising capital to fund the AI data center. How comfortable are you with the values you're creating from the AI data center business itself.

And then as an adjacency, if you don't mind, can you talk a little bit more on the sovereign cloud? Can you discuss how you parlay the AI data center business into being the AI provider for sovereign clouds. And how that should impact your -- the value of -- to Oracle?

Sure. I think we're going to split this ONE up. Mark,

This is Clay. I'll take the first half, and then I'm going to throw it to Mike to talk about some of the sovereign cloud stuff. So look, when you think about the overall profitability of these AI data centers, there's 2 pieces. One is how profitable is it purely on the accelerator themselves. We made guidance in the past that we see gross margin in the 30% to 40% range on that, that continues to hold for us.

And as we continue to get better and better at running these data centers, delivering them more cheaply, optimizing the amount of cost of -- for networking and hardware spend as well as power, we see that continuing to incrementally improve. So we're very pleased with that. The other thing to understand is that in these AI data centers, whether it be for inferencing or for training workloads, only thing being procured is not AI accelerators. There's a lot of general purpose compute. There's a lot of whether it be high performance or large-scale blob storage, there's load balancing, there's identity security products, et cetera, et cetera.

That typically on the order of 10% to 20% of the total spend is going to adjacent services. And when you factor that in, which have higher margins being on the mix of services, the overall profitability continues to improve. And that's without taking into account, as I mentioned earlier, about our multi-cloud database business, that's a much higher margin business, more in the 60% to 80% range. It's growing very, very rapidly. And so when you combine all of these pieces together, the overall margin profile of OCI continues to strengthen and rose rapidly. I mean the thing I would say, the question that I think underlies this that maybe people don't understand is the limitation on the profitability is not on the capacity we delivered.

So let's say that I'm building a data center and has 4 data halls. And I deliver the first data hall that one is profitable. The reason that we're not even more profitable right now, despite the fact that we are continuing to grow EPS, et cetera. is because we have so much under construction at one time, and we have some expenses for those things. Now we're really good at that. We're very, very good at minimizing the time under which that construction is happening. We're very, very good at reducing those costs during that time period, but they're not zero.

And so as our business is going through this hyper-growth phase, that's the only drag on profitability. But thankfully, we're getting -- we're very good and getting better at delivering that capacity that capacity when we deliver it is all already contracted for at a very profitable rate. So when you combine those things together, we're extremely confident in both the capacity we delivered and the continuing to increase profitability of our AI business.

Mike, do you want to talk about sovereignty.

Yes. So sovereignty, as I mentioned earlier, I think we're very well positioned. A year ago, sovereignty was about data sovereignty. And there were some [indiscernible] solutions in the market where there was sovereign data from a primary perspective, but DR and maybe somewhere else, maybe in another country. Of course, that's no longer acceptable. sovereignty is about sovereign data, sovereign operations, and even sovereign contracting. Our Alloy model is perfectly positioned to deliver on all 3 of those things.

And by delivering full stack solutions, again, the big difference between what we're doing with sovereignty and what some of our competitors are doing, we're not simply putting an edge sovereign zone in. We're putting full stack OCI, which had all of our OCI services, and, as you mentioned, margin mix also allows us to run all of our application suite, our AI data platform in that software zone as well. Of course, the margins on some of those are different than our infrastructure margins.

So I think that we're in a very unique position to deliver all that we have at Oracle in a sovereign zone. That sovereign zone can be as small or as large as the customer wants it to be. The other piece is that we have full flexibility as to where we draw the line of sovereignty. We often think about sovereignty in terms of lines of customers, but we also have customers we've been talking with enterprise customers who may operate across multiple countries, let's say, in Europe or in Africa that actually want to have a sovereign zone, a sovereign zone that they control and they operate in their data center, and they're serving customers in a certain vertical industry like health care, for example, or retail, for example, and their sovereign zone is drawn in their alloy across those countries. We can accommodate all of that. We have the most flexibility. We think we have the most flexibility in contract and most flexibility in delivering. And again, the most important thing is that we deliver all that Oracle has in these sovereign zones. It's not a subset. It's not a few edge devices. It's all of OCI.

Our next question will come from the line of Raimo Lenschow with Barclays.

Perfect. I wanted to ask something that we are struggling a lot with when we talk to investors, and that's kind of the theme of SaaS, software application software is dead because AI is going to kill it. Just wanted to hear what you guys are hearing when you talk with customers? Is that like some -- one of these investor things? Is that getting discussed on the customer side as well? And how do you explain it? And I'm just thinking about like what do you guys do with a lot of deterministic rather than probabilistic. So that might probably be the explanation here. But just wanted to hear your perspective again.

This is Mike. I'll take the question. So as far as the customers that I spoke with, I've not yet met a customer who tells me they're ready to give away their retail merchandising system, their core banking system, demand deposit account systems, electronic health record systems and some small cobbling together of niche AI features are going to replace all of that over and overnight. In fact, I hear quite the opposite from the customers. .

What they're asking us is how can we get -- how can we consume as much AI out of the box you're putting into your applications across the board? And how can we get that up and live as quickly as we possibly can because we think that's the best way to actually realize value. But these systems -- what we're running at Oracle, as you know, these are highly complex mission-critical systems with -- we have decades of industry experience, decades of regulatory compliance. And these are the systems that run -- that our customers use to run their business, run their government agency, run their health care organization, whatever the case is.

I really like our position here. As I said, we're leaning very heavily into AI ourselves. So we have 1,000 AI agents already live in Fusion. Our banking alone has hundreds of AI agents just inside our banking solutions. So Yes, we think AI is disruptive. We do, but we think we're the disruptor because we're actually embedding the AI right into our applications full stop, again, at no additional. These are features that come in, in the application suite as part of quarterly upgrades as part of a regular cadence. So I'm actually -- instead of being -- think that AI spells the death of SaaS, at least for Oracle, I think it actually helps our SaaS position and helps us get to market even more quickly. We're thrilled with the results that we have and expect to have a lot more color on this as we go forward.

Our final question will come from the line of Brad Zelnick with Deutsche Bank.

And I'll echo my congrats and also just say that the messaging is very, very clear and very helpful. My question is for Mike and perhaps Larry, and it extends on what Raimo has asked. You've introduced AI agent studio inside of Fusion and we all know that the crown jewels within an enterprise live inside of Oracle Database and Oracle apps. But I'm curious, how do you see Oracle's role evolving in a world where many other players are vying to be the AI interaction layer across multiple different enterprise systems and workflows..

So Brad, it's Mike. So I'll start. So look, I think data gravity matters here. And I think mission-critical data gravity matters even more. So as we said, we've announced the AI agent Studio inside of Fusion. Fusion is a system inside our customers that is the custodian of their operational data, their mission-critical data. So if you're going to build a bunch of AI agents as or your system integrator is going to build a bunch of AI agents. The question I would have is where we'd just start. .

Well, you can start inside the system of record, you'd start inside the system of gravity because that is the data from an inferencing standpoint, from a retrieval augmented generation standpoint is going to be highly relevant and highly specific and add a bunch of context to AI. Now the AI agent studio that we've released in fusion is not just specific -- it's not specific to just Fusion data. You can build AI agents across our industry applications, across third-party applications, third parties can build AI agents in there. So the fact that we're delivering an all-in-one best up solution, a full-scale SaaS application, AI powered SaaS applications and giving you the ability to create your own AI agents either on top of that or next to that in a standard upgraded, quarterly platform release schedule, I think, is going to be quite attractive because this AI agent studio that we built at Fusion, it's part of our quarterly upgrades. It's part of our quarterly regular security patching. So you're getting the best we think of both worlds. You're getting packet SaaS applications. You're getting an agent studio, which is very, very close to the most mission-critical germane data that enterprise possesses and you're getting the ability to create your own custom bespoke agents if you'd like to as well.

Yes. I'll just end with, we provide a bunch of prebuilt agents for all of our applications. But in addition, we provide a development environment, the AI data platforms development environment that allow our customers to easily add their own agents to what we build. We don't think we can build all the application agents for a banking system or all the application agents for a health care system. A lot of our partners are going to do that.

A lot of our customers are going to do that. What the AI data platform does is it provides a complete integrated development environment where you can build your own agents using any AI model that is in the Oracle Cloud and that is basically all of the popular AI models. You can use it for coding the agent, you can use it -- you do multistep reasoning for queries, we plan in our fusion accounting system, for example, we will have a complex agent that does something called the close. So when you close your books with Fusion in the not-too-distant future, it will be an autonomous agent, no human beings involved.

You will close your books by simply telling the AI agent to go ahead and close the books, and then you will get your results. We provide a lot of AI capability built into our applications, but they are open. They are open that allow our customers and our partners to add to that portfolio of agents and we built an entire ecosystem that automates health care, automate financial services automates retail. That is what AI is allowing us to do is to expand our horizons for the scope of the suites of a SaaS software building to automate entire ecosystems. Let me talk about health care. In health care, Epic automates hospitals, acute care hospitals, and in some cases, clinics, but primarily acute care hospitals.

We automate acute care hospitals. We automate clinics. We automate laboratories where we automate the payers, the people who actually maybe we automate the insurance companies. We automate the HCM system that trains their nurses that schedules they're ready to get the right radiologist when an MRI is given. But that automates the hospital's financials but also automates the FDA and that approves the latest drugs that deals with the pharmaceutical companies.

That's the health care ecosystem. It's enormous. And thank God, we have these coding tools now that allow us to build a comprehensive set of software, agent-based software to automate a complete ecosystem like health care or financial services. That's what we're doing at Oracle. That's why we think we're a disruptor. That's why we think the SaaS apocalypse applies to others but not to us.

A telephone replay of the conference call will be available for 24 hours on our Investor Relations website. Thank you for joining us today. And with that, I'll turn the call back to Regina for closing.

This will conclude today's call. Thank you all for joining. You may now disconnect.

Hello, and thank you for standing by. My name is Tiffany, and I will be your conference operator today. At this time, I would like to welcome everyone to the Oracle Corporation Q2 FY '26 Earnings Call. [Operator Instructions] I would now like to turn the call over to Ken Bond, Head of Investor Relations. Sir, please go ahead.

Thank you, Tiffany. Good afternoon, everyone, and welcome to Oracle's Second Quarter Fiscal year 2026 Earnings Conference Call. On the call today are Chairman and Chief Technology Officer, Larry Ellison; Chief Executive Officer, Mike Sicilia; Chief Executive Officer, Clay Magouyrk; and Principal Financial Officer, Doug Kehring.

A copy of the press release and financial tables, which includes a GAAP to non-GAAP reconciliation, other supplemental financial information and a list of many customers purchased Oracle Cloud Services or went live on Oracle Cloud recently will be available from our Investor Relations website. As a reminder, today's discussion will include forward-looking statements, and we will discuss some important factors relating to our business. These forward-looking statements are also subject to risks and uncertainties that may cause actual results to differ materially from statements being made today. As a result, we caution you from placing undue reliance on these forward-looking statements and we encourage you to review our most recent reports, including our 10-K and 10-Q and any applicable amendments, and finally, we are not obligating ourselves to revise our results or these financial-looking statements. Before taking questions, we'll begin with a few prepared remarks. And I'll now pass the call over to Mike.

No, it's actually to me, Ken. Appreciate it. This is Doug. As it relates to the numbers we are about to present, the following apply to both the results for Q2 and to our guidance for Q3. First, we'll be discussing our financials using constant currency growth rates as this is how we manage the business. Second, we'll be presenting our numbers on a non-GAAP basis, except where we indicate otherwise. Finally, as it relates to currency, it had a 1% positive impact on revenue and $0.03 positive impact on earnings in Q2. For Q3, assuming currency exchange rates remain the same as they are now, currency should have a 2% to 3% positive effect on revenue and have a $0.06 positive effect on EPS depending on rounding.

In terms of the results for Q2, we had another excellent quarter of execution. Remaining performance obligations, or RPO, ended the quarter at $523.3 billion, up 433% from last year and up $68 billion since the end of August, driven by contracts signed with meta, NVIDIA and others as we continue to diversify our customer backlog. RPO expected to be recognized in the next 12 months grew 40% year-over-year compared with 25% last quarter and 21% last year.

Total cloud revenue, which includes both applications and infrastructure, was up 33% at $8 billion representing a significant acceleration from the 24% growth rate reported last year. Cloud revenue now accounts for half of Oracle's overall revenue.

Cloud infrastructure revenue was $4.1 billion, up 66% with GPU-related revenue growing 177%. Oracle's cloud infrastructure businesses continue to grow much faster than our competitors. Cloud database services revenue was up 30% with autonomous database revenue up 43% and multi-cloud consumption up 817%. Cloud applications revenue was $3.9 billion and up 11%. Our strategic back-office applications revenue was $2.4 billion and up 16%. As we finish combining our industry-based cloud apps and our Fusion cloud apps under 1 selling organization in each region across the world, we have been seeing increasing cross-selling synergies and that are expected to drive higher cloud applications growth rates in the future.

All in, total revenues for the quarter were $16.1 billion, up 13% and higher than the 9% growth reported in Q2 last year, continuing our trend of accelerating total revenue growth. Operating income grew 8% to $6.7 billion. Non-GAAP EPS was $2.26, up 51%, while GAAP EPS was $2.10, up 86%. We recognized a pretax gain of $2.7 billion in the quarter, stemming from the sale of our interest in Ampere.

Turning to cash flow. Operating cash flow in Q2 was $2.1 billion, while free cash flow was a negative $10 billion and CapEx was $12 billion, reflecting the investments being made to support our accelerating growth. As a reminder, the vast majority of our CapEx investments are for revenue-generating equipment that is going into our data centers and not for land, buildings or power that collectively are covered via leases. Oracle does not pay for these leases until the completed data centers and accompanying utilities are delivered to us. Rather, the equipment CapEx is purchased very late in the data center production cycle, allowing us to quickly convert cash spent into revenues earned as we provision cloud services to our contracted and committed customers.

In terms of funding our growth, there are a variety of sources available to us throughout our debt structure in public bond, bank and private debt markets. In addition, there are other financing options through customers that may bring their own chips to be installed in our data centers and suppliers who may lease their chips rather than sell them. Both of these options enable Oracle to synchronize our payments with our receipts and borrow substantially less than those people are modeling. As a foundational principle we expect and are committed to maintaining our investment-grade debt rating.

Turning to guidance. Let me start with the impact of the added RPO that occurred in Q2 on our future results. The vast majority of these bookings relate to opportunities where we have near-term capacity available, which means we can convert the added backlog to revenue sooner. The result is we now expect $4 billion of additional revenue in FY '27. Our full year FY '26 revenue expectation of $67 billion remains unchanged. However, given the added RPO this quarter that can be monetized quickly starting next year. We now expect fiscal 2026 CapEx will be about $15 billion higher than we forecasted after Q1. Finally, we are confident that our customer backlog is at a healthy level, and then we have the operational and financial strength to execute successfully.

While we continue to experience significant and unprecedented demand for our cloud services, we will pursue further business expansion only when it meets our profitability requirements and the capital is available on favorable terms. As it relates to specific guidance for Q3. Total cloud revenue is expected to grow from 37% to 41% in constant currency and is expected to grow from 40% to 44% in USD. Total revenues are expected to grow from 16% to 18% in constant currency and are expected to grow from 19% to 21% in USD. Non-GAAP EPS is expected to grow between 12% to 14% and be between $1.64 and $1.68 in constant currency and grow between 16% and to 18% and be between $1.70 and $1.74 in USD. And with that, I'll turn it over to Clay.

Thank you, Doug. Our infrastructure business has grown at an accelerating 66% year-over-year. You are well aware of the strong demand for AI infrastructure, but multiple segments across OCI are also contributing to this accelerating growth rate, including cloud native, dedicated regions and multi-cloud. Our diversity of capabilities within infrastructure differentiates us from AI infrastructure NeoClouds.

Our unique combination of infrastructure and applications differentiates us from other hyperscalers -- we have ambitious achievable goals for capacity delivery worldwide. OCI now operates 147 live customer-facing regions with 64 more regions planned. In the last quarter, we handed over close to 400 megawatts of data center capacity to our customers. We also delivered 50% more GPU capacity this quarter than Q1. Our super cluster in Abilene, Texas is on track with more than 96,000 NVIDIA Grace Blackwell G200 delivered. We also began delivering AMD MI-355 capacity to customers this quarter.

Our pace of capacity delivery continues to accelerate. We continue to see strong demand for AI infrastructure across training and inferencing. We follow a very rigorous process before accepting customer contracts. This process ensures that we have all the necessary ingredients to deliver to customer success at margins that make sense for our business.

We analyze land and power for data center buildings component supply, including GPUs, network gear and optics, labor costs for all phases of construction and low voltage work, engineering capacity to design, build and operate revenue and profitability and capital investments required. Only when all these components come together, do we accept customer contracts, having the confidence we can deliver on schedule with the highest quality. As Doug said, this quarter, we contracted for an additional $68 billion of RPO. These contracts will quickly add revenue and margin to our infrastructure business.

We continue to carefully evaluate all future infrastructure investments investing only when we have alignment across all necessary components to ensure profitable delivery for our customers. The holiday season is peak period for many of our retail and consumer customers. It is our responsibility at OCI deliver the most secure, highest performance and highest availability infrastructure to support these customers at the scale they need. Uber has now surpassed 3 million cores on OCI and powering their highest traffic ever this Halloween. Temu scaled to nearly 1 million cores for Black Friday and Cyber Monday. In addition, we also supported thousands of other customers across our retail and other applications through their largest and most successful holiday period yet.

OCI is constantly expanding in features and services. We recently launched Acceleron, delivering enhanced networking to all OCI customers and other services like our AI agent service. However, we cannot deliver everything ourselves, and we rely on our rapidly expanding partner community to provide the best experience on OCI. We added new AI models from Google, OpenAI and AI to ensure our customers have the latest and greatest AI capabilities. Our marketplace consumption has grown 89% year-over-year, powered by partners like Broadcom and Palo Alto. Those same partners drive OCI consumption by building their SaaS businesses on OCI. TaloAlto released their SaaS and Prisma Access platform on OCI and Cyber Region and new fold digital continue to scale their businesses rapidly. These partnerships make our ecosystem richer, which helps our customers, and that growth translates directly into more OCI growth as our partners build their solutions on our infrastructure.

Oracle database services see increasing demand across all clouds. Multi-cloud database consumption has increased 817% year-over-year. We launched 11 multi-cloud regions this quarter, bringing us to 45 reasons live across AWS, Azure and GCP with 27 more planned over the next months. We see increasing customer demand with billions in identified pipeline.

We launched 2 important programs this quarter for multi-cloud. The first is multi-cloud universal credits, which enables customers to commit once to Oracle database services and use them anywhere in any cloud with the same price and flexibility. The second is our multi-cloud channel reseller program. which enables customers to procure Oracle database services through their preferred channel partners. We also launched 9 services across the different clouds, such as Oracle autonomous AI lake caps.

This combination of the best services, universal availability, consistent and easy pricing and procurement and partner support is accelerating the adoption of Oracle database services across our entire customer base. OCI is the only full cloud available to individual customers. We launched Dedicated Reason 25, which provides the full capability of OCI in a tiny 3 Rack footprint. OCI is also the only cloud that enables partners to become cloud providers themselves through our alloy program, and our new footprint is available for all alloy providers. Dedicated region and alloy consumption grew 69% year-over-year. We launched 1 dedicated region for IskaGroup in Oman, and both NTT Data and SoftBank launched on alloy region each this quarter. This brings our live dedicated regions to 39 with 25 more planned.

In summary, the 4 segments of our infrastructure business are growing at incredible rates. This will contribute to a continued acceleration in our infrastructure revenue in the coming quarters. Customers are choosing OCI for its performance optimized architecture unrelenting focus on security, consistently low and predictable pricing and unmatched depth in database and enterprise integrations. Those priorities have been our strategy from the beginning and are the driving force behind this growth. OCI is a cost state of reinvention, and you can see the value that it has for our customers. When you combine our commitment to deliver the best in performance, efficiency and security with the growing customer demand for cloud infrastructure services we are seeing, I couldn't be more excited about what's coming next. And with that, over to Larry.

Thank you very much, Clay. Over the years, Oracle has developed software in 3 important areas: database applications and the Oracle Cloud. We used AI to make our database software and our Autonomous software eliminates human labor and human error. Thus lowering operating costs and making our systems faster, more reliable and more secure. Now with the development of the Oracle AI database, and the Oracle AI data platform, we're bringing all 3 layers of our software stacked together to solve another very important problem. Enabling the latest and most powerful AI models to do multistep reasoning on all your private enterprise data while keeping that data private and secure.

Training AI models on public data is the largest, fastest-growing business in history. AI models reasoning on private data will be an even larger and more valuable business. Oracle databases contain most of the world's high-value private data. Oracle applications also hold huge amounts of exceptionally valuable private data. The Oracle Cloud includes all the top AI models, OpenAI chat GPT, XAI rock, Google Gemini and Metalma Oracle's new database and AI data platform, but the latest versions of Oracle applications enables all of those AI models to do multistep reasoning on your database and application data while keeping that data private and secure.

All our database and application customers want to do this. because for the first time, they get a unified view of all of their data. AI models can respond to a single inquiry, by resonating across all your databases, all of your applications. By treating all of your data holistically, the combination of AI models plus the Oracle AI database and data platform, breaks down the walls that isolate and fragment your data. The Oracle AI data platform makes all your data, all your data accessible to AI models, not just Oracle databases and Oracle applications the data in the Oracle databases and Oracle applications, but data from other databases, cloud storage from any cloud, even data from your own custom applications. are accessible to AI models using the Oracle AI data platform.

Using our AI data platform, you can unify all your data and reason on all of your data using the very latest AI models. This is the key to finally unlocking all the value in all your data. Very soon through the lens of AI you will be able to see everything happening in your business as it happens. Mike, over to you.

Thanks, Larry. As Doug shared, the total revenue growth of 13% in constant dollars. I think it's worth noting that consecutive quarters now where total revenue growth has been in double digits. So a solid quarter and we see even better days ahead of us. So let me break down a few of the numbers a bit further, and I'll do so all in constant currency. Cloud applications revenue was up 11%, and that brings us to about a $16 billion annualized run rate. Within that, Fusion ERP up 17%. Fusion SCM, up 18%. Fusion HCM, up 14%. NetSuite grew by 13%. Fusion CX is up 12%. And in our industry cloud, specifically hospitality, construction, retail, banking, restaurants, local governments and communications, all combined were up 21% in the quarter.

So big growth rates on a big base. In our health care business, we now have 274 customers live in production on our clinical AI agent, and that number continues to rise daily. Also in health care, our brand-new Ali-based ambulatory EHR is generally available and that has received U.S. regulatory approval. And finally, in health care, in Q3, we expect both our bookings and our revenue to accelerate materially.

So overall, Cloud apps were up 11% on a bigger base. This is very meaningful because we see this business as continuing to accelerate going forward. We achieved this growth while we also undertook a major sales force reorg in many regions throughout the world. And this is something we've been talking about for many years, and that is the synergies between our back-office applications and our industry applications.

We're seeing more and more deals where our industry apps are pulling fusion or the fusion apps or pulling the industry apps. And as a result of seeing more and more deals, we're also seeing larger deals with more components. So recently, we combined our industry application sales team and our Fusion sales teams into a single selling organization. This enables our sellers to have more strategic 1 Oracle conversations with our customers to sell higher and to sell more.

And then when you think about our very large installed base of on-premise application customers, these strategic conversations are driving upgrades. And you've heard us say before, just moving a customer to the cloud results in a 3 to 5x annual revenue lift compared to support revenue. Now on top of all of that, top of the sales force, the 1 Oracle go-to-market water go-to-market motions, the industry suites think about combining our AI data platform with this unmatched suite of applications.

This creates an incredibly unique opportunity for our customers to gain value very quickly from enterprise-grade AI. The combination allows customers to unite the industry-leading foundational models with company-specific proprietary data, as Larry mentioned, much of which comes from the Oracle applications. And of course, the AIA data platform also ties in non-Oracle applications, competitive data sources like MongoDB or snowlink, object stores, and even complete bespoke unstructured data.

So we think this allows our customers to very easily build enterprise, AI agents and applications that leverage built in, not bolted on, but built in AI to transform their business. So just to repeat, AI, of course, is a great CI play, but it's also a broader software play for Oracle. It's driving growth in our applications, in our database businesses as well. Let me highlight a few key wins. In communications, the communications industry, digital and Bridge Holdings, selected ERP and SCM at the hot Salom Telecom SCM. Motor oil solutions, ERP, SCM and CX. Team Brazil, who is the country of Brazil is 5G leader just signed a new 5-year expansion to accelerate AI adoption and to transform customer experience at scale, all built on OCI. This actually -- this 5-year deal was an expansion or extension of a partnership that began with Tim Brazil's full data center migration to OCI back in 2021.

So they're now building AI agents that are powering real customer interactions, including the Conta agent which compares bills for customers across months and explains variations automatically already in the pilot, 18% faster issue resolution with the AI agents built on OCI. -- expecting even further improvements as the rollout continues. They've got 24 projects in motion, 7 of them are already in production. It's just a matter of months, 6 more launching soon all enabled by a multi-cloud architecture with Oracle as a key AI infrastructure partner. Already as a result of this initial rollout, customer satisfaction is up 16%, and call center flows are managing end-to-end with 90% accuracy, resulting in 30% faster services times for customers and a 15% fewer network failure interventions, all using predictive analytics and again, AI copilots still and OCI.

So we are the AI engine for Tim Brazil to enable personalization at national scale. In Financial Services, CoreCivic selected ERP, SCM and HCM. And Prime Life Technologies, ERP, Jewelers Mutual Insurance, ERP. In public sector, City of Costa Mesa, ERP, SCM, HCM, the United States Space Force, ERP and HCM and the city of Santa Ana, ERP, SCM and HCM. In the high-tech industry, SolarEdge Technologies, ERP and SCM, Zscaler, ERP, Dropbox, ERP and SEM. I could go on and on about these wins, but I think it gives you the flavor of just a sheer amount of multi-pillar wins and why it's so important that we have so many different options for customers and back office, front office.

In terms of go-lives, we had 330 cloud apps customer go-lives this quarter. That's multiple go-lives per day. Virgin Atlantic Airways went live in September on Fusion ERP and HCM and payroll. Broadridge just recently relaunched our Fusion ERP and EPM go-live. LifePoint Health just went live on their third wave of Fusion ERP, and HCM and Saudi Telecom is live on Fusion SCM, with ERP and HCM to follow soon. DocuSign is now live on Fusion data intelligence.

Again, I could go on about the go lives with 330 of them in the quarter. Gives you a flavor of some of the really important go-lives that we had recently. The cloud application deferred revenue is up 14%. That is higher than the Cloud revenue growth of 11%, just to reinforce my earlier statements that we expect continued apps growth acceleration.

So a solid quarter across the board. We're on the back end of a sales reorg that was focused on unified selling across our applications portfolio. We're seeing a clear AI halo effect for our cloud applications, which is driving upgrades, our AI data platform, combined with our applications is an absolute conversation changer, and it brings the Oracle database and all of our applications into the center of the modern agentic enterprise. Looking ahead, we're executing well on a big and growing pipeline, and I expect revenue and earnings growth to accelerate off an even larger base. With that, Ken, I'll turn it back to you.

Thank you, Mike. Chip, if you could please poll the audience for questions. .

[Operator Instructions] Your first question comes from the line of Brad Zelnick with Deutsche Bank.

Congratulations in a particular shout out to Mahesh and the team for standing up a massive amount of capacity in this quarter. My question is for Clay or maybe Doug. Oracle is clearly the destination of choice for the most sophisticated AI customers, but this is a far more capital-intensive proposition unlike any business Oracle has ever been in before. Very specifically, how much money does Oracle need to raise to fund its AI growth plans ahead.

Thanks for the question, Brad. This is Clay. So look, I'll answer that question in 2 parts. First, let me give you kind of the reason why it's hard to answer that question exactly. So -- the thing I think that a lot of people don't understand is that we actually have a lot of different options for how we go about delivering this capacity to customers. There's obviously the way that people think about it, which is we buy all the hardware upfront. And as we -- as I talked about it in my financial analyst meeting, we don't actually incur any expenses for these large data centers until they're actually operational.

So then it goes on to, well, how do you pay and what's the kind of cash flows look like for the stuff that goes into the data center. Well, we have some other interesting models that we've been working on. One of them is that customers can actually bring their own chips. And in those models, Oracle obviously doesn't have to incur any capital expenditures upfront for that model.

Similarly, we have different models that we're working on with different vendors, where some vendors are actually very interested in the model where they rent their capacity rather than selling that capacity. And as you can imagine, that comes with different cash flow impacts that are favorable and reduce the overall borrowing needs and capital required for Oracle.

So as you can imagine, as we look at all of these kind of commitments, we will use a in a variety of those such that we minimize the overall cost of capital. as well as, in certain cases, we'll be raising our own funds. As part of that, I think it's important that everyone understand that we're committed to maintaining our investment-grade debt rating.

So now to give you some more specifics. What I would say is we've been reading a lot of analyst reports, and we've read quite a few that show an expectation of upwards of $100 million for Oracle to go out and kind of complete these build-outs. And based on what we see right now, we expect we will need less, if not substantially less money raised than that amount to go and fund this build out. So hopefully, that helps answer your question, Brad.

our next question comes from the line of Ben Reitzes with Melius Research.

Appreciate it. Good to speak with you. In light of the answer to that question, the path for OCI margins seems very important to improving the EBITDA and cash flow. So at the Analyst Meeting, you said margins for AI workloads for OCI would be in the 30% to 40% range over the life of a customer contract. I guess my question is how long will it take your AI margins across all your OCI data centers to ramp to that level? And what needs to happen to get there?

Yes. Thanks for the question, Ben. Look, the answer is it really depends. So the good thing is that as I mentioned earlier, we don't actually incur any expenses for the data centers until they're actually built up and running. And then we highly optimized the process by which we actually put capacity in and then are able to hand that over to customers, which means that the period of time where we're incurring expenses without that kind of revenue and the gross margin profile that we talked about is really on the order of a couple of months.

So in that scenario, that time period is not material. So a couple of months is not a long time. The -- what actually matters much more is the overall mix of the data centers that we have online, right, and how they're growing compared to the total amount that we're scaling across the world. And so I think as we go through this buildup phase, right now, we're in a phase of very rapid build-out without the majority of the capacity online, obviously, the aggregate mix is going to be lower. But as we actually get the majority of this capacity online, and that's really our focus, the best way to improve margins quickly is to actually go out and deliver capacity faster. That ends up very rapidly ensuring that we get to that 30% to 40% gross margin profile for all AI data centers.

Your next question comes from the line of Tyler Radke with Citi.

And this question is for Larry or Clay. So Oracle has clearly established itself as a leading provider for AI infrastructure to AI labs and in some cases, enterprise clients. How are you thinking about the opportunity to sell additional platform services such as database, middleware, other pieces of the portfolio? Similar to how we saw cloud providers add that on at the early days of the public cloud space? And what might be some of the similarities or differences that you see with the emerging AI platform as a service market versus the traditional cloud platform as a service market.

Well let me start with traditional cloud and traditional Oracle database. So I think the biggest change we've made there was to make our database available in everyone's cloud. So you can buy the world database at Google or Amazon. -- it's available at Microsoft Azure as well as OCI. So that was the most -- maybe that was the first move we made. We call it multi-cloud, and we actually embed OCI data centers within the other clouds. So they get the latest, greatest version of the Oracle Database.

Second thing we did is we actually converted the Oracle database or added capabilities to the Oracle Database. -- to allow you to vectorize all of your data. So it's a vector database. Some people call that an AI database. So it's designed to make data available to models. You can then -- once you vectorize your data, you can place AI models on top of that. and the AI models can understand what's in the database and reason with the data that's in the database.

So we think that combination of making our -- the data available on our database also accessible by AI models dramatically increases the value of the data. We think that's very -- so far, none of the other large-scale databases have been able to do that. We can do that not only we can do that and keep your data secure. That's 1 of the bigger issues. We have to scale it, keep everything reliable, keep it secure -- and we actually had all of those capabilities and features to the Oracle database.

So that was step 2. First, multi-cloud, second vectorize all the data and make it accessible by all of the popular AI models. Third step. Well, it's great that we're making the Oracle data, Oracle database data available to these AI models. The companies actually have data that's not stored in an Oracle database is not stored in an Oracle application. So we built an AI lake house, we call the AI data platform that actually points to and vectorizes all of your data, whether it's in an object store in different clouds, whether it's a bespoke application whether it's in another database. It really will take the universe of your data, that catalog that data, vectorize it and allow an LLM to do multi-step reasoning on all of that data.

Now the thing that's really remarkable about that is, think about asking 1 query, asking 1 question. And the model looks at all of your data. Normally, when you ask a question, you have to direct it to this database or this application. You can't say, look, I'd just like to know who's next customer I should be selling to. I'm a salesperson in territory. I'm looking -- I want to look at all the accounts in my territory. And I want to see who's the best prospect in my territory. Well, that means looking at contractual data, means looking at publicly available data. That means looking at our sales system, at our support system, all of these separate systems.

Well, suddenly, all of that data is unified. We take all of your data and unify it, so you can ask a single question and the AI models can find the answer to that question regardless of what data store is in. that's really a unique proposition, and we think that thing is going to turbocharge the use of our database and the use of our cloud dramatically.

Your next question comes from the line of Brent Field with Jefferies & Company.

Question for Larry and Clay, on the fungibility of your infrastructure. What would you have to do the converted data center from 1 customer or another, such as if 1 of the larger customers was unable to pay?

Yes. Thanks, Brent. So I think the first thing to understand is that what we deliver for our AI infrastructure is exactly the same cloud that we deliver for all of our customers. We made specific choices at the beginning of OCI around bare metal virtualization and the way in which we do things like secure wipe of hardware, et cetera. But the reason I bring that up take as an example, anyone right now with a credit card to show up and sign up for any 1 of those hundreds of regions that I talked about earlier, and you can spend up a bare metal computer and as quickly as a few minutes. And at the end of that, you can turn it off, and I will recycle that, and I can hand it to the customer in less than an hour. And so when you ask the question of, well, how long does it take transfer capacity from 1 customer to another, it's on the order of hours. And what actually I think is kind of a corollary to that question is, well, how long does it take customers to adopt it.

And thankfully, we have a lot of experience getting more than 700 AI customers on our platform, including the vast majority of the large model providers already run on OCI and when we give them capacity, they typically spend that capacity up in the order of 2 to 3 days. And so when I think about how long does it take for me to take capacity and hand it to the customer, it is not a laborious process. It is not a unique process. And something else I would say that I think a lot of people don't realize about our cloud is this is actually happening all the time.

So we have lots of customers that might sign up for a few thousand of 1 type of GPU and then they'll come back and say, well, actually, I'd like to get even more capacity somewhere else? Well, you take this back. And we do that all day every day, and we're constantly moving customers around and adding aggregate net capacity.

So we have the technology. We have a secure base for which to do that, and we also have a customer base of a lot of demand such that whenever we find ourselves with capacity that's not being used, it very quickly gets allocated and provisioned.

Your next question comes from the line of Mark Moerdler with Sanford Bernstein.

Congrats on the quarter. Doug, you gave some color earlier tonight that I'd like to dig in on. Clay presented a slide at the Financial Analyst Meeting, where he showed the revenue and expenses for a single data center. Doug and Clay, can you talk about the cash flow for that same data center, starting with the commitment for the data center and then the hardware and how that flows into becoming cash flow positive? And then how that rolls up across multiple data centers? Any color would be really appreciated.

Sure. Happy to answer, Mark. So as we talked about earlier in this call, -- it starts with the actual data center itself and the power capacity along with it. And the way in which we structure that is that we incur no cash expenses until that's fully delivered and provisioned and fit for purpose. So then it really comes down to what does the cash flow look like for the capacity that's going inside the data center. And as I talked about earlier, that really depends on kind of the business model and the financial model that we've used for procuring that.

In some cases, customers actually want to bring their own hardware. In which case, we don't have any capital expense, and it's really around the data center itself, possibly some networking gear as well as human labor costs. We have other models where vendors want to rent that capacity in which case, those rent payments start when the capacity is provision for the customer. And so customer cash flow comes in, we're then taking that cash flow and pushing that out to all of the different suppliers.

And then obviously, you've got the model where Oracle takes its own cash, pays upfront for the hardware and then put the capacity in. That's obviously the most cash-intensive upfront and then there's depreciation schedule over the next several years. So it really depends on the exact business and financial model used for each of the data centers.

And then you ask, well, how do they layer together? Well, thankfully. They're not -- we don't need calculus for this one. Basic arithmetic is enough because they actually just layer on top of each other. So if you have a time schedule for 1 data center and a time schedule for a second data center, the cash flows add together. And obviously, if a data center comes in sooner, you'll have the expenses as well as revenue coming in sooner if a data center moves out and the expenses and the revenue also move out.

Your final question comes from Don DiFucci with Guggenheim Securities.

Okay. I won't make you kiss my ring. It's actually John DiFucci. Anyway -- I'm sorry, listen, a lot of my questions on infrastructure. We've already been asked, and they're really important questions because that's a big part -- that's your big part of your growth. I have a question on the Applications business. Mike, you said applications are going to accelerate this year? Why the confidence in this business when all your large SaaS peers are seeing just the opposite where growth is decelerating, especially because we thought that something similar about Oracle's application business last year, we didn't really start to see it until the fourth quarter.

We've heard some things in the field around 1 Oracle, your go-to-market motion where apps and infrastructure are more combined versus separate. And you also talked about combining vertical and horizontal labs teams here. Is that it? Is it mostly go-to-market? Is there something more about the products or something else that we should be thinking about?

Well, yes, thanks for the question, Jeff. First, I think it's a combination of things. But let me just start with what I think is happening in the industry. All of our competitors are largely in the best-of-breed business because they're not in the applications business in totality. They're not in the back office business. They're not in the industry business and they're not in everything in between. They're not in the front of the house, the back of the house in the middle of the house, -- we are the only applications company in the world that's selling complete application suites. Then you add in baked in AI, the AI halo baked in AI right into regulatory applications. So we're over 400 AI features live infusion already.

I mentioned 274 customers live on our clinical AI agent, the go-lives for these clinical AI agents, which are a new SaaS applications for us, are measured in a matter of weeks. So you're looking at an industry like health care, where it would take months or years to get anything done with that magnitude. We -- and by the way, John, the customers are implementing these things all by themselves. They don't need us to help them. You just roll them out and they work. So we're in the applications industry suite business. We're building AI into our back-office applications, our front-of-clas -- and we are building applications that are also AI agents themselves. They are, in fact, IH. So that's why you see the growth rates in our industry applications of 21%. And in the quarter.

Our Fusion ERP is up 17%. SCM up 18%, HCM up 14%; CX 12%. Again, all on a bigger base. Then I think the next piece of this is you're adding the AI data platform. So if you'd like an industry suite of applications, then you'd like to create your own AI agents. You like to create and unlock your own enterprise data on top of it all. We are the only -- all of those ingredients for a customer. And I think as you look at customers of spend on best of brew because the integration costs are so high, and it's hard to bolt AI on all that because you're actually not retiring anything in the process. We're in a very unique position. And I think we're starting to see the numbers too, John, with the deferred revenue for apps growing at 14% now faster than the in-quarter revenue growth of 11%. For all those reasons, I'm optimistic on our apps business going forward. It's a continued growth engine for Oracle.

As you were talking, you sort of clarified in my mind, when I think of an Oracle, I used to think -- I thought of it when I started to hear about it as a go-to-market motion, but it's more than that. It's actually -- it's a product thing, too. It's everything.

Thank you, John.

A telephonic replay of this conference call will be available for 24 hours on our Investor Relations website. Thank you for joining us today. And with that, I'll now turn the call back to Tiffany for closing.

Ladies and gentlemen, this concludes today's call. Thank you all for joining. You may now disconnect.

I'm just a little blown away how big this has all gotten here. I feel like I'm way high this time. If those of you here last year, recall that it was definitely a small room, and it's much bigger this time. Our attendance in this event right here is up 70%. The show itself was way up. And I just have a few slides because I know that as much as you like hearing from me, there's more interesting things coming here.

So you've seen it everywhere. AI changes everything. And it's true. It's in the signage. It's in the presentations. It's everywhere, except for this. So kidding aside, I'm not going to read out the whole slide, but it's important to remember here that we will be making forward-looking statements.

These statements do come with material risks and uncertainties. So please refer to our filings 10-K and 10-Q with the SEC for more information about those. You'll also see us using non-GAAP financial measures. A couple of the presenters today will be using those. So just as a reminder, that they're intended to be used in conjunction with GAAP measures, and those can be found in our financial statements, of course.

And then a number of our execs will be talking about -- for informational purposes about where we're going from a technology and service standpoint. And just a reminder, look, it's not a commitment to deliver any material code or functionality that were being discussed today.

So you're going to see these slides, each of the presenters will have some or all of these slides. They will just make reference that, as Ken mentioned earlier, here's the slides Ken referred to, this is what they're talking about are these 3 slides. So let's talk about the agenda.

We're going to start with Clay in a moment here. And then he will obviously be talking about OCI and Oracle Infrastructure; followed by Mike, who will take a discussion first initially on the application business and then Mike will then continue on with a deeper dive into Oracle Health and our embedded finance businesses. That will take us to lunch, at which point, Larry will take the stage, he'll be talking about AI and database and then Doug will come up with the financial outlook, and then we'll do a Q&A. Larry, Mike and Clay do Q&A. And then so we should end up somewhere around 2:00 this afternoon.

So with that in mind, let me get out of the way and let's bring up Clay.

Thank you, Ken. Ken still has 1 minute and 37 seconds, which I was really depending on. So okay, now my timers -- okay, now I'm late. I don't know what to do. Okay, thank you all for coming. I really appreciate it.

To give you just kind of a quick overview of how I'm going to go. I'm going to talk for what is going to feel like far too long. We'll then have a short section where Mark Hura will come up and talk for another section of what feels like quite some time and then I'll have a bit more details on some financials, and then we'll hand it over to Mike.

So we've got a few minutes. At the end of this, there's a few things that I want you to walk away with. I want you to understand the overall growth trajectory of the OCI business. Obviously, we've got myself, we've got Mike, we've got Seema, we've got a lot of people, Larry's coming.

We're talking about a huge number of our businesses. I'm going to focus right now on the cloud infrastructure business. I'll be talking about the different segments that make up that business. And when I talk about these segments, obviously, they are -- the ways in which we think about them, about how they grow, about who the customers are, about the products that they want, how they consume. So we'll explain that to you.

I'll explain what's driving the growth in each of these segments and also why those customers are choosing to invest more and more with Oracle. The main goal is at the end of this, I think you'll understand why we're so excited about where OCI is at and the great growth prospects that come ahead for OCI. Okay. So these are the slides that Ken talked about. I have to say that you can read them. Those are the slides, and one more, and one more. Now we're -- okay, now we're done.

So I mentioned different segments. So this very simple equation of enterprise plus distributed cloud plus cloud natives, plus AI infrastructure equals hyper growth. And I'll go into each of these segments specifically, and I'll talk about why we break our business up into those segments.

The thing to get at a high level, though, is that all of these actually today have very good growth rates, but they're actually accelerating, right? And a big part of the reason why these different segments are all accelerating is because they're symbiotic. And I'll try to take a minute here to explain what I mean by that.

Let's say that you have a customer that shows up, and they're an enterprise customer. The traditional customers that you think of when you think of Oracle, right? They have a long history with us. They're a database customer. They also have some of our applications. They might move those things to our cloud.

As they do that, one of the things that they want is they want obviously the great functionality that we provide at Oracle. They also want other ISVs. Maybe they want some security products. Maybe they want some new advanced networking features. Well, as Oracle, we don't provide all of that, but we have a great partnerships.

I'll talk about some of those and maybe you have an ISV that shows up. And well, they're much more of a cloud-native company. And those -- the demand from those enterprise customers brings those ISVs to our platform, right? And those ISVs join, and they can drive more and more overall compute and storage and networking consumption growth, and they're happy because they provide services that our enterprise customers consume.

The same thing is true for our distributed cloud customers. The fact that we actually offer not just the public cloud that we have today, but the fact that we go out and we offer our dedicated regions in our alloy business, that actually becomes an accelerator because our ISVs want that reach. So the fact that we have individual customers, say, in Japan that are building their own clouds that they operate and sell to their customers, our -- that also drives things like cloud natives.

When our AI infrastructure customers show up, they typically start by consuming raw infrastructure, but they quickly move on up the stack. They start consuming more of our compute storage and networking products. They also then move on to start doing things like consuming our applications. So as you look through these segments, we're experiencing this accelerating growth across all of them because as each segment on its own starts to grow, it actually helps the others grow faster.

Okay. So let's take a look at the enterprise segment. So I think you already understand what we call enterprises. Here's a great set of logos across the screen. But as I said, traditional companies, most companies have some Oracle, whether it be from database, our applications, our middleware, our industry applications, those customers are very pleased when they run on top of OCI.

So you see the current growth rate, right, 33% year-over-year growth for this segment from Q1 of FY '25 to Q1 of FY '26. However, contrast that with more than 1,500% growth rate in our multi-cloud database business. So the way to think about this is, up until very recently, you could only get the best of our data platform on one cloud, which was OCI. And we like that cloud, we think it's a great cloud. But we actually have extremely popular data platform services.

As of today, you can now get that data platform on all of the cloud providers. And that's why we see this very rapidly growing, right, multi-cloud database business. Okay. A bit more about why there was a margin range here. So when I talk about 65% to 80% as a range for gross margin. The reason for that is actually largely due to the mixture of services that those customers are consuming.

So as you can imagine, some customers end up being very, say, networking heavy. And networking, we're -- we think our networking is quite good, and we have a lot of IP in it, but not nearly as much IP in our networking stack as we do in, say, our database services. And so obviously, based on the differentiation we have in our services, we price them at different margin profiles and so there's a range of this based on the workloads that a customer brings to us.

And then I briefly want to talk about when I say contract to scale. What I mean by this is, there's a process that enterprises go through when choosing a cloud. Typically, they do a POC. They try things out for a while. They then move forward. There's a contracting process. And then once that's done, there's also now an implementation phase, right?

It takes -- people don't move their most mission-critical workloads overnight, right? And so as we see -- the reason I bring up this contract to scale is that we see it move -- the process of moving from initial product launch over to pipeline and then actually to committed contracts and then showing up as revenue, that process takes a certain amount of time for these types of customers, given the criticality of the workloads that they're bringing to OCI.

Okay. So a little bit about why they're choosing OCI. Well, the most obvious reason is because it's a combination of our ecosystem, right? We have cloud infrastructure, but we also have the world's best database with Oracle, autonomous AI database. We also have amazing applications, both horizontal and vertical, and we'll hear more about that in a few minutes.

But even beyond Oracle workloads, OCI is the best platform for enterprise workloads. We built OCI to make it very seamless for you to take an existing enterprise application and move that to our platform. That's very different than building a platform that was designed only for new applications to be written in the cloud.

We've been talking a lot about AI this week. We've been talking a lot about the AI data platform that brings together the great work we're doing with our gen AI models, the work that we're doing with our new AI database and the work that we're doing with our new agent service, all packaged together as part of our AI data platform.

That being available in all of the clouds makes it very easy for customers to pick Oracle as a place to put their data. And then, of course, enterprises do care about performance and they care about price and OCI is by far the best performance at the lowest price.

Okay, so here's an example of a specific customer, NASDAQ. A few years ago, they actually moved CapCloud over to OCI and runs exclusively on there. And obviously, far before that, they've been a database customer for a long time. But then more recently, they adopted Oracle database at AWS, which allows them to then bring their RegCloud to their different environments and be able to rely on X data across all of their different cloud ecosystem.

Okay, so let's now take a minute to talk about our distributed cloud. So when I talk about distributed cloud here, I really mean our dedicated regions and our alloys. So dedicated region is, you can get the entire OCI environment as of yesterday in just 3 racks and you can put that in your own data center.

A great example for that is someone like a Vodafone who bought 6 of our dedicated regions to take our full cloud experience and run it right next to their network on-premise. The other part of this business that we talk about is our alloy. So as an example, take someone like an NRI that bought dedicated region to begin with, but they're a technology company in Japan. They take our cloud, they add to it their differentiated IP, and they actually can go out and operate in a sovereign way and provide that to their customers in Japan.

So as you can see, it's a very good growth rate, right, growing 77% year-over-year already with an average deal size of $67 million. Now here, right, gross margin is in a range between 40% and 60%. And it's not because these things are just fundamentally less profitable, it goes back to what I said about the enterprise section, which is it's all based on the mix of the services that those people are consuming.

So when you have, say, a dedicated region that's almost exclusively used for database, the margin ends up being quite high. But a lot of our dedicated region and alloy customers, they're consuming a lot of essential infrastructure services, and those have a lower margin profile in general. And then when I talk here about how there's 60-plus dedicated and alloy regions globally, we have a very large pipeline of this.

And look, when someone shows up and they want a dedicated region, there's a time period where -- there's a time for us to get it to them, there's a time for them to put in the data center, there's a time to ramp it up. But what we've seen is we're planting a lot of these seeds, and as you can see here from this growth rate, this part of our business is growing very rapidly, and the customers keep coming back for more.

Okay. Why are these people picking our distributed cloud? It's really quite simple. Rather than talking about why our cloud solves their -- technologically, their needs are simple. They're typically in a regulated industry. They need sovereign control or they have things that they either don't or can't move to the public cloud and they want the benefits of the cloud right next to it.

Our distributed cloud offering is actually the only offering that solves these problems, right? If you go to anywhere else, they have a subset of the cloud, you have to pay a lot of money upfront. With us, you can get it in a very small form factor. You sign an universal credits commitment that allows you flexibility on what services you're going to consume.

We can deploy, like I said, in a small footprint very, very rapidly, and no one else offers a turnkey solution. The fact that, as Oracle, we both are in infrastructure and applications. That's something no one else does. And what that actually enables us to do with alloy is we could not build alloy and enable someone else to actually operate and run a cloud, if we didn't have all these great work that Steve and his team had done with Fusion.

So when we talk about alloys, it's not just OCI, it's OCI plus HCM plus our ERP system, plus our coding system plus service cloud. That way, we can actually provide a turnkey solution. And then, of course, another part of the reason why people like our dedicated regions, and you'll hear Mark talk about this with, I think, some customer examples later, is that when they buy our dedicated region or they buy alloy, it doesn't just our infrastructure. You can get a dedicated region and then you can also run our differentiated applications, both horizontal and vertical apps on that same infrastructure wherever you need it.

Okay. So here's an example of e&, previously known as Etisalat. So they started a few years ago with a dedicated region. They then expanded to have a second dedicated region for disaster recovery. They then expanded by adding GPUs to those dedicated regions. And then most recently, they expanded again by contracting for an alloy so they can serve sovereign workloads in the UAE.

That's an example of how these customers start small, right? We make them extremely happy and then they continue to grow and scale with us. And as we continue to add more and more of these customers, like I said, this business starts small, but the customers are seeing so much value, they all keep growing and growing and growing.

Okay. I promise you I'm not that bad at clicking. All right, cloud natives. So when we talk about cloud natives, the way I think about this is these are typically workloads and customers that have relatively few number of applications that consume a lot of infrastructure, right? ISVs typically fall into this business, but companies that have workloads that are very focused on, how do you rapidly scale up and scale down and take advantage of all of the benefits of the cloud.

So this segment of our business is growing very rapidly as well, with 49% year-over-year growth from Q1 FY '25 to Q1 FY '26 with an average deal size of almost $100 million, right? Again, gross margin range of 40% to 60%. A large part of the reason for this comes down to both the service mix that they choose, but also the size and the scale of their business.

So as you can imagine, this group can range from relatively big fish to very, very, very large fish, what you might call whales. And as part of that, with that size expectation, you come different expectations around overall pricing. One thing to think about here, though, is that these customers tend to ramp faster.

Typically, these customers might have some of their own on-premise, they have a platform that they can move very quickly. Zoom was an example. I don't know if anyone remembers, but COVID hit and Zoom actually moved from their existing infrastructure as well as in other clouds, they were up and running on OCI in 9 days. And so these are applications that you can ramp very quickly when there's motivation to do so. And so when these customers find out about us, they tend to be very large and they scale very fast.

Okay. So why are these customers choosing OCI? Well, primarily, it's about best price and performance. And the reason it matters so much to these customers is that typically, if you were to take their IT spend, typically for, say, smaller customers, the largest expense for them is labor costs. And their overall infrastructure spend tends to be a tiny portion.

For these types of customers, by far and away, the biggest expense for them is the actual infrastructure. And that's why having the best price performance and being the most secure matters to them. We also work with these customers through a combination of engineering partnerships. They -- and Mark will explain a bit more about this in a minute.

It's not enough that we just say, "Hey, here's what's on the truck." All of them, given their scale and their unique business, they might have some features they need, they might have some customizations they want of different hardware we haven't considered before. And the fact that as Oracle, we can actually go work with them, figure out how to implement all of that in a standardized way, right? Because what's important when we do this is that we don't fork our cloud, where there's like an UberCloud, which is different than a TikTok cloud, which is different than an OpenAI cloud.

We need to have one cloud that everyone can use, and it's our job to go in and put those features and functionality into the base product. So there's really important reasons why those customers are choosing us. It's really about those deep engineering relationships.

Okay. So an example of one of these companies is Cybereason. As I mentioned, they picked us because of our focus on performance, efficiency and security. Cybereason is a security company that focuses on endpoint protection. They have an amazing platform. They tested our platform, and they've quickly migrated a significant chunk of their infrastructure from GCP over to OCI. The reason for that is that moving to OCI save them more than 40% on their overall infrastructure spend, which directly goes to their bottom line given the type of business that they have and the scale at which they're growing.

Okay. So now I want to talk a bit more about AI infrastructure. So when I talk about AI infrastructure, AI means many different things. Here, I'm really talking about companies that want accelerators for either doing training or reasoning, both fit into this category. We have a lot of these customers, right?

If you take a look here, when we say there's more than 700 of these customers consuming on our platform, that doesn't mean there's 700 customers using AI, just to be clear, right? You hear Mike talk about it. We have thousands and thousands of customers using AI on our platform. This is people who show up and go, "No, I want some type of accelerated computing, whether it be an NVIDIA GPU or an AMD GPU or as we're rolling out different accelerators, any type of accelerated hardware, that's all we're counting for you to fit into this category.

This, as you can imagine, is growing very rapidly, right? 117% year-over-year growth and overall annualized consumed revenue. The margin is different. And the reason for the margin range here is less to do with product mix because in general, these customers actually are pretty well defined on which types of products they want.

A lot of it comes down to the location and how efficient we can be. So as an example, as we continue to be more and more efficient, as we can go in and design better networks, so we can optimize our data center build-out, as we can do things like power capping, that then allows us to have a higher margin.

And in some cases, the other part of the reason for the margin range is, obviously, just like I talked about with cloud natives, there's customers of different sizes, right? Typically, these customers are relatively large to begin with, although there's a lot of very small ones, but they can scale extremely high, as I think all of you know.

Okay. So why they're picking OCI? Well, part of it is, is that we move very quickly. We move quickly when things are stood up, but we also move even more quickly before things exist. So our ability to rapidly deliver the latest and the accelerators.

Networking matters a lot, right? Fundamentally, when people are buying a cluster, like yes, the actual accelerator matters, but all of these workloads are operating in a clustered fashion. That's true across training, especially, but even for inferencing, all of these inferencing is happening at a cluster level.

And then the fact that we actually have extremely efficient data center designs that allow us to optimize our power usage, which then translates into more -- either more power available for them or lower costs. And then it's not enough to just be really good at compute and networking. Storage is also critically important because these clusters are doing something with data, whether that be for training runs or for inferencing, they need access to a huge amount of context for these models to be valuable.

All right. So this is a slide of an example deal. I'm going to walk you through all of the different details. So as an example, let's imagine that somebody shows up and they want to buy approximately 1 gigawatt of GPU accelerators, right? And they sign a contract for 6 years, which would be 6 years at $10 billion a year, it's a $60 billion TCV.

Okay. Now here, I've broken the costs we have into 2 sections. One is what I'm calling land, data center and power. That's obviously the buildings, it's the actual power generation, it's the people cost to run that portion of the business, right? That ends up being about 35% of the cost to deliver that service.

Now if you add up compute, networking and storage, right, all the things that you put inside of that data center, that ends up being about 65% of the cost. Okay, so when -- say this deal is contracted and you have to build a data center from scratch, well, we actually align through the contracts that we do and the way in which we align our overall delivery, we don't pay for land, data center and power until it's actually delivered to us, okay? So let's say, it takes you a year to build a data center. During that year, we're not paying for anything.

Now there's a point at which and that's highlighted in this kind of left section before year 1 here, where the data center is working, and it's our job as Oracle to put all of the compute storage and networking inside of it, all right? During that time period, there's a cost, where we're paying for something, but we're not making revenue yet, right?

And as we can reduce that time, if I can go from 3 months to 2 months to 1 month, the amount of time that I'm paying for something that I'm not able to then provide to a customer, it goes way down, right? So in this example, of this example deal of $60 billion, I've written down here, okay, a 35% margin deal, right in the middle of that 30% to 40% margin range.

But the thing to understand is when I talk about this 35% margin, it doesn't just include each year after it's running. That 35% includes, right, the fact that there's a startup cost upfront. Now obviously, if you can imagine, if you were to put a demarcation line before year 1 starts, we have an expense with no compensating revenue yet.

And that can be happening across many different sites, right? We're building not one of these things, we're building 10, 20 of them at a time. And based on the delivery schedule, right, and our ability to ramp and put the capacity in there, you've got a waterfall across many of these different pieces of infrastructure where some are now starting up, some are now giving you a ton of cash flow off. And part of the job that we do is we're constantly laying those things together to understand how our overall business works.

So key takeaways from this slide. We spend a lot of our energy aligning our revenue and our expenses. The ramp-up time for this stuff has cost, but it's minimal, and it's something that we're continually optimizing. And the expense of that ramp-up is included in the gross margin. So when I talk about, "Hey, here's the gross margin range of one of these customers," we don't go in and assume, right, that, I don't know, some -- you're not accounting for these ramp-up expenses, the gross margin includes the assumptions around those ramp-up expenses.

Okay. Next. I've been talking to a lot of different customers and investors and reporters and analysts about, okay, how do we actually go out and we accelerate and scale this build out? Fundamentally, there is no problem. It's really just an opportunity for us to grow faster, right?

It's not -- there is no issue. There's just opportunities the way I see it. So when I think about what a huge amount of my time and energy goes into, it's about how do we actually enable ourselves and the industry to grow faster, given the unprecedented demand we're seeing for infrastructure like this. And the way that we're growing faster is that we're actually pulling together. The entire industry is coming together across data center providers, energy providers, hardware providers and capital providers. This is already happening all day, every day, right?

You see constant announcements of new business models being created of different ways to go out and minimize risk and grow this business faster. That's actually what's enabling the current growth. And I think you're going to continue to see from us and others, new models created to actually help grow this business faster and faster.

And so to understand the result of what happens when all these partnerships come together, let's take a look at the video to see what's actually possible.

[Presentation]

Okay. So what I'm going to do now is I'm going to kind of just walk you through a little -- a few pictures about the video that we just saw to give you some more context.

So this is a site in Abilene, Texas. It's 1,100 acres across 8 total buildings. So what you're looking at right here that weird kind of quad thing with this thing in the middle, that's actually what we call one building. I know to me, it kind of looks like 5 buildings, but it turns out that apparently, we just don't have enough numbers, we couldn't get to that. Each of those individual pieces or data halls that are all part of one building.

In this location, there are more than 6,400 construction workers on site every day. And once we're complete with this campus, we provide nearly 1,700 jobs, direct jobs and even more indirect jobs. So on site, we have both grid power, and there's an existing substation as well as a new 1 gigawatt substation that's going in.

But in addition to that, there's also 300 megawatts of gas turbines that are installed to provide power to some of the initial buildings and also as partial backup power for the site.

Okay. So we started building this in May of 2024, and the entire campus will be completed in the middle of next year. The workloads for the customer for this site, which is OpenAI in the beginning, went live less than a year after the construction started. Each of these buildings is actually mechanically isolated. They're all networked together. So in that giant campus, all of these buildings are actually through a massive set of interconnections, both within the buildings as well as through very secure fiber between them that actually allows the entire cluster to function together as one giant supercluster. And once this is finalized, as of right now, this will be the largest supercomputer that's ever constructed.

Okay. I don't know how many of you are familiar with the details of data centers, but liquid cooling is something that's existed for a long time, but it's really gone from 0 to 100 very, very quickly. Liquid cooling, so these big pipes are actually the cooling liquid that then go in to cool these accelerators.

Sometimes people ask about, well, what about the water? Does this use a lot of water? The reality is no. This water is in a closed loop system. The actual annual consumption for each one of those buildings that you saw there, which is quite large and is 100 megawatts of power, actually consumes less water per year than a single-family home in the same region. So obviously, you have a lot of water when you first start it up, you have to fill up the swimming pool, but the -- because it's not evaporating all the time, it's a closed-loop system.

Okay. So an example of a customer that is choosing us for AI infrastructure. This is Modal Labs. They provide a developer platform for AI, ML and inferencing. They enable customers to easily fine-tune and do batch processing for AI workloads. They choose us because we have amazing infrastructure. We have the best bare metal compute available and they really appreciate the price performance advantages that we have, combined with our excellent networking.

And then what I want to do now is take a few minutes. I'm going to bring Mark Hura up to the stage, and he's going to talk a little bit about some of our customers and why they are choosing OCI across all these different segments. And at the end, I'll give a little bit more details on where we see OCI going over the next few years.

Mark?

Okay. My job is really very easy. Hi, Mark.

Hi, Clay.

We have mini segments of businesses.

We do.

Let's talk about enterprise. Why are enterprise customers choosing OCI and give us some examples of what it's like working with us.

Yes. So you gave some examples before in terms of why enterprise customers choose Oracle. And that has been our bread and butter for so many years around customers that have deployed Oracle applications, Oracle databases throughout their entire businesses. But then when you think about OCI and how we've defined and developed and deployed a differentiated cloud, that serves many different types of workloads.

And for enterprises, we think about 3 different major categories. One is enterprise customers that have been utilizing Oracle applications and databases that serve either our applications, third-party applications, custom-built applications or large-scale enterprise data warehouses. And those are customers that take advantage of the price performance and capability that we have in OCI, and I'll give some examples there.

The second category is that customers that are looking to exit their data centers and their workloads. And what we have found is that many other clouds requiring you to redefine and build those applications in a new way to fit onto their cloud, so it becomes a constraint for our customers that are looking to take advantage of what a true cloud can do for their business.

And then the third is really bringing pure infrastructure, compute storage networking capabilities to enterprise customers that are taking advantage of raw infrastructure capabilities. Those are 3 distinct areas of why we're winning, how we're winning and how we're engaging with our enterprise customers all around the world.

And if you think about that first category, an example such as Emerson. They look to build and move their entire Oracle EBS infrastructure, while they're migrating to Fusion, but in doing so, they brought the enterprise data warehouse in all of their Oracle databases to our cloud.

In doing so, they also -- what's unique and what we have found, all of the boundary applications that support that supply chain, right, that financial system, all come into OCI, allowing them to exit their data center, get the performance capability while they start to transform to Fusion at the same time.

And so it's a great example of how we support our customers, bringing together the power of the capability of OCI, the performance for them to be able to improve the performance and the speed for their end users around the applications that they're serving.

Claritev is another example, where they were living under the constraint of another provider telling them they had to replatform before they move to the cloud. And the cloud that you designed, Clay, which is bare metal from the core, right, with off-box virtualization and flexible compute for our customers to be able to pick up the VMs that they have and exit that data center without having to replatform, allowing them to focus on the things that are most important for them around delivering services to their customers.

And when we think about just bringing pure infrastructure to our customers, in the enterprise, we have so many examples, but one that may be different and unique is like Goldman Sachs, right? Taking advantage of bringing high-performance computing capability to do risk analysis in their business. And we weren't their original provider, but we are a critical provider of services to them now where they just -- they had to take an opportunity to look at what it meant in OCI and how we are executing and delivering compute capabilities for them and how they were used to doing it before.

And the results were staggering in terms of the price performance and capability that they were able to get and it doesn't have any Oracle databases that they're running in that environment. So there's multiple examples of native VMs and entire data center capabilities, Oracle applications and our entire data platform capabilities for customers or pure infrastructure around enterprises.

Okay. Well, Mark, I agree with you, great examples. I also kind of explained a bit about our distributed cloud customers, and I used a few examples already. But I think there's probably an example that is more surprising and also a bit different than the traditional telco or MSP-type workload here. Tell us more about some distributed cloud customers.

So I mean, I think there's an obvious point, which is sovereign security, highly regulated industries, banking, health care, utilities that can take advantage of a distributed cloud, bringing it to them in terms of the industries that they're serving. But what's also unique about our distributed cloud, it is our full cloud, which means our customers can take advantage of running our full suite of capabilities.

Meta, as an example. Before we get into thinking about infrastructure at scale in training and inferencing, Meta is a user of Fusion, where we've deployed a dedicated cloud at Meta's facilities where they run Fusion to run their back office of their business. And so we have the capability not only to bring infrastructure to our customers, but the ability to run our full suite of industry applications as well as data layers for our customers. So it's a different example, it's an unique example that our customers can take advantage of wherever they deploy.

Okay. So we've got enterprise, distributed cloud. What about cloud natives?

Cloud natives, I mean cloud natives has been a wonderful spot to truly bring the capability of a next-gen cloud to the industry. And you brought up an example around Cybereason. What we have found is these patterns of success that have continued in a variety of different industry segments for cloud natives. Cybereason is one example, but many security companies are running on OCI, right, whether it's Palo Alto, CrowdStrike, Commvault, Sentinel, Fortinet and on and on and on.

And there's a reason why, right? As you mentioned before, there are certain types of workloads where we perform at a higher level than where the industry is. And whether that's the networking requirements from a security company of how they're protecting others or whether it's the compute intensity that they're leveraging to scale up, scale down to run their businesses efficiently to protect what they're providing to their users.

And it's a great example of thinking around what wait a second, if the security companies are running on OCI is the most performing and most secure cloud out there? Well, that makes a lot of sense. In addition to that, we see a lot of ISVs that come to OCI as well. It's an opportunity where we become a critical part of their cost of goods that they provide.

And the technical differentiation that we have and the partnership that we bring to our ISVs is unmatched, where we are working with them to optimize the workloads for OCI to allow them to focus on delivering the services that they bring to their customers so that we are a critical part of their success, but most importantly is it's a highly engineered engagement with the ISVs and it's unmatched in terms of the performance.

Okay. So I also had -- I had Peter from OpenAI on stage with me yesterday. He was very complementary to your team. And then we just talked about Modal. Give everybody here a little bit more perspective on why we're doing so well in AI infrastructure and why those customers are choosing us?

Yes. AI infrastructure, and when we really started in this space many years ago, I recall a distinct conversation when we were working with NVIDIA actually, around NVIDIA needing to run their workloads in a cloud and engaging directly with Jensen and what he said is, "Wow, I'd never thought I would see this, but Oracle, you are fast, you are highly technical, right, and you are incredibly capable from an engineering perspective."

And he said, never lose that advantage. It wasn't necessarily something that always aligned in terms of what he thought in the past. And for us, it's something that we just continue to operate it with all of the customers that are taking advantage of our AI infrastructure that we're building.

Our teams are incredibly fast. We know what we're really good at. We're highly technical and then we provide a white glove service for our customers that is a 24/7 engagement. And what we constantly hear over and over again is that we are the best to work with, and we're the easiest in terms of operating.

And what happens is this is a pretty tight community. And it spreads very quickly. So as the early-stage companies that we started with, they were telling our story to others, and it's just continued to snowball from there. And obviously, we have the largest of the largest AI companies from running infrastructure in the world today. And so again, highly technical. We don't waste anybody's time. We know what we're really good at and the speed is unmatched.

No. Mark, thank you. I think it's great summary of why people are choosing OCI. Really appreciate it. Thank you.

Thank you. All right.

All right. So this is the long-range plan that we provided on our September 9 earnings call, just a few -- well, I guess it was last month technically. The thing to understand about this is that when I talk about this business, all these different segments are included in this. So this projection about future revenue includes all of these business segments that I'm talking about.

The other thing I want you to understand is that when we talk to you about these types of projections, there are projections that we believe in. We have very real reasons across each of these segments that we're across all of these plans that add up together to provide this kind of projection.

The other thing to understand a little bit, and I'll go into it a bit more in a second, is that for some segments of our business, there really are actually supply constrained, not demand constrained, specifically AI infrastructure. And I'll talk about what that means and why you're seeing the changes from us that we're making in the industry.

So to give you some context here. In 30 days during this quarter, 30 contiguous days, we contracted for $65 billion of additional commitment across infrastructure contracts. Now across those, it was across 7 different contracts from 4 different customers. None of those customers are OpenAI.

I know some people are questioning sometimes, "Hey, is it just OpenAI?" The reality is we think OpenAI is a great customer, but we have many customers. One of these customers is Meta. And as I share these numbers with you, this is not all of our contracts. This is not saying, "Hey, this is -- we didn't add up all of the deals." This is literally 7 deals, 4 customers, all of them other than OpenAI, and it shows the diversity of our customers and the size of their interest in our business.

Okay. So as I mentioned before, most people are assuming that we are demand constrained, but we're not. We're actually supply constrained. And the thing to understand here is that a large part of what I spend my time doing is securing what I consider to be good supply. I'll give you an example of an e-mail I received recently. There was a very nice person who lives in Nebraska, who has a corn field.

They said that -- they think that would be a really good place to put an AI data center. You're probably laughing, but I received. So there's -- you cannot imagine the amount of reach outs that come my way in the way of my team.

It's actually quite a hard job to filter through all of the opportunities to understand which ones bring together the energy that you need, the hard work, the capacity that you need, the land that you need and the capital that you need. Our job is to put together all of those pieces. And only then does that result in supply.

And so what we find is that when we actually have good answers around supply in the reasonable time frame, customers then contract very quickly. So when we talk about the business that we're doing, it's not that I have this infinite supply over here of options, and I spend all of my time going to customers asking, "Hey, would you possibly want this?"

Instead, the customers have come to us saying, we would like a lot of stuff. And our job is to go find actual good supply that we can accelerate and deliver on the time line they need. That is what we do all the time. And once we have that done, suddenly, the customers contract very quickly, and we're contracting with all the different suppliers to bring together these pieces, so we can execute as quickly as possible.

So as part of that, we are updating our long-range plan between now and our fiscal year '30. Again, the reason we're updating this is exactly for what I showed you 2 slides ago. As we find ourselves able to execute our supply constraints, as we actually can go in, we have customers that want that. And that's part of the reason to, and you see the change in these numbers, that it's a little bit easier for us to find supply not this year or next year, but in subsequent years. So as we're able to find that supply, customers contract for it, we see immense demand and then we go about delivering that to customers.

Okay. The other thing I want people to understand is that I talked about these different segments. And obviously, the numbers we just talked about was the all up number for OCI. But we're also extremely focused on our AI database and our AI data platform business. We are very confident in our ability to grow this very rapidly. Now there's a few reasons for that. One of them is, up until recently, as I mentioned before, there was only one cloud selling our AI database and our AI data platform, and now we've transitioned to all of them.

The other reason we're very confident in this projection, you'll hear more about that from Larry later today is the huge investments that we're making into AI and to building out our AI data platform only makes the previous investments we have and our customers have made into our database that much more valuable. So as we go through this next period over a few years, we're rapidly accelerating our cloud database and our AI data platform business.

And this is a -- it's amazing for us to see, right, a business that is such a high margin, that's so valuable to our customers grow so very quickly. So with that, I'm out of time, and I'm going to hand it over, I think to Mike, and he's going to talk to you about applications. Thank you very much.

Please welcome to the stage, Mike Sicilia, Steve Miranda and Mark Hura.

Okay. Hello again. So we're going to talk about applications here. As a reminder, we'll do so under safe harbor. We will talk about a road map. We will talk about our future product direction. So you've heard us say throughout the event here that AI changes everything. And I think what's really important to understand is that, for all the reasons Clay just described, for everything that we're doing in OCI, our ability to deliver AI to our customers embedded inside of our application stack and the ability to have just an unbelievable time to value, we think is unmatched.

We've got -- I think we've gotten to a point, Steve, Steve and I have been discussing this in the week, is that in our applications suite across the board, when you think about our Fusion applications, our front-of-the-house applications like CX, our industry applications, we're at the point where there isn't a difference between the AI version and the non-AI version.

It just doesn't exist. And I'll -- we'll go through some examples of some of those things we're seeing. Particularly in health care, we just went generally available with a new EHR. It's all AI. You can't choose not to consume AI to actually run the application. Now if we didn't make all the investments we've made, if Clay and Mark were out there attracting all these wonderful models to OCI, we would not be able to deliver that as a packaged service.

And that's one of the things that I think when we see in the application market, when people say, "Well, maybe there's not an ROI on this stuff." Actually, I think it's because they're trying to stitch too many things together. They've got too many vendors involved, and they get right back into this old mantra of a very large implementation, a very large time and materials implementations, which frankly, customers are losing patient for.

That's not at all what we're doing. We're actually winning more customers because we're shifting our focus to outcomes. We're winning retail customers in Germany. We're displacing them from SAP because we have an all-in platform of Fusion merchandising plus our retail applications.

Our customers are achieving more. Our customers in health care, and I'll go through this example again, it went live with our AI agents. Within 3 weeks, the ROI in terms of return on time spent with the system was 50%, right? It wasn't small. It was by half, an order of magnitude by half. And the fact that we're delivering that as a service has been really just, we think, game-changing. And along the way, as we rely on all of the OCI tech, we're also doing quite a bit with the AI data platform.

And to make a long story short, good things happen when all the data is in one spot. And that has expanded our ecosystem such that we're not just talking about automating entire industries, but actually automating how those industries communicate with other industries, and we'll go through some of those examples in just a bit.

But to dig in, Steve, if you could maybe take us through all the great things in Fusion that we've got with AI to start us off.

I can do just a couple of the great things in Fusion, Mike. Thanks. So first off, let me just give you an update on our customers. So I think you'll find and hopefully, you've seen this week just by walking around, we have a who's who of the top customers, the top brands globally by industry, by what I'll call heritage, meaning people have moved off of E-Business Suite, PeopleSoft, JD Edwards, Siebel, SAP and a variety of other third-party applications.

I used to show this slide really to show some credibility of Fusion and how Fusion is real. I think now it's unquestioned that Fusion is a market leader in terms of cloud-based applications.

The reason why I showed this slide now is twofold. First off, the growing trend of what's happening is customers are going to Phase 2, 3 and 4 of their projects. They started in ERP, they're adding HCM. They started in ERP and supply chain, they're adding CX. They're starting in ERP supply chain and they're adding industry applications or they have industry applications and they're adding others.

So it's a tremendous leverage to us. But more than that, the referenceability and the case studies we have in the customer. So just this morning, I was talking to a major airline who owns a variety of subsystems who have dozens of ERP systems. And what they want to do is they want to consolidate, but they are essentially a corporate headquarters with federated businesses. And they don't want to -- while they want to bring the system together, they still want to have different lines of business.

I was able to quickly give them FedEx did that, except FedEx centralized it all, very similar industry to the airline into a shared service. But they said, "Well, we don't want to shared service." No problem. That's exactly what DWP and the U.K. government did, brought a bunch of government entities together, but had federated ownership. And as it turns out, this airline had already met with DWP at this conference, shared their use case and are going to now connect to them offline and follow that same program. So this is not just a customer base that's buying more products. This is a customer base that's talking to each other and helping us expand and helping others expand and succeed. It's just a tremendous, tremendous success story for all these customers, including AI and including Oracle AI.

Now safe harbor aside, and Mike said, we're going to talk about futures. Everything I'm going to talk about here in Fusion is not futures, is today. So as one example of AI being today, we start first, as always, at Oracle, our own use case. And we have, as you guys all know, a world-class finance organization, and we talk a lot about our efficiency of close and our speed of close and reporting, but we've just made that much better.

Our internal finance group is rolling out our ledger agent. They're rolling out our payment agent. They're using AI to take an already extremely efficient group and fast group and make it more efficient. We've implemented agents across HR. Even in functions, we never really had like HR benefits, goals and functions using AI to improve there.

And then for our support business, handling the support tickets, all of our external customers now have AI agents, both on the front end to deflect SRs and questions and for our assistance on the back end so that our support agents can better find the answers and service our customers.

We've just rolled out those agents already, we're seeing faster time to resolution. less people intervention, meaning less cost to us and more accuracy of the resolution, and it's resulting in greater customer satisfaction on the SR surveys. So we're Safra's manta, we're paying less, and we're doing a better job of it through the AI agents. And those are just 3 examples at Oracle.

Now what we announced -- what I talked about at this conference last year was that we would have 100 AI agents in Fusion. We actually have 600 AI agents, 400-plus in Fusion, 200-plus in the industry verticals. But that is a massive understatement in the AI because what we've built is an AI ecosystem across Fusion.

One are the agents that we built internal to Fusion. Second, however, is we have an agent studio that allows our customers to modify what we give them in agents, to build their own agents, to integrate with third parties. And then at this conference, we just announced an agent marketplace. The agent marketplace allowing you to extend and expand them, and we have over 2 dozen partners part of the marketplace, each who've contributed already about half a dozen agents together. So the 600 are agents that we've built that does not count agents that our customers are building through the agent studio, that does not count what our partners have built.

Steven, I think what's really interesting about that, as we discussed, is that you have kind of this age-old question in procurement cycles of build by partner. I think your answer is, well, how about all 3, right? You can build your own agents, you can buy partner agents or you can buy our agents, but you're doing it in the safety and security of a single platform, right? It's not -- it's still not this idea of stitching a bunch of things together and bringing a bunch of bespoke...

Well, I think that, that last but is a significant difference because in theory, could you use a third-party agentic framework and build agents on top of Fusion? Absolutely. However, when you build Fusion agents on top of the agent platform or use ours, the context is there. So it knows Mike Sicilia, it knows your role within Oracle. It knows what you can see and what you can't see...

I had no access. I tried to...

Now you have all access. But let me pick a different example. When you're Steve Miranda, you have somewhat more limited access. So it knows -- but the benefits agent example. So if I log into Oracle, we have the benefit agent. I want to ask it a question. I'm traveling in Europe. I want a prescription. Is it covered or not? I have someone on my team has a leave of absence. They want to extend it. Do they need to extend COBRA? There's all sorts of benefits questions.

Well, the agents native to Fusion, not only can our customers configure them in all the ways you see here, but it's embedded contextual and secure. So it knows the person's role. It knows their HR status, so it knows their tenure. It knows what health care plan they've selected. It knows what country they're in. So it knows exactly the benefits policy on which to answer the questions. If you had a bespoke or build your own, you could certainly do that. But if you did it totally outside of the Oracle ecosystem, you have to code in the security, you have to code in the context and you have to keep that updated always for every agent you do. It's a much different time to value to your point, and that's where we're seeing the difference in our customers adopting this quickly and moving forward.

And by the way, we now have over 32,000 people, and actually, I think this is -- I know it's already dated. I think it's almost 40,000 at this point, certified on our agent studio. This is Oracle, Internal Oracle Consulting. These are our partners.

To show you a list of the partners, I talked about those 2 dozen agents studio. These are partners, large and small, that have already been certified and already delivered, again, at least half a dozen agents each that have been validated by our development team to the agent marketplace.

Every one of these agents is available to every one of our customers at no additional cost. So these are agents that are contributed to make the implementations easier, make industry functionality easier, make things better across the board. And I said it was a massive understatement. So not only are the 600 that we built doesn't count what our customers built, doesn't count the marketplace.

We had a hackathon on Monday at this event. Mike and I were just talking about it. Mike got a chance to poke his head into it. We had a little bit over 165 different partners and customers and it's all sitting around the table in rooms like this. So last year, for perspective, I promised 100 agents in Fusion. On Monday, not from Oracle, partners and customers in a hackathon, they built 109 agents, which will go into the marketplace.

So when you talk about the speed and progress across the applications and the speed and progress of consuming AI and consuming all the great things that Clay talked about, that's just a perfect example. Okay. So let's take a look about industries.

Okay. Thanks, Steve. All right. So today already, in our industry applications or whether they're edge applications or they're more sophisticated applications like core banking, we have 2,400 customers that are already leveraging Oracle AI, embedded Oracle AI inside these applications today.

They are live across a large variety of industries. And again, as Steve mentioned, we have over 200 AI agents -- features and agents just in our industry portfolio live today. We'll eclipse that, and I'll go through the numbers here in just a bit very, very quickly in the coming months in terms of total agents to market. And our view, and we've heard a lot of feedback that aren't applications in the future, really just a collection of agents. And what do you think about your stance there? And our view is, well, you're exactly right. And we're actually doing that thing exactly.

Our applications stack, whether they're industry applications, Fusion applications, are quickly becoming a collection of AI agents and our ability and our time to market is second to none. It's not just interesting that we have 2,400 customers live or 200 new AI agents.

But if you go back 18 months, both of those numbers were 0. We had 0 AI agents live in our industry applications, and we had 0 industry customers live on any industry AI. 18 months later, we're at 2,400. In very sophisticated, highly regulated industries like health care, so far, the longest go-live we've had with an AI agent has been 3 weeks. That's been the longest. The shortest has been in a matter of days.

The amount of professional services dollars spent by the grand total of every customer. There are over 250 customers just in health care alone live on the AI agents is 0, $0, self-directed, self-implemented. The amount of dollars spent on training collectively by all of those customers is 0. 0 training, works out of the box. And keep in mind, these things are working in highly sophisticated clinical applications. These are running inpatient rooms with no training.

Yes. I think just one point, your point about is SaaS going to change or what it is, especially in industry applications, but even take the finance example, Maria Smith's team at Oracle, their role isn't to use our user interface and type invoices into the system. Their role is to report our earnings to pay our suppliers, to invoice our customers to collect.

Our IP was not our UI to do that. We take great pride in our IP. Our IP is how efficiently we organize that and give data to our customers. What we've done with these agents, the 600-plus is we've allowed customers to do that more efficiently. That's what we've always done, and now we have that in a much, much more accelerated.

Yes. And of course, Steve, it's all much easier when you're a custodian of all the data, you need to build these agents, right? And the fact that all of this runs on the Oracle AI data platform makes this engineering cycle, this innovation cycle just so much faster and the ability to roll this out as a quarterly update, it's -- we really -- we're quite excited by our customers' ability to update this. So as you know, we operate in many different industries, and I'll go through a couple of examples here of agents that I think are really going to be very popular. The first is the embedded AI agent for retail.

So this is an intelligent inventory agent. It's designed specifically to help customers bring together data from CRM, from merchandising, from Fusion inventory, from Oracle Analytics Cloud, from Oracle Xstore, which is the point-of-sale system that we have, from our procurement and order management systems and Fusion.

The agent is not just a point of intelligence. It's not just helping retailers decide which product should I put on the shelf and what is the impact of my forecast demand, my planning buys and moving goods around, selling across channels, my omnichannel strategy, all that, of course, is an output of the agent.

The intelligence is built into the agent to figure that out. But it's also a very interesting point of integration. And if you think about how much money organizations spend today on integrating applications across the stacks, and agents do it mostly for free. I mean the ability to traverse data from multiple sources across multiple systems. And by the way, the CX, the merchandising, the inventory, the analytics, the Xstore, procurement, order management, it doesn't -- they don't necessarily have to be all Oracle systems. That's not how the agents are designed. The agents are designed to pull from multiple systems and put together points of intelligence to give, in this case, the retailer very high visibility so they can flag potential inventory issues and translate complex data into very clear, transparent recommendations. I'm sure those of you who have covered or understand retail, inventory management is one of the costliest and most difficult things and if they can avoid liquidation of inventory, it's real money back. So a lot going on in retail.

I did mention a bit about health care. It's important to know that we've got dozens of AI agents live across our health ecosystem today, many more planned. We're looking at chart review care navigation, clinical decision support, patient risk predictions, preventative care and many more. In fact, our next-generation AI EHR is now generally available. It is generally available and it is also approved by the regulators. This is also another interesting AI story.

We've got customers running this in beta. Seema is going to talk a little bit more about this as we go forward and why we're so excited about this. And the feedback has been absolutely tremendous. You remember that these customers are coming from a -- literally a Windows 95 Citrix experience to agentic AI. That's the leap of technology that we see in health care.

And the feedback, it's not just a great business, and it's not just going to be a great growth engine for applications, but it's actually an emotional piece as well. I mean, the feedback that we get from providers and patients is, "This changes my life. It changes the way I practice medicine because you've actually given me a tool that helps me not burdens me."

And we've got a lot of competition in that market. But if you break it down and you think about AI as powering EHR, the one question I ask our competitors is well, see how many fuel cell power plants are you building on site right now? Because you're not doing that, probably not going to have as good of a chance to be closely provisioned to a large language model and apply reasoning models and all the things you actually need to work -- to make this work at scale to automate an entire hospital.

It takes usually -- the regulatory cycle for approval for electronic health record software today is about 2 to 3 years. So it takes about 2 to 3 years to get through a regulatory cycle, and they get the self-fulfilling prophecy. Because the cycle is so long for approval, people end up running very old technology for a very long period of time because the pain of switching is very high.

We were able to get through the full regulatory cycle in the United States for this in 6 months. The reason we got through the regulatory so quickly is because we actually used AI to generate a lot of the documentation. Now we cross-checked it. We have a baseline. We make sure it's right. We make sure it's accurate. But it's actually saving us a tremendous amount of time and putting us in a position to be able to bring these products to market, particularly in heavily regulated industries like health care and banking and utilities, where there's a very large burden of documentation to actually bring products to market because in health care, you've got federal government regulations, but you also have state-specific regulations. California is different than Texas, for example, in terms of some of the things you need to do for Medicaid compliance.

We're actually able to automate all of that process across our industry applications and bring these products to market. And in such, we think that regulators are actually happier with that as well. They're actually happier to receive more timely information. They're actually happier to reduce the burden on them in terms of having to review an incredibly complex set of documentation.

I'll switch gears now to banking, which is a really, really interesting business. So Steve, I mean, I know the Fusion applications are a very popular choice among big banks. The core ERP systems, HCM systems, and the banks have largely had an appetite to move that to the cloud over the years.

Now when we talk about core banking systems, for lack of a better way to put it, the stuff that moves money around inside the bank. There's still a lot of mainframe, still a lot of mini. There's still a lot of very big iron running these applications. And we are definitely -- if you look at core banking applications move to the cloud, let alone move to AI, okay? Just think about just the first step. You're in single-digit percentages in the terms of that, that has been moved to the cloud. But as we said this week, AI changes everything.

So the future of what we're building and helping banks move away from large on-premises bespoke applications is with that agentic approach. And we see this as a tremendous opportunity to unlock our core banking installed base and actually to win new customers in the shift to the cloud.

Again, very, very difficult in a highly regulated, very sophisticated application space like core banking, unless you're doing all of it. It's going to be very difficult for competitors to come in, we think, with piecemeal applications and stitch stuff together to make a core banking AI solution.

So just in the next year at most, just in the next year, just in the core banking space, we'll have 125 agents live. And I'm going to go through some of those here in just a minute, just in the next year. Steve last year made a projection about 100. We ended up with 400. I think we have a good chance of being in a very similar situation here where we're even surprised as to how fast we can go. So these are across a wide variety of banking and insurance spaces. They're going to include things for human interactions. We'll go through some of those a bit. Domain agents and actually those that are really focused on improving accuracy and processing of time.

Our agents and features are designed to automate the entire bank or the insurance as a KB. We're talking about corporate banking, retail banking, revenue management, billing, insurance policy administration and just got so many terrific things coming out.

So I'm going to start with one of them, which is a huge pain point. Again, I'm not going to go all through all 125. I'm just going to pick 2 in the interest of time here. But financial crimes and compliance. So this is a huge pain point for the banks as we talk to them. So many of you may know and your colleagues, for those of you that are with the big banks, spend an incredible amount of time. It's very costly. It's very time consuming to investigate financial crimes. Unfortunately, financial crimes are not going away. And selling it last year, I said the firm spent $155 billion in 2024 just on the investigation process.

Tier 1 banks have thousands of people, staffing investigative teams and are spending hundreds of millions of dollars a year. Leveraging AI, we can improve accuracy and drastically reduce the investigation time. So if you took a look at these 3 representative banks here, names would tell, but these banks in the chart look how much they've disclosed that they're actually spending on this. We believe that our agentic approach, and we've been beta testing this with the banks can save up to 60% of the work. Just in Phase 1 of the agent, we should be able to knock out 60% of the work. So I'm going to take a look at some of it here.

So this is the output of a case that's already fully being investigated by our financial crimes AI agent. And with the agent -- without the agent, the investigator would have spent hours, days, weeks opening this case, digging through the details, understanding the parties, their professions and whether they had prior activity with the bank.

With the AI investigator, the initial groundwork, it's already done. Before the investigator even opens the case, it's done. And we can immediately see -- hopefully, you can see on the slide that this individual is, in this case, a retired nurse on a fixed income and that she's had prior cases where suspicious wires were sent to unknown parties and they were blocked.

Now the AI investigator is going to give the investigator a total overview of the case. It's showing patterns. It's showing activities and typologies. And things like funnel behavior require repeated wire deposits, withdrawals of high-risk jurisdictions. And then I have the ability to see all of that over time. So the investigator runs through this, decides and whether or not they agree with AI with the findings.

And if they want to dig deeper, they're certainly able to dig deeper manually. But the interface is going to show them all the relevant risk factors, and it provides a very clear narrative of what was observed versus what wasn't observed. And if the investigator agrees with the AI agent, they can finalize decision. If not, they can intervene. System is going to guide them through all the various views so they can quickly see what happened and why.

Then the UI is built for the investigator to feel confident about decisions. This is just exactly the same way our AI works in health care, not replacing the doctor, not taking away from the provider, but actually showing the provider, in this case, showing the investigator exactly how the decisions were built, exactly where they came from and the investigator is always in the loop. So the person still signs off on this, but if you could do your job 60% faster as a result of this, I think we'd all be happy.

So this is a really powerful shift. We think it's a perfect application of AI today in large banks. This is all generally available and really much easier for our customers to absorb because we're supplying as a service here, the whole stack. It's OCI. It's the database. It's the core banking applications. It's the analytics. It's a generative AI data platform, large language model integration, which wrote all the narratives here. In this case, again, if you're the custodian of all the data, it's so much easier to bring these type of agents to bank very, very quickly.

All right. So I'll show you another one of the -- #2 of the 125 is our retail banking agents. So we're going to use AI here to improve both customer and bank experience and streamline a very critical business process.

So ordinarily, the process of providing a personalized service to an individual customer inside a bank could be quite complex. I mean banks like all businesses want to provide as much personalized service as they possibly can. But with AI, we think we can actually help the banks get to that more personalized consumer experience when they want to position exact services like loans or even before the customers may know they need them.

So we've got a video here that we're going to play to show this agent.

[Presentation]

Okay. 123 more of those agents -- such agents are planned for the next year here. So we'll have a lot more to say about banking. Again, it's an industry where I think AI is just going to be -- agentic applications are going to be a tremendous wedge to move a very large, very dated infrastructure base, both applications and infrastructure as well to our cloud and AI solutions.

So you see some of the stats across other industries. Again, I won't go through all the industries. I spoke about the clinical AI piece for hospitals. In hospitality, we released an upsell AI agent, which helps hoteliers understand where the upsell wheel maybe, what's the sweet spot for engaging with customers, how many weeks prior to check-in, what are the services these people like and how do we understand how to position.

Just in year 1 of rolling it out, which was actually on just a small bit of our installed base for hotels, hoteliers generated $350 million in upsell. That's what they reported back to us in the upsell wheel from this AI agent.

I spoke about the financial crimes and AI investigator. We just rolled out a brand-new agent for energy optimization in our Opower stack and very early days, but we got an immediate feedback from one of our very large energy customers that it saved them $2 million instantly and reduced calls to their call center. So this story goes on and on and on across the board.

What's really interesting about the banks, too, if I go back to that before I ask Mark for his comments, is the whole stack advantage, I think, is really interesting in banks. We're having conversations not just about how do you take all these very old custom bespoke applications and how do you take everything that's bolted on to them and how do we use AIs and unlock mechanisms to move them.

But what are you going to do with all of your infrastructure? And because of our dedicated region in OCI, we're actually having an entire stack conversation with these customers. And we think it's different than our competitors. Now you've heard us say many times before that we have the widest portfolio of applications of technology services, of cloud services, GPU, CPU on the market. And that's been something that we've all been quite proud of and that we've been able to be good in all those industries. One of the things that sometimes is a little suboptimal in that market is that you have a lot of salespeople. You have a lot of people who are calling on customers in many different pieces.

And then in the on-premises days and even in early cloud days, that worked because a lot of the buying was by pillar. But now we're seeing a shift in customer buying patterns. And we're seeing customers as we position the whole stack, really wanting to talk about outcomes and less about all the parts that make up the outcome. In other words, we're selling and positioning to the customer and outcome. We're selling a 70% reduction in financial crimes. We're selling a 49% reduction in paperwork for health care.

And in order to do that, we've made a lot of changes about our -- with our go-to-market strategy. And I think that those changes are just as important as everything we've done in the product stack so that we can accurately communicate that to customers and engage with customers at the senior level.

So Mark, I'm going to ask you to kind of walk us through what we've done to address this whole stack advantage.

Perfect. Thanks, Mike. I think AI changes everything. It changes how we develop products. It changes how we deliver products. It changes how our customers consume our AI agents embedded in our applications and our AI database and the capability that exists. It also changes how we go to market.

It's given us an unique opportunity to really transform how we engage our customers to bring the power of our entire portfolio to the industries and the customers that we serve all around the world. There's 3 key things that I want to make sure that we all can walk away with is that we are making Oracle easier to understand, easier to do business with.

We have simplified and unified our teams of how we go to market and how we bring the capabilities of the industry suites of applications that have embedded agents and the ability to build and extend on our agent platforms, the ability to deploy our database anywhere and allow our customers to take advantage of the AI database that has AI native capabilities in our data platform or take advantage of that infrastructure.

And we've done that in a way that allows our customers to work with us in a more seamless fashion. In addition to that, as Mike was describing, the capabilities that come across the entire platform, when we work with our customers around bringing our infrastructure, our data and our applications together, that one Oracle advantage, when we do that at scale, the opportunity is incredible.

For our customers to be able to consume and bring outcomes that deliver real value, where they're not focused on integrating solutions, they're not focused on deploying differentiated capabilities, they're not worrying about where their data is moving and how is it secured or not.

When we bring that entire value, we do truly bring an advantage to our customers to focus on the things that matter most to them, which is the services that they are providing to their customers or their employees. We also -- it also gives us a unique advantage to help our customers accelerate on their AI transformations with the capabilities that we have throughout the portfolio.

So let me start with the simpler unified approach to the go-to-market. At the top level, it's bringing the Oracle brand to our customers. It's Oracle to our customer, it's Oracle to governments, and in some cases, it's Oracle to countries around how we bring the full capability. And that may be in certain instances, bringing our cloud to the customer with the full stack capabilities that exist.

If we really think about our commercial teams, and AI didn't allow us to say we're going to reduce the commercial teams, it allowed us to deploy our resources so that we actually have more sellers in front of our customers that are bringing meaningful solutions for them.

In our applications teams, we bring together our industry suites and our fusion suites together for our customers. If we're going into a retailer, we bring the entire suite of merchandising, right, capabilities around inventory management, the back office around financial consolidation, close, human capital.

But also, we have the ability to bring other solutions, a retailer that may be online but also stores that are deployed, not only in the United States, but all over the world. Our engineering and construction platform because they need to build out those capabilities and need project management resources. We bring some capabilities around financing solutions, loan and lease applications that these customers need.

Also, they may have a call center. Well, guess what, our communications business has call center capabilities to allow that customer to take advantage of the AI inherent capabilities. So when we bring one resource to our customer around our full application suite, they can take advantage of the integrated capabilities that exist across that entire platform, which is an unique differentiation that there is no other competitor in the industry that has that capability.

In addition to that, when we think about our AI data platform, and you'll hear from Larry later around the innovation around native AI capabilities in the Oracle AI database, and we announced and launched the AI data platform to allow our customers to take all that enterprise data. Our applications are data generators. They bring immense value around incremental data that is generated around a customer buying something, around an employee and how they work. Well, that data needs to go somewhere.

And in that data platform, the ability to bring the latest large language models to inference off of that data, where that customer wants that data to be securely in that environment allows us to help customers transform. Well, in the past, when customers bought on-premise, we had a person that was selling to them on-premise, we had a person that was providing them and selling them hardware, we had a person that was trying to sell cloud databases and capabilities.

Well, no longer it's one person that brings the entirety of our data platform to our customer. But giving them choice where they want to run, database anywhere, on-premises, in OCI, in our multi-cloud partner capabilities, but also cloud at their facility as well.

And in our infrastructure business, it's not just good enough to bring those Oracle workloads to OCI. We've developed a differentiated cloud. We've gone from a disruptor in the industry, right, from an underdog to a disruptor and bringing compute storage network capabilities at a differentiated level to our customers allows us to be the destination cloud for many customers that have critical capabilities around high networking, high compute-intensive workloads.

But also we become the destination cloud for training and inferencing, where we now have the ability to bring those models to the customers wherever they choose to, to securely and inference off of their private data, but again, the opportunity to bring the entire stack to our customers is differentiated that no one else has the capabilities of this entire stack.

We've never truly taken advantage of bringing it all together, and we are now. And this is an opportunity for us to continue to execute that scale across the globe of how we engage with every customer across every industry that we serve. But when we bring the one Oracle to our customers, we have incredible scale and advantage. Not only do our customers benefit from the integrated solutions and the capabilities, but Oracle also benefits.

Let me give you some examples. If our customer base, if you can see here, 55% of the customers own a single product pillar. Let me explain that. And let's just call that 1x of spend. That may mean they own our database capabilities or they may have a Fusion application. They may have ERP or they may have HCM.

They may be an infrastructure user and not amongst the other products. But that doesn't mean they're using all the products in that pillar. They spend 1x with Oracle. Well, as customers move down the path and have 2 pillars of those 3, where they may actually be using our database on OCI or they may be using database with a cloud at customer.

There may be a Fusion user that's using OCI for integration capabilities to other applications. When they use 2 of those pillars, they spend 8x more than the 55% of that customer base. As you move down the path and go to 3, where they're utilizing HCM and our data platform with a Cloudera customer as an example, they spend 25x more the average base.

And that doesn't mean they're using everything out of the suite. It doesn't mean they're using the retail solution with merchandising, and warehouse management and fusion across HCM and ERP and CX and supply chain. It just may mean they're using one of those products. But truly, as you go up the scale where customers are truly taking advantage of the entire suite, where utility is taking advantage of our operational solutions and capabilities that cut across their customer care and meter data management, they're operating the grid, they're taking advantage of our energy efficiency solutions, managing their financial capabilities, their customer experience, their human capital running it on OCI and taking advantage of our data platform across their business. 2% of our customers in that category, 150x on average. That's on average.

That means customers are spending 300, 400, 500 times as well when they truly take advantage of the entire suite. That is why we have changed our go-to-market to simplify how we're engaging with our customers and executing. And it's not only how we're engaging, but we're also transforming how we're allowing them to consume our products.

We started this with OCI, and we have an universal credit because customers don't know exactly what they're going to use out of the products. They may flex in their compute. They may need more network. They may have some different types of storage. Well, we just announced our multi-cloud credit capability for our customers to deploy their database anywhere, it's one price, no matter where they choose to deploy.

And we're going to continue to modernize and do the same with our applications for our customers and make it easier to consume and move up the stack for them to get more value out of our solutions. And there's just some examples...

Sorry, I think, Mark, just to clarify, too, the customers are actually spending less. They're spending more with Oracle. Because what Mike talked about, the single stack advantage and all the engineering and putting all the pieces together, instead of customers having to stitch it together and spending the money between that, they get 2 components, 3 components and more that Mark shared.

And it's not -- and in any one of these categories on a stand-alone product basis, Oracle is still, right, rated as a top product and capability in that area. So bringing it together does not sacrifice the customers' capabilities, but it truly takes advantage of all of the integration that exists across the platform, where customers can blend together different solutions and take advantage of capabilities across industries and whether that's in financial services, along with our Fusion platform and our data platform.

It allows them to spend their time, energy and effort focusing on the things that matter most of them around their transformation, around their AI deployments and not worrying about integrations, not worrying about patching, not worrying about continuing to make sure things are going to work together. What cloud is it in? Where is it in? How does that need to work?

We're allowing them to take advantage of the capabilities and the work that we are doing. And here's just a smattering of examples that you see here, which it doesn't cover every industry that we're in, but you could see, it covers a variety of industries, whether it's in the energy industry and utilities, transportation, logistics, health care, financial services, hospitality, communications, high tech. It cuts across every one of these industries.

And whether customers choose to start with the entire platform like Exelon, and Mike, you had the CEO of Exelon with you on stage on Monday; or it's AtlantiCare that's focused on deploying our latest AI-enabled health applications to have a data strategy running on OCI, where they're focused on providing outcomes to patients and not trying to worry about securing all of their applications and managing that infrastructure.

Or whether it's maybe federal, that's deploying the SaaS core banking solutions as well as our Fusion suite, but their CEO talked about the need to have a foundation of data across their entire business running at on OCI or Avis. Avis has deployed Fusion across their platform. They recently brought their entire data stack from another proprietary environment into the Oracle AI database, running on OCI, 1 trillion rows of data.

Now they're deploying AI agents embedded infusion and extending those with the agent platform. And in a very short period of time, realize the importance of how they can quickly tap in to that trillion rows of data that they have in the Oracle AI database to deliver outcomes for their operations, their employees and their customers to transform their experience, all running on OCI. And that is the Oracle Advantage.

When our customers take advantage of every layer of the stack and whether that's an OCI, where we've gone from an underdog to a disruptor, where we train and inference on the large language models in the world, having the most secure, scalable and deployable performance platform in the industry to our AI data platform where we have the AI database with AI capabilities native to allow our customers to transform all of their data to be enterprise ready for AI capabilities and our most complete suite of industry applications.

It is a simpler, easier approach to the marketplace. It is truly differentiated that no one else has to allow us to accelerate our growth, but more importantly, is to help our customers transform on their AI journeys. It truly is an unprecedented time, and AI changes everything of how we go to market as well.

Mark, a quick question. Early days in rolling this out, what's the feedback been from our customers? Steve, I know you mentioned you heard some of it as well in terms of our new approach to customer engagement.

The feedback has been amazing. The conversation quickly transforms to I didn't know that you had all of those capabilities because it wasn't necessarily brought to me in the past because there was so many different people now I can see where you've developed, how you develop, more importantly, how you can help me drive outcomes.

It's amazing to see our customers, even many of the customers that were here speaking this week, learn so much more about what we have to offer and by doing it in a much simpler approach. And it's -- we're seeing our pipeline grow. We're seeing customers choose more of the full suite of our capabilities, and they're selecting to run it on our infrastructure as well.

Perfect. Mark, Steve, thank you very much.

And the feedback of simplicity, too. So yes, that's great.

Great. Thanks very much, guys.

Thank you.

Thank you.

Okay. So we're going to take a deeper dive into health and then also at the end, talk about what health has to do with our banking business, which I think is quite interesting.

So to talk more about health care, I'm pleased to welcome Executive Vice President and General Manager of Oracle Health globally, Seema Verma.

Seema, welcome. Thank you.

So Seema, health care continues to be front page news everywhere in the world, usually not for great reasons. You have an interesting perspective as the former administrator of the largest payer in the world, which is the United States government for health care.

And for several years now, you've been engaging with both our commercial customers and our government customers around the world. What's going on? What's top of mind for everybody in terms of the challenges and problems with costs and all the other things we're reading about?

Yes. So I've been in health care for a long time, but I think that this particular period, I think, is becoming more challenging, and let me just start with some high-level issues that I don't think we contemplated not in this country or anywhere in the world. First that starts with just the aging population and people are living longer, which is a good thing. But we really haven't prepared the system. We don't have the workforce. So that's becoming very challenging.

And I think most of us would say, hey, it's hard to get a doctor's appointment, so just getting into the system. And then the other part of it is we're seeing more disease. We look at in terms of mental health, we're seeing cancer happening in younger ages. So the system in and of itself is very stressed. At the same time, we're seeing costs just continue to rise.

And for governments, this governments around the world, this is just becoming unsustainable. So if we look at health care costs for states and the state budget. Medicaid is actually their largest budget item. For the federal government, it's 1/3 of the federal budget, and these are entitlement programs. So costs are continuing to go up. And I think there's a fatigue from government around how do we solve these problems.

And if you look at the history, and just in the United States, the history of health care legislation and how government has tried to solve these problems as they've thrown money at the problem, right? So they've expanded programs. They've added more services. We think about the Part D program, the prescription drug program, adding more people to our existing entitlement programs.

I think we're getting to a point kind of a tipping point of fatigue with how the system operates. And I think for the very first time, which you saw with the legislation that passed with a Big Beautiful Bill, we said, this is kind of a pushback, right?

So for -- in a long time, we've been adding -- and this is the first time the sort of a push back where I think the government is saying there's fatigue. If we look at -- in the U.K., the same thing with the NHS, right, so we can't continue to sustain these programs. And so we think about what's going on and what is the problem here?

Like what is the underlying problem? And there's lots of reasons of what's driving cost. But one of the things that I think we could all agree on is that there's a lot of inefficiency in the system. And that inefficiency kind of gets tied back to we have a lot of people. This is a very people-oriented business, doing a lot of work. And a lot of that work is very manual, it's very redundant.

So there's -- and if we go back to sort of say, why is it manual, why is it redundant? I would venture to say a lot of this is a data problem. And health care is inefficient because the data that we have is not -- sometimes it's too much, sometimes it's too little, and we don't have data at the right place and the right time.

So I'll give you a couple of examples, like prior authorizations, right? So in this country, you want to get a service, you need to have this -- the insurance company agree that this is an appropriate service and the back and forth that goes on, it's just a lot of time and a lot of energy.

And so we're spending about 25% to 30% of our health care dollars just on administrative costs. So it's just be getting to a point where it's unsustainable. This is passed, I think we're hearing that premiums are at an all-time high in the United States, one of the largest increases. So for all of the money that government has put into it, we're still seeing poor health outcomes. We're seeing more disease and the system is just becoming more and more unsustainable.

Well, lots of problems.

Yes, yes.

Long list of problems. But as I say, you can't fix the problem unless you understand the problem. So maybe you can give us a little bit of inspiration and dive into how you think AI both in health and life sciences, I mean, I think because both have some similar challenges and why you're so excited in the work that you've been leading for our team around AI? Why you think it helps solve these problems?

Yes. So that is kind of a grim view of what's going on in health care. But I'll tell you what I couldn't be more excited from where we are in health care, and that is the opportunity to use AI to help solve some of these challenges. I think we're at a point now where government can't do it, and they really do need the private sector. So we think about great moments in history, right?

So we think about what shapes history? A lot of it is government, a lot of it is war, but it's also technology, and technology has an opportunity to really reshape the future here for health. And I can't think of a better industry that can really benefit from AI because AI can help.

We talk about a lot of the people problem, the data problem and AI can really come in and help with a lot of those tasks, right, with the agentic AI, but even just the understanding of the data and having that real-time data and not just a ton of data, but understanding what it means, right? So one of the issues, if we think about it in terms of health care costs are drug costs, right?

So expensive because it takes -- it's very difficult to conduct a clinical trial, and we have, what, 1% to 2% of the population even participates in clinical trials, getting people, finding people, matching them to a trial is an enormous task, it costs the system a lot of money. And that's why drugs cost so much money.

There's a ton of money in development. But what if we could use AI to match patients, right? So the doctor knows -- you look at the patient, the doctor knows right away this patient qualifies for a clinical trial, and we can easily enroll the person in a clinical trial. Same thing with prior authorization, the doctor decides they want to prescribe a certain test or a medication, that whole process can be completely automated.

So I think there's this incredible opportunity to automate a lot of the manual work and also to make sense of the data that we have and bring intelligence to the bedside.

Yes. No, I certainly share your enthusiasm. I mean I've had the great honor of being with you in many of these discussions with governments throughout the world as well as large health care -- well, large health care organizations and frankly, small health care organizations, too, in the United States. And I think you're right. I think there's -- at first, there was some skepticism around AI, as there always is in clinical settings.

But now just with the proof points that we've delivered and the things that we've delivered even in early days, the line of people lined up to consume the AI is longer than I think that we could align this form more quickly than we could have ever imagined. And maybe you could talk specifically about some of the products that we're bringing to market and again, why we think they're so to have such a dramatic impact on this huge inefficiency and cost problem in health care.

Right. I think you're exactly right though, right? People -- I think in the health care system, they're getting to a point where they understand that the problems that they have are not going to be solved by government anymore, right? There's not going to be this big bolus of cash that solves the problem and that they need to become more efficient. And there's a lot of enthusiasm around AI and what AI can do.

I think there's also recognition that the technology that they've had in the health care system isn't going to cut it, right? You can't take a 1990s database and think that you can bolt-on AI and that it's going to work well. So the industry has been so excited about all of the applications that we're bringing to the market. In our recent health care conference, we had the CEO of the Mayo Clinic, the CEO of the Cleveland Clinic and just a lot of enthusiasm about what we're building.

By the way, they're our competitors, two biggest. Some of our competitors the biggest...

That's exactly right. I think that our competitors' customers are starting to understand the power, the value of AI and that they need a platform that can allow them to use AI to build their own agents, to bring in other agents. And the current products that are out there in the market are not quite cutting it, and they're not able to do it.

I think the other thing that I really appreciate, and we can talk about the products in a second. But I think that our approach and Oracle's approach to AI and how we are using AI in the last -- just in the last few months here has become very apparent that what we're bringing to the market is way superior. So you can probably tell them more about how we're building the AI and how that's better.

Yes. Well, we're going to show a demo here maybe in just a minute.But as I said earlier, you've got to have the data in one spot to actually have to have any kind of an AI strategy, let alone in the clinical setting. And what you said here was really important. I think it's important to underscore that there actually isn't more money to put into the health care system.

This is true globally. I mean governments and payers are running out of money to put into the health care system. At the same time, we need to deliver higher-quality services to our customers and the providers need to deliver higher quality, better outcome services, better outcome-oriented services to their patients. But they got to take a bunch of money out of the system at the same time. It's an incredible challenge. There really is no practical solution to that except for AI.

I mean we're hundreds of thousands of people short in terms of clinical providers just in the United States. We're not going to manufacture with the birth rate where it is. I don't think we're going to manufacture enough people very, very quickly here to solve that problem. So we've been thinking long and hard about how do we solve this problem? How do we deliver better outcomes? And how do we actually take cost out? How do we help our health care customers spend less at the same time? And that's an important part of what we do.

So we started -- I'm going to show a couple of examples, both the patient example as well as the provider example. So this is a patient example. So here I am as a patient, and I just got my cholesterol report. And what's the first thing you do when you get a lab in isolation, you get your doctor, you get a test, it comes back. First thing you do is paste it into some search engine or pace it into GPT or some other large language model and say, what does this actually mean?

And that's the first mistake, right? Because the first mistake is you're only pacing that particular lab and that particular metric in isolation and none of your longitudinal record. So the lab result, while it may not be normal, might still be okay clinically when you consider everything else you have going on and you deliver -- you consider your factors over time. So what we're doing is we're actually putting patients at the -- and this is another problem. This has gotten even worse because -- and a lot of states have these prompt notification laws where you actually get your test results sometimes before your doctor gets, which can be a little scary, right?

Because if you're not a doctor, you know how to read this, it's a little scary. And then we got the situation where people are going outpacing this stuff. And what happens is that the doctor's call centers are completely overwhelmed as a result, they're completely overwhelmed. So when we're actually doing this, and this is based on our integration, we announced this at our health conference recent in Orlando.

This is based upon coming back to what Clay said, it all starts with OCI. We didn't have OCI and we didn't have the large language models actually running in very close proximity to our health applications, we wouldn't be able to do this. But when you ask a question and you receive a diagnosis, we're actually considering your entire longitudinal health record. We're considering all everything about you, and we're giving the providers guidelines as to what they're comfortable and not comfortable communicating to customers, and it's all in plain English. None of it is in medical jargon.

So you can see that with these embedded engagement tools, the patient portal really becomes a basic hub for health care services. It's a trusted partner in care because it's automated using large language models. It's automated using arbitral augmented generation by taking your longitudinal health record and making the searches contextual, but it also has all the guardrails of the clinician in it so that we're not going to actually get into a situation where we're communicating something clinically that a doctor would not want to communicate themselves. And I can speak to it, I can talk to it. It's all completely AI-powered. And you can use -- you can see here, I started to type, I started to talk, and I'm asking very simple questions. How does this compare to my last test? Questions in English, not questions about LDL versus HDL versus VLDL and all these other things. And what does this metric mean and what does that metric mean? Just speak in English.

This is very helpful for patients, but it also saves the providers a ton of time and a ton of money in answering questions, direct questions from consumers in isolation. As I move forward, the integrated clinical assistant is also going to guide the patient on the content that matters most. So you get this, first, you want to figure out what's wrong, if anything at all.

The next question you had, what should I tell my doctor? What am I supposed to ask? You ask a very simple question. What should I ask my doctor? And the generative AI is going to help the customer understand and help the patient in this understand and actually make recommendations to say, this is how you should communicate with your doctor. These are the questions you should ask.

Particularly helpful sometimes for patients who -- you have a situation where you go into a provider and the provider was 10 or 15 minutes and you get out, you say, I forgot to ask that. I meant to ask this, I meant to ask that. So all of this is done before the visit. And then actually drafts a note and said, would you like me to draft a note to the doctor, so you don't forget about this when you go into the provider.

When you go into the patient -- when you go into your next appointment, the provider has all of your questions ahead of time, has all of your contextual information and patients are in a state of far less worry because they've got this whole process automated. Again, we rolled that out. And without all the infrastructure and without the GPU infrastructure and the large language models running on top of it, we wouldn't be able to deliver that whole thing as a service.

Yes. And I think it's an important piece, right? So we're looking at this portal, and let me kind of go up one level higher, right, which is to say the engagement by us bringing these tools together, by bringing AI directly to the patient, right? What we're going to see with our portal is engagement in a level that we don't normally see with most portals.

Today, most of us have 4 or 5 portals on our phone because our health care data is in many different places. So the point that you made about bringing together the longitudinal record, that is not happening today. We have our data in all these different places. So by Oracle bringing it all together, and then also bringing AI, what we expect to see in our patient portal is a high level of engagement. And then we don't want to just keep that to Oracle and our providers' customers, but we want to open that up as well.

So we're going to allow the health care ecosystem to be able to connect directly with our portal as well. So we sort of see in the future that this portal becomes this is your go-to place for all of your health care information. You should be able to connect it with wearables, different devices, maybe new, maybe weight watchers, whatever that is, that you want to connect your health care data with.

So it kind of becomes a one-stop shop for the customer. And we think we've been talking to a lot of the other health care companies out there, and they want to partner with us because they want to be able to have a place where they know patients are going to be engaging.

A lot of -- one of the biggest challenges for health care providers is that patient engagement. So I think that's the power of the portal that we're bringing.

Yes. It's been stunning to see the engagement before and after engagement with this new breed of AI technology. And on the same token, we're rolling out the same thing for providers. So if we flip the script now, and you're into your doctor's office, and one of the first questions that doctors say, "Well, what can I help you with today? "

You sort of feel like saying, "Well, aren't you really supposed to know? I mean, shouldn't you kind of know that before we show up here?" And oftentimes, the doctor has the same problem. There's an overwhelming amount of information. They can't read everybody's chart. They can't read everything and they actually want to have that same consumer-like experience with the application.

They want to ask a very simple question. What should I know about William before I walk into the room. And that's the simple question that they ask. And now they get a summary here of the patient's last visit, things that may be happening with social determinants of health, not just medical records, but everything we know about that patient, briefly summarize, so the doctor walks in to a far more engaging interaction with this customer.

And this is a preview of -- well, this is actually generally available. Now this is our new electronic health record. And you can see, just like we saw in the financial crimes investigation agent, where you have these time lines over time that at any given point, I can click on this and see what's happened with this patient over time in a very easy-to-use screen.

This is all runs on tablets. It runs on mobile phones. So the form factor automatically resizes itself. And again, it's all built with AI built in. There is not a non-AI version of this. The AI is the underlying mechanism that powers the entire EHR. We just went through, as I said, went through regulatory approval.

And you're going to see the same benefit that the consumer gets, the doctor gets the same as well. Doctor wants to know about A1c trending over time. They have the same problem where you got to stack a PDFs and a stack of papers that are very difficult to get longitudinal information about, and therefore, very difficult to have an intelligent, contextually aware conversation with patients.

All of this is now available. And if you'd like, as a provider, all the screens are really not necessary because all of it can run in the background with a listener, and that's how most of our providers have actually chosen to update the new technology. You simply place a listener in the room, the listener can be a mobile device. And the whole thing is automated.

The whole thing is automated. The whole conversation between patient and provider is automated, the orders, the labs, the summaries, the patient discharge notes. And just like our financial crimes investigation AI agent at the end, the doctor still has to sign off on it. It's just a -- it's orders of magnitude different experience for patients and providers with health care systems, and we're just hugely excited about it.

Yes. I think it's a game changer for the industry, right? So let me just dig in a little bit deeper. You talked about the patient summary, right? And that's a short summary. But think about what it's like for the doctor, let's say, they're dealing with a 70-year old patient, right? There's multiple comorbidities.

We think about the veterans administration that we're working with and they're going to extend the Oracle EHR across their entire enterprise. And this is really important because you can think about the complexity, and they're having to go through sometimes thousands of pages of health care information.

And a lot of times things get missed or we're not addressing something. We don't -- the data is there, but the physician may not see it, it's a very short period of time. So the ability to have that summary is a game changer. And like you said, the SDOH, all of those different things.

And the automation, I think the other piece that's important is the autonomous coding, right? So the other thing on the screen there is that it'll recommend which codes. And this is an important piece because if we think about it for the health care industry, we're spending 25% to 30% on administrative costs, a lot of that is around billing. And the cost to collect for providers is about 5%.

So all of that money, just to be able to get the bill paid, and we're automating that entire process. So if we think about it, the whole billing process, at least in the United States, was built off a paper system, right? And then when we went to digital recorders, we just basically took that process and digitized this process. But with AI, we don't need to do all those things.

We don't need all of the different middlemen to be able to get this claim paid, right? We can actually do all of that with AI. And so our EHR sort of starts as, it's almost a misnomer to call it an electronic health record because it's really a system of intelligence. And it goes way beyond just the patient-doctor interaction. We started there. But we're trying to address all the problems across the entire ecosystem.

So that autonomous coding piece is really solving a lot of the friction between payers and providers. Not only are we going to try to make claims payment easier, but even just the prior authorization, all of those things that happen between the payers and providers, we're introducing those solutions.

And it's ironic that it's actually the payers that are coming to us and they're saying, look, after the big change event, there's a lot of concern about security and so being able to have the Oracle Cloud in the middle of this, helping with those transactions is really important to the industry, it's safer, it's more secure and also we can use AI to automate a lot of those processes.

So that's one piece on the payer side. And then there's a whole other piece with clinical trials, right? So if we think about it, most clinical trials and a lot of research is only happening a big, large academic centers. So if you're lucky to go to one of those, you may get connected with a clinical trial. If not, you never know what's going to happen.

And so what we're able to do with AI is we're able to do that matching inside the clinical trials. But where we're going is to be able to use that EHR not to actually do the clinical trial right inside the electronic medical record. Today, with clinical trials, what's happening is that they have to go to an entirely different system. So they actually physically move the data from the electronic health record into the electronic data capture for the clinical trial.

So you can imagine a lot of the back and forth work and the retyping and you got to have somebody to check in all the information. So our electronic health record is not only addressing, it's actually providing payer solutions and also the ability to help clinical trials and research on that end of the spectrum. So it's doing more than just helping the doctor and the patient, it's a much broader view.

Yes. Well, that's terrific. Thank you for giving us hope and for sharing the solutions because I really do agree that bringing this whole consumer-like experience to health care, both for patients and providers is going to be game changing. And obviously, your passion and knowledge in the space has been a big driver in helping us get to this point.

Something else interesting happened while we were bringing all this data together for health care providers. And that is we actually started to realize that there was more utility in the data in the collection of data on the AI data platform that we've used the power of these health care systems that just for the providers themselves.

But actually, as we become intimately close to these providers, we realized, particularly in the United States, that a lot of providers, a lot of health care providers have cash flow problems. And for those that are publicly traded, when they report out, they report out their cash flow in days -- report out their available cash in days, not weeks, months, years, but days. That's how much cash they have on hand.

And they've got to make decisions around payroll, ICU -- equipment for the ICU. These are real decisions that very big health care systems have to make. We looked at the data that we had. And we thought this might be -- and we asked questions about how do you finance these systems, how do you -- banks don't really know a whole lot about what we do, they don't know about a receivable flow.

And it's not that they had a receivables problem, but they actually just have a cash flow problem. It's not that they're not going to get the money, but when they're going to get the money is an acute problem, no pun intended for health care systems.

So we see the same kind of transactional flow happening across the board in our whole applications. You can see some of the stats there. And -- but we particularly leaned into a couple of industries, and we're going to talk about a few of them here, where the traditional financing models are very complex.

And the financing models really are complex because the banks, the financer in this situation does not have access to real-time data and the data that they do have access to has very little qualitative information about it. Very little qualitative information associated with it. So we, through the course of conversations with the health care providers, construction companies, we're going to talk about a few different examples here and the banks, we actually found out that the AI data platform is a wonderful vehicle for banks to learn a lot more about their customers and for customers that have a much better relationship with their banks.

So we're going to talk about embedded applications that benefit -- why it benefits the banks and also why it benefits the business or, in this case, the health care provider.

So with that, I'm very pleased and honored to have Lia, who is the Managing Director of Global Head of Payments and Embedded Finance at JPMorgan; and Geoff, the Managing Director and Head of Global Lending for Trade and Supply Chain Finance at Bank of America to come talk with us more about this exciting platform.

So thank you, Lia. Thank you, Geoff, for being here. I gave a very brief example, a very brief introduction to embedded finance and a little bit about health care, but I know it's more than the health care. Geoff, maybe just go down a little, I'll start with you. Could you give us a little bit of a preview of why this is so exciting and what we're actually doing here together to better serve our collective customers?

Yes. So Mike, as you alluded to before, so if you think about health care providers, they are our clients as well. And for years, they've had working capital issues and they come to their bank and say, look, we've got a working capital issue. We look at the model and we say, okay, it's not a payer problem. There's no shortage of customers. And what you realize quickly is what you have is just a working capital problem or a cash flow problem. And if you think about the way that banks help our clients solve these things, each industry becomes kind of a little bit different because it's got its own characteristics. And we think about health care.

Health care, the receivables can be hard. So the banks can do certain things, certainly of credit lines. You can, in some cases, hospitals can access the municipal bond market. But some ways that we traditionally help to free up cash flow and accelerate working capital are difficult for the banks because we don't have what we need in terms of like, okay, who's going to pay this and when and how much.

And for years, we chased that data, right? And Seema talked a little bit earlier about the data that's being followed to get a better health outcome, well, we at the banks, we're sort of chasing that same data. Somebody could think, well, what is the bank care about a certain procedure and how much of that gets paid and when.

And the reason that's important to us is because if we've got this receivable and we're trying to accelerate cash, we need to know what's going to be paid and when. And it became clear to us as we went through this process. And Mike, you alluded to this earlier that some of the data that you had aggregated, consolidated, and then we're very good at being connected to was exactly the same data that we would need to provide additional working capital options.

And as we thought about that, we said, okay, well, if we could do that, then we could layer in some of the AI predictive models on top of that to be able to say, "Okay, well, now we think we know with a bunch of procedures that happened what the expected payment will be and when, right?

And any of you that have experience with health care know that the original invoiced amount has very little relationship with what's actually paid at the end of the day. So we needed a way to get from point A to point B. And it became clear to us also as we saw that the availability of data that was coming in that Oracle had access to, not only that but also you were very good at layering on that AI at the end of the process to say, we're going to give you a predictive model that tells you what's expected to be paid.

And that solves our big problem, which is we don't know how to get from point A to point B in terms of what's invoiced and what's paid. And by doing this, we're able to help accelerate cash.

Thank you. Your perspective, Lia.

Yes. First, great to be here. So at JPMorgan, we are partnering with Oracle to embed payment services and financial services directly into the Oracle ecosystem. So that in Oracle solution like industry applications, Oracle can provide those payment services and relevant financial services to their clients directly in a very seamless, integrated and scalable way.

And I think for the banks, in my view, is a very meaningful evolution of how financial services are accessed and distributed, right, really through a platform approach. And by doing so, I think there's tremendous benefits to the -- our clients and to Oracle's clients as well, right?

I think of it as a very streamlined automated workflow so that the end clients have ultimate data and visibility and transparency of where the payments are. And then to Geoff's point, also the data plays a huge role here in delivering those benefits, especially in our strategic partnership, right? As I think about at JPMorgan, we -- in the Payments division, right, we processed about more than $10 trillion of payments every single day.

So we have a tremendous amount of data around who pays whom and when and how and where and in which currency. And that, coupled with the enormous amount of data in the Oracle industry applications. With that partnership, we know a lot about every single client. And I think that is a very powerful way to serve our clients to have that intelligence to serve our clients.

Yes. That's very helpful. Thank you. Geoff, when we started -- first started talking about this partnership, you gave me some stats about how much money was sitting on the sidelines in each one of these industries waiting to invest. But the problem was that the people willing to invest didn't know enough about the industry and couldn't get enough of the qualitative metrics to feel comfortable about this investment.

So this just changes that, we think. But ultimately, how does it benefit you as a bank? What's the ultimate benefit to the bank? How does it change your unit economics? How does it change your investing philosophy?

So we can kind of look at it along 2 vectors. So the first is we've got clients who are looking for better working capital management, so we can help them with that, right? I think that's the first thing, right? As we serve our clients, and we think that's a good thing because as you both have noted, you've got hospitals in the United States alone that are struggling financially, right?

And it's not because of lack of customers, it's not because of lack of high-quality payers. It's for free cash flow and working capital reasons often. So we can help.

The second thing is part of our business is connecting our clients to investment options, right? As we think about this, we think about the opportunity to take some of what is out there, predictable receivables in a way that hadn't been there before with high-quality payers, we can translate that into investment opportunities for the capital markets.

Now I can tell you the capital markets have, to your point, Mike, been waiting on the sidelines for great investment options that are always looking for things that are all over the risk profile. And for us, to have ways to be able to wrap things like receivables and say to them, look, not only do we have this investment option for you, but it's coming from a bank in a regulated environment with a risk management profile under a regulated environment.

And we've got it, and we're putting it out into the market, that means something to an investor, right? It means, okay, so there's been work that has been done behind the scenes to get us to the point where we think we have an investable asset. And that work, again, I think the reality was 5 or 6 years ago, there just wasn't enough out there.

We were chasing data in a 1,000 different ways. And people underestimate, I think, what I always call mechanical side of that. We can do credit analysis pretty easily. But what we can't always do is say, like, how will the mechanical parts of this? Where is the receivable being created? Where is it housed? How will it get from wherever it sits to the bank systems so that we can then do something with it and wrap it and send it to the market, all of that stuff is impactful to an investor, right?

A smart investor looks at all of that and says, look, I don't want to take my chances that this receivable becomes something else in the course of its lifetime. And I think as we track just historical data, performance data, all of the data that becomes available, all of that starts to become solved, and then we've got investable assets. So we've got on one side of it, we're helping our clients be better payers and on the other side of it, we're creating investable assets for our clients.

So are these investment products that you can create as a result, are these products that you previously couldn't create or you had to create with a different rating or different risk profile associated to them?

It's a really good question. So we're in the business of pricing risk, right? So part of that is, as I said, the credit risk, which is easy, but the other part of it is just as important, and that's -- people underestimate that side of it. And I think the reality is if the receivables were really hard the banks just kind of sat on the sidelines. So as you think about just some industries like health care, like construction, where there was uncertainty of payment from a bank's perspective because remember, by the time the data got to us, we only had access to small portions of it, right?

So all of it just looked like it was very unreliable. So in the context of a regulated risk managed bank environment, there just wasn't a way we could get comfortable with the predictability of outcomes. So for us, we put -- park that part of our toolkit and then we moved on to what we could do, right, create a facility or a municipal bond market or whatever it might have been. But that's the difference between what we had before and...

Yes. I think as we've talked, Geoff, it's been quite interesting to understand the bank's philosophy and all this is it's not that -- and we're going to talk about construction and retail and more than health care in a second. It's not that the bank necessarily is interested that Medicare is going to reimburse the provider for a hip replacement, but what you're more interested in is to make sure that, that person doesn't get readmitted to the ER in the next 30 days with an infection as a result of that because that actually changes the reimbursable for Medicare in a value-based arrangement.

That's the level of data that you need real-time. And obviously, for 1 patient easy, millions, hundreds of millions of patients, we'd have to scale that over time, very difficult to do without an AI data platform, kind of constantly monitoring and scanning for anything that could change the risk profile as well.

And this is to your point, people might think to themselves, well, would a bank be interested in that? We're very interested in that because it's predictability of outcome. It's the predictability of us getting repaid and an investor getting repaid. So it's very important to us.

Might just going to add one more thing, right, before we lead health care, right, is that the government in and of itself is now moving more towards those value-based payments, more capitated payments where they're not -- providers are not getting paid on more real-time. It's more after a year if you've taken risk. And so for providers, the need for cash flow is even -- has really increased dramatically. So I know our providers are very happy about this solution as well.

Yes. So many -- we could go on for a very long -- but just on health care alone, there's just so many interesting dynamics. And Lia, I'm going to shift gears for a little bit. Retail, restaurants, construction in all of these other industries where we're managing all these transactions, either at the point of sale or payment or money changing hands using the Oracle platform technologies today across so many different industries. How do you see the different needs in embedded payments across these different verticals industries, how do you see that evolving? And why is the automation so important to you?

Yes. No, great question. I think across all the industry verticals, I think there are common themes no matter which industry vertical, we want the payments to go out in a very secure way. I think we want to know or Oracle and the clients want to know the transparency where is my money. If something goes wrong, they want to know where the payment is stuck, right?

So I think there are foundational basic needs of security, transparency, right, and the optimization of the payments. Those are universal. Then to your point, Mike, for different industry verticals, they're nuances. We talked a lot about health care. Seema, you mentioned, it's a very complicated ecosystem with payers and providers and regulations and HIPAA compliance. So that really implies a lot of the payments handling, right, for data privacy and whatnot.

You think about another vertical consumer retail, you mentioned is really fast evolving into already omnichannel in person, in-store experiences with in-app purchases with online, like how do we deliver that payment experience in a holistic way. And today, we talked a lot about AI and people talk about agentic commerce. And in the future, maybe all the agents will do all the purchasing and whatnot on your behalf.

Then we need to think about fraud and authentication and liability shift and all those very interesting future business models or current business models that are evolving very rapidly I think with the power of data combined in our partnership and with the AI models on top of the data, that's where what I mean earlier, like the payment intelligence, right?

How do we deliver that insight into Oracle clients so that they know the fraud prevention, they know how to position cash, how to really mobilize all the transactions to add value to the ecosystem.

Yes. And so in some ways, this is automated embedding into the applications it becomes another distribution channel for the bank -- compliant distribution channel. We know a lot about the customers, we can automate KYC, all the things that need to be done. We've got that data. All of that data available as part of our industry applications just becomes a wonderful distribution channel and frankly, far easier for consumers to consume because it's just built -- it's built into everything that we do.

Totally. It's like a prebuilt enterprise-grade or banking and the payments core that's already compliant and extensible. So I think that is really where we're headed in cocreating this model.

We appreciate it. So I'll ask you both the same question. We're not the only technology vendor in the world. Why partner with us, why Oracle? What's special about this relationship between Oracle and the banks?

Look, you go with what works, right? And so there is -- as I said, we spend a lot of time chasing data to try and sort of predict outcomes, right? As I said, we're in the business of pricing risk and predictability is a big part of what we do. And I can tell you, we've burned a lot of calories on this in the banking world trying to sort of figure out how we can get from point A to point B in certain industry verticals.

One of the things about, as I said earlier, when you -- when each sort of financing opportunity grows up in a certain industry, it becomes bespoke to that industry to a large extent. And it's hard to move from one to the other, right? You have to have a level of expertise in each industry.

And what we realized, I think about Oracle is you're doing a lot of the really hard work. We don't always need to know when we talk about health care, like following data to get to a better health outcome is harder than what it is that we need to do. We're just sort of looking for -- at the end of the day, we need predictability of payment. That's a part of it, but it's not the hardest part of it, right?

So you guys are doing the hardest work, but a lot of that is stuff that's really, really useful to us. And what we found is it doesn't look that different in health care as it does in construction, as it does in retail. So now not only have we been able to solve and unlock some of these industry verticals, we've actually been able to migrate from vertical to vertical because we realize like predictability of outcome in an Oracle built model often looks the same across industries, but it's the same usefulness to us.

I think it's a terrific point, Geoff. We're not building a financing vehicle that is specific to an industry. We're building a platform that has industry context built into it, it's going to give you the data that you need, but actually scales across all the industries that we serve, right?

And that's not what we expected at first, right? We expected to kind of stay industry by industry. And it was only after going down this road a bit that we realized, this is more scalable and valuable than we thought it was, which I think is going to be a really good outcome.

You're right. I think for us, the partnership with Oracle is really critical. In terms of we share a lot of common values, right? I think Mike you talked about in your keynote just at the beginning of the conference, right, scalability, security, full stack advantage. I mean those are things that we look at our JPMorgan's platform, we share the same value.

And then we think by partnering with Oracle, that's a force multiplier, right? With the scale is just going to be that much more even more global, more scale, more extensible. And the security is really a critical, critical component. I think you also mentioned, right, whenever you add in a new component into the system, there's risk by integrating the 2 global platforms in a very digital right API-driven way, in a modular way, I think that creates this tremendous platform and infrastructure layer to serve our clients even better and more seamlessly.

Yes. Well, that's much appreciated. And we do resonate with the shared value. So I visited both of your organizations in person many times now and even -- but on the trading floors. And I can tell you that the passion is very contagious for this access to real-time data. So much appreciate it.

And to be clear, we're going to market together right now in construction in restaurants and health care and certainly plan so much more to come. So I'll finish where I started is that good things happen when the data is all in one place. And our ability to aggregate operational data, both qualitative and quantitative data and to deliver that as a service to you so that you can deliver that as a service back to our collective customers, we can't thank you enough for being a part of that. We can't thank you enough for pushing us and inspiring us and helping us to find this product, and I'm so excited about what's to come. Thank you, everyone.

Thank you, guys.

Thank you.

We will now take a short break. Lunch is located next door. Our programming will resume momentarily.

[Break]

Please welcome to the stage, Larry Ellison.

Hi, everybody. Let's see. Okay. So everyone is very excited about AI. It does extraordinary things. In fact, my son and I were just consulting it on a legal matter this morning because there was -- anyway, there was an interesting dispute between a couple of investment bankers as to certain rules about how much stock foreign corporations could own under certain circumstances and quickly ran off and asked a multimodal AI model and got an answer almost immediately. They're quite extraordinary. They certainly know all the laws in the U.K., all the laws in the United States, the rules of the New York Stock Exchange, the SEC, all of that, is trained on all of that data. And all that data is publicly available.

What is less common and what people want to do, but really can't do very easily right now is use their AI models, ChatGPT, Grok, Llama, Anthropic, all of them, to reason not on publicly available data where the models have been trained on the Internet, all of the Internet, but private data, the private trading records of an investment bank on Wall Street or the private genomic data of a genetic engineering company. that's analyzing genomes. And genomes, it's an enormous amount of data, and they'll do gene sequencing on plants. I'm actually going to -- I'm going to show it in a minute because I think it's such an interesting example. We're working on a lot of plant genetics, and I'll describe a couple of the projects we have ongoing.

But people don't realize how big plant genomes are. The human genome is right around 3 billion base pairs -- actually a little short of 3 billion base pairs. The wheat genome is 15 billion base pairs, 15 billion base pairs. Because wheat has been around a lot longer than human beings, and wheat has been evolving over time. And when you sequence wheat, you get this massive information. And by the way, there are all these different varieties. I didn't know this a year ago -- 2 years ago.

There are all these different varieties of wheat grown all over the world based on variations in soil and variations in climate. And they all do things slightly differently. They do photosynthesis slightly differently. The genomes are different. And what you want to do, if you want to optimize wheat, if you want to increase yield or make it drought resistant, you want to really look around at all the varieties of wheat and analyze and understand the genomes of wheat. And you've spent a lot of money to sequence all those wheat plants. And that's your proprietary information. And your business is to produce this new variety of wheat that is going to be drought-resistant, higher yield, and that's your business, and you don't want to share that information with other people.

How do you do that? How do you do that? Oracle ran a project inside. First thing we did, we took all of our customer data, all of our proprietary customer data, and we have a lot of customer information. We have a lot of people using Oracle in the cloud every day where we keep track of what they're doing. We're curious about what features of Fusion applications they use more frequently, what features they tend not to use as frequently, what features require -- what features are not so easy to use. People tend to make mistakes on that. So we monitor all of this, and we find someone's making a bunch of mistakes using a particular feature, there's probably -- the feature is not easy enough to use, and we want that insight, and we want to go ahead and fix it.

But how do you get the -- back to the basic problem, the basic problem that has not been generally solved is how do you take these fabulous reasoning models and allow them to reason on your private data, whether it's plant genomics or customer usage in the cloud and what features are easy and what features are error-prone. How do you take your private data, make that available to AI models while keeping that private data private? And can we make that easy to do? Because everybody, everybody wants to do that.

And we've been working on this problem for some time. And we call it -- and we have a new version of the Oracle database called the Oracle AI Database. And we didn't name it the AI Database just because AI is fashionable, as I said the other day. We did it because the AI -- the new -- our new database has a lot of new features to solve this problem, to solve this exact problem, solve this one problem. How do you make all of your private data accessible to AI models for reasoning while keeping that data private without compromising data privacy in any way. And we've done that, and we call it the AI database.

And by the way, just the latest version of the Oracle database. This is not an all-new database. This is not an AI database in the sense of Pinecone is where -- it's just a vector database for AI. By the way, we added vectors to our database, but it's still the full Oracle Database that has all the Oracle security features, all the Oracle high-performance features, all the recovery features, all of that. But we've added all of this AI capability to the Oracle Database. So it's highly secure, highly reliable, highly scalable, very fast. And -- but it makes your private data easily accessible by the AI models, okay?

And again, we also decided that if we're going to do that, if we're going to make the data -- we actually should have -- make it easy in the Oracle Cloud for users to pick the AI -- their preferred AI model. We use a variety of AI models. I mean, again, Anthropic tends to be pretty good at code generation. So if you're doing programming, Anthropic is pretty good at that. ChatGPT is a phenomenal legal expert. I can just go into this. These models are somewhat different. And depending on the application, you might use one model, or you might use a different model. So we decided to make all of the popular AI models available inside of our cloud. So Oracle, if you go to OCI, you can get ChatGPT 5.0. You can get Google Gemini. You can get xAI Grok 4 Heavy. You can get the latest versions of Llama from Meta.

You can -- and because the Oracle AI Database, to be an AI database, to do reasoning. Oracle doesn't do the reasoning. You need the model on top of Oracle to do the reasoning. So when you configure the Oracle AI Data platform, you pick one of these models, and we then give you a private version of that model, sitting on top of the Oracle Database with your private data in it. And so it's all there. You just configure the model you want. I mean basically click on the name of the model you want, we configure it and put it on top of your database.

And then the Oracle Database will make all of the data -- all of your private data that you authorize, it will make it all available to the AI model. How does it do that? Well, it uses a technology called RAG, retrieval-augmented generation. It simply allows the AI model to read the database. There's a server, there's an MCP server. The AI models are designed to be able to go and read anything on the Internet that's publicly available, but they also can read private data, whether it's in a file system or in a database, different kinds of databases. It can do all of that. It can go out and look at that data, read it and actually understand it.

So that's what we did. I'm going to go into a little bit of detail on how it works. The -- what we added to the Oracle Database, obviously, we had a RAG capability. But right now, the Oracle Database can vectorize all of your data in the database. It can vectorize text, it can vectorize images, it can vectorize videos. And what AI models understand -- the format of data that AI models understand are vectors. And for example, a very famous AI search is, "Vectorize a particular movie and then show me movies that are similar in content to that movie." And that's a vector search. "Find me a movie where the vector is similar to the other movie or other movies somebody else watched." They use it for recommendation engines.

But also for gene sequencing. For gene sequencing, there are genetics -- there are gene sequences to do with photosynthesis. So you find a set of sequence -- gene sequences that are doing photosynthesis, and you say, "Find me all -- and all the other genomes that I have, find me the gene sequences that are involved with photosynthesis." And we can find that part of the genome, searching through the 15 billion base pairs, we can immediately zoom in on that part of the genome. And then you can say, "Show me the differences between how this plant does photosynthesis and how this other plant does photosynthesis." And you do that with something called vector search. And there's a whole bunch of vector mathematics that has vector spaces and vector distances and all of that.

But that's what we do. That's what we've done to make this an AI databases, is allow you to vectorize all of your data, all the different types of data. Once you've done that, if you look at what we were able to do, as I said, the first project doing this inside of Oracle, the first of our private data that we decided to make available to an AI multimodal model was our customer data because I'm not sure, for us, there's anything more valuable than our customer data. The -- and then we started asking questions once we vectorized our customer data.

And for example, you'll see what I mean by valuable, what Oracle customers -- this is my second group -- under the second bullet -- first line under the second bullet. "What Oracle customers are likely to buy another Oracle product in the next 6 months?" We'd like to know who they are. And we'd like to know -- second line, "What Oracle product are they most likely to buy?" And those are the kind of questions you can ask using a reasoning model. And you get -- then it tells you.

And then you can ask, but you can also have agents associated with this. It doesn't just have to be ask a one-step question. You can actually ask us, okay, well, let's send e-mail to all of those prospective buyers and let's show them the three best Oracle references in your -- for you, in other words, in your industry, in your country that are -- bought the same product and used it successfully. So as you look at this, I mean, you would build -- if you were a company building CRM software, or sometimes called CX software, customer engagement software, the ability to enable your customers to ask these kinds of questions, to build an application -- CX application suite where you could do this kind of AI marketing, if you will, AI reasoning on top of the customer data and qualify leads and then pursue leads with agents. You qualify leads through the reasoning process. And then through the agent process, you would go ahead and pursue those leads. That's how you go about building the next generation of CX applications, and that's what we're doing.

And we do all of that while maintaining the strict privacy of your customer data or your genomic data or whatever. Medical genomic data is very, very sensitive. So you have to keep it private. So the Oracle Database security model, which we've worked on for decades, is what we rely on to keep this private because the vectors are inside the Oracle Database. And we use a security model, again, that we've been working on for a long period of time to make this -- keep your data private while making it accessible to AI models.

Next slide. I'll press my clicker button. Great. Okay. So as we're building the next generation of CX products, we're not -- we are the owner of Java. The Supreme Court might disagree with that. But -- and Google probably would, too, because they won the case. But we bought Sun who developed Java. We are the primary maintainer of Java, and it is the world's most popular programming language to this day.

However, in the age of AI, you can use something like Anthropic or ChatGPT or Grok to write code. And you can just declare your intent what -- in other words, say what you want the program to do, and the AI will generate the step-by-step process to do it. So it won't be conventional coding. It's called vibe coding, an interesting term, which is just, I guess, [ steel ] vibe, very modern term for coding. But it's really just what do you want the program to do? You just tell me what you want the program to do and don't worry, I'll generate it.

Now we have been doing this for a while, doing code generation for a while. And there's a big debate inside of Oracle, and there are people on both sides of the debate. I'll tell you what side I'm on, which is programming in English. In other words, declaring what you want the program to do in English. English is a notoriously imprecise way to communicate. It's not like mathematics. There's a lot of ambiguity in English. To make English perfectly clear, perfectly precise is very difficult.

So we think you could have an alternative declarative programming language, which was designed to declare intent of what a program should do. You could create that language with great -- and then with great precision, you could generate the code. Now the jury is out. People are programming and declaring intent in English and generating code, and people are declaring intent in specialized languages, to declare intent designed to be precise and designed specifically for code generation. I'm -- having done both, I'm a great believer in a more specialized language. I think once you learn the specialized language, you're much more productive in generating code. And our experience is -- I mean, these are huge, huge improvements in productivity, 10x productivity if you generate the code rather than writing the code.

One of the reasons we felt we could take on something like Cerner, and we knew we were going to have to rewrite all the Cerner code. We have to rewrite it all. And it took decades for Cerner to write that code. And we thought we could rewrite it all in a few years because we weren't going to rewrite it. We were going to generate it using AI. And that's what we've done and built very complicated and interesting agents using AI. And in fact, in the future, what are computer programs? Computer programs are a collection of agents connected by workflow. That's what they'll be.

Okay. This is a big leap from code generation and making your private data accessible to AI models. This is a greenhouse that we have developed. It's version 3 of a greenhouse that we're developing. Actually, it's Danny Hillis' team. Oracle made an investment in what's called Applied Inventions. Danny Hillis, by the way, invented thinking machines when he was undergraduate in MIT.

Actually, Danny and I used to be competitors. I was working on a computer out of -- with the Caltech team doing something called nCUBE, which was massively parallel computing. It was parallel computing. It was, if you will, the precursor to NVIDIA and vector processing in computing. It was doing a lot of calculations in parallel, which obviously is very important now. Unfortunately, as Danny and I discussed because we both failed, nCUBE failed and thinking machines failed. We both lost a lot of money and a lot of time on this. We learned a lot. But we were about -- I was going to say 20 years too early, but it was more than that. I don't want to talk about it. I'm going to do the calculations.

But anyway, Danny is now -- we bought his company. And one of the things that we're working on is -- and he's really responsible for, amongst other things, robotics inside of Oracle because robotics is a very special case of AI. The leading AI model for robotics is very easy to figure out. It's owned by Elon Musk, but it's not Grok. It's the AI model he built at Tesla. And he -- the first popular common robots in the world -- I mean, yes, there are robots -- there are general-purpose robots. Yes, there are robots that assemble PCs and laptops and desktop PCs and a lot of electronics and iPhones and all of that. There are tons of special-purpose robots.

But the first, what I would call kind of general-purpose robot is the self-driving robot that Tesla has created. And then Elon is working on his second group of robots, which are humanoid robots. So he's going from 4-wheel robots to 2-legged, 2 armed robots, general-purpose robots using full AI model, real-time AI models. And we're doing -- we're not creating the AI models from scratch. We're using those models to automate a variety of things. And this is a greenhouse, by the way, that has no people in it. That yellow thing over there, they're very large.

The reason there are no people in it, there are a few reasons. One, is the atmosphere inside of this greenhouse. By the way, there is no structure. This is -- this building is held up by air pressure. So there's positive air pressure inside of the building, and that holds up the roof, which is ETFE, which is a kind of plastic that lets through -- it's the most transparent material in the world, let's through more light than any other material, more light than glass or other forms of plastic.

And if you talk to Danny and said, what do greenhouses do? Well, basically, they convert sunlight and CO2 into food. And we don't let people in here because people can contaminate the plants and people aren't going to like all the CO2 and all the humidity we have in here. And the robots will move the plants from the growing area into the harvesting area. And AI decides when you harvest the plants, there are cameras, AI decides we can grow different crops in here.

The only greenhouse -- you have lots of different crops inside the same greenhouse because it's completely computerized. And the nutrition, if you're growing strawberries is very different than the nutrition when you're growing lettuce. The heating requirements, we heat the plants from the bottom. We don't heat the whole building. Anyway, I'm not going to give you all the details. But each area is carefully climate controlled, atmospherically controlled. The hydroponic nutrition is all computer-controlled. And it's designed to produce food at a much higher quality and a much lower cost than currently other forms of vendors are growing.

By the way, when you grow indoors, you use 90% less water, which is truly incredible. By the way, this is also a habitat from Mars. So if you think about an interesting use for a greenhouse, I don't think it's a huge market. I don't think our Martian market is going to be gigantic. It's not in our numbers, by the way. None of the Martian consumption of our technology is in any of our numbers. That's all upside.

So the -- what does the greenhouse do? The greenhouse grows food. And if you're going to have people living on Mars, you need to grow food. But it also converts CO2 into oxygen, doesn't it? So if you're going to have people living on Mars, you're going to have to -- they're going to have to breathe, and you can either ship the oxygen from earth to Mars or you can create the oxygen from CO2 on Mars. And the people can create the CO2 that the plants will consume. So this is a kind of a combination. Its intent, by the way, our first market, we really have focused on earth. But just notice that this thing would be a perfect habitat for Mars or other places you want to go. As I say, it's really -- it's not in our business plan.

This is what the whole building looks like from the outside. The green areas are the harvesting areas. The robots will move the plants into the harvesting and packaging areas. And again, there are no people in the growing areas at all. It's kind of interesting.

As long as we're talking about plants, I want to give you an example of the kind of things we're doing with our database, to make it work better with AI. And in addition to being able to vectorize all of the data, which makes the data easy for AI to consume, we've also created special data types during the -- let me give my verb tense correct, are in the process of creating special data types for DNA. And say DNA search is -- a DNA search -- by the way, that is wrong, and it's my fault. It says, "The wheat genome is 15 million base pairs." It's 15 billion base pairs. Human being, 3 billion base pairs.

And it's got all of the historic genes, all the genes that became obsolete during the hundreds of millions of years of evolution of wheat on the planet Earth. It's just a grass. Wheat on the planet Earth is recorded in that genome. It's why it's so large. We don't delete when genes suddenly stop being used, they don't get deleted. So there's a whole history of how the wheat evolved. That's very interesting that you capture when you gene sequence wheat or you gene sequence plants or animals for that matter, you capture their evolutionary history.

And the things we're doing -- another team is doing, not Danny's team, is looking at wheat photosynthesis. And if you can improve wheat photosynthesis, and they have -- they have to improve wheat photosynthesis using AI. They've changed the -- they've used CRISPR-Cas9, a gene editing technology, to improve photosynthesis. When you improve photosynthesis, you convert more CO2 and more sunlight into food. So the yield per acre of this wheat we created -- they created. I didn't have much to do it -- I didn't have much to do with it. Actually it has 20% more yield per acre. So we're producing more food in the same space.

Another thing that we're looking -- and that's already working today. So the increased yield through -- using AI to figure out how to improve photosynthesis. And once we figure out how to do it, you use CRISPR/Cas9 to go ahead and do it and actually do the gene edits. The thing we're looking at right now is, not just converting CO2 into food, but also converting CO2 into calcium carbonate. I'm sure all of you will be fascinated by the fact that the coral reefs are made up of calcium carbonate. The plants that live in coral reefs actually -- they actually secrete calcium carbonate, their own skeleton. Well, we have skeletons too, right? They also have a lot of calcium in them.

We are now engineering a version of wheat that will convert CO2 out of the atmosphere into calcium carbonate, an inert version of CO2. Therefore, you can take vast -- you can have varieties of wheat that take vast amounts of CO2 out of the atmosphere and deposit it in the ground as little, tiny microscopic pieces of sand -- calcium carbonate sand. And you can manage the level of CO2 in the atmosphere pretty much to whatever you want it to be at no cost, at no cost.

And the world right now is working on this problem of figuring out what to do with CO2 in the atmosphere, and they have all these interesting ideas, which -- including getting rid of all fossil fuels, which is very difficult. And -- but there are other ways to tackle the problem that might be much easier and much more cost effective. And then, by the way, if you look at the markets, you can actually figure out if you're a farmer, you can use satellite imagery, AI and satellite imagery, to look at your wheat fields and figure out exactly how much CO2 you're taking out of the atmosphere, and then you can get carbon credits. You can trade it for carbon credits, okay? And kind of an interesting approach to -- and a much simpler approach to managing CO2 in the atmosphere than what we're currently doing.

Another one -- another interesting example that we're working on, again, with plant genomics using DNA data types, using vector search, for using all these AI capabilities. It's -- right now, we fertilize. Almost all the plants that we fertilize in North America and Europe throughout, we fertilize them with nitrogen fertilizer, huge amounts of nitrogen fertilizer. Because -- even though there's tons of nitrogen in the atmosphere, nitrogen is the most common element in our atmosphere. We breathe it in. We don't use it, but we breathe it in every time we take a breath.

But some plants, soybeans, for example, actually get their nitrogen directly from the atmosphere. Well, we can engineer corn or wheat or other plants to get their nitrogen from the atmosphere, so we don't need fertilizer. And fertilizers do incredible environmental damage. During rains, it runs off into rivers, a bunch of nitrogen runs into the rivers and into lakes. You get these big blooms of plants in the lake and in the rivers. And plus, a lot of farms can't afford to buy fertilizer. Therefore, their yields are half what the farm yield should be. Well, we can fix that. We can fix that by having versions of corn, version of these grains that fix the nitrogen from the atmosphere.

So that's my -- that's my presentation on what we're doing with the Oracle Database. The primary thing, make it easy for people to use AI models on top of their private data while keeping it private. And then we have all of these advanced technologies that we're adding right now so that as -- if you are in the business of genomics -- and by the way, that's every pathology department, that's all of medicine is in the business of genomics. All of agriculture is in the business of genomics. These 2 enormous businesses, you have to fully -- your database have to fully understand DNA and have operators that can operate on these enormous data types -- these enormous genomic data types. That's what we've been doing. That's what we're currently doing.

I think the -- I think, again, fully ignoring Mars, there's enormous upside to our database business over the next 5 years. We think it's going to be one of our fastest-growing businesses. And we don't think -- and we're happy to talk about this later when we -- in Q&A. We don't think we have a lot of competitors left in the database business. It's fascinating. Our primary competitors in database, let's say, the Databricks, you could say, is new -- or it's kind of the newest one. Snowflake, they don't even do transactions. They're query-only systems. There aren't a lot of new database technologies being invested in.

We're kind of the only game in town. We think that gives us an enormous opportunity to increase our franchise in the database business over the next 5 years and the dawn of the AI era if we can merge our database technology with the latest AI technology. Thank you very much.

Please welcome to the stage Doug Kehring.

I think Larry used the magic word, upside. So I'll be presenting our updated financial outlook, which I know I think everyone has been anxiously waiting to hear. But let's start off with the usual exciting stuff that Ken went over earlier. It's funny now that they're asking me to certify the financials. This actually means a lot to me. So please pay attention. Okay. This one reminds you about our forward-looking statements. And this one is reminding you that we'll be using some non-GAAP financial measures in my presentation.

Okay. You've heard the strategy throughout the day from Larry, Clay, Mike and the rest of our management team. We have an unbelievably strong and deep set of enterprise technologies for both cloud and AI. And we have the vision, leadership and experience to execute this. It's now time to see how all of this impacts our financials.

Building on our long-term expectations can be boiled down to a few simple steps. First, we work extremely hard to turn the customer momentum we are seeing as evidenced by the amazing enthusiasm you probably -- you've seen here at AI World into an accelerating RPO backlog. Second, we then deploy our operational expertise to provide capacity to customers that help us turn that backlog into accelerating revenue growth. Third, we leverage this growing footprint, scale and the utilization rates of our data centers to turn this larger revenue into profit growth. And the result is that we are raising our long-term financial outlook again.

I'm going to quickly walk you through the process of how we arrived at these new figures. The best way to think about Oracle these days is as a hypergrowth company. Our remaining performance obligations, or RPO, is the clearest indicator of the revenue that's about to come. As we announced on Q1 earnings, we highlighted two things on this topic: First, that our RPO balance now exceeds $455 billion, up 359% year-over-year; and second, that we expect RPO to likely exceed $500 billion. In fact, already through the first 1.5 months of Q2, we've signed several additional large contracts, as Clay mentioned during his presentation, which put us over the $0.5 trillion mark. The demand we are seeing is really hard to comprehend. To put it in perspective, our RPO balance is up nearly 10x since fiscal year 2022. Clearly, the customer demand is strong, but it's also enduring.

The biggest impediment to growth right now isn't so much finding customer opportunities, but rather executing on these opportunities by converting that demand into revenue as soon as possible. As our data center operations engine has revved up and we become more experienced at it, we are bringing on capacity faster and faster and with more efficiency, as Clay discussed earlier. The ramping has already started, as you've seen over the last couple of years, with cloud growing as a percentage of total revenue from 20% in FY '20 to 44% in FY '25.

As cloud crosses the 50% mark as a percentage of our total revenue, the revenue growth rate is further accelerating as evidenced by the forecasted 16% growth rate for fiscal year '26. To put this growth rate in perspective, the last time Oracle grew this fast organically was over 15 years ago. As well, when you look at the expected revenue growth rates over the next 12 months, for companies in the S&P 500 with more than $50 billion of revenue, there are less than 5 companies growing faster than Oracle, and we aren't even close to seeing the peak growth rate yet.

As our revenue begins to accelerate, so does our operating income growth. The reason is that our pricing discipline, coupled with scale efficiencies enable us to gain significant profit leverage as our revenue grows. As the utilization rate of each data center increases, they contribute more profits. And the rest of our operating expenses have grown much more slowly than revenue, also helping further our profit growth.

Now before I turn to the updated financial outlook, I wanted to revisit the figures that we presented at last year's Financial Analyst Meeting. As you may recall, we announced, last year, our expectation to reach over $100 billion in total revenue by FY '29, nearly double the revenue from FY '25, while simultaneously accelerating our profit growth. But that was last year.

As we dive into this year's outlook, I want to start by explaining what is guiding us as we work to deliver the financial outlook that I'm about to share. First, every customer is put through the prospect of the lens of both a revenue opportunity and a profit opportunity. I've read a lot of stories that are speculating that Oracle is chasing revenue for revenue's sake. But let's be crystal clear. We only pursue opportunities where we have a clear line of sight to attractive margins that reward us for our intellectual property and the activity we bring to customers.

Second, as we work to build capacity, we are pursuing a range of financing options to support our growth. We pay careful attention to our cash flow, our debt ratings, our debt capacity and the various funding mechanisms that are at our disposal. These all factor into how we strategically grow revenue and profits faster. Third, we are working diligently to constantly match our expenses as our revenue ramps in our data centers. This operational discipline is critical to be able to deliver profits to our shareholders faster. And fourth, our overall cost focus on every aspect of the company will help us deliver higher profits from our revenue base. That has not changed, and it will not change. And finally, if all of this works in harmony as we expect, the result is superior investor returns for our shareholders.

So here goes. Clay showed you earlier our updated infrastructure revenue targets, which are even higher than what we shared on our Q1 earnings call. Building on that, along with the RPO backlog that we've already signed, the customer opportunity pipeline and the strength of our competitive differentiators, we see much more revenue upside in the next 5 years than just a year ago.

Our updated revenue target is to reach $225 billion by fiscal year 2030. This represents a CAGR of over 31% over the next 5 years. Our pipeline is very deep, and we could see more large-scale opportunities signed over the next 12 months, which could change this forecast and outlook further. And in terms of profit growth, we forecast reaching $21 of EPS by fiscal year 2030. This represents a CAGR of 28% over the next 5 years, consistent with my comments that revenue and profits are symbiotic. These figures are stunning with both revenue and EPS growing nearly 4x over the next 5 years.

Now before I turn it back to Larry, Clay and Mike for Q&A, it's important to note that these figures are as of this moment in time. If we see additional demand that enables us to grow revenue and profits faster, we will accelerate near-term investments in order to capture additional market share. As Larry recently said, "AI is a much bigger deal than the industrial revolution, electricity and everything that has come before."We are extremely well positioned for this opportunity, and we will pursue more growth so long as it fits our profitability expectations. Thank you."

Please welcome back to the stage, Larry Ellison, Clay Magouyrk, and Mike Sicilia.

Give me a moment to recover from those numbers. Okay. Oh my God, this is nuts. Okay. No questions. Thank you all very much.

Jackson Ader at KeyBanc Capital Markets. Yes, this is great. All right. Let's start with those numbers. I guess, first, I'm curious, Clay, when you gave the gigawatt illustrative example, right, like we're going to make $60 billion, it's going to cost $39 billion, something like that. Is that purely illustrative? Is that an average? And then your largest customers, whether it's Meta, OpenAI or what have you, are they even close to that type of gross margin? Or do they come in below that?

Sure. So first, it is illustrative. But the reason why it's more illustrative than exact details is because a gigawatt is changing very quickly, right? Is it a gigawatt of H200s? Is it a gigawatt of GB200? Is it a gigawatt of GB300? Are we talking about MI355? What's the mixture of the number of GPUs to the amount of storage compared to the amount of general purpose compute? So obviously, it doesn't sound like a big difference, but that can be plus or minus 10% to 20% of revenue both ways.

In terms of the margin profile, no, it's very illustrative of even the very largest customers. So we are very committed, like when I gave you that range of margin, it wasn't like this is the margin and there's an asterisk and this is only for like the customers that aren't driving all of the revenue growth. That would be counterproductive for me, it'd be counterproductive for you. That absolutely is illustrative of even the very biggest deals that we're doing. I feel like you're like confused.

Microphone back. No. Okay. No, that is very helpful. And then I'm curious, Larry, maybe this makes more sense for you. Just if we think -- and I'll stand up again, I'm sorry. If we think about what AI can do to your internal operations, we're looking at $21 in EPS in 5 years. What does that imply your internal usage of AI does for your operating expense growth in that kind of time frame?

Yes. I think -- I mean we've underestimated how the positive impact of AI for internal use. I don't think we've -- but I'll let the guys comment on that. But I don't think -- we do have estimates. I mean, clearly, our program is going to be more productive. We're going to produce more product. We're going to be more ambitious. We're going to write more programs. We're going from industry suites to, if you will, entire ecosystems where you look at the health care ecosystem, it's hospitals plus government regulators, plus pharma companies, individual patients, it's entire ecosystems. And we'll be able to do that. So a lot of the productivity will be -- we'll have more comprehensive suites of software. But I would still say we've not fully accounted for the scale of the productivity gains, but I'll let these guys respond to that because they actually run the business.

Yes. A couple of examples, I think, that are relevant. Just internally, I shared earlier in the health care section. Usually, these regulatory cycles take 2 to 3 years to get through approval. We got through a regulatory cycle in 6 months because we're able to automate all of the documentation required for the regulatory. Now it doesn't really displace any people, but allows us to get to market far more quickly and capture revenue much earlier in the cycle than we would have had.

To Larry's point, we start to look at certain functions across the company in our engineering space. And we do have some internal estimates that we've been going through. And frankly, they keep changing for the better. They actually keep getting better. Cogeneration, QA, support ticketing and support tickets. I mean, all these operations are large global scale operations. At this point, we're really focused on productivity enhancements. How can we make people more productive so that we don't have to bring on a lot more labor to get the same thing done.

Early days in terms of where we'll go in terms of the labor force in general. But I would say we're certainly optimistic that particularly for -- at least for industry development, I can tell you very specifically, that we hit a baseline in cost that scales dramatically, and we don't need to add additional cost from a labor perspective to get not just the same amount done, but actually about twice or 3x as much done. That's the kind of scale factor that we're looking at.

Let me add one more thing. We really can't just look internally if how AI is going to change Oracle and make us more productive. It's also going to change our customers. It's going to change the FDA. What happens if the FDA can get through clinical trials in half the time they be used to. What happens if pharma companies can design drugs in half the time and 1/4 of the cost than they used to do it. It's not just us changing internally. The entire ecosystem called the Planet Earth is going to start becoming more efficient. And it's -- there are going to be these incredible effects on the entire economy. We will be a much more prosperous and wealthier world because of artificial intelligence, because of robotics, because of drug design, because of government agencies and regulators making use of these technologies. As they say, AI changes everything. It's going to make us much more efficient across the board.

Rishi Jaluria, RBC. Really appreciate the session. A lot of great detail and obviously, amazing to see these sort of numbers. One question that I think a lot of us have been kind of weighing on the architecture side is -- and maybe it's an overly simplistic way, but what goes into as we think about the balance of power shifting from training to inferencing, fine-tuning, reasoning, et cetera, how easy is it to repurpose that architecture and really eke out the most efficiency out of all that you're building? Maybe if you could walk us through, that would be helpful.

Sure. Well, look, in my keynote yesterday, Peter from OpenAI, I think, did a good job of addressing that exact question. So for an even better answer, we can send you the video link. But one of the things he described is that especially for these large model providers, it's critically important that they have flexible architecture. And the reason for that is that there's this concept that, hey, models get trained, then they get copied off somewhere and then reasoning happens. But actually, models are constantly being updated. And also, as a provider, you need the flexibility to shift back and forth, right? You're doing research on new models. You're doing training updates on existing models, and you also have customer demand.

Let's say that you're a provider and you have a viral event where suddenly you get a huge amount of demand, you need the ability to say, okay, let's not do that training run and use that for inferencing. But -- so obviously, it's better if the infrastructure is flexible. Well, it turns out you can build it flexibly, and that's exactly what we're doing, right? So you build it for the maximum requirements, which is really the highest kind of demand training workloads. And if you do it well, it's a huge amount of effort around optimizing the power of the data center design, optimizing the networking design, then fundamentally, that infrastructure is both very capable of doing the reasoning as well as it's also very cost effective at doing the reasoning.

Great. Tyler Radke from Citi. Congrats again to you, Clay and Mike on the well-deserved promotions. Clay, I wanted to go back to one of the slides from your opening presentation around the AI database forecast. I think you're expecting $20 billion of revenue in FY '30. And I was just wondering if you could kind of unpack the -- or stack rank the key drivers of that? Is that kind of the traditional database migrations to the cloud? And then what is it going to kind of take for the AI start-ups that are buying your infrastructure to start using your databases?

Well, I don't know if you saw it, but I signed this up for $20 billion.

I saw it.

Okay, good. Well, honestly, I say, Larry, why don't you start? I think it's actually a great question of why are you so excited about people adopting our AI data platform?

Yes. I think -- again, I think there's a lot of -- it's an incredible number. The $20 billion is a pretty solid growth rate over the next 5 years. That said, we think everyone is going to want to do reasoning on top of -- by the way, reasoning and inference -- I'm using reasoning and inferencing the same way. And inferencing is kind of the -- AI models do more than inferencing. They do deduction, they do inferencing, they do rules, they do mathematical calculations. So reasoning might be is a little more modern word than inferencing, but it's applied AI. It's actually using the models to reason. I don't know who's not going to do that.

Now the question is how quickly can we deliver it, transfer the technology to our customers. Again, it's the same -- it's not going to be a demand problem, that is for sure. It's going to be how quickly can we get -- transfer this technology to our customers and help them be successful and get started using this technology. Now the good news is there are a lot of our customers -- well, customers are very familiar with the Oracle database. And it's not that they have to relearn the Oracle database. They just have to take advantage of these new features of the Oracle database. And we've got great distribution of the Oracle database right now. You can get it -- obviously, you can get it in Azure. You can get -- by the way, that alone, just multi-cloud alone is going to drive a huge amount of adoption.

Remember, it wasn't long ago. The only cloud you could get the Oracle database in was OCI. And OCI is a great cloud, but not everyone is in OCI. So the fact that we're building all of these data centers, we started with Azure, and we have quite a few data centers in Azure. And if you look at the numbers, if you unpack our numbers, that 1000% plus growth rate we have in multi-cloud is being driven by Azure, who were the first ones that signed up from multi-cloud, and we got the data centers all built and running for them. We're not anywhere near at scale at Google, who came second and then Amazon who came third.

So just scaling out those data centers, just the normal migration, what you asked for the normal migration that's going to be enabled because of multi-cloud is going to, I think, easily get us to $20 billion. I'm not trying -- I don't want to raise those numbers up any at all. But Clay said it's $20 billion. I'll take it. But I think we get -- I think multi-cloud alone might get us there. Well, multi-cloud alone gets us there, what about the use of -- it is a vector database for AI. It's very hard to extrapolate because we have no points. We really don't know how big that's going to be, but I think it's going to be all of our -- it's going to be everybody, right?

I think it's not just an accelerated business on its own as a platform service. But keep in mind, it's the same platform that we build all of our applications on top of as well. So the $20 billion is just the revenue that we'll collect from that platform service alone. We're also very optimistic that this continues to propel our application growth because all the Agentic AI agents that we spoke about earlier today with Steve and Mark and Seema, all of that's built on top of the AI platform as well. And if not for that, we wouldn't be able to go as fast and grow our applications business as quickly as we are, too. So it all sort of works -- it all works together.

Yes. I think that's so interesting with what Mike is saying because you said who are the first users of the AI data platform? It will be our internal application groups or our first users. And one of the advantages I've always thought Oracle had, and you could also say one of the disadvantages Oracle had, and I'll come back to that, is that we do -- we do applied technology. We do applications. We use our tech, and we also build the tech. We build data centers. We train models. We build databases. We build all of this -- we build code generators. We do all this stuff to enable the creation of applications and then we actually create the applications.

None of the other 3 big cloud vendors do that. They are primarily tech platforms. They are not -- they don't build large-scale applications. We get all of these insights from building these large-scale applications where Mike's team will say that, gee, it would really be nice if I could do this more easily. And not that these guys ever -- not that anyone ever asked for more around here. But the database guys get asked for additional features, more ease of use, more automation and backups. That's how we got so much better at the tech.

We had the -- our captive -- very large-scale captive applications team that was putting demands on the tech and helping guide some of the new features that we developed. And as we -- so as we build the AI data platform, we also test the AI data platform on our own applications. And then what our customers get is something we've already used successfully rather than them being the guinea pigs very, very early on. So -- and then we constantly -- then it's continuous improvement. We constantly make it better because we improve it, we use it, we get insights, we improve it again, all of this other stuff.

So yes, the database is now inextricably linked to our AI strategy and inextricably linked to our application strategy. So all the pieces at Oracle now are fitting together. Why some people thought it was negative? People said Oracle is trying to do too much, right? Oracle really needs to focus either on tech and spin off the application business. I remember hearing that a long time ago or they should just do applications and not try to do because no one can do both. Well, so far, we're the only one doing both, but one is more than zero. So we think it's working out very well for us.

This is Patrick Colville from Scotiabank. Great to be here and really exciting time to be part of the Oracle story. I've got one for Clay, please, and one for Larry. So Clay, I think one of the big standout announcements this week from AI World was the AMD partnership. So my question is, I guess, what was the logic there for that AMD partnership? And then also, what does it mean for Oracle's NVIDIA partnership? Because that's been just tremendous for both firms. And then, Larry, if I may, love the long-term targets. One of the questions we get from investors is what happens beyond fiscal '30? And I guess the reason we get asked...

I missed the funny part. What was -- we couldn't hear it up here.

He said what's beyond fiscal '30? So if you could just make up like a '31 projection and '32, he would love it. It'd be...

Yes. Well, wait, we talk about 40 NAND. That's going to be awesome. I'm looking forward to that.

So do these AI labs, do the Tier 1 AI labs beyond the midterm start in-sourcing infrastructure? Or do they lean even more heavily on neoclouds and can Oracle accelerate share gains beyond the midterm?

Sure. Okay. Well, let's start with the question. So look, I'm also very excited about the AMD announcement. And you said, okay, so what does it mean for AMD? What does it mean for NVIDIA? What does it mean for us? Well, first, look, we have an amazing relationship with NVIDIA, right? NVIDIA, I think everyone would agree, did a really good job of getting this industry started. And we -- as a tech industry and as AI, we wouldn't be where we are today without the work that NVIDIA did both on the accelerator side and the networking side.

Now I'm glad that AMD is doing a good job because I think that we don't have a shortage of demand. We have a shortage of supply. And so the reality, I think the biggest thing that I find myself having conversations with people across all areas is they're using a scarcity mindset in a world that doesn't have that problem. So for me, it's like, oh, well, if AMD does well, does that mean that NVIDIA is not going to do well? Or I look at it like imagine there was infinite AI demand. What if we had options and choices that allowed us to scale even faster, right?

So our AMD partnership is amazing. We have -- the reason we're doing it with them is they've been a great partner across our CPU business, our networking business. They make a good product, and our customers want it. So it's just good for all of us. But then you asked kind of the second question, and I'll let you finish, Larry, you can make it think longer about 2031, 2032. The reality is that I think that there's this perspective that these AI companies are only coming to us because they're temporarily out of luck and they don't want to be doing that.

We do a really good job, and we do a good job that actually complements and supplements what these people are doing. If you go -- I would advise you to go talk to these people. When I -- you saw Peter -- go watch my keynote from yesterday with Peter on stage, he doesn't need more problems. He's got a lot of work to do to be able to give the infrastructure he needs to all of his researchers and all the demand of this technology.

Everyone wants help. So I don't think that there's a shift at 2031 or 2032, where suddenly people go, "Oh, there's no more AI infrastructure business for companies like Oracle." I think it's more interesting to figure out what -- like we just talked about how we have so much unexpected demand from this inferencing and reasoning and how that then goes out and makes everyone's lives better. Those are going to need computers, too. So no, I don't see like a step shift. The reason we give the forecast out is we can only see so far.

I think I'm going to let Mike talk about the future because I think the future -- to understand what the world is going to look like, you have to understand the automation in the different -- how the world is going to change. This is back, okay, Oracle using AI, how is that going to impact our financials using it internally. Well, let's look at our medical business, and then I'm going to let Mike go into detail about that. We need to work with the FDA to make it more efficient to a drug that works to get it approved more quickly.

We need to do a better job of delivering health care all over the world with modular -- working with companies building modular hospitals all over the world. And I know in some countries, health care is considered a human right. Other countries, health care is hardly existent. As our -- we have the chance to democratize a lot of these technologies, make governments more efficient. Sometimes governments are -- big corporations might not be very optimally efficient.

AI is going to help that. Governments will be more efficient. Poor countries will get food supplies, they will get energy supplies. They will get hospitals. And -- but they'll get this next generation of smart hospitals. Mike, maybe you can go into, again, that the world -- how we benefit as the world gets better.

Well, I think what does it look like beyond 2030? I think we're rapidly heading towards self-implementing self-learning, self-healing systems. You can certainly get your head around what that means for the GDP -- if you just look at, to Larry's point, something like health care and hospitals, and I'll expand on that in a second. What does it mean in terms of Oracle? It's hard to say. I can't imagine that there's anything negative to say about it because I like our chances of being able to deliver that full system ecosystem.

But in terms of -- let's just talk about clinical trials for a second and compare and contrast. COVID-19 was not that long ago. We were involved as the technology provider and for basically every COVID-19 vaccine and every therapeutic that was coming to market. And here's how it worked. we were shoulder to shoulder with our customers. And what that meant was we had people that would help gather the documentation at the pharma, and they literally rode buses, caravans and buses to Washington, D.C. with stacks of paper. The bus was half filled with people and half filled with paper. People had to read the paper on the way to make sure that it was complete. And that's how it worked in COVID-19. It's exactly how it worked from the most sophisticated pharma to the start-up biotech pharmas, all of whom were involved in trying to create vaccines and therapeutics.

Now fast forward to now, and we're working with regulators across the world, not just the United States to actually accept electronic documentation, accept electronic documentation so that we can take proof points, we can take efficacy points, safety points from these clinical trials, and we can get drugs to market far more faster. What that means is they're actually cheaper because there's a lot of time and money that's wasted in very long latency processes around documentation, around reading around all this stuff.

The next phase of it is, and we're not that far away from this, is that how do we have clinical trials that are reliant on just real-world data. Right now, the gold standard for clinical trials is double-blind, placebo-controlled trials. And there's not a bad thing. That's not a bad thing. It's led us to some wonderful therapeutics and some wonderful vaccines along the way. But what if you could capture data in real time from everybody who's taking a particular pharmaceutical. Not only do you change how fast you get to market, but you actually have a clinical trial that never ends because right now, clinical trials are thought of as a project.

It's a project and it ends. And once the project ends, we actually stop tracking those people. We actually don't know what the counter-indications are of those people when they start to take other medications or they start to develop other preexisting conditions down the road. I think that, that's an example of not just changing the way an industry works, but actually benefiting humanity in a way that is really just -- it's hard to put a monetary value on what that means just yet. But I would tell you that we are very, very quickly approaching the spot where the technology is not the barrier to make any of that happen.

But there's one key. You have to have all the data in one spot. You have to have an AI data platform. to actually make that work. You've got to be able to get all that data very quickly. You've got to be able to collect it from multiple endpoints in real time, and I like our chances to do that. That's just one example, Larry, one industry. I can go on about all 22, but...

No, the fact that what you've done, I mean, the real -- it impacted us. The fact that hospitals suddenly have access, they're connected to the bank, the banks. And we've connected hospitals to banks. Now the banks can look at if you will, a bunch of receivables and see are those receivables likely to -- is the payer likely to pay, the insurance company likely to pay the hospital? And can I safely make a loan to the hospital. So again, these ecosystems as they get connected, and the liquidity you're providing to the medical business, maybe just a couple of minutes on that. And I love the example -- there's an Oracle example that did directly affect our quarter.

Absolutely.

No. It's worth mention. It's so interesting.

Can I give that? I give that.

Yes, you got it.

All right. So we had the banks here earlier. We had JPMorgan and Bank of America here earlier, and they went into quite a bit of detail about the ecosystem. So I think what's -- to your point, Larry, we're not just talking about automating the health care system. We're actually talking about creating the ecosystem that didn't exist before. That ecosystem is the automated real-time autonomous connection between the banks and the providers. We had -- as we mentioned, lots of very big hospital systems have cash flow problems, not necessarily a long-term receivables problem, but a point-in-time problem, which is they don't have the cash to pay certain bills, including sometimes our bills at any given point, money that they owe us.

So we actually connected one of these banks that was up here today with the hospital system. And based upon the real-time debt -- based upon the real-time access to the receivables position, the quality of the payers, all the things I talked about earlier, how many readmissions we have for hip infection and all that stuff, based upon all that data, which they never had access to, they actually created a municipal bond offering for that particular hospital. And actually, we're able to, for the first time, rate it -- actually rate it.

So one of the hard problems that banks have is exactly rate the debt here because it's a little bit of a crystal ball to say, what's the timing? What's the repayment factor and what's the risk profile associated? That customer in the quarter was able to pay a bill that they owed to us. Now fortunately, what's more important, they were able to pay the doctors and nurses and everybody else that was in there. But as a side effect, they were actually able to pay us a bill that was, shall we say, rather late. So these type of ecosystems benefit everybody and the bank actually got a customer that they didn't have before. So that's the kind of situation that works out in terms of these automated data platforms.

Yes. I mean the fact -- the liquidity that information is able to provide that the banks can use AI to consume that to come up with bond ratings to decide to make -- take -- what the interest rate should be on the receivable, figuring all of that out. With all the information, you know the receivables safer, you can charge a lower interest rate because it's more likely to be paid. I showed you a picture of greenhouses for the food business, but we got really interesting modular hospitals that, again, we're not doing these, but partners are doing them.

And the modular -- people of the modular hospitals are coming to us and saying, we'd like to incorporate all of the Oracle technology into these modular hospitals, which we will be building all over the world. And people all over the world, it's a huge -- health care is a huge market. A lot of people can't afford it right now. As the world gets wealthier, more people will get good health care. The global economy is going to get much bigger, help us make our numbers.

Siti Panigrahi from Mizuho. Larry, I want to ask you about enterprise application software. How is that going to evolve in this AI world? Means do you expect this to be rewritten like we see in prior architectural safe like client server, 3-tier cloud applications were rewritten. Do you think in this agenting world, enterprise applications have to be rewritten again? If so, how Oracle is positioned?

Yes. We're constantly rewriting our applications. Steve, I remember I get this every 3 months, we come up with a new version of our applications, which is just unbelievable. If you think about what it used to be like and SAP customers would upgrade every 20 years even if they didn't need to. Well, actually, that sometimes they wouldn't upgrade in 20 years. They just -- 30 is fine because it's -- you know what some of those SAP implementations in excess of $1 billion to put in SAP. You don't want to do that every 3 months. So we give we're continuously rewriting our applications. Michael?

We have 600 AI agents live right now across our portfolio of applications, and we have 2,400 customers that are already consuming those AI agents. That's across the industry applications and the Fusion applications. As I mentioned earlier, in banking alone, within the next year, we'll have an additional 126 agents live, just in banking, not counting everything else that we do. And by the way, if history is any teacher for us -- in the AI world, history is not that long ago, Steve, right here on this stage last year predicted that we would create 100 new Fusion applications over the last 12 months, we actually created 400 and had 400 go live. So that's the pace and scale that we're moving.

Will applications become a collection of AI agents? Yes, and that's exactly what we're doing. That's exactly what we're doing. And certainly, again, at the expense of repeating myself, much easier to do when you've got the underlying infrastructure and the AI data platform on top of it, and we are the custodian of the world's most valuable data. So our agents are incredibly rich, and they're very easy to adopt. They're just part of what we do in our quarterly release. There's not a special AI agent release vehicle and a special implementation plan. You just take it out of the box and it works.

Larry, you've historically seen value in software companies when the market has been skeptical and you've never been afraid to price software for the value it delivers. Now we're in a time when the market is questioning the terminal value of application software and is worried about the seat-based model. Do you think AI will shift value away from the application layer to other layers? And how do you think software should be priced?

Interesting question. Do I think the seat-based model works? Yes. I mean I think it's a combination. There really are 2 models for pricing and applications right now. There's kind of CPU consumption. I mean how much -- and then there's -- so there's the -- on the supply side, how much CPU did you use? And then maybe storage, but primarily CPU, how much CPU, GPU did you use on the supply side? And then on the consumption side, how many different people used it are the 2 models. And I think we'll continue oscillating between those 2 models. I think for certain kinds of AI, very complex reasoning, I think we're going to go by the supply side model, you're going to pay for GPU usage.

For some of the more agentic things, automatically fill out my expense reports for me, something as mundane in that is that. Give me a couple of choices for a doctor's appointment is like in the NHS in the U.K. Give me -- show me the soonest appointment I can get and show me with my preferred doctor, how soon I can see my friend out to decide between those or you figure out and decide between those and schedule it for me. That might be down to -- still be down to how many individuals are using it versus what the costs are of actually supplying the service.

And I think, again, we'll be -- we've been constantly adjusting that. This is nothing new. We've been adjusting between these 2 models for many, many years. What I don't think makes sense, by the way, is people saying, well, here's my application, and here's the really cool AI stuff you buy separately. I wouldn't even know how to build an application like that. If you don't buy the AI stuff, your application won't work at all. I mean it is your application.

So I don't understand the separate charge stuff where -- because the AI is going to be so dominant. I don't -- I think that's a transitory model. That model will disappear from most of the application companies who are using it right now during the introduction of AI. We're more in the middle of this converting our applications to AI, where they don't work without the AI parts.

It's John DiFucci from Guggenheim. And Larry, thanks for the answer. That was actually my question. The last part, too, I really appreciate. But -- another question I had was AI changes everything. In Oracle, some of the stuff you showed us today is changing the world of health care and maybe agriculture. I mean these are things that Oracle has always run a business for profit. But I see something more here like to me anyways, having covered Oracle for 26 years, it's Oracle sort of doing good for the world, not that you haven't before, technology itself does it.

But I guess going back to running a business for profit because there's a lot to do with Cerner. You guys -- you acknowledge you have to rewrite everything and you're doing that. But when should we expect something like Cerner, which is huge, to get to the profit margins we'd expect for a software company, if ever? And how should we think about that?

Well, I'll give you a very high-level answer because I work at a very high level space. He's the guy who builds those applications. I'll let him go into the detail. But I think we will have largely finished the rewrite of all of Cerner next year, everything will be new. And we will have a comprehensive -- and it's not just Cerner that we rewrote. We're writing agents for payers. We're writing agents for clinical trials and all of these other things.

So it's going to be much bigger than Cerner ever was. Cerner automated hospitals and clinics. This is automating the entire health care ecosystem down to individual patients and individual doctors, individual nurses. The code will be in place next year. And I really don't understand how a small company can compete with what we're doing in health care. But Mike...

So in terms of doing good for the world, look, you're absolutely right. I mean, I'll tell you that there's nothing more inspiring and there's nothing that will tug at your hard strings more than walking around a VA hospital and looking and working with people, which we do every day, who are caring for those who serve our country in the United States. And that's a mission for us that will continue to be a mission and where we will deliver for all of those people and actually all hospitals worldwide.

So you're absolutely right. There is something here that's a mission, and we've got people rallied around that mission. In terms of thinking about how the business evolves profitably, I don't think we should stop and just think about Cerner. I think we should think about the entire ecosystem of what we built. We had customers 2 of Epic, our primary competitor in the EHR market. Two of their largest customers in the world, the CEOs were on stage with Seema just 3 weeks ago at our Orlando conference, and they were talking about the AI data platform. We're not just talking about the EHR. We're talking about all of it. And it's going to take all of it to fix the -- at least the American health care system. The American health care system, there's no more money to go into the American system.

As we said earlier, we actually have -- to be successful, we have to take money out of the American health care system, but I think we can make more money doing that because the only way to take money out of the American health care system is to be able to automate the entire process. Just the EHR is not enough. It's got to be supply chain. It's got to be HR. It's got to be finance. It's got to be the banking relationship. It's all of that. And when you look at our Fusion business, health care is actually one of our most popular businesses for HCM and ERP applications.

In fact, we're doing wonderfully well in the health care business with our Fusion business. When you look outside the United States, we are the largest provider and our growth profiles are excellent outside the United States in just a pure EHR market before you even count all the rest of it. We just won a deal yesterday in the Middle East that was signed here, displaced SAP with Fusion, with our Oracle Health EHR, with OCI for all their bespoke workloads. People are buying all of it. So we no longer think about the Cerner margin, so to speak. In the first couple of years of acquisition, obviously, we keep those things separate. But we don't think about it as Cerner anymore. We think about it as the Oracle Health ecosystem. And when you put that all together, we're actually already operating at a clip, which is not diluting the company, not dilutive to the company margin.

And I think to the point that Larry made, I just don't see how the competition is going to keep up with what we're doing because I go all the way back to sort of first principles with -- and this is what I said to some analysts at our health conference -- how many fuel cell power plants is Epic building? That's the first question I asked, right? How many large language models are -- generative AI models are positioned on their cloud stack? Second question I asked. Then I stopped and I said, you get the point, right? Unless you're going to do all of it, you're actually not going to do that.

And I think the health care system in the United States is tiring of point solutions. They're tiring a very large system integrator. They don't have the money, just don't have the money. The only answer is complete system automation. And for those reasons, I think we are going to be just fine, if not wonderful in health care. And to your most important point, going to do well for the world in the process.

Let me just add one thing that I've said several times before, that is what is our model for all of this? My model, at least in my head, it's Musk's law. I don't think Elon ever created Musk's law. I mean he just -- he didn't write it down, but he just did it. And if you look at Tesla, how did you -- he built -- had to build an electric car ecosystem. He couldn't -- the problem was not building an electric car. I mean if you build an electric car, you've got to be able to recharge that car in Sweden and Norway and Vietnam and all these other places. How do you build a global charging system? How do you do that? How do you build -- building electric car is really easy. Rivian can do it. A lot of people can do it. I can build an electric car in my basement. But how do you build 10 million of them, 20 million of them a year. How do you build those factories?

The largest building ever built in human history was in Austin, Texas, it's the Tesla factory filled with robot. You got to build robots. The robots he's building the Optimus robots, the first inspiration, the first use of those Optimus robots are in Tesla factories. He had to create the back of the model Y is one piece of steel. It was -- it's lighter. So he had to create all new stamping machines for doing this, different machines, different factory automation, supply chain, battery technology, battery science had to do the whole ecosystem for an electric car to change transportation but he did it, and it works pretty well. Those cars are rather inexpensive considering they have these incredible computers in them and drive them and he had to solve one of the hardest AI problems of all time, which is self-driving and real-time AI.

So you take what he did at Tesla, and that's how at least we look at the ecosystem for health care. It's not simply the hospital or a clinic. It's the entire the patients, how do patients make appointments? How do payers decide to authorize that hip replacement or not? What's the process they go through. How does a regulator approve a new drug? How does a pharma design a new drug? That's the entire -- the health's ecosystem, by the way, a good deal more complicated than and larger and a bigger opportunity than the electric car market, even if the electric cars are fancy electric cars and fully robotic.

So the way we're approaching these problems, the way Mike's teams are approaching these problems is to look at automating the entire ecosystem. Our HR teams changed when we bought Cerner because we now had to train doctors, train nurses, schedule people differently. It was partially -- they were partially gig economy, partially hospital employees, supply chain, inventory, all the shipping, keeping track of inventory in hospitals unbelievably complicated. I'm going to go into all of the details, but we have to design RFID tags and RFI readers. There's security in hospital systems. It's doing the entire ecosystem. If you do the entire ecosystem, you get a much better result, and it's a much, much bigger opportunity.

I just wanted to say congrats to Mike and Clay. Obviously, you guys are going to be great partners with Larry, but it's the only time I'll say this, but I kind of miss Safra a little bit.

She's here.

Larry, Mike Clay, unless you want to keep going, this will be the last question. But if you want to keep going, they'll stay here all day with you. So your call.

It's up to these guys. I've got nothing to do. I'm retired. Actually not true. It's really not just Safra was just shaking your hand. No, no, no. No, no. Okay. So you get to retire, I don't. Okay. Something is very wrong with that. That doesn't mean -- I know you're not retired. But you're sitting there and somehow I'm still hitting here. Okay. All right. I'll figure it out in time. I think you guys want to keep going a little bit up to you guys.

Awesome. I'm over here, Larry. Brad Zelnick with Deutsche Bank to your left. I actually have 2 questions. Larry, my first one, I've been coming here for many years. And every year, I look forward to seeing you and the team every year, I get a year older, and you seem to stay the same age.

If it were only true.

My first question, if you have 1 or 2 tips for how we all stay young, I would love to hear it. My second question for you and for Clay.

I'll send you the research papers.

Okay. As we think about Stargate, can you talk about how strategic AI is to governments around the world and how Oracle is working with them hand-in-hand to make it all a reality?

Yes. Well, I mean, I think if you imagine for a minute that you were a government, and let's imagine that what we've all been saying is at least mostly true, you would want to make sure that you have that technology available inside your country into your citizens. Otherwise, I think in the same way that people have -- for the past few hundred years, well, how do I manage my food supply? How do I manage my energy generation? How do I make sure I have the right telecommunications, right? AI is technology that we think is even more valuable than that. You need to have access to it. And so we're in constant conversations with different governments in multiple different areas.

I think one aspect of it is around how do you make sure that your citizens are getting access to the AI, right? And that's about bringing access to those models to those different geographies. There's the conversation about, well, how do we get things like sovereign AI. Part of the way in which we solve that, if you think when I was talking about our distributed cloud situation, we have a set of technologies that enable us to actually deploy regions globally to enable them for individual customers, but also for local sovereign operators.

So as an example, the one I used earlier about [indiscernible], they not only have a sovereign alloy that they're using to serve the needs of the UAE government, they also have access to the latest and greatest GPUs inside that environment. So I think it's critically important for all of these governments to have that technology, not just from an infrastructure perspective, but then also making sure that -- because if you don't have the infrastructure, how can you suddenly then deploy the latest and greatest EHR on top that makes use of all that AI.

And as Larry said, we don't have a non-AI version anymore. Now specifically to the question about Stargate, I think lots and lots of countries are very interested in how do we get that sovereign AI. And we have good relationship with OpenAI as kind of, I think, due to our flexibility and our speed of being that partner of choice to go out and then deploy local sovereign AI infrastructure facilities for them. And so there was UAE Stargate. There's things that we're working on right now in Africa. I just had a conversation with a customer yesterday out of Latin America that's very interested in the same thing.

And part of what I think makes us a great partner in those sovereign AI conversations is that we can scale up and we can scale down. Not all countries need a 500-megawatt OpenAI deployment. Some of them might be very well served by 5 megawatts. Well, we can do that, and we can do it quickly. And our relationship with the top leading model providers actually works to our advantage because then we can go deploy the infrastructure and then suddenly the AI is available, both in terms of sovereignty as well as in local secure access. So I think Oracle is extremely well poised across all the different layers of the stack for sovereign AI and enabling all the world's citizens to get access to that technology in a secure and controlled way.

Alek Zukin with Wolfe Research. Thank you for an amazing day, truly unbelievable numbers. I have maybe just a quick 3-part question, which is...

Only 3-part? A quick..

Is there a prequel?

I have a really fast 9-part question. Sorry, I am enjoying myself up here.

As we look at the pacing of reasoning versus training, when does that inflection point happen? Like in 2030, what does that workload ratio look like? Because it's important for margins. Interesting. You talk about demand and supply. How do we think about the pacing of CapEx to facilitate that? And then any back story on how you're able to kind of over jump or jump over some of these other hyperscalers to win and become the preferred vendor for some of the model companies as they scale beyond what we all thought was possible in a given time scale.

Those questions have almost nothing to do with these. I feel like -- would you like a scoop of ice cream on the salad with some bread. I've already forgot. Did you remember the first part of the question? I'll start with the back -- the last part, why are people picking us? At least I remember that part of the question, I can answer it quickly. Look, the reason that customers are picking us is for the reasons that we say customers are picking us and for the same reasons that those customers also stand up and tell people like you why they're picking us.

And like I said, Peter from OpenAI was on stage yesterday explaining exactly why they pick us. It's because we are extremely fast at getting things done. Everyone has excess needs. If you can be somebody that has an answer that says, okay, you need capacity. If you go 2 years, 2 years, 2 years, 8 months, okay. I like 8 months better than 2 years. Okay. If you have unique requirements where you show up and you're like, "Hey, I have custom requirements around some of my networking or the way in which I want to power cap this infrastructure, I want to codesign it with you or perhaps I want to deploy it in a location that might have lower cost, okay, no, no, no, yes.

Okay? So if someone is saying yes to the things that you need and they're delivering very, very rapidly, and they're extremely high performance and very cost efficient, that's why people are picking us. And that's why it's not just one customer, it's all of the customers, both very, very large customers and small customers, right? When I showed the details of that 700-plus AI infrastructure customers, that's why they're all choosing OCI. Now not 100% of the market, like we do definitely have competition.

But as we continue to execute and as I think Peter said yesterday, this community is very small. As this information gets out there more and more, all of those people are coming to us with more and more capacity bands. Now you're going to repeat the other 2 sections because I already forgot. Larry, do you want to talk about training versus reasoning and how you think it changes?

Well, everyone is going to do reasoning and very few people are going to do training. So it will definitely cross over. I think I wish -- I think it's a great -- and actually, it's quite a fabulous question because I don't think we have enough data points yet to be able to figure out when it's going to cross over. I think what's going to help is the AI -- believe it or not, I think it has a lot to do with how good our data platform is. If our AI data platform is as good as we think it is, it's going to help people get to inferencing/applied AI/reasoning, all the same thing, using the models as opposed to training the models, which is, again, everybody.

So we have to make it easy for everybody to do on their own private data. 5 years from now, will we still be spending more money on training than on reasoning? I don't know. Anyone -- Clay, do you want to take a shot at that?

Well, I'm going to have a complicated answer, and then Mike is going to have a useful answer. I think people also -- I think it's an interesting question. I think it's also a question that's unanswerable. So look, if you think about what training does, we take together all of the information that humans are willing to share and then we train something on it. Eventually, you run out of data, okay? Well, how does data get created? Well, people use reasoning to create data and they write it down. If you actually look at where AI models are going, much of the data that they're going to be training on is actually data that they're doing their own reasoning on.

So I think this question, if you actually look at what I think in 5 years, what most training of AI models looks like, it looks like AI models thinking -- and then AI models using the results of that and going, oh, is this better or not? If you look at how RLHF, right, reinforcement learning with human feedback works today in these AI models, what's happening is you train a model and then you give the results out to human beings who score it and then you feed that back into the training data. Guess what happens as the models get really smart. You don't pay a bunch of people to score the outputs. You have models score the output.

So I think that this -- that's kind of what I think Peter yesterday was trying to say was like, hey, you think of these as different things. They are the same thing. And in the same way that we don't think of our brain is like, well, are we using our reasoning function right now? Are we learning? You're constantly reasoning. And as you're reasoning, you're also learning. It's going to be this iterative cycle.

Let me reframe the question. So I think it's really simple -- in a way much -- forgive me, I think simpler to understand. When will AI -- the people creating AI models be taking in more money than they're spending?

Well, I think that actually depends less on how much money they're taking in because that's growing very, very rapidly. I think it's when do we reach diminishing returns on spending extra money, how it makes it more valuable. And if I had the answer to that, we would do a different company.

Yes, even the simpler version of the question is still hard -- it's the same question and still hard to answer, still hard to answer. I think the problem I have, and I agree with everything Clay said, yes, it gets very blurry because every time you ask AI a question, it's learning something. It's training itself. So what is training? What is inferencing/reasoning? How fast does -- let's just pick one example, OpenAI or Grok, how long are they in hypergrowth mode where they're spending money faster than their revenue because that's very common in the early -- obviously, railroads obviously spend a lot of money.

When does it cross over? When do passenger tickets exceed laying of track? I don't think we have quite enough data yet on inferencing and the speed of inferencing growth. Though one of the most interesting calls we ever got was, hey, do you guys have any capacity? Where? Anywhere? Well, how much are you looking for? All of it. You want to buy all the capacity we're not using everywhere in the world? Yes.

Okay. I've never heard that one before, but I haven't heard a lot of this stuff before. This is very strange. I remember I called Safra, but someone just called and said, felt like what the hell is going on. I mean it's not like I was trying to sell anything. They called us. I just wanted to buy everything we had. So I don't think we have quite enough data. I think we'll understand it a lot better about -- believe it or not a year from now. I mean it sounds funny, but I think Mike?

It's really a philosophical multi-parter. It's a great question. How does synthetic data feed into this? Is it really retraining? Or is that inferencing? And then the next frontier or 2 is how does private data fit into all this? And do we take some of that scale and create very rapid small language models, too for enterprises. I think that's going to continue to evolve over the next, I would say, 2 years, but probably not even at the rate we're moving. We're going to have clarity on that. But I think either way, we're going to be in a good position to serve those markets.

I'm going to say this is the last question for me. You guys keep going. That's fine.

If he leaves, I'm leaving. Mike, okay. We work together as a group, okay? You can't just walk off and leave me and Mike out here.

Watch me.

All right. And then Matt is getting us off the hook as well. So thank you all very much.

Are we done?

You are done.

All right. Thank you.

Please welcome to the stage, Ken Bond.

All right. Okay. Thank you all very much. That concludes the show. Thank you very much for coming out. Really appreciate you taking the time out. Look forward to it. If you have any questions, just follow up with us. Take care. Safe travels home. By-bye.