Marathon Digital Aktie

Great overview, Fred. Thank you for that. Russell, Core Scientific was the first bitcoin miner to secure the first large-scale AI colocation deal think over 2 years ago now. If you could walk us through that first contract and how you've seen the market evolve since then, that would be great.

Yes. So our first deal actually was in February 24. It was a small air cooled site. It was a site that we leased, and then we converted it. So it was to kilowatts or we converted it to 17 to 20-kilowatt structure on 100 200. That was like a 45-day conversion. The next deal we worked out with Corio, we actually signed it in June of '24. And it seems like 1.5 decades ago. So if anything moves as fast as Bitcoin, it's actually the AI-place because -- in that 2 years and 3 months, we've turned on and we're building like 450, 500 megawatts of that right now, which is a good place to be in. But all those other megawatts had to be built from the ground up. And that's basically because direct liquid cool and air cool are just not the same thing. So the legacy data centers that were air cooled just -- it made no sense to try to convert those. We had to build everything from the ground up.

Ask about some of the first deals versus right now. I remember very distinctly when we were pricing the first deal, it was like in late '23, we had good hard estimates of around $4.5 million a megawatt to build the direct liquid cool stuff. By the time we got around to ink in the deal in June, that same facility didn't -- is just under $8 million a megawatt. Now the things that we actually turned on this year, the other sites, they were floating around $10 million to $10.5 million a megawatt. And then to give you a little bit of idea, the stuff that we have turning on next year stuff is turning on is between $12 million and $13 million a megawatt.

Now most of that increase is in labor and then in a lot of the gears actually coming in the switchgear, the transformers and stuff. But to give you an idea on labor, -- if you have a kid graduating high school right now, they need to become an electrician. So journey electrics right now on 1 of our sites make about $250,000, $350,000 a year in master electricians, are bringing in $750,000 a year. So labor has gone up, and it is literally one of the key points right now for being a roadblock. So it's like, is there a power at the site, is there NIMB at the site? And then do you have your long lead equipment, transformer switch or that kind of stuff? And do you have your labor? And those are kind of the points and every day, that's what we debate now. So the things that were the issues have just gotten bigger. And as Fred just said, Bitcoin miners are in a unique position because we have a lot of those things. We've been doing a lot of those things. So I'm not sure if I answered your question, but I tried to give you a little background on the SP999 History there in this first sites.

Spot on. Thanks for the history there, Russell. So there are 2 primary business models to monetizing AI infrastructure deployments. We have colocation and AI cloud services. Sam, welcome to the panel. Good to see you.

Thanks for having.

Great to have you. So this 1 is for you. At White fiber, you're actually pursuing both strategies, right? You're doing colocation, you're doing some cloud services. If you could talk to us about the differences between the 2 models and why a Fiber has elected to pursue both strategies.

Well, WhiteFiber actually started in the cloud business. We're, I think, 1 of the first in the sector in the Bitcoin mining sector to have that cloud contract, and it was essentially a catalyst for us to spin off the business. We felt instead of turning it Bit Digital into an AI infrastructure platform that it would be better, and there are a couple of reasons for it to just spin out the business, which is now today called White Fiber, which IPO-ed last August. And there are a couple of technical reasons for that. One of them is financing. We were dealing with a lot of conservative banks who wanted to finance our AI infrastructure business, but we're uncomfortable with the fact that there was some crypto in the business. And one of the things we were able to do to provide comfort was just start over, just spin out the business and just make it a pure-play AI infrastructure business with absolutely no crypto. So that way, these very conservative credit committees we get comfortable on establishing some credit facilities to develop our AI infrastructure.

Another reason why we decided to spin out is just to have a more institutional shareholding capital structure. So when we did the IPO for the -- for WhiteFiber, we were talking to and long-haul like bonds, we're talking to a very institutional blue-chip shareholders. And these are not the type of shareholders that would be interested if we had crypto. They were interested in AI infrastructure, not crypto. And so -- that's -- those are the 2 main reasons why we decided to IPO the business. And we started with this cloud -- this very large cloud deal and we felt to be AI infrastructure, you should really have both the cloud side and the colocation side. And we felt that we did not have the chops , the technical chops to really create an AI campus, creating a bitcoin mining facilities is not the same as creating a data center for AI infrastructure.

So what we did was we acquired a company called novum a couple of years ago. That company -- that team has been doing AI campuses for many, many years in a retrofit format. They worked for hyperscalers, such as Amazon and Microsoft. And that team has been doing that for so long, and they just knew how to do that retrofit format. And we felt that the skill set to create a data center for AI customers, just so different from a Bitcoin mining facility that it would make sense to acquire that company. So that company became our colocation business, and then we have our cloud team, and those are 2 separate teams and very, very different skill set. And there are some people here on the panel who may disagree with my position that it's just a very different skill set, creating a big coming from transforming to campuses. But I can tell you that we've done it now. We've -- we've taken over mattress factories. We turned that into data centers.

We did that for Cerebrus that recently IPO-ed -- we've taken over a 1 million square foot facility in North Carolina, and we did that for end scale for an $865 million contract. And they are now up and running. They're up and running. And we've done that within 6 months. And that was not because of our abilities as former bitcoin miners, but because we acquired that particular specialized team that does that day and have been doing that for many, many years. So that's sort of a meandering way of answering the question that this cloud business is -- was the catalyst that started this whole company that's called WhiteFiber today, and it's divided into 2.

There were technical reasons for it. The first reason was shareholding in the second was financing. And with respect to the colocation business, that's very separate from the cloud business. It's a very different skill set. And we're very proud with what we've built in just basically 1 year.

A mattress factory to an AI data center in 6 months, pretty impressive. fabs. Good stuff. Thank you for that, Sam. Michael, you're also scaling both business models over at bit tier. Anything to add?

I just think that it is a very different process to do both of them. Because of the size and scale that we are in approaching the business, we decided for Tieto Norway, it would be better to do a colocation deal. So we signed a deal with Volta, which is providing GPUs for their end customer. It's about 121 critical IT megawatts. And we're building out the data center itself and then the GPU is being handled by our tenant. That way we can learn how to scale and build large data centers to the current technical specs that are required while somebody else is taking care of the GPU and the GPU financing. So we're pleased that, that deal went through and working hard to get the buildings and everything finished for the end of this year and the end of Q1 of next year.

At the same time, we feel that, one, we have some smaller sites that don't make as much sense to do colocation alone. It's just not the scale that the big players at 1 would like. and two, to take the land and power that we've worked so hard to build up and lock it all up -- all of it for 20 years. It's almost like becoming a read, and it sort of takes the optionality of what we could do with this land and power ourselves. So we're also starting a neo-cloud or a GPU as a service business, mostly starting in Malaysia right now, our headquarters in Singapore puts us in the Asia market and lets us contact some leading Asian customers. So starting off with about 10-megawatt deal cloud site in Malaysia, which we have 2 customers for and then we're going to be rapidly -- we have another 20 megawatts in Malaysia, and we just announced this morning another 65 megawatts. We have good customer pipelines there to fill them up. In the U.S., we have a 10-megawatt facility in Washington, about 50 megawatt facility in Tennessee and about another 40, 50 megawatts left in Tito, Norway that we could do something with as well. So I think we're well positioned to do neo-cloud at smaller sites where we can scale and learn the skills and get the software all perfected. And we have 2 large sites in the U.S. with Rockdale being the easiest, we just did a large land purchase for that, the size that we could build an AI data center may be too big for us to do everything to build the data center and to do the financing. So we'd look at doing colocation as a possibility. But again, maybe we don't want to lock up all of the available power in the sites that we have so that we're not locked up our best asset for the next 20 years. So that's the business we're working through right now.

And for those on the panel who are doing only co-location today, would you consider procuring your own GPUs on the future deployments? Why or why not? And I'll leave that one open for the group.

Yes. One of the reasons we took majority control of the French company, Axion, which was a subsidiary of EDF was specifically because that's their specialty. We did that deal closed in March of this year, and they have already started taking contracts. Interestingly enough, hosting U.S. inference and training loads in Europe, because the delays in getting stuff online in the U.S., there are now people seem very happy to take European capacity, and then we'll bring ion's expertise in running these mission-critical data centers and building them onshore in the U.S. as we continue to develop the smaller sites we have, which, to your point, a lot of times, they don't make sense for hyperscaler or another neo-cloud to take them. They're better off just using these inference sites.

All right. So we talked about the history, how the market has evolved, the broader market opportunity and ways to monetize these deployments. I'd now like to focus on recent deals that each of you have secured some of the specifics around those contracts. In addition to a question we get a lot from investors sites within your portfolio that you believe are closest to lease execution and why? And we'll go through the whole panel for this one, Matt, why don't we start with you? I think Clean Spark entered the game coming out became strong with $6.6 billion deal a couple of months ago. .

Yes. Thanks, Mike. So Clean Spark has amassed a portfolio of 35 different sites around the country. We're in a number of different states. And we made the decision about a year ago to start to evaluate the highest and best use for the electrons that we currently have energized. And what is a massive differentiator is having substations built and live power today because as we've seen, the barriers to entry and really the bottleneck for development of data centers really is land and power. So our first site that we leased is in Sandersville, Georgia.

We've been operating there for almost 4 years with 11x Ashton Mining, but we built a substation. It's fully energized it's 250 megawatts of power. So we met with our tenant and came to understand their basis of design and the reference architecture that they wanted to incorporate into the data center. And then we went to the community. -- we said, "Guys, we need more land. So they assisted us in procuring 122 acres of land, driver 9-iron from where our existing site was. They also assisted us in the power pathway to pull the power over with the right of ways in the easements. And that enabled us to go to this 20-year triple net $6.6 billion lease with a high investment-grade tenant. So what was important to us and a lot of really valid points talking about owning GPUs and doing GPU as a service or cloud spinning up your own cloud for us at Clean Spark it was about stability. So getting our first lease, $6.6 billion, which equates to $330 million in annual revenue and being triple net, which means that million annually is almost 100% to the bottom line gives us the stability and flexibility to move into other opportunities that we see fit.

Importantly, as we represented the relationship with the community to our tenant, they were impressed with the flexibility that the community showed and their willingness to assist us. So that same tenant asked to get exclusivity on certain other components of our portfolio. So they've also entered into a short duration exclusivity on up to 885 additional megawatts in Texas. Important to note that of that 885 megawatts, 585 megawatts was recently announced to be batch 0 base load. So the time lines for energization and the ability to deploy a data center there meets the expectations of our tenant.

From a standpoint of construction, I've listened to a lot of my peers talk about the risk, and that's a very real thing. So for us, our tenant actually came with a specific request for an EPCM and that EPCM has built for them before, and they also manufacture a lot of the mechanical, electrical and plumbing components internally. So it enabled us to get certainty on supply chain. It enabled us to push some of the potential risks and liquidated damages over to the construction site. So we entered into a lump sum turnkey contract that contemplates the first data hall will be delivered in December 2027. So with that, I think one of the things that it's important to note, and this was brought up as we've investigated the financing of this lease long term. And that is in the beginning the analysis for high yield or investment grade was who's the guarantor? What's the credit wrap, Who's going to give certainty that the rent payments are being made.

That evolved to who's the guarantor and what's the certainty on construction. And it's now gone beyond that to who's the guarantor, what's the certainty on construction? And what do the political headwinds look like? So we've really done everything in our power to mitigate that risk. We've got a guarantee from the parent, high investment grade, of which there are a handful in the world. Obviously, that will become public as the financing is completed. The ability to build that quickly has been derisked largely because of the fact that one -- that lump sum turnkey contract at the request of our investor. And the last point about political headwinds as we've begun to grow the portfolio, we reached out to some of the communities that we operate in, and we've asked them for letters of reference.

And to give you a specific example, Jimmy Andrew is the mayor of Sandersville, wrote a letter of recommendation for CleanSpark for other cities that we may operate in, talking about the community partnership and the win-win spirit that we've operated within those jurisdictions. So -- it really has knocked down a number of the hurdles or the barriers to entry, and we feel very confident that the next -- the exclusivity period and the option for that will ultimately be executed, and we're seeing price go up in parallel with demand.

Thanks for that, Matt. Fred will go on to you. MARA has guided to 2 leases by year-end, and I would love to get your thoughts on that.

So we don't announce LOIs or exclusivities around leases, but we feel very confident about signing the 2 leases by year-end. I think if you look across our portfolio, the other part of your question was where do we expect to see stuff happening first. I think you're going to see it both in the small and the large sites because there's a wide appetite for sites today. Hyperscalers, frontier model providers, neo-clouds. There are different size requirements. In some cases, it's 400 to 500 megawatts in other cases. It's 50 to 150 megawatts. And so you can optimize your pricing model to the credit profile of the prospective tenant to the duration of the lease to the use case for the site, really to optimize the portfolio value. And you look at a large site like the Medicare side or the Long Ridge site, where you have the opportunity to scale significantly north of 500 megawatts, you have the ability to have multiple tenants and optimize the value of the portfolio.

So if you have a hyperscaler in a triple-net lease where you're talking about a yield on cost of you may have a neo-cloud at 15% yield on cost, and you may have enterprise clients at an even higher level. So building a portfolio that maximizes the profitability of the electrons is kind of very much our focus.

And John Salon is taking an interesting angle with renewable energy assets. If you could share where you guys are on your , I believe, LOI most recently.

Yes. Just for the benefit of the room, if folks don't know about Salon, we do something unique in that we go to existing renewable energy power plants, both wind and solar that already have substations that are connected to the grid, but they have a problem that they need to solve. And that is about half of their power never makes it to the grid. So they have power that needs to be monetized. They don't have a way to monetize that energy. We bring the monetization or the load to that wind and solar plant, and then we build a data center campus in that location and go vertical there. For the last 5 years, we've been doing this in the U.S. We have over 200 megawatts of that spinning. And on the Bitcoin side, we've built data centers that are then leased by very large mining companies.

In the last year, we've been transforming the company during -- taking the same model where we interconnect to the power plant, get access to its power, power from the grid, and we can also build on-site generation to build these unique campuses and essentially have been building out a new series of facilities that will be AI focused. This model has allowed us to amass over 6.3 gigawatts of power assets over 30 projects. We have about 1.6 gigawatt of that, that we're actively developing. And there are 2 projects that we've announced most recently, one is called Cai. It's in the southeast part of Texas. It will be a 350-megawatt campus. -- we are actively designing and preparing to start construction on the first 100 megawatts of that. And we did sign an LOI this year and negotiating a lease for that. So that will be our first announced lease for that campus. And then the second 1 is Dorothy 3, the northwest part of Texas, and that sits across the street from an existing bitcoin farm. It also sits across the street from a wind farm that we acquired this year. So we took 1 step further in our thesis that power assets and compute assets will become over time, 1 in the same thing. So Soluna is building a power compute integrated platform. to support the future of AI.

So Dorothy 3 will be our second campus. We're building that to be about 300 megawatts. Part of the power will come from moving the power from the Bitcoin side, and then the rest of the power for both D3, Dorothy 3 and Cody will come from something we call clustering because our pipeline is so big over 6 gigawatts now in each site, we have multiple wind farms or solar projects that can source -- we can source power from that -- that are within 5 to 8 miles of each other. And so we can pull power into a campus and build a very large footprint. Cody 2, for example, will benefit from clustering to get it over that 350-megawatt structure, and Dorothy 3 as well, we'll use clustering to bring in over 200 megawatts of CIT.

So we are in the midst of our transformation. We're very excited about what we've done thus far. There's been some discussion about the risks in building out these new technologies. We focus on infrastructure and power because that's our core expertise, and we're doubling down on that capability. And we're adding to our existing strength execution capability. So we've hired an incredible leader from Microsoft, who will lead up our construction and operations and Well, I'd see not only the very large campuses, but I also see small difference sites being in high demand right now, especially from your labs that just need some of that inference load. And that's what we're at right now.

Thank you, Russell. Now Sam, white fiber has had a very busy summer. It feels like the last few months, you've announced an accelerated number of cloud services deals. Would you like to talk about those?

Yes, we just announced about $500 million of cloud service deals, which is actually even surprising for me, our cloud services team is been really at it. So we're really proud of that. And we see a lot of stuff in the pipeline as well on the cloud side. So that's been surprising for management and how hard working our cloud team has been producing results. But just going back on the colocation side, I think it's always very tricky to like to build and spend all this CapEx on a facility without a contract. That's something we try not to do.

We have now 6 facilities that we could speak of. The first 3 facilities in Quebec, Montreal 1, Montreal 2, Montreal 3. Let me just give some color on that on when we have 20 customers in that smaller facility. The second facility was a pharmaceutical capsule production factory, which we took over. The third facility, as I mentioned, was a mattress factory, which we turned into an AI campus for Cerebras. That's fully dedicated to cerebri and we got that up and running on time and within budget, Cerebrus has been a great client of ours, and we look forward to potentially working with them again in the future since we have served them well.

And now in the United States, we've been focusing on North Carolina and the utility company there is Duke. We have great relationship with them. the NC1 campus, the 1 million square foot, that's just online is just basically -- we're only halfway there with 50 megawatts, and I'm rounding some numbers up right now. And Duke has -- there is a path towards 99 megawatts associated with that facility. And over the next few years, potentially another 100 megawatts in NC1. We finished -- we're now on -- we first -- we have the first 50 megawatts online for NCI for. So we started billing, which is great news for us. And there's tranche 2 for NCI, which is another -- so that is something in which we have a very pregnant pipeline of customer interest, particularly investment grade, looking to get that second tranche.

They've seen what we've done with britey've seen what we've done. So they like the execution story, and they are looking at the second tranche C1 in order to execute for them and get that up and running. We -- as you mentioned, you didn't mention this, but we just acquired we just put deposits on NC2 and NC3, which is about 60 megawatts. And again, because of the reputation we've acquired on executing on time. We have investment-grade clients who are looking to -- basically, they want us to build it out for them, on NC2 and NC3. So it's -- these are good problems to have, but I do sympathize with the heavy CapEx and trying to match that in time with your contract revenues. That's always a very stressful period.

So multibillion-dollar deals secured, and it sounds like several billion on the way with this group here. So you did touch on this, Sam, I want to kind of go a little deeper on to this. So CapEx intensity of AI data center builds certainly quite after. There's no secret there. John, maybe you can discuss the sources of funds, cost of funds and the cost per megawatt to build out these data centers.

Yes, sure. We -- our model is probably no different than others on the panel. We look at a project level financing that will be primarily debt, 80% of the cost being debt, high yield, attached to the asset and the remaining 20% through equity, most likely through the TopCo pushed down into the project. To give you a sense, just on a theoretical project, call it, 100 megawatts, call it $1.2 billion to $1.4 billion to build. So it's somewhere about $10 million to $12 million per megawatt. This depends on where you're building it. We tend to be in remote locations. So cost is higher. Russell's point about labor cost and access to certain critical folks raises the cost to some extent. And so when you do that 80-20 split, you're talking about a few hundred million dollars of equity that has to be invested in our first, let's say, 100 megawatts.

But when you look at the returns on that, you're talking about upwards of $10 million to $15 million net NOI. I think we have a slide in one of our investor presentations that teases this out. The yield on cost is going to be in the high teens. And at some point, the cap rate is in the single digits. And so you get somewhere on 100 megawatts, $2.4 billion of asset value added to the company. So you compare that to our current valuation and asset base, the 6.3 gigawatts that we have is a very valuable asset if we can convert those assets to spinning data centers.

Our average construction time is going to be much like anyone else here so you can map that out to what that looks like from a revenue and growth perspective for the company. What I'd like to do sometimes for fun is take that framework and then apply it to the 1.6 that we've got spinning and then apply it to the 6.3 to see what the potential value is. And then the rest is really about execution, making sure you put together a bankable lease can take it to market. So the high-yield folks like what they're going to get and feel secure in the asset -- and then, of course, you need to execute and actually deliver the data center.

Thank you, John. Anyone else have anything to add on CapEx or funding for these projects?

So I agree with everything that everybody has talked about. And 1 of the constraints, obviously, is labor. And a lot of these data center builds, the GCs use travelers. So guys that live in North Dakota that are spending 10, 12 months in West Texas to build the data center. We took a bit of a differentiated approach with our EPCM and the fact that they build the vast majority of the data center in a factory. So it becomes an assembly line process rather than a bespoke one-off deal where you've got a number of different trades trying to work together. And that has simplified the delivery for us, but it's also with a fixed price lump sum contract given us certainty. So the way we talked about it is our expectation is that $10 million, $11 million a megawatt complete RFS on the data center, and we expect similar to what John is talking about, to finance that at the project level through the debt markets. We're targeting right now high-yield investment type -- excuse me, high-yield investment bond opportunity. But CleanSpark has -- we've used our bitcoin as the means to prevent any equity dilution on the company for that build.

So we went into the project, signed the lease on the very first day we did our equity contribution. So it is fully funded from our side of the table, and having that certainty of the space in the assembly line and the certainty on the supply chain really derisks that. So in our conversations with lenders that's really what they're focusing on is what are the potential risks that get in the way. So as you think about CapEx and you think about the finance on these, what you also have to consider is the liquidated damages if you fail to deliver on time. and that becomes an existential risk for some folks in this business because if you make promises and there are GPU purchases that can be many times the price of the data center build.

Now you've got some real exposure for liability. So getting the certainty there and being able to laser focus in on what that build cost and the time frame looks like is a differentiator when it comes to financing these projects.

Great insight, Matt, Anyone else have anything to add here?

Just I'll pile on a little bit. I think similar to Matt's concerns regarding risk of delivery. It was 1 of the reasons why we chose to partner with Starwood we have such a big portfolio to do this ourselves, especially not being an expert in building data centers for hyperscaler demands. We want to increase certainty and reduce risk Starwood has built over 70 gigawatts of data center capacity for hyperscalers. They have a captive EPCM within the worldwide digital who's built a lot of these sites. And what we're finding is because of the structure of our model, our cash-on-cash returns are significantly greater than the traditional model because we leverage our site as a contribution on the equity stack.

And if you use an 80-20 model the amount of additional capital we have to include over and above the value of our site in these joint ventures is fairly de minimis, which means that the cash-on-cash return is much greater. It also means we can do more sites in parallel simultaneously. -- and greater capacity without it having to all be credit provided by us. So we feel very optimistic about that. It's going to allow us to move very quickly as we really get started here, and we're super excited about it.

Yes. If I may add, we do something similar. We have access to power, infrastructure. And in some cases, we'll have power plants. But that land once it gets fully developed and you're ready to go vertical becomes much more valuable than what you pay for it, and that can be used as a contribution on the equity side. On the execution side, being a smaller Bitcoin infrastructure company, we were thinking about the risks with execution around these very complex data center projects and -- we approached it in 2 ways.

One is on our initial project, we partnered with a development company that is a group of AWS folks who have experience developing, constructing and operating these facilities. That's allowed us to design and develop and market these sites in a way that customers would expect. And then this past quarter, we started building our own in-house capability, hiring the first major role to the company, a gentleman that is run and built over $10 billion worth of data center projects for Microsoft. I think you built their largest AI facility. -- and hiring is tea here over the last couple of months. So building out our own execution capability to deliver these sites to the end user customers at the level of quality and timing that they expect.

All right. Let's move on to a controversial topic to spice things up a little bit. Some headlines out there would suggest that we are in an AI infrastructure bubble. I'd be curious to get the group's thoughts on this, and I'll leave it open for whoever would like to start here.

I mean I think if you talk to the people who have a need for the compute, they would tell you there's no bubble. They just need a lot of compute and they don't have the ability to deliver on it. I think there are 2 things going on. One is, is there a bubble in the hyperscalers and this $1 trillion spend? Or is there a bubble in the AI compute world, 2 very different things you're looking at. AT&T recently announced that over 40% of their workloads are being done on open weight, open source models. They are not using frontier model providers for that. They expect that number to grow to 70% over the next 2 years.

One of the biggest challenges today is cost per token. And about 60% of workloads done by AI today are fairly mundane. They can be done by open source model or an open weight model on de minimis amounts of infrastructure. So there is a huge need for inference the question is how much of it will happen in the hyperscaler cloud and with the Frontier model provider versus how much of it will happen in the enterprise and with the let's call them neo-clouds for now, but for the neo-clouds providing a full stack of software services that allow people to run and load whatever models they want to run in fully air gapped environments.

One of the reasons we when after the Axion opportunity was, we wanted to have the capability to run fully airgap, fully sovereign increase at scale for enterprises because if you look at what it costs to run in AWS today with dedicated hardware, dedicated resources versus running in private cloud, it is a 10x higher cost. And companies are looking at AI as a cost item, right? It has become part of their OpEx budget. And if they are going to pay huge fees to the frontier model providers versus training their own models and paying once for a model and then running it proprietarily.

You're going to see a lot more inference move towards either behind the firewall or near prem, on-prem inference providers who can allow people to run sovereign models. And I think that is going to be by far the biggest thing moving forward in this industry. And you'll still see a lot of revenue going into the frontier model providers, a lot of revenues still going to the cloud providers. But the first in cleaning of this, I think, is how Google reorganized their AI business. Who is it that runs Google's AI business today. It is the guy that operates their cloud business? It is not their AI team.

Yes, to echo Fred's comments, I certainly -- from CleanSpark's perspective, we certainly don't see it as a bubble. Russ and I had breakfast this morning, we were talking about the increase in rates for these long-term leases. And the -- how rapidly they're appreciating and what we really see is that there are 3 constraints. Capital. So money is flowing where there's the least amount of risk and the highest credit certainty energy availability and those energized megawatts to have conversations with hyperscalers that they're actually contemplating behind-the-meter investments and building their own power plants in order to support the data center, the need for compute that they have. gives you a pretty strong indication that this is certainly not a bubble. And then the last thing is delivery certainty.

As we've entered the conversations for credit assessment of the project and working with the high-yield market. they talk about the fact that certainty of execution is now the predominant factor in determining not only the availability and the willingness for the high-yield market to provide that credit, but also drives the rates in a very strong way. So I see that if you have energized megawatts of power, and you have those -- those exist in areas that don't have strong NIMBYism. I mean, let's be honest, if we sat here a year ago, I don't think anybody on this panel would have assumed that the Governor of Texas would come out and put a moratorium on data centers for certainty of execution. This has become a political hot mess. And both sides of the aisle are mutually opposed to it. But now we find ourselves with the labor unions in agreement with President Trump that it's an absolute necessity to build these sites here for 2 reasons that number one, just like with Bitcoin mining. If you drive Bitcoin mining offshore, it will continue. It will just continue somewhere else, and that doesn't give certainty on block space for U.S. investors.

But number two, these models that are being trained are being trained outside of the U.S. And so that presents an additional level of risk. So we really feel strongly that bringing the right package and the right assets in the right communities and doing an educational process in that is going to be a differentiator to solve for the bubble.

In Texas, we worked with our EPCM provider for our Georgia site, and we did a roadshow. We set it up like a trade show with boots addressing everything from dark sky issues and water usage to ambient noise and impacts on utility rates. And if you can lead with education and facts rather than just hype in theory, it's a huge differentiator that we believe solves for a lot of that issue, but as to whether or not there's a bubble, hard note from my perspective.

I have ever contrarian perspective, and I think this is a better way to look at the question. I would say that whether it's a bubble or not is kind of a moot point because if you look at technology, development and infrastructure build-out for the last 50 years, every single major improvement in our capabilities as a humanity businesses, technology services on a global basis has been preceded by major investments like this. The last 1 was 30 years ago, where people were sort of scratching our head. What is this Internet thing?

And let me tell you what happened. We built out the most advanced telecom infrastructure capability around the globe. We're all super connected. We then added to that mobile and content platforms and social connectivity software, which eventually delivered to us. What everybody is itching and looking at in their hand, where you essentially have a compute content bundle that gets delivered to you effectively for free. And that took billions of dollars of build-out to deliver that type of capability. AI is another one of those. It is the greatest general purpose technology we've developed in the last 30 years. And I can tell you, I've been a technologist for that time. And I've was trained to see these types of waves. This is another big tech wave. You can tell it's a major tech wave when you have infrastructure investments that hit the core elements of the base infrastructure in any given epoch, if you will. And this is the first time we're touching the energy infrastructure to power this technology. So like most of my colleagues are talking about hyperscalers are looking to build out energy.

They're looking to solve the pain, which is speed to power, if you will. That's because the ultimate new bundle is going to be intelligence. We will now have the ability to have intelligence walk around in our pockets at the greatest level we've ever seen, and that's going to be made possible by multiple hundreds of billions of dollars per country. So that's trillions of dollars on a global basis that will be built out to deliver this technology. We cannot stop. As a country, if we stop doing this, we will fall behind because it's so powerful to technology that is now almost at a military level. So I think we were too where we built out roads to sort of protect the country in that sense.

That's the power of this technology. And the reason the companies on stage here are so valuable is because we are part of helping to create that new technology infrastructure. So bubble or no bubble, you can only see it in hindsight, connect the dots, but I'm telling you the dots point to a new form of infrastructure that integrates energy, compute connectivity and telecom into one bundle to deliver tokens or otherwise known as intelligence framework or intelligence infrastructure.

You already have it. We are not in an AI infrastructure bubble, at least according to those -- Sam, did you have something to add here?

Yes. Just very briefly, obviously agree with that. I would say that there are 3 catalysts for the Industrial Revolution. The first 1 was electricity was changed industry than the Internet, it's change industry and now AI, which is going to change the industry. So that -- there's clearly no bubble. But just with respect to the over -- there's a lot of pushback on data center development. And I think there are many communities these days who are upset that some of us are perhaps taking farmland or building these on-site data centers in their communities where it just used to be greenfield. And that's one of the main reasons why white fiber is retrofitting. These are -- we're taking over basically facilities that are left for dead. And we have community days. So for example, in North Carolina, we had a community day where we explained that we're using 95% less water than the facility that was being used before we took it over, that we're creating a lot less noise than the facility that we're taking it over from.

So this retrofit format really solves a lot of the tension that's happening for -- against -- which I think is probably not a great idea, doing greenfields are kind of -- you're going to get community pushback, especially if you're taking up farmland. So this retrofit format is a way to sale from that community pushback that we've been that we've been able to solve with our Community Day with the local communities where we're building these retrofitted data centers.

Mike, could I add a quick note I wanted to add on to something that Fred touched on earlier about the enterprise client and with the question being, are we in a bubble. The answer for Big Digital is now as well. I have -- I spent 20 years in health care. Just 1 small sector of this great economy we have in the U.S. And in the last 2 months, I've had the opportunity to sit down with 2 founding CEOs of health care-focused and drill down to long-term care focused software companies, building AI models to serve that sector. And I heard the same thing from both of them. we have the product, we have the client.

Our biggest fear as we scale up and out is that we can't secure the level of compute that we'll need going forward. And if I could leave anything with the investors in this room is we're always thinking about the big 5, right? We're not thinking about the tens of thousands of enterprise-level companies out there that need the same level of compute and they need it secured and available so that they can grow their businesses. And so that's something that we are at Big Digital and very focused on is the enterprise level clients more specifically with my background in health care, I'm very focused on how can I serve the health care client. But I would say the current narrative whomever was made up by, I'm sure they had their agenda, but it's completely false. -- let me tell you, there is a bubble.

Let me tell you where it's at. So there's about 5 gigawatts of rack space being delivered every year, and there's about 15 gigawatts of chip demand every year. But in Texas, what the governor did was reasonable because there was 474 gigawatts of request in for data centers. We don't have enough foundry silicon foundries on the planet to use that much space. And that's what's scaring everybody. It's because there's 5x the request in Texas that the whole grid commits to. So the issue is not the guys that were rack space or the guys living the chips -- it's every farmer in rancher in Texas or Oklahoma or any state that has a transmission line or a gas line running through their site that says, "I'm going to put a data center there. we get people telling us, "I've got a 3 gigawatt site, the next cobest says, I got 3 gigawatts stand the goose says, "I have 3 gigawatt site. Well, they're all counting the same gas line and they're all count the same gas. And that's the issue that the Governor's office in ERCOT had in Texas.

It was how many of these data center requests are actually real. So there really is 15 gigawatts a year of demand, and we really are -- we're not able to fulfill it. And it's going to go on for as long as we can. What there's not is nearly 500 gigawatts of demand in Texas alone. And that's where I think -- I actually don't think it was a bad thing the governor said, "Hey, wait a minute, how much of this is real. Who has a substation, who has a PPA, who has power, who has your behind the meter plan together." And I think it's reason for any state to ask that question because all of these are not real. We saw the same thing in Bitcoin mining, some -- I know some of you guys were there. Texas ERCOT told us -- and by the way, Core Scientific is 1 of the largest users of power. We talked to ERCOT multiple times a day, all the time. ERCOT told us there was 20 gigawatts of Bitcoin mining demand in Texas. There wasn't. And there's like a bit of that. What it was, was the same guys chasing the same -- everyone wanted to sell me some stuff, they want to sell freshest, -- they want to sell mats.sthe same as people just brokering these deals or putting deals together. So that actually is a bubble. There's not tens of thousands of gigawatts of demand in every single state in the country. So I think what needs to happen is some regulations on, hey, what's real and what's not real. I think that's an unbelievably reasonable request for anybody to make.

I agree with that. Just to level set, it takes 5 to 6 years to build a new gas-fired facility. And while solar and wind or grade, it's intermittent. And until we manage the orchestration issue. So if the U.S. Sun Duke did the study last year, the U.S. has 72 gigawatts of power available to it based on current generation today. you want to add anything to that? -- it all takes a couple to 6, 7 years and billions and billions of dollars to do it.

Yes, there's a dislocation. All right. Last question. We do have 3 of the largest publicly traded bitcoin mics on the stage here. So I'd be remiss to not ask a question about Bitcoin and Mining. So specifically for MARA, CleanSpark in bid tier if either of you had something and, specifically, how do you plan to leverage your mining assets in Bitcoin Holdings to really fit into your longer-term strategy as you continue to execute on the AI data center side?

Well, Bitcoin was obviously higher a year ago, and it was obviously lower a month ago. And so we've developed an institutional-grade trading desk to manage our Bitcoin. We have -- if you go to Bitcoin Treasuries, I think we're 8 or 9 on that list. Fred, I think it's but we have over $1 billion in Bitcoin, and we have $400 million in largely untapped going back lines of credit and we have an institutional grade trading desk to help generate yield on that $1 billion plus a bitcoin that we hold. So we see it really as a capital tool.

Additionally, we've prevailed on land and power sites because of Bitcoin. In Cheyenne, Wyoming, as an example, 110 megawatts from Black Hills Energy, they put out an RFP to acquire that, and there were 2 bidders, One was a $1 trillion hyperscaler and the other was CleanSpark. And we prevailed not because we have a better balance sheet or were more handsome but because of the fact that we were able to deploy immersion cooled.

From our perspective at Saluna in this topic, Bitcoin is a core part of our business, we're not 1 of the largest big coin miners, we're not a bitcoin miner at all. We're 1 of the largest bitcoin hosting companies that host some of the largest miners -- and that business approach allows us massive amounts of flexibility, especially as it relates to partnering with these large power plant owners. They have a problem, they have wasted energy that they need to monetize very quickly to return these projects to profitability. So we can deploy Bitcoin mining to some of these sites, especially the smaller ones, we can continue to grow our hosting business, build a relationship with these power asset companies, which allows us to then monetize that asset while we're developing the AI project, which tends to take longer and also show them a quick win for monetizing that power asset, which unlocks for us a lot more projects that are even bigger. That's how our pipeline has grown so significantly over the last few quarters because we've developed this very strategic approach to monetizing their power fast and then opening up the door to a longer relationship where we can build much larger scaled operations on the AI side. So Bitcoin has become a very strategic platform for us as we transition into.

All right. With that, that will conclude our panel discussion for today. Thank you all for attending.

Greetings, and welcome to MARA's Second Quarter 2026 Earnings Conference Call. [Operator Instructions] Please note, this conference is being recorded. I will turn the conference over to your host today, Robert Samuels VP of Investor Relations. You may begin.

Thank you, operator. Good afternoon, everyone, and welcome to MARA's Second Quarter Fiscal Year 2026 Earnings Call. Thanks so much for joining us today. With me on today's call are our Chairman and Chief Executive Officer, Fred Thiel; and our Chief Financial Officer, Salman Khan.

Today's call includes forward-looking statements, including those about our growth plans, liquidity, and financial performance. These involve risks and uncertainties, and actual results may differ materially. We disclaim any obligation to update these statements, except as required by law. For more details, see the Risk Factors section of our latest 10-K and other SEC filings.

We'll also reference non-GAAP financial measures like adjusted EBITDA, which we believe are important indicators of MARA's operating performance because they exclude certain items that we do not believe directly reflect our core operations. Please see our earnings release for reconciliations to the most comparable GAAP measures.

We hope you've had the chance to read our shareholder letter and look forward to your feedback. We'll begin with some prepared remarks from Fred and Salman. After their comments, we will open the call to Q&A. I'm going to turn the call over to Fred to get things started. Fred?

Thank you, Rob. Good afternoon, everybody. Thank you for joining us. For much of the past 2 years, the AI conversation has focused on models, chips, and capital, but underneath all of that is a more basic requirement, power. That is becoming the central infrastructure challenge of the AI era. The market has no shortage of ambition or investment. What it lacks is enough energized, permitted capacity in the right places available on a time line customers can use.

So the question is no longer who can fund the next wave of compute, it's who has the power. That question goes directly to MARA's strength. We did not arrive at this opportunity by chasing a new trend. We arrived here after more than a decade of solving the same operating problem at global scale, securing power, deploying compute, and running infrastructure efficiently around the clock.

Through Bitcoin mining, we built one of the world's largest distributed compute platforms, spanning 19 data centers across 4 continents. Along the way, we accumulated strategic land and power assets, deep technical expertise, and a disciplined framework for deciding where each megawatt can create the most value. Today, MARA is applying that foundation more broadly. We own, develop, and operate digital infrastructure across power, land and compute.

Depending on the opportunity, we can convert electricity into higher-value compute ourselves or provide infrastructure to customers who need it. That flexibility matters. Many companies entering this market are still searching for sites, power, and operating capabilities. We have spent years assembling them. Our move into AI infrastructure is therefore, not a break from MARA's history. It is the next use of the platform we created.

The second quarter marked another important step in that evolution. We advanced the Long Ridge transaction towards closing. And after quarter end, we announced that we acquired the rights to a strategically located power site in Matagorda County, Texas, with the potential to support approximately 2 gigawatts of future capacity upon ERCOT and interconnection approval.

On completion of the pending transactions and required approvals, we expect our power portfolio to reach approximately 4.8 gigawatts, which would more than double our current capacity. We believe that would establish one of the largest powered land portfolios in the industry and create a significant platform for long-term shareholder value. As the opportunities become clearer, so has our focus.

MARA operates one integrated digital infrastructure platform built around power, land, and compute. Digital infrastructure is our primary growth focus. That is where we are developing campuses and pursuing long-term customer relationships at scale. Exaion and our technology initiatives add targeted capabilities around that core. They help us address specialized customer needs, improve the utilization of our infrastructure, and extend the value of the assets we own.

We manage all of these capabilities as one platform, and we allocate capital across them using the same filters expected returns, customer demand, execution risk, and contribution to long-term shareholder value.

Let me spend a few minutes on digital infrastructure, our primary growth focus. The demand picture is straightforward. AI infrastructure investment is accelerating while the supply of power-ready sites is not keeping up.

Industry estimates suggest that the 4 largest hyperscalers alone could invest approximately $725 billion in AI infrastructure during 2026 and that annual capital spending could exceed $1 trillion by 2027. The power requirement is rising just as quickly. U.S. data center electricity demand is expected to grow from about 31 gigawatts in 2025 to 41 gigawatts in 2026 and 66 gigawatts in 2027.

New generation and transmission are not coming online at the same pace. That imbalance is increasing the value of infrastructure that's already energized or can be delivered with greater certainty. Our strategy is designed for that environment, to own scarce powered assets and create as much long-term value from them as possible. The Matagorda County site is expected to add approximately 2 gigawatts in one of the country's largest power markets.

Just as important, it's expected to provide enough wholly owned capacity to support our transition away from hosted mining as existing agreements expire. That should increase our operational control, improve unit economics, and give us greater flexibility in allocating capital. The pending Long Ridge acquisition is equally important. We believe it will transform our existing Hannibal campus by adding adjacent land while contributing positive EBITDA at closing.

With more than 70% of Long Ridge's power output contracted under long-term agreements, we expect the transaction to enhance earnings while significantly expanding our AI infrastructure opportunity. Together, these transactions reflect our infrastructure investment model. We acquire scarce powered assets, enhance their strategic value, develop high-quality digital infrastructure, and secure long-term customers. As those assets are developed, they can become durable cash flow generators that remain in our portfolio or can be monetized, allowing us to recycle capital into future opportunities.

Speed, certainty, and reliability are principles that define how we invest, how we build, and how we aim to serve customers. Speed matters because customers cannot wait years for power. Our portfolio of energized sites can support earlier in-service dates than many competing developments, giving customers access to capacity when they need it.

Certainty matters because infrastructure must be delivered on time, on budget, and to specification. We believe our development strategy, our relationships with utilities and equipment providers, and our partnership with Starwood will give customers greater confidence in execution. The Starwood partnership also gives us the ability to scale with proportional capital support.

And reliability matters because mission-critical AI workloads require experienced operators. MARA brings years of experience designing, owning, and operating large-scale compute infrastructure. Starwood adds engineering, procurement, construction, and development capabilities backed by more than 7 gigawatts of delivered infrastructure for many of the world's leading hyperscalers and frontier AI companies.

Together, we offer a combination of operating experience, development expertise and capital discipline that we believe few competitors can match. Commercial momentum continues to build. Our goal is a diversified customer base across hyperscalers, AI native cloud providers, silicon vendors, and enterprises with the right balance of credit quality, returns and long-term portfolio value.

Working alongside Starwood, we are progressing lease discussions across multiple sites, and we remain confident in our ability to sign at least 2 leases before year-end. The objective is not merely to sign tenants, it is to establish durable customer relationships that maximize the value of our infrastructure over decades. Within that integrated platform, Exaion gives us a targeted capability in sovereign AI infrastructure.

The customer need is becoming clearer as AI moves from experimentation into day-to-day operations. Once AI becomes mission-critical, enterprises care much more than raw compute. They also care about where their data sits, which rules govern the infrastructure, how resilient the service is, and how much control they retain. That is the market Exaion was built to serve.

As a European company, Exaion can provide private cloud infrastructure governed under European jurisdiction. For enterprises and public sector organizations operating within the EU regulatory framework, that is a meaningful advantage. Customers can deploy advanced AI workloads while keeping control of their infrastructure, data and operations. For critical infrastructure, regulated industries and government adjacent services, that level of sovereignty is moving from a preference to a requirement.

The addressable market is also much larger than new AI applications alone. Roughly 80% of enterprise data still sits outside the public cloud. As organizations modernize that data and infrastructure for AI, they will need providers that can meet demanding standards for security, compliance and operational resilience.

Exaion already has credibility in those environments.

It operates critical infrastructure supporting EDF's nuclear reactor operations where reliability is simply nonnegotiable. And for those who are not aware, EDF is one of the largest operators of nuclear power in the world. Its selection for the AION Consortium, an EU-backed initiative targeting approximately 3 gigawatts of AI-ready data center capacity provides further validation.

We are also advancing opportunities outside Europe, which supports our view that sovereign AI infrastructure is becoming a global requirement, not just a regional trend. Our technology initiatives are another targeted capability within our digital infrastructure platform. It takes operating knowledge developed inside MARA and turns it into technology that can improve our assets and serve outside customers.

Running a large distributed compute platform has taught us a great deal about power management, infrastructure optimization and digital asset management. Some of the tools we built for ourselves now have clear applications beyond our own fleet. Vertebr.AI is one example. The platform manages power allocation and infrastructure performance in real time.

In our mining operations, it has helped us add computing capacity with the same electrical footprint. In other words, more output without needing more power. As power becomes more valuable, that capability should matter well beyond mining. AI data centers, independent power producers, and other energy-intensive businesses face the same need to improve utilization, operate more efficiently and lower costs.

The second platform is Hashrate Under Management or HUM, our blockchain financial infrastructure platform. This is the first time we're discussing HUM publicly. We're doing so from a position of demonstrated commercial traction, not simply future potential. Both HUM and Vertebr.AI reflect the same principle. Innovation should increase the value of the infrastructure we own and create value for customers at the same time.

That brings me to Bitcoin mining. It remains an important part of MARA, not because it defines the limits of our future, but because it continues to strengthen the broader platform. Mining gave us the foundation, strategic power assets, experience operating large-scale compute and the capital allocation discipline we use today. In that sense, mining was never the final destination. It was a platform we could build from.

It still plays 3 important roles. First, it generates cash flow that supports investment across the business, while we continue to operate with one of the industry's lowest cost structures. Second, it gives us flexibility. We can deploy mining equipment quickly at a newly energized site and begin monetizing the power while an AI facility is being designed, permitted, and built. When customer demand is ready, that same site can transition toward AI or high-performance computing without leaving the infrastructure idle in the meantime.

Third, mining is still one of our best sources of operating insight. The work of optimizing power use, improving compute efficiency, and managing mission-critical systems at scale directly informs how we approach AI infrastructure.

We will keep improving the mining business through disciplined fleet modernization and intelligent power management. As more efficient machines replace older equipment, we can increase compute within the same electrical footprint and improve the economics of the operation. So we do not see Bitcoin mining and AI infrastructure as competing businesses. They are different applications of the same underlying asset, power.

The capital allocation question is, therefore, simple. Where can each megawatt create the most value? In one market, the answer may be Bitcoin mining and in another, it may be AI infrastructure, sovereign cloud, or enterprise computing. Our advantage is that we have the assets, expertise, and flexibility to make that decision dynamically as market conditions change. We believe that flexibility is a meaningful competitive strength and an important driver of long-term shareholder value.

The first half of 2026 was about expanding and transforming the platform. We grew our portfolio of powered infrastructure, advanced transformational transactions, strengthened the commercial pipeline, and continued investing in the initiatives that can drive MARA's next phase of growth. The second half of the year is about execution.

Our focus is clear: convert infrastructure into long-term shareholder value by signing customers, bringing assets online and demonstrating the earnings power of the platform we have spent years assembling. Over the coming months, we expect to complete the Long Ridge acquisition, advance lease discussions across the digital infrastructure portfolio, expand Exaion's international presence and continue commercializing our technology initiatives.

Most importantly, we expect the investments we have made over the past decade to become increasingly visible in our financial results. The foundation has been built. Our focus now is monetizing it. Later this year, we look forward to hosting our Investor Day. We plan to provide a deeper look at our strategy, showcase our infrastructure portfolio, and demonstrate how the pieces of our business work together to maximize the value of every megawatt we own.

So let me come back to the question I raised at the start. Who can power, build and operate the next wave of compute? MARA has spent more than a decade building an answer. We have assembled one of the industry's largest portfolios of powered digital infrastructure and developed an operating experience to put those assets to work. The opportunity in front of us is to turn that foundation into a broader platform for the next generation of compute and to do it with the same discipline that built the company.

Bitcoin mining provided the foundation, digital infrastructure. Exaion and our technology initiatives expand the value we can create from that foundation. Together, they position MARA across multiple layers of the AI infrastructure value chain while maintaining discipline in how we allocate capital.

Ultimately, our shareholders should judge us not by our vision, but by our execution. The AI infrastructure market is moving quickly and credibility will be earned by consistently delivering results. Power is becoming the defining resource of the AI infrastructure market. Our objective is to convert the power, assets and expertise we have assembled into durable value and to establish MARA amongst the leaders of that market.

Thank you for your continued support and confidence in MARA. With that, I will turn the call over to Salman.

Thank you, Fred. Good afternoon, everyone. Two things defined Q2 for MARA. Bitcoin prices created a challenging revenue environment, and we used the quarter to fundamentally transform our power portfolio and capital structure. That context matters as I walk through the numbers.

During Q2, we made meaningful progress in building MARA's digital infrastructure platform, taking actions that we expect will expand our total power portfolio to 4.8 gigawatts, nearly 2.5x its size at the beginning of the year, and secure our position as one of the industry's largest holders of digital infrastructure power capacity.

After quarter end, we acquired rights to 1,200 acres at a strategically located powered land site in Matagorda County, Texas, representing up to 2 gigawatts of potential capacity subject to ERCOT and interconnection approvals. Our confidence continues to be reinforced by encouraging interest from prospective tenants. We have also advanced the Long Ridge acquisition by securing approval from holders of Long Ridge's senior secured notes to assume the notes at closing. The transaction will close after FERC approval, which we expect to occur soon as guided previously.

Subsequent to quarter end, we further advanced the Long Ridge acquisition by entering into 2 Bitcoin-backed credit facilities with Coinbase and Two Prime at a weighted average cost of debt of 7.56% for incremental borrowings under these facilities of $600 million. In addition, we refinanced our existing $150 million facility with Coinbase and consolidated it into the new Coinbase facility. This borrowing originally due in Q1 of 2027 will now mature in 2 years, along with the incremental $600 million.

These financings strategically activate a portion of MARA's Bitcoin reserves as a nondilutive funding source while preserving our exposure to Bitcoin's potential long-term appreciation. The facilities will be used towards funding the cash consideration for the acquisition and together with the assumption of certain of Long Ridge's existing indebtedness, provide funding towards completing the transaction. To be direct, we are funding a $1.5 billion enterprise value acquisition through a Bitcoin-backed debt and assumption of Long Ridge's balance sheet, all nondilutive financings. This is the capital discipline we are committed to.

Once completed, the Long Ridge acquisition is expected to contribute approximately $144 million in annualized EBITDA and durable free cash flow with roughly 70% of its output secured under long-term contracts. These contracted cash flows will diversify our revenue base beyond Bitcoin mining, while our capital-light partnership with Starwood will preserve balance sheet flexibility as we develop AI and high-performance computing opportunities across our own power portfolio.

Together, these attributes strengthen our financial position and reinforce our disciplined approach to capital allocation. None of this happened overnight. It is the result of deliberate work across the organization, and that work continued to gain momentum this quarter.

With that context, I will turn to Q2 financial performance, capital allocation, and balance sheet activity. The Bitcoin price environment remained challenging, reflecting broader pressure across risk assets, driven by macro uncertainty, tighter risk appetite and continued pressure on mining economics. It is important to view this alongside the substantial progress we are making to build a more diversified digital infrastructure platform.

Revenues during the second quarter of 2026 were $174.9 million compared to $238.5 million in the prior year period. Bitcoin production contributed a $7.2 million increase year-over-year, though this was offset by a 28% decrease in Bitcoin's average price, which reduced revenue by $65.9 million. Other revenues declined approximately $4.9 million, primarily reflecting lower revenue from other digital assets and elimination of our hosting services compared to the same period.

During the quarter, we mined 2,422 Bitcoin or 26.6 Bitcoin per day, approximately 64 more Bitcoin than the prior year period. We won 700 blocks, up 1% year-over-year and up 8% from Q1 of 2026. We held a total of 35,577 Bitcoin at the end of the quarter, valued at approximately $2.1 billion at a $58,524 spot price, down from 49,951 Bitcoins held a year ago.

Of the total Bitcoin held, approximately 26% or 9,270 Bitcoin were loaned or pledged as collateral. Of that, 4,742 Bitcoin were loaned under our digital asset management strategy, generating approximately $4.3 million of interest income during the quarter. We delivered an energized hashrate of 70.3 exahash per second, increasing 22% from 57.4 in Q2 of 2025.

Sequentially, hashrate was down modestly from 72.2 exahash as we phased out legacy miners. This reflects continued fleet optimization and opportunistically upgrading our infrastructure by phasing out legacy miners to boost our total hashrate. Our share of available mining rewards reached 5.9%, up from 5.5% in Q1 of 2026.

Approximately $343 million of our net loss this quarter was driven by unrealized mark-to-market fair value adjustment for digital assets, a direct reflection of the drop in Bitcoin price during the quarter. In total, we reported a net loss of $611.3 million or negative $1.60 per diluted share compared to net income of $808.2 million or $1.84 per diluted share in the second quarter of 2025.

As a reminder, every $10,000 change in Bitcoin price results in an approximate $350 million impact on the fair value of digital assets on our income statement, which is an unrealized noncash adjustment. Accordingly, adjusted EBITDA for the quarter was negative $360.9 million, similarly dominated by Bitcoin mark-to-market change compared to $1.2 billion in the prior year period. We use adjusted EBITDA as a supplemental measure of operational performance and a full reconciliation to net loss is included in our shareholder letter and earnings deck.

Our daily cost per petahash per day improved 4% year-over-year to $27.7 from $28.7 in Q2 of 2025. And over the past 9 quarters has improved by 27%, which we believe remains among the lowest at scale in our sector. That is the cost structure behind the efficiency Fred referenced earlier, and it is the cost structure we expect to bring to every megawatt we convert to AI infrastructure.

Our cost per kilowatt hour was $0.04 for our owned sites in Q2 2026. Purchased energy cost per Bitcoin for our owned mining sites was $38,690, up from $33,735 in Q2 of 2025, primarily due to higher network difficulty driven by growth in global hashrate. Our own efficiency metrics improved. The per Bitcoin cost increase is entirely a function of rising global difficulty, a market dynamic outside our control.

Despite the increased difficulty levels, Bitcoin production at our owned mining sites increased 2% over the same period. Looking ahead, our most significant third-party hosting arrangements are set to expire beginning in the third quarter of 2027, with all the arrangements concluding by the first quarter of 2028, at which point in time we expect to eliminate third-party hosting costs and improve our cost per kilowatt hour.

General and administrative expenses, excluding stock-based compensation were $69.5 million for the quarter compared to $40.1 million in the prior year period. The increase reflects the scaling of our operations, higher personnel costs associated with headcount growth from the prior year period, and administrative fees in support of our expanded global footprint.

Acquisition and integration costs burdened our G&A by $15.4 million, and we also incurred a $10.2 million litigation settlement, representing the amount paid in connection with the final resolution of a patent dispute. Excluding both items, underlying G&A was approximately $43.9 million and more comparable to the prior year.

Compared to Q1 of 2026, G&A benefited from lower headcount costs related to the previously announced reduction in force. We expect our quarterly G&A run rate, excluding stock-based compensation and acquisition and integration costs to continue to trend lower as these savings are realized over time.

Now let me turn to the balance sheet and liquidity. We ended the quarter with $421.3 million in cash and cash equivalents and approximately $2.5 billion in combined cash and Bitcoin. Our capital allocation strategy remains disciplined and focused on supporting long-term shareholder value. Following the expected close of the Long Ridge acquisition, we anticipate assuming approximately $900 million of Long Ridge's debt.

In addition, as I mentioned previously, MARA has recently added $600 million in borrowings, which are secured by our Bitcoin holdings. As a result, 54% of our Bitcoin holdings have been pledged as collateral under our borrowings. These financings strategically activate a portion of MARA's Bitcoin reserves as a nondilutive funding source while preserving our exposure to Bitcoin's potential long-term appreciation.

With that, I will turn it back over to the operator. Operator?

[Operator Instructions] The first question comes from Greg Lewis with BTIG.

I did want to touch a little bit on Long Ridge. I guess a couple of things about Long Ridge. Just the first is around the potential timing. Has there been any feedback from the federal, local or state levels about the potential closing? Are there hurdles that need to be done? I guess, that's my first question.

We haven't received any feedback yet, and we don't think there's anything at this stage that's going to block the approval. If you look recently, WULF just got approval for one of their acquisitions, and we expect FERC to respond to us definitely before year-end, but much sooner than that.

Okay. Super helpful. And then just as we think about the opportunity set in Hannibal, like as we -- as you're negotiating with potential HPC customers, is a little bit of chicken and egg where really until the deal goes through and the land that is required to build this out is in place, we're kind of in a holding pattern. Is that kind of a fair way to think about it?

Well, we're in a holding pattern from signing a lease. That being said, we are very actively engaged with prospective tenants in evaluating exactly what they're going to build, how they're going to do it, how the fiber is going to be laid, et cetera. So we're moving along at about the same pace as if the deal was already closed.

But if you're familiar with how these lease discussions go, from when you have kind of a letter of intent from a prospective tenant, it can be 60 days, sometimes plus just talking about design and how you're going to permit and all that. There's nothing holding us back now other than closing really the transaction, but things are moving along at a very good pace with the tenant.

The next question comes from Paul Golding with Macquarie.

Congrats on the announcement of the new site in Texas. I wanted to ask with the recent developments around the Texas audit process tacked on to the batch study process. Are you getting any additional interest? Or is there a market repricing on your existing energized sites that are available? And then I have a follow-up.

I mean, there's broad demand across a number of sites with multiple tenants in discussions on multiple sites. The Matagorda site in and of itself is a hugely attractive site for tenants. And everybody assumes that this is going to eventually go through, meaning the Batch 0 and Batch 1 processes that Governor Abbott has delayed.

If you look at these requests and the queues for a number of years, it's been pretty prevalent that people have submitted requests and because they haven't had to put down huge deposits. There are a lot of phantom requests in the system. And so by going through the audit process that Governor Abbott has requested to be done, they will flush out a lot of those.

We're very pleased that for our particular site, there's no infrastructure improvements that the utilities have to do to bring power to the sites. We already have multiple transmission lines coming into the site. We're very close to the power generating source in one of the biggest power markets in Texas. So we feel very confident that as this audit process progresses, we'll progress through the queue as well. And just a question of things happening in the right way.

And then maybe just a follow-up on the Matagorda County site itself. It seems that the purchase structure is potentially favorably set up in terms of milestone payments relative to approval. Could you just give some more color around how that was struck and some of those details just working off of the presentation with the result?

Yes. I think the way to look at it is this, is that, obviously, -- because when we did the transaction, there wasn't 100% certainty about Batch 0 approval. There are contingencies that drive the deal. And so the idea is that from our perspective, we have the ability to wait until Batch 0, but there is, at some point, a terminal point in time where we either have to close or step away. But it is structured in a way to benefit us.

And just to add to that, as what Fred mentioned, this is attractive for us and our shareholders as it aligns our counterparties' interest together with the development of the project. And as we progress, everyone progresses.

Maybe just -- I'll try to sneak a third one in, sorry. But does this fall within the Starwood partnership if you were to do a deal and have a capital partner through them for development of that site?

Any deal has the opportunity to fall within the partnership, the deals that were already allocated to the partnership are those that were in the portfolio at the time we signed the deal.

The next question comes from Chris Brendler with Rosenblatt.

I would love to hear more about some of the -- how we should think about the revenue opportunity from Exaion, EDF as well as HUM. I don't know if you talked about it kind of listening to calls at once, but I'd just love to hear how we should think about it as the revenue base -- if the revenue base will diversify with these initiatives.

Yes. I mean the HUM has now contractual revenues. They're not going to be material to the overall total revenues in the near term. HUM is essentially an instrument that allows Bitcoin mining pools to gain a little bit more certainty in how they're paying out their fees. So it's a type of -- you could think of it as a way to leverage our hashrate and take part of our Bitcoin and use it as a way to provide more stability to pools.

And so there's a certain limit to where it can grow. But I mean, it is definitely an 8-digit a year business at -- on an annualized basis. So that will be not necessarily material, but it will still contribute nicely. As we look at Exaion, Exaion is just coming out of the fold, if you would. So Exaion's revenues this year will be in the low 8 digits, most likely, and we expect them to continue to grow as Exaion diversifies the customer base.

Having been captive with EDF, it takes a little while to kind of go from building interest with new customers to closing contracts, but we're already seeing good traction there. And we're very bullish long term on Exaion, especially when you think about how the continuum from powered shells, which is essentially what we're doing together with Starwood.

And as you go to campuses where you're moving from just a powered shell to potentially colocation type contracts to then moving to potentially Platform as a Service or think of it as the GPU rental to then fully managed. And an important thing to think about is that -- and this has been reported in the news by analysts more and more frequently now is the frontier models are great for certain work, but they are hugely expensive to use for doing the bread-and-butter work that many people use AI for, such as analyzing e-mails, writing materials, coding, things like that.

And so you're seeing a growing interest in open source models and open weight models that people want to deploy. And a lot of the forecasts are showing that these models will grow as a percentage of the overall amount of compute that's deployed. And this is Exaion's forte. One of the things they do for EDF historically has been manage a portfolio of models where somebody who's going to run a particular task can choose the model they want to run, and it's all running on their infrastructure.

And so with the combination of the needs for sovereign compute, the needs to keep your data within your own firewalls, and the desire to lower token costs substantially, we believe a lot of these open source models will gain significant traction, which is very additive from an Exaion perspective. So we think -- if you think of the market, we're addressing it kind of from 2 ends and they will come and merge towards the middle.

My second question was on Starwood and just thinking back to a very helpful meeting we had in Vegas where you provided a lot of color on not just the timeline, but also the numbers. I am just wondering potential -- I guess, what the numbers could look like. I just wanted to see if you -- you're getting closer to actually executing on one of these transactions, any changes in your view, more bigger confidence, pricing, demand and timelines? Are they all sort of as expected or any changes to that over the last 3 months?

I think we've been surprised by the demand being greater than what we initially expected, at least the response from the tenants, but that's obvious given the fact that there's not a lot of available power, and we just happen to have a lot. I think the expectations regarding the quality of discussions that Starwood was going to bring us into have been definitely exceeded.

I think the professionalism of the team, how they drive conversations regarding build and design because we're kind of at that stage with a handful of these things have been very good. And financially, there's no difference in the calculus. So I think we're very, very pleased with how things are going with Starwood and the team has been great to work with. So we're very pleased with that. And I think they are quite pleased with kind of how things are shaping up.

Looking forward to that first transaction.

The next question comes from Michael Donovan with Compass Point.

Hybrid energy storage prototype system with TAE Power Solutions. What performance and economic thresholds does it need to meet before MARA considers broader development?

So we're in the process -- so we're currently running [ Vertebra ] which is where that partnership fits within our solution set. We're running Vertebra and we developed originally Vertebra for -- to operate at our wind farm because we needed to be able to do load following or rather follow the amount of energy being generated by the wind farm and operate our compute to maximize every electron that wind farm is generating as opposed to having a substantially lower threshold and just running it at that level.

We then adapted the technology so it can now do the same thing regarding load following. So if you have a system that has a varying demand, we can take the other side of the power equation and then build -- load the batteries and at the same time, load -- create a load that uses that excess energy.

And the last area where Vertebra is actually very exciting in the data center world is when you fire up a new data center, you have to use load banks to simulate load. And Vertebra as a solution together with mining is a perfect load bank for start-up of data centers. And so we expect to see some great utilization there. So that's where the TAE relationship works. It's their technology that we've worked with them on to integrate into Vertebra that allows us to do what we do so effectively there.

Appreciate that, Fred. And then just to switch over to global blockchain side. This past week, there's a lot of discussion with Slipstream and Coldcard wallet. So looking at Slipstream, how do you guys think about monetizing that further?

Slipstream was a platform we developed originally back in the time when ordinals and these nonstandard kind of payloads were very attractive and there were great transaction fees. And over time, demand for ordinals has declined. When the Coldcard incident happened, we made a decision that we were basically going to allow people to use Slipstream as a way to move their coins in a way that the hackers wouldn't be able to take advantage of it.

And so longer term, I think Slipstream is a tool that our foundation, the MARA Foundation, really controls. It's -- the MARA Foundation's efforts are really around supporting Bitcoin and ensuring Bitcoin's continued development and safety. We're very active in the quantum resilience piece of what's going on in the Bitcoin market there. And Slipstream is really viewed as kind of a utility that we're providing to the market there.

So I wouldn't allocate significant revenues to it at all. It's more sort of a tool that we believe is really better for the community to utilize. Over time, if Bitcoin begins to see increases in the need for payloads that are not just traditional transactions, then Slipstream could very well be a great product and service in that area to generate revenue. But currently, we don't see a lot of demand in that area, but you never know.

And one more, if I may. Obviously, adding Matagorda increase your portfolio quite a bit. But looking at the non-hosted capacity, I believe you mentioned before you're considering about 90% of -- or evaluating 90% of that for AI. How should we think about that percentage now after you -- as you guys are looking at this more deeply?

So if you think about our overall power portfolio or our Bitcoin mining portfolio, about 30% -- a little less than 30% is hosted today. The rest is owned and operated. And so the 90% relates to that, 90% of the 70%. So as the 30% approximately comes off of contract, we could run that at Matagorda while sites are being developed for AI.

So you have to realize that as power comes on in Matagorda, it may take 18, 24 months, whatever the build time is for tenants. In the meanwhile, we can monetize all that power using Bitcoin mining if we want. And again, the power doesn't come on all at once. It comes in gradually over the 2 years' time.

But what that gives us is the ability to not lose that hashrate when the hosting contracts come off, but rather just reallocate a portion of Matagorda to that for a period of time. It serves 2 purposes. One is it lowers our cost to mine quite significantly compared to what we were paying in the hosted environment. And it allows us to use power and have an offtake for that power. We can monetize the power while the data centers are being built. So it's a very symbiotic kind of shift.

At this time, I would like to turn the call back over to Mr. Robert Samuels for closing comments.

Thanks, operator, and thank you, everyone, for joining us today. If you do have any questions that were not answered during today's call, please feel free to contact our Investor Relations team at ir.mara.com. Thanks very much, and enjoy the rest of your day.

Thank you. This does conclude today's teleconference. You may disconnect your lines at this time. Thank you for your participation, and have a great day.

Greetings, and welcome to the MARA First Quarter 2026 Earnings Conference Call. [Operator Instructions] As a reminder, this conference is being recorded.

It is now my pleasure to introduce Robert Samuels, Vice President of Investor Relations. Thank you. You may begin.

Thank you, operator. Good afternoon, everyone, and welcome to MARA's First Quarter Fiscal Year 2026 Earnings Call. Thanks for joining us today. With me on today's call are our Chairman and Chief Executive Officer, Fred Thiel; and our Chief Financial Officer, Salman Khan.

Today's call includes forward-looking statements, including those about growth plans, liquidity and financial performance. These involve risks and uncertainties, and actual results may differ materially. We disclaim any obligation to update these statements, except as required by law. For more details, see the Risk Factors section of our latest 10-K and other filings.

We'll also reference non-GAAP financial measures like adjusted EBITDA, which we believe are important indicators MARA's operating performance because they exclude certain items that we do not believe directly reflect our core operations. Please see our earnings release for reconciliations to the most comparable GAAP measures. We hope you've had the chance to read our shareholder letter and look forward to your feedback. We'll begin with some prepared remarks from Fred and Salman. After their comments, we will open the call to Q&A.

I'm going to turn the call over to Fred to kick things off. Fred?

Good afternoon, everyone, and thank you for joining us. Q1 2026 was a redefining quarter for MARA, not an incremental one. This was a quarter where we executed deliberately across multiple fronts at once and move the company decisively forward.

During the quarter, we moved the Starwood joint venture from announcement to execution, closed our acquisition of a majority interest in Exaion and retired about 30% of our outstanding convertible debt, all while realigning the organization to fit the business strategy. Shortly after quarter end, we announced a definitive agreement to acquire Long Ridge Energy & Power from FTAI Infrastructure. These were not isolated events. They are connected pieces of a strategy that is now fully in motion. That strategy starts with a single conviction, the next phase of digital infrastructure value creation will be shaped by the control of power where it is located, when it's available and how it can best be monetized.

AI adoption is accelerating faster than power can be brought online to meet demand. That is not an opinion. It's the defining constraint of this market. Available connected energy is debottleneck on AI compute growth. The ability to source control and dynamically allocate that power is a structural advantage. And the lack of available power will negatively impact semiconductors related to AI, if there's not sufficient capacity to absorb ship supply, some semiconductor vendors are investing directly and locking up demand as evidenced by NVIDIA's recent investments.

MARA has positioned itself squarely in the bull's eye, with already energized power to enable hyperscalers to in the near term energize compute with our previously 1.9 gigawatts of power capacity. And now with the addition of Long Ridge, we're having advanced conversations with multiple prospective tenants across multiple sites.

Let me start with Long Ridge. We view Long Ridge as a land and power acquisition to develop our premier compute campus. It is a strategic enabler for our [indiscernible] operations, adding to the site 1,600 acres with the path to grow the existing 200 megawatts of power to over 1 gigawatt. It will establish a leading AI HPC data center campus in the PJM Interconnection, one of the most active data center and power markets in North America.

In a market where power and infrastructure constraints take years to solve, Long Ridge gives us exactly what is needed to build our value to shareholders upon closing. This is not a greenfield site. It is a site that is already operational, already generating cash, and that gives us immediate access to the infrastructure, interconnection and physical footprint required to scale to over 1 gigawatt. The power is there. The land is there. The water is there. The fuel supply is there, and the interconnection is there.

The centerpiece of the campus is an approximately 505-megawatt nameplate combined cycle gas turbine, one of the most efficient in the entire PJM Interconnection. It generated $144 million of annualized adjusted EBITDA in the second half of 2025 with 76% contracted capacity. This is stable visible cash flow from the moment we closed the transaction.

Beyond the power plant, the campus consists of over 1,600 in U.S. acres and includes the 200 megawatts that mark existing capacity at Hannibal. As a signing, we have already submitted plans to augment the Hannibal Interconnect and we will move quickly post close to further expand power capacity at the power plant.

At close, we plan to retain Long Ridge's skilled team, consisting of about 25 full-time employees that have deep operational knowledge of the facility. They will supplement our existing energy asset operating expertise.

Here's what matters most about the scarcity of this asset. If you try to build this from scratch today, the land, the power, the permitting, the water, the interconnection, you're looking at $2 billion to $3 billion of capital and 7 to 10 more years of development time. We're stepping into a platform that is already built, already operational and already generating cash flow. Assets like this are very hard to come by. Some might even call it a unicorn. So when they do, you move.

In total, Long Ridge gives us over 1 gigawatt of total potential capacity and a path to scale to 600 gross megawatts of [indiscernible] in critical IT load over time. This transaction increases our owned and operated capacity by approximately 65%, taking us from about 1.3 gigawatts of energized capacity today to roughly 2.2 gigawatts by closing and including expansion capacity to 2.4 gigawatts.

We've been actively engaged with multiple top-tier potential tenants around this asset. These conversations are now accelerated since announcing this transaction, and the current plan calls for an initial 200 megawatts of AI build-out with construction beginning around the first half of 2027, initial capacity coming online in mid-2028.

The power plant is not the [indiscernible], it's the enabler. It provides reliable control over an increasingly scarce input at a cost of approximately $15 per megawatt hour. This is a cost position that very few can match as well as a positive cash flow tomorrow. And to be clear, our existing bitcoin mining operations at Hannibal will continue without interruption until such time as the data center campus needs the power. MARA does not expect to reduce Long Ridge's current supply of power into the PJM grid. As we develop compute capacity behind the meter, we will pair that demand incremental generation over time.

Our goal is to continue to operate Long Ridge Energy and ensure that consumers continue to benefit from the reliable power they have been accustomed to. Taken together, Long Ridge gives MARA a scaled power advantage platform, immediate and durable cash flow and a clear path to build one of the leading digital infrastructure campuses in this market.

Next, I'd like to talk about our strategic partnership with Starwood, which has made meaningful progress during the quarter. We moved from announcement to execution, advancing permitting and site preparations across our portfolio and entering active tenant discussions with multiple counterparties, including hyperscalers and across 90% of our existing owned and operated sites, including the Long Ridge campus. I want to take a moment to explain why the structure of this partnership matters because it's fundamentally different from a traditional lease and that distinction has real economic benefits from our and its shareholders.

First, Starwood is a trusted institutional counterparty with global investment expertise and a dedicated data center development platform. Their team has developed, built and put into operations more than 7 gigawatts of definer capacity worldwide for premier tenants. This means Starwood is a trusted counterparty, having negotiated multiple leases with premier tenants, which we believe accelerates the time line for site evaluation and lease signing, something we have already seen.

Second, Starwood brings captive development and EPC capabilities. They lead design, development, construction and facility operations, giving MARA an experienced execution partner without having to source and manage third-party contractors.

Additionally, their prior experience for constructing sites for premier tenants provides an enhanced certainty regarding their ability to develop on tenant time lines and technical requirements. This trust factor provides prospective tenants more confidence that their time lines and specifications will be met. Our peers who have never done this before still need to build trust with prospective tenants because they lack a proven track record.

Third, the structure is capital efficient. When MARA contributes to sites, its value is determined using pre-agreed site-specific economics tied to power, land, interconnection and development attributes. That value gives more equity credit in the project before joint venture cash contributions are required. To put this in context, on an illustrative 200-megawatt project, MARA could generate approximately $50 million to $100 million of net annualized stabilized cash flow based on a 9% to 15% yield on cost range with little to no incremental equity required beyond the value of the site we contribute.

As projects scale the structure naturally evolves, MARA set contribution is fixed. So for larger developments, the incremental gross capital becomes more proportionate between the partners. At that point, the funding model starts to look more like a traditional data center development structure, including the use of construction financing that can support roughly 80% loan to value. The key point is that this model allows MARA to monetize the value of its powered land portfolio, preserve significant upside in long-term cash flows and manage capital exposure in a disciplined way. Most critically, this is not designed to be a onetime transaction. As we continue to aggregate land and power assets, our goal is to contribute sites into the structure repeatedly. Starwood is a capital-efficient engine for turning MARA's powered land portfolio into contracted institutional-grade digital infrastructure at scale. We expect to sign multiple tenant leases by year-end, as the pipeline converts, we'll disclose contracted megawatts.

While the Starwood joint venture addresses the large-scale hyperscale end of the AI infrastructure market, Exaion addresses a different but equally important segment. sovereign, enterprise and private cloud AI compute. Together, they give MARA 2 distinct pathways into AI, both grounded in the same foundation of energy-backed infrastructure, both serving real and growing demand.

Governments and enterprises, particularly across Europe and Canada are increasingly unwilling to rely solely on hyperscale platforms for their AI infrastructure due to data sovereignty cost. They want control over compute, data autonomy, jurisdictional compliance, security and independence. This is not a niche requirement. Its AI policy evolves and data sovereignty standards tightening, a meaningful share of AI workloads will require infrastructure that is compliance-ready jurisdictionally controlled and trusted. Exaion is built to serve exactly that demand.

We continue to build on our proven success in UAE, Finland and our recent launch in Oman. We are in active discussions with major energy companies in France, Brazil and Saudi Arabia across energy-rich regions, where reliable, scalable power supports long-term digital infrastructure development. We are still early, and we will share a more detailed road map as this effort develops.

The simplest way I think about it is a Starwood and Exaion are different expressions of the same thesis. The JV pursues large-scale colocation for hyperscalers. Exaion pursues private cloud, sovereign AI and enterprise deployments in regulated markets where these are critical criteria. Both depend on MARA's core capability of controlling the monetizing energy-backed infrastructure. Together, they expand our addressable market across 2 large and growing segments of the AI infrastructure opportunity.

Finally, bitcoin mining is the operational foundation we're building from. Our strategy is to co-locate new infrastructure with our existing mining operations. This allows us to monetize power assets immediately while building on the operational discipline and infrastructure expertise at mining demand. Mining generates revenue to date. It preserves the option to redirect capacity toward AI and critical IT loads as those opportunities mature on the same sites. That flexibility is deliberate. It's not incidental to our strategy. It is essential to it and that it allows us to best monetize our power and compute.

We continue to believe bitcoin is supported by institutional demand. In our view, that creates a constructive setup over time with a bias to the upside of institutional buying continues and retail demand returns. We continue to believe bitcoin will appreciate beyond its current levels.

We also took deliberate steps to strengthen the balance sheet during the quarter. We retired about 30% of our outstanding convertible debt at a discount, reducing potential dilution and increasing our financial flexibility. This was a decision to reduce the capital structure's drag on equity value and give us greater capacity to pursue the highest return opportunities across the business with discipline and without being forced to dilute shareholders.

With that, I will turn the call over to Salman to walk through the financial results.

Thank you, Fred. Good afternoon, everyone. Before I walk through the quarterly results, I want to briefly frame the first quarter of 2026 from a strategic and financial perspective. This was a quarter in which we strengthened the balance sheet, reduced potential dilution from convertible notes by as much as approximately 46 million shares or 9% on a fully diluted basis and continue to align our capital allocation with the strategy. As Fred outlined, we are converting MARA's digital infrastructure with lower cost, large-scale energy capacity into AI and critical IT. Two initiatives are central to that strategy.

First, our recent announcement to acquire Long Ridge adds one of the most efficient energy-backed compute campuses with existing cash flows, own generation, existing interconnection, low-cost vertically integrated power generation complex and a significant development opportunity over time. This acquisition is next to our existing bitcoin mining site and is expected to expand our AI expansion in an AI-rich corridor. As we pursue the regulatory approvals and seek consent from Long Ridge debt holders, we believe Long Ridge will provide near-term diversified financial performance while unlocking significant long-term contracted digital infrastructure revenue.

Second, the Starwood joint venture gives us a capital-efficient path to monetize the value of our sites by converting them to AI, HPC and critical IT workloads. It is important to note that in our JV structure and Starwood -- our joint venture structure with Starwood, MARA contributes a site into the joint venture once the tenant is signed for which we receive credit based on the site's power, land, interconnection and development attributes at predetermined value, as Fred mentioned earlier. That is the power of the joint venture model. It allows us to convert the embedded value of our existing infrastructure into meaningful ownership in large-scale digital infrastructure opportunities while significantly limiting the incremental capital required from our balance sheet, giving us a higher return on capital than our peers.

Now let me turn to bitcoin price in Q1. This was a challenging quarter for the bitcoin price and reflected broader pressure across risk assets. The decline was driven by a combination of macro uncertainty, tighter risk appetite and continued pressure on mining economics. For MARA, that backdrop reinforces the importance of operating discipline. We remain focused on fleet efficiency, cost control and capital allocation rather than pursuing growth for growth's sake. Since quarter end, bitcoin has rebounded meaningfully, increasing approximately 20% from its March 31 closing price. While volatility remains inherent to this asset class, the recovery reinforces the value of maintaining bitcoin as both a reserve asset and a source of strategic financial flexibility.

With that context, I'll turn to our Q1 financial performance, capital allocation and balance sheet activity. Revenues in Q1 of 2026 were $174.6 million compared to $213.9 million in the prior year period. The decline was primarily driven by an 18% decrease in bitcoin's average price, which reduced revenue by $33.1 million and to a lower extent, impacted lower production, which accounted for approximately $2.5 million. In addition, other revenues declined approximately $3.7 million, primarily reflecting lower revenue from other digital asset hosting services compared to the prior year period.

During the quarter, we delivered record energized hash rate of 72.2 exahash per second, increasing 33% from 54.3 exahash per second in Q1 of 2025. This growth reflects continued fleet optimization and the deployment of approximately 2.4 exahash of new generation ASIC miners at favorable pricing during the quarter. Our share of available mining rewards reached 5.5%, up from 4.8% in Q4 of 2025.

We mined 2,247 bitcoin or 25 bitcoin per day in the first quarter of 2026. That is approximately 39 fewer BTC than prior year period, reflecting a higher network difficulty level, partially offset by our higher hash rate.

We reported a net loss of $1.3 billion or $3.31 loss per diluted share this quarter compared to net loss of $533.4 million or $1.55 loss per diluted share in the first quarter of 2025. Approximately $1 billion of this net loss for the first quarter of 2026 was driven by unrealized mark-to-market fair value adjustment for digital assets, a direct reflection of the drop in bitcoin price during the quarter. A reminder that based on our current bitcoin holdings, every $10,000 change in bitcoin price results in an approximate $350 million impact on fair value of digital assets, which is an unrealized mark-to-market adjustment to our income statement.

Adjusted EBITDA for the quarter was negative $1 billion compared to negative $483.6 million in the prior year period. Similar to net loss, this figure is denominated by -- or dominated by the bitcoin mark-to-market change. We use adjusted EBITDA as a supplemental measure of operational performance. A full reconciliation to net loss is included in our shareholder letter and earnings deck.

On the cost side, our cost per kilowatt hour was $0.04 for our owned sites in the first quarter of 2026. For context, we believe this remains among the most competitive in the sector at a larger scale. Our purchased energy cost per bitcoin for the quarter for our own mining sites was $40,047 in Q1 of 2026 from $35,728 in Q1 of 2025, primarily due to higher network difficulty driven by growth in global hashrate. This resulted in an 8% decline in bitcoin production at our owned mining sites compared to the prior year period.

Our daily cost per petahash per day improved 3% year-over-year to $27.6 from $28.5 in Q1 of 2025. And over the past 11 quarters has improved by 42%. We believe this remains among the lowest at scale in our sector.

In the first quarter of 2026, general and administrative expense excluding stock-based compensation, was $57.7 million compared to $36.9 million in the prior year period. The increase reflects the scaling of our operations, higher personnel costs associated with headcount growth from the prior year period and administrative fees in support of our expanded global footprint through acquisition and integration costs. Acquisition and integration costs burdened our G&A by $11 million for the first quarter of 2026.

As part of our strategic shift towards AI and critical IT we have realigned our business operations and reduced workforce by 15%, providing combined annualized savings of $12 million. This was a difficult but strategic decision.

In addition, we incurred a restructuring charge of $45.9 million due to elimination of certain business initiatives and realignment. The organization focused on scaling bitcoin mining is different from the one required to build a digital infrastructure company. This realignment positions the company to pursue AI opportunities as Fred discussed earlier. Following this restructuring, we expect our quarterly G&A run rate, excluding stock-based compensation and acquisition integration costs to trend below the Q1 level as these savings are realized over time.

Now let me address deleveraging our balance sheet and our recent bitcoin sales. During the quarter, we retired approximately 33% of our outstanding debt, which included 30% of convertible notes reduction at a discount. This reduced potential future dilution, lower leverage and improved our ability to allocate capital towards higher-return strategic opportunities. We've funded a portion of this debt reduction through bitcoin monetization, the client is only a reserve asset -- it's not only a reserve asset on our balance sheet. It is also a source of strategic financial flexibility. We will continue to deploy thoughtfully when doing so creates measurable value for shareholders and intend to use it selectively to strengthen the balance sheet and fund strategic priorities.

During the quarter, we sold approximately $1.5 billion of bitcoin. These funds were used to repurchase at a discount over $1 billion of face value of our 2030 and 2031 notes and reduced our line of credit by $200 million. In addition, we refinanced $150 million of our line of credit at a 7% interest rate versus 10.5% previously.

I also want to highlight that we have not used our at-the-market equity offering program, or ATM since the end of the third quarter of 2025. We have funded operations and balance sheet actions through bitcoin monetization, not equity dilution. We think this is an important data point for shareholders as we continue to allocate capital towards the highest return opportunities.

Now let me discuss our bitcoin holdings. We held a total of 35,303 bitcoin at the end of the quarter, a decrease of 12,228 from the previous year. Of the total, approximately 28% of the holdings were activated and loaned or pledged as collateral. The loan bitcoin generated approximately $6.4 million of interest income over the first quarter of 2026.

Finally, I want to provide additional clarity on the pro forma capital structure we expect to have in place at Long Ridge upon closing the acquisition. Long Ridge's $400 million term loan is expected to be repaid at closing. We are also currently conducting a consent solicitation to waive the change of control provision in Long Ridge's $600 million secured notes, which would allow the notes to remain in place. The $115 million Can-Am facility is similarly expected to remain in place. As a result, total pro forma debt at Long Ridge is expected to be approximately $900 million, down from $1.1 billion previously with approximately $185 million of tack-on secured notes expected to be issued. We expect to fund the remaining consideration through a combination of cash on hand, borrowings collateralized by bitcoin and potentially proceeds from the sale of bitcoin, depending on the market conditions at the time of closing.

We have also secured a $785 million of commitment letter backstopped by a bridge loan from Barclays in case needed. We have a plan in place to finance this acquisition, and we are very excited about the Long Ridge what it will bring to MARA and our stockholders.

With that, I will turn it back to our -- to Fred.

Thank you, Salman. The actions we've taken so far this year were purposeful and they were interconnected. The Starwood joint venture creates a capital-efficient path to convert our power portfolio into AI infrastructure ownership. Long Ridge adds a differentiated power advantage platform for premier AI and critical IT campus anchored by our existing Hannibal operations. Exaion gives us a second pathway into AI, sovereign and private cloud, domestically and internationally. Balance sheet actions reduced dilution risk and increased our financial flexibility, and bitcoin mining remains our foundation.

We recognize that the market is focused on demonstrated execution, signed contracts, contracted megawatts and tangible proof that this strategy converts into shareholder value. MARA is redefining itself as a digital infrastructure company, controlling and monetizing electrons to their best value across multiple compute markets. This transition is already underway. Q1 2026 was a meaningful step forward.

With that, I'll turn the call over to the operator to open it up for questions.

[Operator Instructions] And your first question comes from Paul Golding with Macquarie Capital.

Congrats on all the progress this quarter. I wanted to ask, as you think high level -- and maybe this is for Fred. As you think high level about the approach to expanding on the HPC strategy, on the one hand, you've got multiple tenant prospects across the portfolio of existing sites and through the Starwood JV. And on the other hand, you also did an opportunistic deal to acquire the Long Ridge asset, how should we think about your broader strategy between commercializing existing sites and adding capacity through these opportunistic deals? Was Long Ridge sort of a one-off because of the relationship and it coming to market? Or is this potentially a simultaneous approach that we should see unfold between assets coming to market that you would look to acquire versus the existing portfolio?

Yes. Great question. So the Long Ridge deal has been in the works for quite a long time since we acquired the Hannibal asset originally actually. The site, obviously, Long Ridge provides us with the land that we need to be able to build a true premier campus. And the original intention with the Hannibal site was to build a much bigger data center facility. And it just took a long while for the respective parties to reach agreement on the deal here, and they obviously had to take it to market through a process to ensure that they were doing the right thing for their shareholders. But that has been a deal that's been in the works for quite a long time, actually.

I think going forward, what you should see is you can think of us as doing focusing on the combination of small sites, which are perfect tuck-ins. We recently added a smaller site earlier at the end of last year, for example, which is now operational as a mining site, which has the opportunity to potentially convert into a smaller token factory facility if we wanted to do that with that site. At the same time, we're going to continue to look for larger lands and power opportunities where we can build significant campuses together with Starwood.

We really have the best of both worlds here because the large-scale opportunities, have you started as a partner just really a wonderful job of de-risking the whole process and ensuring that we're able to execute properly. At the same time, at the smaller scale sites where we know how to develop smaller sites, especially as you start looking in the world of inference, where a lot of this is moving to ASIC technologies away from NVIDIA's traditional GPUs, those facilities now are able to operate more modular data center formats, which are more akin to what we've been doing all along with bitcoin mining, where all our sites operate as kind of modular data centers. And so we believe building this duopoly, if you will, to be able to develop smaller sites that specifically service inference needs for a variety of potential tenants or customers as well as doing the larger sites with Starwood is a way that we'll be able to build a beautiful portfolio of assets that will provide long-term value to our shareholders.

And maybe just as a quick follow-up. I was wondering if I could pull on that inference versus training thread a bit. Are you able to just share any detail around the general mix of interest right now from these prospects? Is it indexing more towards the inferencing use cases? Or is it more towards training or equally distributed?

Sure. So when you generically use the word hyperscaler, you're typically talking about somebody who has large amounts of data that they have collected that they train a model on that they then use that model to do things. Amazon, Google, Microsoft use data they have to essentially do inference train a model and then do inference on that model. So you have a lot of those sites are a combination of training in inference. And if you've been following with Jensen and NVIDIA's been talking about his belief is that these training sites will, over time, do more and more inference. I think the models going forward, you're going to see a need for sites where people can deploy models that they have done in-house and run them. And these are these token factory type sites, which I think we're going to see a lot more of where essentially somebody needs the ability to run a handful of megawatts of model scale.

We're starting to see already financial players, meaning nondata center players wanting to now have data center capacity that they can use for -- it could be financial trading, it could be health care data, it could be other things where the ability to develop models and run your business using these models has become mission-critical and therefore, you don't want to put it up in the cloud, you want to do it in your own private cloud. And so this is where Exaion marries to this model very attractively. We're able to engage with kind of any tenants across whether they want your traditional large hyperscaler site, which is training in inference together typically or somebody who just wants proprietary air gap capacity for either training and/or infants typically together or just inference and just want essentially a token factory. They want to run a coin model. They want to run an open weight model, and they literally are just looking for kind of this mix of lowest cost per token with best quality of service. And so as an example, if you're a financial trading company, you may do model development at a data center where latency is not important because you're really trading our model. But once you deploy that model to actually run it, you're going to run it somewhere on or near prime where latency is next to 0. So that's a quality of service. So you're willing to pay more per token if the quality of service suits exactly your needs.

And if quality of service, meeting latency speed and connection time isn't important, then you can run it at a token factory that's more remote. So we believe the market is going to consist of a variety of those tiers, and we're already in talking with enterprise customers is doing part of our market research.

What we're finding is there are companies who, their public cloud bills, if you would, their invoices have gone from hundreds of thousands of dollars a month to millions of dollars a month because of the fact they're running models in the public cloud, and they're finding it's just financially not an option. You're also seeing, however, the large model providers needing more and more capacity to run their models. And as they develop more and more tools, and I'll use as an example, Anthropic has just released these new tools for financial analysts for investment banks. They're doing all of these vertically designed genetic frameworks. These are systems that consume huge amounts of tokens, but they still are running essentially on your data, but it's still claw that's running in the background. So there's a need to be able to run across a huge infrastructure of sites globally to be able to operate these things.

And so I think you're going to see a pretty -- you're going to see inference growing, but trading is still going to be growing for the foreseeable 6, 7 years, I think, but you're just going to see inference volumes increase dramatically as more and more agent technology comes to play. We're seeing thousandfold increases in agentic in token consumption when somebody moves from chat to using a Cowork or Claude Cowork, for example.

Just talk to any CIO and ask what their token bills are lately. And they'll share with you the token maxing is not something they want to really incentivize people to do.

Your next question comes from Chris Brendler with Rosenblatt Securities.

I wanted to ask on the G&A line, has seen a pretty significant increase over the last couple of quarters even if you back out stock-based comp and call out on acquisition expenses. And just sort of trying to reconcile that versus the headcount reductions. I know that you're probably more forward-looking, but can you talk about a little bit about some of the investments you've made in what areas? And I was struck by your comments about sort of repositioning the organization. As you outsource more and more stuff to Starwood, I would think the organization may not be as large in the future as more and more bitcoin mining folks are repositioned. Just can you talk about the path of expenses? Because it seems like it's a little elevated still in my mind. Salman, do you want to do that?

Sure, Chris. Thank you for the question. As you know, we've said this before, we've been growing over the years. And we've -- as part of our announcement in Q1, we looked at our organization and reorganized ourselves more focused on what is in the pipeline in the future. And as we discussed in today's call, we've got the Long Ridge acquisition, we've got Starwood, which we're very excited about. What you have to remember is that we are still very good at what is MARA good at? MARA is extremely good at securing low-cost power and scale, $0.04 per kilowatt hour at 2 gigawatt capacity that is not many people can claim that they have that on of power. So we have the operations to manage that from a bitcoin mining perspective today. And that capacity that we have and the additional capacity that we plan to acquire gets dropped into our joint in certain cases, for example, at Starwood because we maximize our return on that by not having to invest incremental dollars because we get credit for the assets that we drop into the partnership, our dilution compared to other miners is much lower.

So I'm just going back to the structure long-winded answer, MARA historically was a pure-play bitcoin miner. We were growing and we had certain technology initiatives. MARA going forward continues to remain a technology company that happens to be surrounded by energy in the middle and AI and critical where we expect to monetize and generate free cash flow from a long-term perspective. So we looked at the overall structure and whether what are the skills that we're missing that we need to add to get there and what are the skills that we don't need for the growth of our previous [indiscernible] bitcoin mining business, and that's what resulted in.

In terms of the cost structure, the cost structure for our company size, I would expect, as we have stated in our prepared remarks to be lower than what we incurred in Q1, but that to also realize that when we are having these transformative transactions and acquisitions, there are costs associated with those. And as we have disclosed in our adjusted EBITDA disclosures, we will continue to disclose this and isolate those costs so that you can see what are the recurring nature of costs and what are nonrecurring nature of those and that way it helps modeling the cost better.

And then, Chris, the other thing is, obviously, the transition takes time in the sense that if I sign a lease tomorrow, that site is still mining bitcoin for another 18 months, maybe while the sites being built. So it's not just we're going to let go of a whole bunch of operations folks just because we're transitioning the strategy. It takes time.

Yes. Okay. My follow-up question was on the funding plan, a lot of [indiscernible] have sort of started shying away from the ATM. You mentioned you haven't used ATM since September. As you think about your, you have a lot of cash in picking the balance sheet for Long Ridge, but are you striving to be more of an investment grade credit and use more traditional equity funding methods in 2026? Or is that more of a longer-term plan?

Yes. So a couple of things to think about, Chris. The transactions that we're talking about, and we have, you can look at the examples of what other miners have disclosed with HPC conversions. We expect to -- as we've stated previously, we expect to have a few announcements around the tenant in the second half of this year. And usually, these transactions are either with an investment-grade counterparty or backstopped as other people have announced. So from a financing perspective, yes, those financings are considered investment grade from a finance, project finance standpoint.

And when you talk about our profile and our cash flows our goal and intention is to enter into these -- to create this vehicle where we continue to acquire these low-cost power sites and drop them into the partnership with a little amount of capital needs depending on the size of the project. and continue to have those triple-net lease revenues flowing through our P&L. So that generates a you get to have more predefined future free cash flows generating for income, your cash flow statements, then obviously, our balance sheet position improves our overall, whether you want to call it investment grade or non-investment grade, then you can have a better conversation around what your balance sheet looks like.

As you know, historically, the sector has not been valuated much or paid attention to when it comes to credit rating agencies. But as with 2 gigawatt capacity power and so many opportunities to generate free cash flow from a long-term perspective, 15-year risk-free or low risk rate of return type projects is certainly that the attention from a rating perspective.

Okay. Thanks so much for the color Yes. Next question comes from Brett Knoblauch with Cantor Fitzgerald.

Fred, on the Long Ridge acquisition, you outlined like a path to get that to maybe, I guess, 600-megawatt AI campus. Could you maybe help put a time frame around that? Like where are they in terms of that extra 200-megawatt kind of grid connect that they're pursuing now? And then how long would it take maybe to expand the generation capacity what approvals would you need?

Sure. So the behind-the-meter expansion is already in process so on a shorter-time frame than the grid expansion, and the grid expansion application submission process is what it is. But we figure that, as you look at the development time frame, from a 200-megawatt facility will likely take 18 to 24 months before it comes online by that time, an additional 200 megawatts behind the meter should be available. And shortly thereafter, we expect the other 200 -- the remaining 200 megawatts to come online from the grid interconnection. So the key is getting the first site up and running for the tenant and then just having the power in the queue and ready to go. But the behind the meter, additional capacity is what will come on soonest of the 2 additional capacity increases.

Awesome. Helpful. And then maybe just as a follow-up, I think you guys kind of reiterated expect first lease was the Starwood get signed at some point this year. I guess what's giving you the confidence that this could be executed as quickly as you're expecting?

Competition amongst the prospective tenants to get into the site. We have -- as I think we said in our prepared remarks, multiple tenants looking across multiple sites that make up 90% of our capacity today. And so as this market -- the demand in this market isn't decreasing. People are getting ever more antsy about getting more capacity. I mean you can just see what some of the model players have been doing just to garner more capacity out there. And you are -- as a model provider, you are directly limited in your ability to grow capacity by the amount of compute you have because you can only have so many clients hitting your model. before your compute runs out of gas. And then the only thing you can do is yield management and raise your prices. And so if you look at what some of the model providers have been doing recently is they have essentially gone from, oh, that $200 a month all-in things, guess what we're going to have to put a capacity cap on that and you're going to have to pay higher fees because I don't think anybody fully expected the explosion in demand for tokens that has happened once agentic technologies started to be introduced open claw, open the floodgates for people to start really looking at how to do this.

And this is not just an enterprise play, and it's not just a consumer play. There are, across the full spectrum of users. People are starting to build agents. People are starting to use tools like cloud Cowork for example. You have Google is about to release Remy, I think, is the name of their agenetic helper if you're in the Google ecosystem, which is a huge part of the SMB market, with Gmail, Google Calendar, et cetera. That's an agent that will do essentially what Cowork does, but do it in the cloud, whereas Cowork is local machine-based. And so that's just going to drive more and more and more demand.

And so as you add customers, you need to do more inference. And at the same time, your model sizes are growing. Look at what Anthropic said about ethos. Is it 10 or 100x more compute it needs than the prior model. And so when you look at that increment in both model size and compute requirement for both training and operating plus the inference side of it, there's pretty much just a huge demand for capacity today. And so we have, as I said, multiple prospective tenants fighting over the opportunity to get into some of the sites we're super excited about that. And we're happy to be in the position we are with the amount of capacity that we have and with a partner like Starwood, but we're able to take advantage of that.

Your next question comes from Ben Sommers with BTIG.

So you mentioned conversations with both hyperscaler and enterprise customers. Curious if there's any preference there from your side? I guess also if there's any difference in the conversations than kind of how you think about the customer mix longer term as you build out the HPC business?

Yes. I mean from a per megawatt basis, the hyperscalers are going to dominate by a large extent, just because of the sheer capacity they need. A single enterprise, 25 megawatts would do provide a huge amount of capacity for an enterprise customer. So I think it will be in the near term 90-10, and over time, maybe 60-40, but that really is going to be dependent on how enterprises decide to do things. If they decide to do it on-prem private cloud and we have the Exaion solution to service that need, and we're able to go in and help them operate that. If they want private cloud in a near-prime or remote solution, we can do that as well. But we think that the hyperscalers are going to be the first sets of tenants that we scale with. And then over time, you'll see the enterprise kind of customer mix increase.

Got it. Super helpful. And then I know you mentioned scaling the Starwood partnership with new sites. I guess just kind of curious since announcing that partnership, if there's been any kind of change in how you're thinking about developing the power portfolio moving forward kind of if their long-tenured expertise in the power market just kind of help potentially scale what MARA currently has in the power portfolio.

I'd say that the -- what is the beauty and the partnership is that we are really good at building pipeline, of sites, acquiring land and power at power attractive prices and paying the right price, if you would, for land and power. They are really good at finding tenants getting sites designed that built and operational. And so it's a perfect complement. There's no real overlap in that regard. And that's what really makes this relationship work as well as it does. So we're just very focused on continuing to fill that funnel of prospective sites such that we continue to be viewed by the prospective tenants as a reliable source of capacity going forward.

just one other comment I'll make is that a key difference between our model and what most of our peers are doing is they are typically starting with 1 reasonably sized site, 250 megawatts, 300 megawatts, something like that. And then they have to sink all their attention to capital into getting the upside done, and then they go to the second site, and then they go to the third.

With the Starwood model, we can just go out and acquire and then they can do their part of the deal, and we can move at a much faster pace and scale much faster than we were trying to do this all on our own. And I think that's where, while we may be late to the party, I think we're going to catch up quickly. And I think we'll go to scale past what some of our peers are doing because of the value of this partnership.

Thank you. And that's all the time we have for questions today. So I'll hand the floor over to Robert Samuels for closing remarks.

Thanks, operator, and thank you, everyone, for joining us today. If you do have any questions that were not answered during today's call, please feel free to contact our Investor Relations team at ir.mara.com. Thanks very much. Enjoy the rest of the day.

Thank you. All parties may disconnect.

Greetings, and welcome to the MARA conference call. [Operator Instructions] Please note that this conference is being recorded.

I will now turn the conference over to Robert Samuels, VP of Investor Relations. Thank you. You may begin.

Thank you, operator. Good morning, and thank you for joining us on such short notice. With me on today's call are our Chairman and Chief Executive Officer, Fred Thiel; our Chief Financial Officer, Salman Khan; and Chief Growth and Strategy Officer, Duncan Dickerson.

Today's call includes forward-looking statements, including those about our growth plans, liquidity and financial performance. These involve risks and uncertainties, and actual results may differ materially. We disclaim any obligation to update these statements, except as required by law. For more details, please see the Risk Factors section of our latest 10-K and other SEC filings.

We also reference non-GAAP financial measures like adjusted EBITDA, which we believe are important indicators of Long Ridge Energy's operating performance. Please see our press release and investor presentation for reconciliations to the most comparable GAAP measures.

We'll begin with prepared remarks from Fred and Salman. After their comments, we will open the call to Q&A. I'm going to turn the call over to Fred to kick things off. Fred?

Thank you, Rob, and good morning, everybody. Thank you for joining us on such very short notice. We're excited to announce our definitive agreement to acquire Long Ridge Energy & Power from FTAI, a transaction that we believe is both highly strategic and financially compelling for MARA. Salman and I will walk you through the presentation, and Duncan Dickerson, our Chief Growth and Strategy Officer, will join us for Q&A.

Let me begin with how this transaction fits into our broader strategy. We've been very deliberate in evolving to maximizing long-term value by expanding into Energy and AI/HPC. This has included reallocating capital towards these higher return opportunities, including entering our partnership with Starwood Digital Ventures as well as reducing costs in areas that do not support the strategic direction.

At the core of that strategy is a simple idea. Power is the most important input, and we want to deploy it across the highest value applications. That includes AI and high-performance computing, critical IT infrastructure and flexible compute, including Bitcoin mining.

Power is at the core of that model, owned, operated behind the meter and grid connected, and our objective is to dynamically allocate that power to maximize returns over time. Against that backdrop, the agreement to acquire Long Ridge Energy is a significant step forward in executing our optimized digital infrastructure strategy and increasing our market position in AI/HPC infrastructure.

The assets we will acquire will establish a leading AI/HPC data center campus in the PJM Interconnection, one of the most active data center and power markets in North America. In a market where power and infrastructure constraints take years to solve, we believe Long Ridge Energy will provide exactly what is needed to immediately deliver value to our shareholders upon closing.

This is not raw land with a long development time line, but rather a site that is already operational and that gives us immediate access to the infrastructure, interconnection and physical footprint required to support expansion. The power is there, the land is there, the water is there, the fuel supply is there and the interconnection is there.

In total, the site supports more than 1 gigawatt of total potential power capacity across generation and load with up to 600 gross megawatts for AI/HPC and critical IT loads. Assets of this scale and quality are increasingly difficult to replicate given the time, cost and complexity required to secure power, land, permitting and interconnection in today's market.

We are acquiring assets that would take up to 10 years to assemble on our own. These assets are also strategically located and that they are anchored by our existing footprint at our Hannibal site, where there has already been interest from investment-grade prospective AI/HPC tenants. Taken together, this transaction will increase our owned and operated capacity by 65% while adding $144 million of contracted EBITDA based on Long Ridge Energy's annualized second half 2025 results.

It is also worth noting that MARA does not expect to reduce Long Ridge Energy's current supply of power generation into the PJM grid and does not anticipate any impact to consumers. As we develop additional compute capacity behind the meter, we expect to pair that demand with incremental generation over time.

Turning to Slide 7. You will find specifics on the assets and the team that will become part of MARA at close. Long Ridge Energy campus includes a 505-megawatt nameplate combined cycle gas plant. It is one of the most efficient combined cycle turbines in the PJM interconnection and provides stable cash-generative operations. It generated $144 million of annualized adjusted EBITDA with 76% contracted capacity based on second half 2025 performance.

In addition to the combined cycle gas plant, the campus has over 1,600 contiguous acres and 200 megawatts of existing MARA capacity already operating at Hannibal that I mentioned. Following the closing of the transaction, MARA plans to retain Long Ridge Energy's team of 25 full-time engineers and employees. These team members have operational knowledge of the facility and will supplement the MARA team's existing expertise. When you compare this site to other digital infrastructure campuses, what stands out is the completeness of the platform illustrated on Slide 8.

You have colocation generation, vertically integrated fuel supply, firm grid access and proximity to core AI demand at the Northeast and Mid-Atlantic. That combination is rare and increasingly hard to find in today's market. To put that into perspective, if you look at Slide 9, if you were trying to build this from scratch today, you're looking at $2 billion to $3 billion of capital and 7 to 10 years of development time. So we are effectively stepping into a platform at closing that is already built, already operating and already generating cash flow.

From a capacity standpoint, this is a very scalable platform. We start with 505 megawatts of generation, plus 200 megawatts of existing MARA capacity and then have multiple paths to expand from there via grid expansions and on-site power generation. That gives us over 1 gigawatt of total potential capacity and the ability to scale to 600 gross megawatts of AI and critical IT load over time.

What makes this especially compelling is the flexibility. This is not a single-use asset. It is a multi-revenue platform where we can monetize power through long-term AI/HPC leases, flexible compute operations and wholesale power generation. That flexibility is critical because it allows us to allocate power dynamically based on where we see the best returns.

On the demand side, our Hannibal site has received inbound interest from multiple potential investment-grade AI/HPC tenants. In terms of the path to getting additional capacity in service, this is where our partnership with Starwood Digital Ventures plays an important role. We expect this campus to be developed as part of our broader platform alongside Starwood, leveraging their development capabilities and tenant relationships to build capacity to market faster. We expect an initial 200-megawatt AI/HPC build-out with construction beginning in and around the first half of 2027 with initial capacity coming online in mid-2028.

Looking at our owned and operated power generation capacity, this transaction is a step change. It will increase our owned and operated capacity by 65%, taking us from 1.3 gigawatts today to 2.2 gigawatts with the addition of the Long Ridge Energy Power Plant and our planned expansion. This will be a meaningful increase in both scale and capability. We are excited about what this enables to MARA and the value it will deliver to our shareholders. We look forward to updating you as we move forward through the closing process and begin executing on the development vision at Long Ridge Energy.

I'd like to turn it over to Salman now to go over the financials. Salman?

Thank you, Fred. Before I get into the transaction economics, I want to reinforce a few of the strategic points Fred made. First, Long Ridge is a current operating asset generating free cash flow, not a development bet. The Long Ridge Energy Power Plant currently generates approximately $144 million of annualized adjusted EBITDA with about 76% contracted, which gives us stable visible cash flow from day 1.

Second, this acquisition is a bolt-on to our existing Bitcoin operations. Our Hannibal site is already operating at this campus. This is an extension of what we are already doing, expanding MARA's footprint on infrastructure we know well. Third, the strategic upside here is significant. The acquired asset positions us to develop critical IT and HPC infrastructure in one of the world's largest AI corridors. The combination of location, existing infrastructure and scalability is what makes this especially compelling.

And fourth, the vertically integrated natural gas and power plant structure is a competitive advantage. Operating at less than $15 per megawatt hour, all in is a low-cost position that few can match. We believe that cost structure will directly translate into MARA's ability to operate an HPC complex profitably. It is the economic foundation that makes this platform so compelling.

With that context in mind, let me walk you through the financing discussion. On Slide 13, I will review some transaction details and financing. Under the terms of the agreement, MARA will acquire Long Ridge Energy for approximately $1.5 billion, including the assumption of at least $700 million of debt backstopped by a bridge loan from Barclays.

We plan to fund the equity portion of the transaction through a combination of existing cash and Bitcoin-backed financing capacity. As you can see in the slide, we have over $500 million of restricted and unrestricted cash on hand, approximately $2.4 billion of [ BTC holdings ] based on a Bitcoin price of $70,000 per Bitcoin and more than $2 billion of near-term liquidity.

This gives us line of sight to completion -- to complete the transaction in a disciplined way with non-dilutive financing sources that we have available to us. At a high level, this transaction is expected to be immediately additive to EBITDA upon closing and will provide stable free cash flow that we expect will help support the build-out of our AI/HPC platform without relying entirely on external capital.

With that, I'll turn it back to Fred. Fred?

Thank you. Yes, thank you, Salman. So with that, we believe this transaction will give MARA a scaled power advantage platform, immediate durable cash flows and a clear path to build one of the leading digital infrastructure campuses in the market.

It will strengthen our ability to maximize the value of every megawatt we control, which remains our guiding direction strategically. We're excited about what this enables for MARA, our shareholders and all our stakeholders.

And with that, operator, we're ready to open the line for questions.

[Operator Instructions] Our first questions come from the line of Brett Knoblauch with Cantor Fitzgerald.

Congrats on this transaction. It's awesome to see. Curious with the initial 505 CCGT. How much of that is contracted to third-party or a grid? Or can you use, I guess, for captive use for your own data center build? And is there any regulatory process you have to go through in order to use that for a data center? Or can you just maybe just walk us through how that would work?

Yes. Why don't we have Duncan do that for you. He has the detail of his fingertips.

Yes. And thanks. Great point. I think the current plant just is at -- operating at a 485-megawatt interconnect. So they're selling that power into the grid, and it's about 76% hedged under a financial swap. And that power is incredibly valuable to Salman's earlier points about operating cash flow.

And we don't currently intend on taking that behind the meter for projects. And we also believe that the line of sight for grid expansions and other on-site power give us plenty of growth paths for AI and critical IT.

So the 505, we're going to keep as effectively just kind of selling energy and then kind of just use this one combined campus to co-locate your Hannibal data center with Long Ridge Energy. Is that the plan?

Yes.

Awesome. And then the other capacity, I think you noted Long Energy has in PJM. How does -- are you acquiring those assets as well? Or do they just have rights...

We already have 200-megawatt load interconnect at the site, which makes it such a strategic bolt-on. And this acquisition gives us also the land to deliver the campus. And so really unlocks a lot of the potential across our existing portfolio.

And then in process, we already have permits and line of sight to another 200 megawatts of grid capacity there. And then the Long Ridge team also has permits in process and all the necessary supply for another 200 megawatts of on-site generation. So all of that's expected to be delivered at the latest by 2030 and -- but it's anchored with our existing already operational 200 megawatts of capacity.

Our next questions come from the line of Jason Mandel with RBC Capital Markets.

Congratulations. I was wondering if you could just clarify, I think on Slide #13, just to be clear, I think as I understand it, the debt that's being repaid as part of this is the existing term loan that exists in Long Ridge and then the bonds remain outstanding. And then the financing that's potentially being done regarding the Bitcoin that would essentially sit outside of Long Ridge and be part of MARA secured financing on Bitcoin. Is that all accurate?

The way to view this is that the existing debt stays with the -- which is asset-backed debt, the senior secured notes, the term loans. And then we have a backstop as we get to the closing. Obviously, we -- this is an announcement today for the signing of the transaction as we go towards the closing, there are a couple of steps before we get there.

So as a backstop, we have $785 million secured through a backstop with Barclays. And then the rest of the equity check that comes from a non-dilutive source, as I mentioned earlier, we have option to do financing against Bitcoin. Our thesis around Bitcoin remains bullish from a long-term perspective. And we -- from a capital stack standpoint, we are in lots of gratitude that we are in a position where we can use nondilutive sources as an option to finance against Bitcoin as an example.

Okay. And maybe just one quick clarification point. I take a quick spin through the purchase agreement. And it does, if I'm not mistaken, appear that there's a requirement to deliver a payoff letter of the term loan, but it sounds like you're saying the term loan remains outstanding. Just to help clarify.

Yes. The term loan will be repaid at close, and we have the incremental bridge from Barclays to both handle that, but then also to reflect any available for incremental capacity beyond the existing bonds and CanAm loan that are on the facility.

Our next questions are coming from the line of Michael Donovan with Compass Point.

Duncan and Rob, congrats on this deal. Just wanted to get clarification on the build-out for AI workloads. So would Hannibal be converted over to HPC? Or would you first work on the initial 200-megawatt grid expansion?

The first target is going to be Hannibal, of course, because that's ready to go from an interconnect perspective and load perspective. So that's the first target. And as you know, these campuses are done in phases typically. If you're looking at, say, 200 plus an additional 200, that first 200 phase may take 24 months potentially to build.

And as that's being done, the second 200 megawatts will become available and then you roll into the second phase of the campus to build the additional buildings for that part. So what we really love about this is, a, we have capacity we can directly apply today for use for AI/HPC, which is why we already have tenants looking at it. And we really just had to wait until we could announce the deal to say we actually had the additional land capacity under control. And now that we've done that, these conversations will advance fairly quickly, we're certain.

But this unique combination of grid attached power with an opportunity down the road to have behind the meter potentially and all this land in this one location just make it such an attractive opportunity. And we think that obviously, the fact that it's already cash flowing is very additive for MARA's shareholders as well. So we think this is a win all around, and we're super excited about it.

Appreciate that, Fred. Now can you talk more about the inbounds you're receiving in terms of the types of AI workloads? I know in the past, you mentioned sovereign and enterprise HPC. Can you get a little bit more color there?

Duncan, do you want to handle that?

Yes. And I think that the thing that I would start with on that, just to qualify, we have no preconceived notions about the workloads that a lot of the customers expect to run the site. And I think that we've had interest from AI workloads and critical IT workloads equally, especially given where the site is located and really a premier market.

It really opens the alternatives for both cloud and -- or traditional workloads and AI and inference at this site. So we've had really a diverse inbound of potential customers and tenants that I think has been one of the things that's given us such conviction to really move on something like this. But of course, you can't lease what you don't own. So that's one of the main reasons to do this transaction, highly strategic in that way.

And then -- but it's been -- it's been a fantastic time working with the site through the -- with our partnership with Starwood, getting it perfected for leasing. And just to be directly clear to your question, we've received inbound interest from both the investment-grade hyperscale tenants that you would expect to see from some of the larger deals and then also the kind of the neo-clubs with shadow investment grade or other alternatives too. So it's been a comprehensive view of demand.

That's helpful, Duncan. One final question, if I may. On Long Ridge Energy's adjusted EBITDA estimates based on 2H '25 results, is there a degree of seasonality here?

Yes, there is always, especially in this area. But the key thing to keep in mind is the vertically integrated and owned gas supply. And then also the fact that 76% of the capacity is hedged eliminate that a little bit. I would -- I think that, that gives us so far tracking north and strongly versus the last year's numbers, but the seasonality there is largely moderated by the vertically integrated benefits of the plants and the fact that it's hedged.

We have reached the end of our question-and-answer session. I would now like to turn the floor back over to Robert Samuels for any closing comments.

Thanks, operator, and thank you, everyone, for joining us this morning. We look forward to speaking with everyone again soon.

Ladies and gentlemen, thank you for your participation. This does conclude today's teleconference. Please disconnect your lines at this time, and enjoy the rest of your day.

Greetings, and welcome to MARA 4Q '25 Earnings Call. [Operator Instructions] As a reminder, this conference is being recorded.

It is now my pleasure to introduce your host, Robert Samuels, VP Investor Relations. Thank you, Robert. You may begin.

Thank you, operator. Good afternoon, everyone, and welcome to MARA's Fourth Quarter and Fiscal Year 2025 Earnings Call. Thank you joining us today. With me on today's call are our Chairman and Chief Executive Officer, Fred Thiel; and our Chief Financial Officer, Salman Khan.

Today's call includes forward-looking statements, including those about our growth plans, liquidity and financial performance. These involve risks and uncertainties, and actual results may differ materially. We disclaim any obligation to update these statements, except as required by law. For more details, see the Risk Factors section of our latest 10-K and other SEC filings.

We also reference non-GAAP financial measures like adjusted EBITDA and return on capital employed, which we believe are important indicators of MARA's operating performance. because they exclude certain items that we do not believe directly reflect our core operations. Please see our earnings release for reconciliations to the most comparable GAAP measures. We hope you've had the chance to read our shareholder letter and look forward to your back.

We'll begin with some brief prepared remarks from Fred and Salman. After their comments, we will open the call to Q&A.

I'm going to turn the call over to Fred to kick things off. Fred?

Good afternoon, everyone, and thank you for joining us. Before we get into the results for the quarter, we're excited to discuss our just announced strategic partnership with Starwood Digital Ventures, the data center development platform of Starwood Capital Group and one of the premier data center developers and operators in the world. This joint venture accelerates MARA's expansion into AI and high-performance compute and represents a meaningful step forward in the evolution of our platform from a pure-play Bitcoin miner into an energy and digital infrastructure company.

Alongside other actions we have taken, including closing our investment in Exaion, we are strategically positioning our platform to support a broad range of AI deployment requirements from large-scale cloud environments, to private enterprise and sovereign deployments, where AI inference operates closer to its contextual data with reduced latency constraints and enhanced operational control.

Through our partnership, MARA and Starwood will jointly develop, finance and operate next-generation digital infrastructure capable of meeting growing demand from enterprise hyperscale and AI customers across MARA's existing power rich portfolio. MARA will contribute dedicated energy, advanced data center sites, while Starwood Digital Ventures will lead, design, development, tenant sourcing, construction and facility operation with Starwood providing investment expertise to support enhanced project level economics.

We have the option to retain up to 50% ownership in the joint venture positioning us to participate in future cash flows while capturing long-term value creation. The joint platform is expected to deliver more than 1 gigawatt of near-term IT capacity with a pathway to more than 2.5 gigawatts. This JV structure accomplishes several things at once. It accelerates speed to market and introduces institutional grade development and tenant relationships. Importantly, it also allows us to leverage the wealth of power capacity embedded in our existing energized sites in the near term. These assets were built around power and hyperscale cloud remains the fastest path to monetization that power at scale today. At the same time, the structure allows us to continue mining through a lease arrangement while accessing excess power at preferred prices during lower hyperscale utilization. That flexibility improves economics and smooths load across the site.

Now let me address directly why we chose to partner with Starwood. Enterprise, hyperscale and AI customers are inherently risk averse when selecting infrastructure partners. They require certainty of execution deep development expertise, balance sheet credibility and a proven track record of delivering mission-critical facilities on time and on specification. While MARA brings the power, the sites and the operational expertise, Hyperscalers typically do not award large-scale AI workloads to first-time developers without institutional backing.

Partnering with Starwood ensures that we're not asking customers to take that risk. Starwood has decades of experience as a real estate asset investor and developer established long-term relationships with hyperscalers and enterprise customers. and a proven ability to finance and deliver complex data center projects globally by aligning with an experienced tenant first developer we expect to increase execution certainty and accelerate our ability to secure institutional grade tendency. This is about optimizing probability of success and compressing time lines, not simply proving we can build stuff alone. Demand signals are already strong. MARA and Starwood have been engaged in at discussions with hyperscalers and leading HPC tenants reflecting meaningful early interest in Power Advantaged AI read capacity across our sites.

In parallel, design, permitting and commercial leasing processes are well underway, with applications submitted in electric markets to support accelerated delivery time lines. In other words, while we are formally announcing this partnership today, we are already well down the path towards securing a [indiscernible].

We also announced that we closed our investment in Exaion, acquiring a 64% stake and expanding our enterprise-grade AI and HPC capabilities. Through Exaion, we can deliver Infrastructure-as-a-Service and edge inference solutions for large energy and industrial customers, particularly in environments where requirements around data locality, latency and operational control how compute is deployed.

Importantly, Exaion fits in a broader international strategy Building on our proven success in the UAE and the recent launch of our pilot site in Oman, we are accelerating conversations with energy majors in France regarding global opportunities, including in Brazil as well as domestic energy producers in Saudi Arabia. These initiatives are all part of a deliberate strategy to expand our global footprint across energy-rich regions, where access to reliable, scalable power supports long-term infrastructure development.

Starwood [indiscernible] complementary elements of the same strategy, where Starwood partnership accelerates our ability to serve hyperscale cloud customers. Exaion strengthens our ability to deploy private enterprise and sovereign cloud environments. This is especially important in international markets where Exaion already operates data center infrastructure and provides a foundation for sovereign grade AI and high-performance compute deployments. Together, Starwood and Exaion give MARA multiple proven pathways to deploy the same assets, power sites and infrastructure in ways that maximize long-term value as demand evolves.

Now I'd like to take a step back and put this strategy in context. Jensen Huang said something on NVIDIA's earnings call last night that captures exactly what we are building towards. He said simply, compute equals revenues. His point was that in this new AI economy, the ability to generate tokens to run inference is the direct input to revenue growth for every enterprise and hyperscale customers in the world. That compute requires power. Power is the scarce input, and that is precisely what MARA controls. Our sites were originally developed to mine Bitcoin efficiently, but they were built around power.

As we continue this transition and as demand for AI and HPC at our sites accelerates, the economics of our sites will increasingly reflect long-term infrastructure characteristics. When the site supports contracted AI or HPC workloads, the underlying drivers of value shift, cash flows become longer duration and more predictable, execution risk is reduced and the operating profile increasingly resembles infrastructure rather than pure compute. We believe the same underlying assets can support different economic outcomes depending on how they are deployed. That is why optionality matters.

Bitcoin mining allows us to monetize power immediately and flexibly while AI and HPC workloads and when demand supports them, monetize that same power through longer-term contracts and higher-value use cases. Our responsibility is to allocate capital where the return profile justifies conversion and to manage our sites in a way that maximizes long-term value across market cycles.

This quarter, we've also advanced our strategy in other important ways. We increased our Nebraska footprint through the recent acquisition of a 42-megawatt data center adjacent to an existing site, expanding the campus by approximately 40%. With below-market power rates, this lowers our average cost to mine while strengthening operational efficiency. That same site also provides option value for AI and HPC workloads over time.

Lastly, we doubled our NGON gas to power operations from 25 megawatts to 50 megawatts during previously flared gas into some of our lowest cost mining power. Given the recent decline in Bitcoin price and considering the potentially accretive impact of the Starwood JV, we are adopting a capital allocation priority to focus on the highest-value near-term opportunities. While we are continuing to advance discussions with MPLX regarding development of integrated power and data campuses in West Texas. This is a longer-term project with significant capital expenses. The scope under consideration has evolved from the initial letter of intent, and we remain engaged in evaluating a transaction structure that aligns with our capital allocation priorities.

All of this is designed to expand margins and be accretive to NOI over time. Now Bitcoin remains a core pillar of our strategy. Despite a proven sell-off and continued volatility and we increased energized cash rate from 53.2 ex a hash to 66.4 has during 2025. We deliberately chose not to pursue projects that failed to meet our return thresholds. Capital discipline remains central. Historically, we retained the majority of the Bitcoin we mined as the long-term strategic assets.

Beginning in the second half of 2025, we began selectively monetizing Bitcoin to support operations. Given recent weakness and volatility in Bitcoin price, we have -- that have impacted both sector sentiment and elements of our trading performance, believe maintaining flexible -- sorry, financial flexibility is particularly important.

Looking ahead, we expect to continue taking an opportunistic approach using Bitcoin to enhance financial flexibility where appropriate. As always, these decisions will be guided by market conditions, and our capital allocation priorities with a clear focus on strengthening the balance sheet and enhancing long-term shareholder value. While the timing of a recovery in Bitcoin prices is difficult to predict, our long-term conviction in the asset class remains unchanged.

Let me close with this. MARA is no longer simply a Bitcoin miner. We are already well down the path of building an energy-dominant digital infrastructure platform. Starwood accelerates hyperscale development, Exaion strengthens our enterprise AI layer. Digital infrastructure and Bitcoin mining provide the economic engine and power ownership provides a strategic advantage. Every decision we make is guided by one principle to maximize the long-term value of every megawatt we control. We believe this strategy positions MARA to deliver durable, compounding shareholder returns.

I'll now turn it over to Salman to discuss Q4 financials.

Thank you, Fred. I'd like to begin by highlighting the strategic and financial significance of our partnership with Starwood Digital Ventures, a global leader in data center development and operations, as Fred mentioned earlier. From a financial perspective, we expect this joint venture to generate meaningful net operating income or NOI and free cash flow over time by reducing earnings volatility relative to a pure Bitcoin mining model. Importantly, partnering with Starwood enhances our access to institutional investment-grade capital as we jointly develop and finance utility scale AI and HPC infrastructure across our power advantage portfolio.

With a pathway to more than 2.5 gigawatts of potential capacity that could be allocated to AI HBC over time, we believe this partnership will materially improve MARA's long-term NOI profile, cash flow visibility and overall valuation framework for our business. We are also pleased to have completed our acquisition of a majority stake in Exaion which we expect will further diversify revenue as it expands its sovereign cloud and enterprise AI compute offerings. Together, these initiatives reflect this strategy focused on expanding free cash flow generation and driving long-term shareholder returns.

During the quarter, Bitcoin price volatility was the defining market force. Bitcoin began the period at roughly $111,000 and reached a new all-time high near $125,000 in early October. However, an overnight liquidation event compounded by broader negative market sentiment drove a sharp reversal with prices falling to roughly $87,000 by quarter end. This nearly $40,000 swing created one of the most challenging macro environments we have faced in recent periods and served as a significant headwind to our financial performance. Against this backdrop of falling Bitcoin price, global hash rate increased modestly as miners remain disciplined and cautious in deploying additional capacity amid the volatility.

Now let me give an overview of our key financial results and operational highlights which are still quite sensitive to fluctuations in the price of Bitcoin as well as the total network hash rate, which affects the total amount of Bitcoin you mine. For example, every $10,000 change in the price of Bitcoin results in approximately a $538 million change in the value of our Bitcoin holdings.

Revenues in the fourth quarter were $202.3 million compared to $214.4 million in the fourth quarter of 2024. For 2025, revenues grew 38% to $907.1 million from $656.4 million in 2024. Although Bitcoin's average price increased year-over-year, contributing $24.8 million to our 2025 results. Production volumes were lower throughout the year. We mined an average of 21.9 Bitcoin a day in Q4 compared to 27.1 Bitcoin point in Q4 2024, resulting in approximately 481 less Bitcoin mine this quarter.

Q4 marked exceptional operational performance across our core owned mining sites with several operating at or near 100% uptime. The decline in production, however, was primarily driven by higher network difficulty level due to rising total network hash rate. While we had the opportunity to deploy additional exahash more aggressively, we chose to remain disciplined and measured in our expansion given broader market uncertainty.

Despite the increasingly competitive operating environment, we continue to grow both our compute capacity in Bitcoin Holdings. Between Q4 2024 and Q4 of 2025, our Bitcoin holdings increased by over 20%, growing from approximately 44,000 Bitcoin to nearly 54,000 Bitcoin. Over the same period, our energized cash rate increased 25% from 53.2 exahash to 66.4 exahash. We reported a net loss of $1.7 billion or $4.52 negative per diluted share last quarter compared to net income of $528.3 million or $1.24 per diluted share in the fourth quarter of 2024. It's important to note of this net loss for the fourth quarter 2025. But due to the decline in the price of Bitcoin, we booked a [ $1.5 billion loss or $1.5 billion loss ], which was due to a change in fair value of digital assets, including Bitcoin receivable.

For the full year, we recorded a net loss of $1.3 billion compared to net income of $541 million in the prior year period. During the quarter, we also recorded a non-cash goodwill impairment charge of $82.8 million following our annual impairment review. This change is entirely non-cash and had no impact on liquidity, operating performance or cash flows.

On the cost side, our cost per kilowatt hour for our own sites were $0.04 in 2025. Our purchased energy cost per Bitcoin for the quarter was $48,611 compared to $31,608 in Q4 of 2024. Importantly, our daily cost per petahash per day improved 4% year-over-year to $30.5 from $31.7 in the fourth quarter last year and over the past 11 quarters has improved by 36%. We believe this remains among the lowest at scale in the sector.

MARA is the one of the largest corporate public holder of Bitcoin, and we actively generate returns on our holdings. The Bitcoin on our balance sheet strengthens our debt profile, reinforces resilience and provides flexibility to pursue disciplined growth opportunities when they arise. I would like to remind everyone that we are not a digital asset treasury company. MARA is an operating company, not a passive Bitcoin balance sheet vehicle.

During the quarter, we mined 2,011 Bitcoin and purchased an additional 1,670 as part of our trading strategy. As part of our digital asset management strategy, we aim to deploy Bitcoin holders through risk-optimized trading initiatives, lending arrangements and collateralized borrowings under credit facilities. As of December 31, 2025, we held a total of 53,822 Bitcoin, an increase of 8,929 over the previous year. Of the total 15,315 Bitcoin were [indiscernible] actively managed or pledged as collateral.

9,377 Bitcoin coins were known to counterparties, generating approximately $32.1 million of interest income during the year. We also pledged 5,938 Bitcoin to access financing, supporting liquidity while minimizing dilution. In total, approximately 28% of our total holdings were activated through our digital asset management strategy as of the year-end.

Now let's turn to our balance sheet. I want to address our debt maturity profile. $925 million notes due 2031 and $1 billion notes due 2030 have a [ clip right ] exercisable on June 4, 2027 and December 1, 2027, respectively. These represent a meaningful cash obligation that could come due in 2027, and we are proactively planning for that scenario today. I want to be clear about how we think about managing these obligations.

First, our Bitcoin holdings at current market price represent approximately 2x of these puts. Second, the [ 0 coupon ] structure on the vast majority of our notes means we have no material ongoing cash interest burden related to these notes eroding our liquidity between now and those [indiscernible]. Third, we have a history of prudent balance sheet management with our previous converts. The path we are building through the Starwood JV and our infrastructure transition is specifically designed to generate contracted cash flows that diversify our liquidity resources -- excuse me, liquidity sources beyond Bitcoin and load. We are not managing this balance sheet reactively. We are managing it with full visibility into every obligation on the horizon.

Now historically, we held the Bitcoin we produced as a long-term investment. In the second half of 2025, we began selling Bitcoin to fund operations. In 2026, we expect to continue to monetize Bitcoins opportunistically to enhance our financial flexibility, including to provide liquidity or to fund capital projects and other initiatives that we believe enhance long-term shareholder value. subject to market conditions and our capital allocation priorities. In response to the more volatile pricing environment, we elected to suspend news of our ATM at the end of third quarter of last year and instead funded operations through the sale of a portion of our mine Bitcoin.

Notably, Q4 marked the first quarter since 2022 that we did not utilize our ATM program. By shifting to operational funding through Bitcoin sales from production, we strengthened near-term cash flow while maintaining a disciplined and flexible balance sheet strategy.

With that, I will turn it over to the operator to open it up for questions. Operator?

[Operator Instructions] First question comes from the line of Paul Golding with Macquarie Capital.

Congrats on the partnership announcement. I wanted to ask Fred, you noted that the partnership gives you the opportunity to retain a 50% stake in these projects. Could you give some more color around the financing dynamics around the 50%, whether there's an opportunity to contribute the powered site in exchange for other forms of consideration outside of the JV? Or what you meant by that comment in a bit more detail.

And then as a follow-up, you also made a comment around -- it seems like load balancing across mining and HPC at certain sites that are part of this partnership going forward. Could you speak to the technical requirements of that and how we can think about mining versus HPC as you progress with this partnership?

Sure. So thank you for your question. When you look at the JV structure, essentially our initial contribution to the JV to each specific -- for each specific site would be the asset itself. And then additionally, we would capitalize a share of the -- we would provide capital for our share of the development costs. where we could retain up to 50% of the JV. And hopefully, that clarifies that part of it. There are mechanisms within the agreement that allow us to essentially decide not to fund our portion and [ there are ] methods within the agreement that allow us to essentially be liquidated, if you would, at an attractive prices or option or the -- if we don't fulfill our obligations and we decide to drop out, if you would. And I think we can provide more detail on that later.

The key thing regarding load balancing is a combination of technologies that we've developed by leveraging special battery technology. We've announced previously a partnership with TAE batteries, which is a very advanced battery technology that can switch at subsecond rates such that we are able to essentially balance load within data centers. If you think about the data center development project, our ability to be able to retain Bitcoin mining at the site while the project is being developed and then even retain a portion of the power at the option of the tenant allows us to act as a load balancer.

Depending on the type of compute load that's executed at the site, there may be for example in the case of inference loads, a variation in that load over the course of a 24-hour period or even over a period of a week, where the inference demands on that site may decrease at night, for example, or over weekends. And having Bitcoin mining at the site allows us to, again, based on the arrangement with the tenant mine whenever power is available that isn't being used at preferential prices. Hopefully, that answers your question.

It does. And maybe just as a quick follow-up, at those sites would the partnership then benefit from any revenue generated from the Bitcoin mining that happens when the load balancing is occurring? Or is that something that would be retained entirely by MARA?

Yes, it's primarily by MARA.

Our next question comes from the line of Reginald Smith with JPMorgan. .

Congrats on the announcement as well. I guess you guys are the last major Bitcoin miner to make the switch, so I guess, welcome to the party. Question, you mentioned Starwood and the fact that having a partner, you're not going at it alone. And obviously, using other Bitcoin miners spine deals. So my question is, should we expect the time to sign a deal to be shorter because just Starwood, along with you, does that alleviate some of the risk and maybe collapse the time line? And then I have one follow-up.

Yes. I think as I said in my opening remarks, this isn't a relationship that just starts with the signing of this agreement. It's been developing for quite some time. And we've been very actively working with prospective tenants. As I mentioned, the permits have been submitted already for some sites. And we are actively engaged with tenants. And the idea with working with Starwood was, again, many fold.

On the one hand, having a partner who has relationships with the tenants and that the tenants already trust, if you think about Starwood, they have built and operated sites for, I believe, 3 of the 4 Tier 1 hyperscalers. And having those relationships and the trust that exists because those relationships dramatically reduces let's just say, the courtship period that a newcomer to the market like ourselves would have to do. And so because of that trust it's easy for Starwood and prospective tenants to have a very accelerated time line on the process of evaluating a site, submitting permits, getting things going such that we can work towards getting leases done in a more accelerated fashion than if we had done it ourselves.

Additionally, Starwood's captive EPC capabilities again, dramatically facilitate the ability to build and execute these things in a very efficient and timely manner. The single biggest challenge today for hyperscalers is the ability to have certainty about power availability. And Bitcoin miners provide a lot of certainty because of the fact that we're currently consuming the power. One of the advantages with some of what we've done also is that we have been already operating inference on one of our sites in a containerized fashion. And if a tenant were to be interested in a modular approach versus a traditional box approach, if you would, of building a large building, especially for inference applications.

And just as a reminder, Jensen Huang in his comments that inferences where the revenue is in AI and inference is becoming the most important part of AI deployments today. And we believe that we're going to see a lot of advancement in the side of how these sites are designed, how these sites are constructed, which will align very well with the experience that we've previously developed in this area.

That makes sense. We've historically told clients and investors that take 9 to 12 months for a deal to sign, but it sounds like you guys are on an accelerated time line.

And second question, it sounds like you're putting the MPLX deal on the back burner for now. My question is, are you still in the market for sites that are already powered or have been approved? Is that something that would also be more of a longer-term investment or we view those types of opportunities. So a site in some other city in Texas like would that be something that you may pursue now? Or is that also something that would be viewed as less immediate and probably on the back burner?

No. So you have to look at what the market needs are today. If we're talking about power that will be available after 2030, there's not an aggressive demand for that capacity. The hyperscalers themselves have many efforts around developing sites, building their own energy generation for that time period, that is 4-plus years out from now. What they desperately need now is power that's available today where they can quickly get a site permitted, build and deploy and be turned on in a short time as possible. So we're prioritizing the ability to deploy capital where we can most readily convert it into those types of opportunities. So yes, we are still pursuing sites both domestically and internationally.

As we mentioned in our prepared remarks, we have spent a lot of time working with large French energy majors who are global leaders in energy, especially in the U.S., Latin America and the Middle East. And we're actively engaged to develop sites over time with those partners in regions of the world where we believe it will be very attractive to develop sites that have the ability to be used maybe initially for Bitcoin mining and then convert it over time into HPC AI or other enterprise workloads.

Our next question comes from the line of Greg Lewis with BTIG.

Fred, I did want to touch a little on -- you've been kind of alluding to the fact that this has been ongoing and we've been kind of building towards this for a while. I can appreciate it. That being said, there's definitely probably still some work to do to get the pen of the paper. But as we kind of look across your own portfolio, obviously, you have the nice presence in Texas, or Ohio, [indiscernible]. I mean you've e's presence across the mid part of the United States. Are there any areas that are kind of gaining more interest as you and Starwood continue to look out to look to kind of onboard that first customer?

I think typically, what tenants are looking for is, as I said earlier, the power has turned on. It's easy and fast to build and there's access to Internet and if there's need for [ water, water ] at the site. And a number of our sites fit that profile, obviously. If you look at other locations across the country, it's a question of really triangulating high-speed Internet, always on power and access to water. And all of our peers, ourselves, the [ Neo ] cloud providers and many others are all chasing opportunities around the country.

And obviously, our focus is to initially monetize the sites that we have because the power is already on, if you would. And we don't have to do any building to any greater extent to convert those sites. There may be upgrades to substations, things like that, that have to happen before somebody comes in and starts converting the site building buildings or doing whatever they're going to do for their specific needs. But the goal the fact of the matter is the power is the important thing, right?

Okay, super helpful. And then just realizing probably not as familiar with Starwood Digital Ventures as I'm probably going to [indiscernible], but kind of curious, clearly, they have a presence already in Europe with data centers. You mentioned some of the things you're looking to do. As we think about this relationship going forward, could this be an opportunity for kind of clearly where Marathon already has owned infrastructure that's an easy lift for getting more involved in some of these projects. But as we think about the next, I don't know, maybe not the next 12 months, but the next 3, 4 years, do you see an opportunity for MARA to kind of build with Starwood beyond just what you have is your own infrastructure?

If you mean by our existing infrastructure, absolutely.

Our next question comes from the line of Chris Brendler with Rosenblatt.

Congrats on progress here. I'd like to ask on sort of a trays question on the portfolio of data centers today and locations. And I noticed there was a comment in the deck that you have a gig of critical IT available today, that seems to include some of the sites that you're currently in hosting mode. So I was just wondering if you think you can potentially either acquire or run AI data center loads at these other sites that you're currently in hosting arrangements? Or am I reading that wrong?

Yes. I think good question, Chris. I think you have to also not look at those sites, the hosted side so much as while we operate a certain amount of load at existing sites, there is additional capacity available or expansion readily at hand with some of these sites, the substations, for example, are ready for expansion, additional load can be made available. and the name take capacity of the sites and the number of cases is greater than that, which we operate today.

Makes sense. Follow-up question would be just there wasn't a lot of numbers in this relationship presentation. And obviously, it's a huge pivot and exciting development. Just wondering like what kind of size projects are you thinking of in this sort of -- I don't think we're thinking about 200-, 300-, 400-megawatt sites like some of your competitors have. It sounds like it's going to be smaller. I know there's always been some question to the time line, but just thinking about sort of the numbers impact, how much do you expect -- I guess it depends on your ownership, but how much would you expect in the short potty envelope the economics on this joint venture compared to current Bitcoin mining operations at today's prices.

I'm not going to go into the economics today. I think more information about that will become more evident as we actually start speaking more about the specific sites and specific tenants because, obviously, who the tenant is and the economics will vary because of that. But to your question on size and scaling, you could look at a site, for example, like one we have in Texas, which currently operates over 200 megawatts of capacity in that site can be converted directly to a hyperscaler site. So are we looking at doing 40 and 50-megawatt sites? No. We're looking at doing much bigger things.

Okay. Great. And I guess you said that you'd have more detail as you make progress. Is that going to be -- we have to wait for contracts to be signed? Or are you planning on disclosing and presenting more information even before then?

So [ take ] it this way, you have to look at this almost like a real estate development project. It all depends on who the tenant is, what has to get built what the economics of the development costs are going to be, what percentage of the JV we're going to have with that specific project. Again, we will retain up to 50% of projects. or ownership in projects. So -- and then you have to look at, at the end of the day, what the lease rate is going to be. So every project will be different. So -- it's a little hard today to say, "Well, here's the number."

Our next question comes from the line of Kevin Dede with H.C. Wainwright.

Could you offer a little color or deeper color maybe on time lines on these project enhancements with Starwood? When do you think shovel start hitting the ground and when do you think they might actually start running for customers?

So you don't turn a spade until you have a tenant typically signed because that tenant has a specific use for the site. So it's not -- we're not building on spec here, right? We're not doing with some of our peers have done where we're going to build a powered shell and we're going to either fill it with somebody's compute or just build a powered shell and see who comes.

Starwood's expertise is that they've worked with the top tier hyperscalers and they understand what they want and a tenant looks at a site and looks at what can I do to that site to have what I want. They work with Starwood to design, get it permitted and built as opposed to building and hoping somebody will come like some of our peers have done. We are very focused on rapid execution with a lot of certainty. And one of the key reasons we chose Starwood is dramatically increasing certainty of execution and by partnering with somebody as all the relationships that are required on the hyperscaler side, has EPC capability in-house and it has the ability to have very good credit profile to be able to ensure these projects get funded and built.

And I think if you look at -- a key signal will be we have signed a lease for the tenant. And at that point, there's a clock that starts ticking to ensure that permits have been approved and then spades can start digging in the ground. But as I said in my prepared remarks, we have had very fairly advanced conversations with tenants, permits have already been applied for at sites. And our expectation is that this will be an accelerated process and that we will see updates regarding leases in a time period that I think most people will think is pretty accelerated.

Would you mind taking a minute or two to sort through Exaion, it's not absolutely clear exactly how MARA intends to lever that. I little digging, I understand it operates under MARA France, and I think you have 3 board seats of the 8. And I'm kind of scratching my head on how you intend to leverage that deal?

Okay. So Exaion was developed within EDF, which is a French state-owned energy companies operates one of the largest fleets of nuclear reactors in the world as well as huge renewables. I believe they're one of the largest electrical energy producers in the world. Certainly, the greenness with about 70% plus of their energy generation being green between hydroelectric capacity, other renewables and nuclear.

France, especially EDF has huge needs for AI. They have very large needs around private cloud because if you're operating nuclear reactors and you have the plethora of data that's coming off them and you're running AI models to ensure that they're operating up just safely, but you're optimizing how they're running. They did not want that as an outsourced service to a third-party provider. And so they built that competency in-house.

There came a point though where they believed it was better to take in external capital to advance the funding of that team and take what that team had built. And in is, I think today, about 90 people. They have built infrastructure tech stack and systems and services to be able to operate the data centers on behalf of EDF. They also operate a quantum computer in their Montreal facility, for example. And they are a team that has built systems specifically for private sovereign cloud type operations of data centers where you're running inference loads where data security is the absolute top priority, and they today operate 4 data centers, including Tier 4 data center capacity.

With the mix of AI loads traditional CPU load storage. They also have built infrastructure around blockchain. And for example, they provide for a major French bank, the underlying infrastructure that manages stable coin issuance for another French entity also real-world tokenization financial assets. So they have very advanced technologies. This is a group that is primarily engineers and technicians, operators, if you would. And our investment in Exaion will own 64 -- we now own 64% of Exaion. It is really focused on being able to leverage the infrastructure as a service technology, the platforms and service technologies that they've developed and deploy that in data centers around the world.

One thing you have to realize is geopolitically, we live today in a multipolar world. Gone are the days where U.S. companies could dominate the data center operations around the world. Countries do not necessarily want U.S. hyperscalers subject to the Cloud Act to operate in their countries where their sensitive data may be subject to U.S. government control. And this is especially true in Europe. What does that mean? It means that Europe is erecting walls where potentially it will be harder for U.S. hyperscalers to provide the types of services that they provide to enterprises in the U.S. to key strategic enterprises within Europe.

Our investment in Exaion is specifically targeted at 2 things. One, getting the technology platform so that we can deploy highly sophisticated private cloud with full security and data integrity globally in data centers for enterprise customers as well as within France and across Europe, provide private cloud infrastructure and services to leading enterprises where they would prefer not to make that data available to hyperscalers who are subject to the U.S. Cloud Act.

In a way, it's almost a market where there's an advantage to being outside of the U.S. And we are very bullish on the opportunities that Exaion's going to provide us. They know how to service energy majors very well. And we believe that our relationships with the other fringe energy majors will be an advantage there as well as in other countries around the world. And one use for our smaller data centers is specifically as private cloud operations. The vast majority of corporate data today resides not in the public cloud, but in private cloud or behind the firewall of the corporations -- financial services companies, health care companies, drug research companies, defense industries and other strategic industries do not put their data in the public cloud. People will put e-mail, they'll put other general-purpose data, but they will not put their core operating data into the public cloud. And if you want to run AI, it means you have to run it in the private cloud.

We have spent a lot of time leveraging one of our key board members [indiscernible] George, who today runs a large part of the efforts at Mastercard and previously did so at Intel and at Oracle. And we've spent a lot of time to really understand the needs of enterprise customers, not just in the U.S. but internationally and understand where their key pain points are.

When you look at the hyperscalers today, and Jensen Huang said this in his earnings call yesterday, the majority of the inference that's being done by the hyperscalers is in improving their own search products, their own product selection and recommendation engines. And the tools that -- and services, they are already used for selling to customers and improving their advertising basis, Google, for example.

What you're starting to see now is corporations wanting to start deploying vertical AI solutions for example, production optimization, fraud detection, things like that, all of which they're going to run behind their firewalls or in full private cloud. And companies that want to run that are going to need infrastructure. and the infrastructure they're going to need has to have the ability to be fully secured, belly private cloud and ensure that the owner of the data has full control over the data regardless of who operates the data center. And these are the core technologies that Exaion brings us, and we're super excited about this because it really allows us to totally differentiate ourselves from people who are just offering basic services to [ Neo ] clouds or even [ Neo clouds ] themselves because it's a much more sophisticated infrastructure that we believe will generate much more value per megawatt, which again is our core metric and is much stickier.

[indiscernible] that you'll be allowed to take that technology to geographies outside of Europe.

Yes. So we own -- we have majority control of Exaion. You asked a question about the Board seats. There are 8 members of the Board, 3 from EDF, 1 seat is the CEO of Exaion, we have 3 seats and then a French technology entrepreneurs, [indiscernible] has 1 seat. And we're very excited to have him on board because of his background. You guys can do your own research on who he is. He invested personally in MARA France, which is our holding company through which Exaion is owned.

There are no further questions at this time. I'd like to pass the call back over to Robert for any closing remarks.

Thanks, operator, and thank you, everyone, for joining us today. If you do have any questions, that were not and during today's call, please feel free to contact our Investor Relations team at ir.mara.com. Thanks very much, and enjoy the rest of the day.

This concludes today's teleconference. You may disconnect your lines at this time. Thank you for your participation.

Greetings and welcome to MARA's Q3 2025 Earnings Conference Call. [Operator Instructions]. Please note, this conference is being recorded.

I will now turn the conference over to Robert Samuels, VP of Investor Relations. Thank you. You may begin.

Thank you, operator. Good morning, and welcome to Mara's Third Quarter 2025 Earnings Call. Thank you for joining us today.

With me on today's call are our Chairman and Chief Executive Officer, Fred Thiel; and our Chief Financial Officer, Salman Khan.

Today's call includes forward-looking statements, including those about our growth plans, liquidity and financial performance. These involve risks and uncertainties, and actual results may differ materially. We disclaim any obligation to update these statements, except as required by law. For more details, see the Risk Factors section of our latest 10-K and other SEC filings.

We'll also reference non-GAAP financial measures like adjusted EBITDA and return on capital employed, which we believe are important indicators of operating performance because they exclude certain items that we do not believe directly reflect our core operations. Please see our earnings release for reconciliations to the most comparable GAAP measures.

We hope you've had the chance to read our shareholder letter and look forward to your feedback. We'll begin with some brief prepared remarks from Fred and Salman, after their comments, we are going to be conducting an analyst interview with management.

Today's session will be conducted by Reggie Smith, analyst at JPMorgan. And with that out of the way, I'm going to turn the call over to Fred to kick things off. Fred?

Thanks, Rob, and thank you all for joining us. This quarter, we continued to evolve MARA from a pure-play Bitcoin into a vertically integrated digital infrastructure company, one that converts energy into both value and intelligence. At the heart of our strategy is a simple belief, electrons are the new oil, energy is becoming the defining resource of the digital economy, powering everything from Bitcoin mining to artificial intelligence. And we believe those who control abundant, low-cost energy will shape the future of both finance and intelligence.

Bitcoin has now entered its institutional phase. We're seeing financial leaders such as BlackRock, Citi Corp and now even JPMorgan integrating Bitcoin into traditional frameworks. And we're seeing the establishment of strategic Bitcoin reserves by corporations and governments alike and even the -- Treasury has posted positive notes of up Bitcoin on X.

What miners have always understood is now being recognized by global markets, Bitcoin has Digital Energy, a mechanism for story and transmitting value. As one of the largest bitcoin miners in the world, MARA sits at the center of this shift. Our NP value infrastructure allows us to convert raw power directly to Bitcoin that we hold on our balance sheet, a distinct advantage that grounds our broader mission, transforming energy into intelligence.

Every electron has potential value and artificial intelligence represents the next frontier events of the transformation of energy into even higher value. We believe that inference AI with the value of AI is actually created and drive and not treating in foundational models is where the industry will create the greatest amount of value over time.

Every insight produced by an AI model has a cost per token, driven by the cost to build and operate the data center, of which the energy cost ticks up a major component. Over time, compute and the cost to build the data center will drop as technology advances such as HODL ASICs open-source models and the ability to operate in less sophisticated and less costly data centers drive efficiencies resulting in rapidly declining drops in cost per token, making the AI data today unable to compete on cost per token over time about significant technology refreshes, requiring even more and higher capital injections.

We believe energy not compute really becomes the primary constraint on AI growth. We are already seeing the alternatives to GPUs enter the market and open source AI is making it far easier and much less expensive for companies to deploy advanced AI systems directly in their own private cloud environment.

In the past, most models were only available through public cloud APIs. That meant enterprises had to send data offsite and pay high per token fees to access AI capabilities. But today, many of the world's most capable models like LAMA straw and others are available in open source form, giving companies full control to run AI more cost efficiently and fine-tune their models privately.

This is a major inflection point for enterprise computing and the shift that plays directly to our strength as we build out low-cost, high-efficiency compute powered by our own energy infrastructure. we believe we're positioned to provide the kind of private, scalable environment enterprises need to deploy these open models securely.

MARA has positioned itself at the nexus of these 2 AI trends, open-source AI expanding the addressable market for private cloud compute. We believe that the future in structure will be able to serve that demand efficiently and profitably. This is where MARA's expertise in securing and operating low-cost power gives us a distinct advantage. Just as we optimize for the lowest cost per petahash and mining, we're now optimizing for the lowest cost per token in AI inference.

Our long-term vision is to integrate these 2 energy pathways, Bitcoin and AI into a single platform. Bitcoin mining monetizes underutilized energy and stabilize grids, while AI inference transforms that same energy into intelligence and productivity. By bringing Bitcoin and AI together, we seek to maximize the value of every megawatt hour we manage.

We've already begun executing on this strategy. This quarter, we installed our first AI inference rack at our Grand Berry site within a modular nonwater-cooled containerized data center. This site currently has 300 megawatts of nameplate capacity with potential opportunities to expand our growing AI inference business in combination with our Bitcoin mining operations at the site.

This milestone marks a significant step forward in proving out our AI infrastructure and next-generation interim hypothesis. It also demonstrates the versatility of our platform, underscoring the potential flexibility of our mining sites to support AI workloads along with Bitcoin mining.

Two major initiatives this quarter are propelling our strategy going forward. First, our pending acquisition of Exaion, a subsidiary of EDF in France. Once regulatory approvals are completed and closing conditions have been met, Exaion will expand our capabilities through enterprise-grade, AI-optimized private cloud and HPC infrastructure. We believe this will position MARA as a credible -- for enterprises seeking secure localized inference capacity.

Second, today, we announced an initiative with MPLX, a separately traded public company formed by Marathon Petroleum Corporation the largest petroleum refinery operator in the United States to develop and operate multiple integrated power generation facilities and state-of-the-art data center campuses in West Texas.

Under this initiative, MPLX will provide long-term access to lower-cost natural gas at scale where Mara will develop and operate on-site power generation and compute infrastructure. The initial capacity is expected to reach 400 megawatts with the option to expand to up to 1.5 gigawatts across 3 plant sites. We are also evaluating additional prospective sites to support modular AI and HPC data centers alongside mining operations, creating optionality future AI inference workloads. MARA's approach is to deploy smaller modular facilities directly at lower-cost power sites instead of building hyperscaler campuses. We believe this distributed model will enable us to capture value at the inference layer while continuing to monetize mining and grid sales. This modular structure also gives more the optionality to shift capacity towards HPC over time. As an -- economics and infrastructure maturity support greater AI utilization.

We believe MARA is positioned to capitalize on a key structural advantage as power becomes the primary key strength in AI growth. Together, Exaion and MPLX connect the 2 sides of our AI and data center business, energy and compute and strengthen our ability to control both cost and performance from power to inference. Internationally, we're deepening relationships across Europe and the Middle East, where we see significant opportunity to deploy our integrated energy and compute model.

Our pending Exaion acquisition exemplifies us, and we're honored to welcome Gerard Mestrallet, President Macron Special Energy Envoy, as an adviser tomorrow. His expertise strengthens our global strategy as we pursue our goal driving 50% of revenue from international operations by 2028.

On the financial front, we continue to operate with discipline and transparency. We ended the quarter with 52,850 Bitcoin having mined over 2,100 BTC during Q3. We remain focused on improving free cash flow through ongoing cost optimization level efficiency gains and disciplined capital allocation. We have begun opportunistically monetizing Bitcoin from production to fund operating expenses and aim to limit reliance on our ATM to support growth initiatives, helping to mitigate shareholder dilution.

As I spoke about last quarter, Bitcoin prices have consolidated within a range of at least Q2. With intermittent volatility, we view this as a healthy period of equilibrium characterized by institutional inflows into ETF and balanced by long-term holder liquidation activity using [ Dority Visitors ] IPO analogy Bitcoin is going through an IPO where early investors in BCs are exiting and institutional investors are coming in forming a new base and foundation for growth.

Meanwhile, broader macro trends, including rate cuts, and expanding liquidity suggest improving condition for risk assets. Regardless of short-term volatility, our long-term trajectory remains unchanged, building enduring value through energy ownership, operational excellence and strategic execution.

Finally, I want to provide an update on TPIC. While we continue to recognize the long-term potential of 2-phase immersion, Its practical broad application is still a few years out. And directorship cooling remains the preferred COO methodology of data center operators and compute OEMs.

We have exited near-term investment in 2 phase immersion to focus resources on opportunities with a more immediate and higher return potential. In closing, MARA is evolving from a Bitcoin miner into a digital infrastructure leader, combining energy generation, Bitcoin mining and AI compute under 1 scalable platform. Our guiding metric is simple profit per megawatt hour. It measures how effectively we convert energy into value, whether in Bitcoin, AI inference or grid stability. As we continue to execute, we believe the market will increasingly recognize the strength of this diversified model and the strategic importance of energy ownership in the digital economy.

I want to thank our employees for their exceptional work this quarter and our shareholders for their continued support as we build Mara into the world's leading digital energy and infrastructure company.

With that, I'll turn it over to Salman to review the financials.

Thank you, Fred. During the quarter, global hash rate grew by roughly 20%, with the hash rate and network difficulty both hitting new all-time highs by end of the quarter. Bitcoin's price remained relatively range found, trading between $104,000 to $224,000, closing the quarter with a modest $7,000 gain. It was one of the most competitive mining environments in recent times and a difficult backdrop for our performance this quarter.

Despite this, Q3 was the highest revenue and exahash quarter in the company's history. Our focus on operational and financial discipline over the past year is reflected in the substantial growth of our compute capacity and deploying holdings. Between Q3 2024 and 2025, our Bitcoin Holdings expanded by over 98%, growing from approximately 27,000 to nearly 53,000 bitcoin.

Our energized hash rate also expanded, increasing 64% from 36.9% to 60.4 exahash per second. Bitcoin price appreciation resulted in approximately $4.3 billion or 256% increase year-over-year. While Fred spoke to our vision and strategy, our vertical integration and capital allocation strategy is reflective on our financial results.

That balanced execution allowed us to expand our holdings and take advantage of favorable market conditions, while maintaining liquidity and flexibility. We mined 2,144 Bitcoin and purchased an additional 2,257.

The impact on our financials is evident in the results we achieved. Let's dig in. Revenues increased 92% to $252.4 million from $131.6 million in the third quarter of 2024. Bitcoin's average price increased 88% over that time period, which contributed $113.3 million. We mined an average of 23.3 BTC a day throughout Q3 compared to 22.5 BTC in Q3 of 2024 which resulted in 74 more bitcoin mined this quarter.

Our strategy to deploy exahash responsibly resulted in growth of our BTC mine despite a significant growth in global hash rate and with network difficulty levels. We reported a net income of $123.1 million or $0.27 per diluted share last quarter compared to net loss of $124.8 million or negative $0.42 per diluted share in the third quarter of last year.

We also booked a $343.1 million gain on digital assets, including Bitcoin receivable during the third quarter of 2025. The reflecting the positive impact of the Bitcoin holdings on our balance sheet. Now let's talk about our cost structure.

Our purchased energy cost to Bitcoin for the quarter was $39,235, and our daily cost per petahash per day improved 15% year-over-year, which we believe at scale is one of the lowest in the sector. This improvement is directly tied to our growing inventory of owned and operated sites, which now account for approximately 70% of our nameplate megawatt capacity.

That transition supports our vertical integration strategy, but also pays dividends, both financially and operationally. Since we do not control the price of Bitcoin -- mine, minimizing the cost of inputs like energy are critical to the financial resilience and long-term success of the company.

Next, I'll provide some insights into our Bitcoin Buildings and digital asset management strategy. MARA is the second largest corporate public holder of Bitcoin, and we seek to generate returns on our holdings as Bitcoin price appreciates. Our approach combines the potential for long-term discount appreciation with disciplined efforts to generate return while managing risk.

Additionally, we have also used Bitcoin as a collateral to borrow under lines of credit. As of September 30, 2025, we held a total of 52,850 Bitcoin, including 17,357 Bitcoin that per loan actively managed and pledged as collateral.

As such, approximately 1/3 of our total holdings were activated through our digital asset management strategy. In Q3, we issued $1.025 billion or $1.025 billion of 0 coupon convertible notes due 2032, extending our maturity profile and increasing balance sheet optionality.

With additional liquidity, MARA gains strategic flexibility to act on opportunities, whether that's acquiring morbid coin, funding acquisitions balance sheet management or general corporate purposes. We have positioned MARA to act in response to market conditions in order to maximize long-term shareholder value. As of September 30, 2025, we held over $7 billion in liquid assets, giving us the flexibility to fund domestic growth and pursue international expansion.

To streamline our communications starting in Q4, we will share our production on a quarterly basis. Investors can continue to monitor our monthly MARA pool production in real time on the manpool.

As we have stated previously, electrons are the new oil, and we are laying the groundwork for 2026 and beyond. We're executing on a pipeline of energy infrastructure projects both in the U.S. and internationally, and we expect these investments to expand our capabilities while keeping costs low.

With that, I'll turn it over to Reggie Smith from JPMorgan. To begin our management interview. Reggie?

I appreciate you selected me for this call here. obviously very big announcements this morning. I guess kind of help me interpret this morning's announcement versus, I guess, kind of your prior strategy. Like what's being emphasized, deemphasized? Maybe talk about that from the -- like what's the most that you're placing on the business and maybe the lease because there's a lot going on here, certainly relative to the other Bitcoin miners, I know the other guys is either deploying mining or kind of colocation. You guys seem to have a lot more live mobile and maybe talk through those differences.

If you think about the deal we announced today, it's about getting access to low-cost energy that is reliable, available 24/7, because we are the generator, it provides us with a very low cost. If you look into the details of the announcement, you'll see that the pricing on the gas is amongst the lowest in the market.

The other thing is that it gives us now the capacity to add potentially up to 1.5 gigawatts of data center capacity if we want to -- which gives us lots of flexibility. A lot of our Bitcoin mining sites are very attractive to use for inference AI, as we discussed earlier. We talked about what we're doing at Granbury and what we'll be able to do at some of our other sites in a similar fashion where we can blend inference AI in Bitcoin mining.

But the relationship with MPLX and the opportunities it provides give us a much broader canvas that we can paint on, whether that is traditional HPC, like some of our peers have done or whether we want to build it out as hybrid AI, inference AI and Bitcoin mining sites. So it gives us a lot of flexibility. And we believe controlling and owning power is the core part of any company that operates in the digital infrastructure space.

When you look at the spending that's going on. And I think Sachin Ardell said this in a recent podcast that was quoted where he was quoted as saying that compute isn't the constraint, energy is the constraint. And so access to energy, we believe, is critical.

We think inference over the long term is where all the value is going to get created in this space. But we believe that Bitcoin mining has a very important role to play in which is balancing grids, but providing a flexible load when mixed with AI, such that AI can begin to operate in more places than it does today.

And the last thing I'd say is that we believe that the technology curve is going to move so quickly in this space. Because you have to realize that just like in Bitcoin mining, where cost per petahash is the most important metric that drives profitability in the AI business, unless you are in the application layer.

In other words, running -- owning the data and the application that is generating value for the enterprise in health care, owning the health care data, running the actual analysis. The only thing hosting providers and model operators provide our tokens in the sense of we need lowest cost per token if we're going to use that service. And using the APIs when the cloud providers is a very expensive way of running AI and most enterprises today are being confronted with the fact that the cost per token is too high using existing systems, and they want to move to lower-cost systems, and we're going to start seeing that we already are seeing ASIC-based solutions coming open source models, all of which will allow enterprises to build their own and operate their own private cloud or use those services from third parties, allowing them to drive value from AI.

So I think for a lot of the big guys, the challenge is they are doing deals with colocation partners where they are not taking on the debt. The debt is being laid on the joint venture or the SPV related to that colocation facility. And that correlation partners having to deploy a lot of capital to build the sites and equip those sites and you have technology obsolescence over the course of a 10-year lease, you will have to upgrade the hardware in that location. And you have to estimate that in the cost of where it's going to be to build and operate. And I think there's a risk potentially, that $1.4 trillion of data center contracts signed by OpenAI over the -- that will have to be operating in the next 5 years according to what was recently reported in the press, that some of that may not actually be able to come online and generate revenue. So I think our approach is much better, more prudent, certainly much more capital efficient.

And by being at the end of the spectrum where we're vertically integrated and able to operate at lowest cost per token and deliver lowest cost per token, we will have a significant advantage in the marketplace.

And Reggie, just a reminder, we -- today, we control approximately 2 gigawatts of capacity. And this added capacity is incremental to that, that takes us to close to 3.5 gigawatts over a period of time.

Got it. Understood. I'd like -- I appreciate the color there, and I was doing some kind of light math this morning. And when I think about, I guess, kind of AI and HPC, you made a comment in your shareholder letter about the price of power and the price to compute. You made some parallels between Bitcoin mining and HPC. And I was looking at the numbers, and I think they may be a little bit off, but directionally, this is like a fair statement.

When you look at Bitcoin Mining, the price of power and the price of the actual ASICs, you think about depreciating per hour are about the same, like a 1:1 ratio there. For GPUs, that ratio is more like 1 per 10 GPU, some depreciation charge and not really depreciation is super high. So you talked about ASICs and somehow, I guess, driving the cost of the hardware, am I thinking about that right, like what are you seeing and kind of where do you see the role going there?

Just think about it this way. When Bitcoin mining started, we were running CPUs, right? Then we went to GPUs, then we went to FPGAs, then we went to ASICs. And when you look at the amount of compute power, for -- think of it as the number of terahash we could produce for a jewel of energy. It has dramatically changed, so you are now processing many more calculations at much lower cost of energy. And in our business, we depreciate the compute over 3 years.

So if you're a hyperscaler and you're signing a deal for 10 years, some of these are 15 years and the depreciation schedule is 5 years for the machines. Does that mean they're going to have to replace those machines 3 times in that cycle, right? And to your point, GPUs to power is most probably a 10:1 ratio.

And as you get to ASIC, that starts dropping and power starts becoming an even more important component. And when you start looking at the end cost per token, at that point, the model cost also comes into play. And so if you have open source models, if you have low-cost hardware, that's energy efficient, you're operating in data centers that don't cost you $10 million of megawatt to build. You start getting to economics, would start resembling Bitcoin mining overturns.

Understood. Now help me understand this. I want to understand or make sure I'm hearing you correctly. When you think about kind of the investment risk and the CapEx risk within the chain. Obviously, you've got guys that are building data centers, you've got people that are buying like GPUs and hardware and then you say, obviously, you got the model guys as well. I guess your comments on kind of where the CapEx risk is greatest. Are you suggesting that the people that are buying the machines are taking on the most risk? Or do you think there's still substantial risk in building big dissenters. And I ask you that in the context of you guys just, I guess, bought a few GPUs yourself, like help me square all of this together to understand kind of what your view is there?

Right. So part of the question is, are you in the business of being a bare metal shop, right? You're providing essentially hosting and GPUs. Looking at what iron is doing, right? Bare metal, somebody has to load their software on it, but they're renting capacity on GPUs effectively. And that deep cloud as it's called, that case, the owner operator is funding the GP purchases, right?

In the case of a colocation, there are some deals that have been done where the operator is funding the GPUs and there are other deals where the lesser of the space, if you would, is bringing the GPUs, and they are the buyer and operator. So if Microsoft comes and is going to contract with you to just buy capacity from you, they're going to bring the GPU, so far, you would hope at least, and they're taking that risk.

But there are lots of different models out there being operated by people. What we're doing with inference at the edge is much more around providing inference AI, which is not running on GPUs. We're running on ASIC type solutions. And so it's a very -- it's a different model from a hardware cost perspective. It's air cooled that's not liquid cold, for example, which means your infrastructure is much less expensive. You're not having to spend many millions of dollars per megawatt on building infrastructure, specialized cooling infrastructure. And all of that adds up to the economics of what you can do.

But inference is also done at smaller volumes, right? You don't have to do 100-megawatt sites yet. Most of the needs for inference still are quite young. It's early in the market. But if you believe what Gartner and the Analyst Day, over the next 3 to 5 years, the inference will be the primary generator of revenues and value creation within the AI space. So that's where we're swimming.

Understood. You've got to skip around a little bit here. I wanted to talk about Exaion, I think kind of looked back into the broader discussion. But obviously, you guys have enough set of acquisition. Help what they do today? Maybe talk about the scale of their operations, like are they running data centers today? And if so, what's the size of those what do they do exactly?

Yes. Exaion is today until we close a full subsidiary of EGF that operates EDF data centers were all of the data for the nuclear fleet operates in this process. So they run EDF AI and traditional data centers across the EDF enterprise. They have about 4 data centers today, 3 in France, 1 in Canada. They also operate quantum technology in the Canadian data center, which is made available for research purposes, and they have built a whole set of software solutions that allow you to operate the data center, store data in full private mode, meaning the user's data fully encrypted. Exaion doesn't have the keys to that data. And so we're that data center to be broken into, if you say, of somebody were to steal data, the data in the data center is encrypted. So the customer holds the key to that data.

And so it's a way to build private cloud solutions that are fully secure. And so the whole reason for making investment in Exaion is it gives us access to a team and an existing set of data centers that are Tier 3 and Tier 4 already. They know how to operate the most sensitive data, data to protect it. They have existing customers. So they have experience and we are going to leverage their knowledge, their experience, their technology and their platforms to expand what they do on a global basis.

Got it. So they're asset like. They're more of a service layer, they're engineers or software, things like that, like they don't actually own any data centers. It's really running that data center securing data, is that the right way...

Yes.

Understood. Okay. Is there a way to frame it maybe early their revenue run rate? And interestingly, about that transaction, I think the first transaction, you bought them for $168 million. The next 11% will be at a much higher rate. Like what was the thinking there...

I think you can think of how many times deals like this are structured, you're paying for a portion of the business based on where it is today? And then the growth opportunity for the existing investors is in executing on a plan to help grow the business and therefore, you're going to pay a higher multiple for that. That's how you should think...

Okay. Now on the tile. So the MPLX transaction, real quick on that, does it require any like ERCOT approval like these guys have the natural gas, you guys would make the power plant or from the generation assets and the data center but is anything needed from ERCOT? Any road blocks there? And how quickly could you -- data center up in kind of running?

Yes. I think you have to think of it more as the first thing we're doing is building a power generating station, which will be gas-fired power plant. So you have regulatory requirements around air permits, for example, which in the current political environment should not be exceedingly difficult to acquire. We feel fairly confident that we'll be able to get those without much problem.

So once you build a power plant, then because you are not directly bird attached yet you then have to apply to attach to the grid and be a provider to the grid. So there's a regulatory process for that. Meanwhile, you can be producing energy and operating data center fully behind meter. And it's -- ERCOT gets involved when you connect to the grid or the utility stuff. When you connect to the grid.

And the goal here, what's really important to remember about this MPLX relationship is it gives us the ability to own and operate gas-fired power plants with very low cost gas with the ability to colocate large-scale data centers with reliable 24/7 power is a very attractive part of the marketplace. And so it gives us a lot of control to really drive our growth in a very cost-effective way. And I think it positions us very well come what may in this HPC AI market and give us a lot of opportunities. to really operate and continue to generate a lot of value for our shareholders.

No, I agree. It's been thinking about this idea of like vertical integration, and I didn't know if it was going to be a tower company acquiring data center capabilities and the other around. So this is very interesting. If I could dig in a little bit more. So I think you talked about [ 400 ] megawatts of capacity to start. Should we think about like the minimum effective dose kind of get started. So I don't know if you want to commit to 400 megawatts were out the bad, is it at -- megawatts and how quickly can set like this come together? And then I know it's early days, but like we've heard estimates of up to $10 million per megawatt to build out a data center, like what are you thinking about from that perspective as well?

So you don't build a power plant in 20-megawatt increment. You build it right to a certain size at each site. So there are 3 sites will likely think of it in 100-megawatt increments initially, but you have the ability to scale these plants, much larger.

As it relates to the data centers, we have the optionality. We can build these as traditional Bitcoin mining data centers that are fully containerized at somewhere around $1 million including hardware costs for compute. If you then want to look at going the AI route, if we're doing it similar to how we're running the inference AI, we're running today, the actual infrastructure cost is very similar. It may be on a little bit more expensive, depending on the cooling technology, if we use direct-to-chip cooling or we continue to use air cooler. And if you use direct-to-ship cooling, your cost of infrastructure will end up somewhat higher.

But the key is we're not building buildings that take 3 years to build. We're doing these as modular container solutions, which gives us full flexibility to reconfigure a site depending on whatever we want to do at it. And I'm a big believer that you will see very high-performing HPC capable modular solutions on the marketplace within the next 2, 3 years, where you will be able to deploy the same sophisticated solutions you're building in these very sophisticated data centers, where people can run some of the most sophisticated they need to.

Remember, there are not many customers in the world who need data centers of the scale that open AI needs it, right? OpenAI needs a lot of compute capacity because of the breadth of data and the breadth of solutions their models operate.

If you remember what DeepSeek did and how deep sea create the turn in the market, it's because instead operating with a broad foundational model, they only load into memory, specifically the model segments that they need and the data to solve the query, which means you now don't need all of that scale. So what I think will happen in the marketplace is that you're going to have efficiencies and models going to open source, clients developing their own models and training their own models because the clients don't want to give the data to OpenAI.

And I'll give you an example. I was at FI last week in Saudi Arabia, and I was sitting with the Head of Strategy for Aramco on a panel. And they don't put their seismic data in the cloud. They're not going to do that. What do they do? They build their own models. Other companies do the same thing. Look at what Lockheed just did the deal the -- with Google right? It's an on-prem solution. You are not -- I'm not going to put my data up into your cloud, Google. You're going to build a cloud instance on-prem, on my site that is airgap from your systems.

That's what corporations want. They want data sovereignty. They want private cloud. They don't want to run up in Meta's cloud, Amazon's cloud or open AI systems. 70% of corporate days today is still not in the cloud. There's a reason for that. And I think when you look at inference, inference is driving insights from the data that runs your company, right?

If you're in the health care business, doing drug discovery, it's all the patient data, the lab samples, et cetera, all that data you're driving insights from it, right? And if you are doing building airplanes, it's all the design data and the manufacturing data. If you're running a factory, it's the operations data of the factory, right? If you're running a power plant, it's the operations data of that power plant. You don't want to run that off site. You want to actually run it on site because as those systems become mission-critical and actually operate the resources and operate parts of the business, you can't take the risk that you have a system failure that brings your whole business down just because you lose a lead to a cloud or Amazon goes offline like it did the other day.

So I think people really have to understand that there is a limit to what data and how much risk people want to do in putting their core critical assets into a cloud operated by a third-party. And if they can solve the model issue and do it at lower cost, near prem or on-prem in a private environment, they will do it. And I have been speaking with the heads of AI for major corporations in the financial market today, who tell me that they are relocating AI systems out of the cloud back to near-prime on-prem private solutions because it is significantly less expensive to operate than doing it in an Amazon cloud or other places like that.

And I think that the analyst community really needs to do a much better job of talking to the enterprises who are the users, these are the people who are actually going to pay the money that will allow open AI to be successful or not, that will allow Microsoft to be successful or not. You can talk to your blue interface with the people building these things, but it's like building railways. If there isn't passenger traffic and there is any cargo, the rail line fail. So that'd be down on this, but this is an important thing that a lot of people aren't doing. You need to talk to the customers, who's going to pay for this stuff.

And I want to make sure I'm hearing this right and connect these debts. So I think you mentioned kind of a smaller kind of, I think, a 1 megawatt -- what do you call it, I guess, kind of like a sample or a small -- data center. If I'm hearing this right, are you saying that like that could become like the prototype for enterprises having their own on-premise like AI...

Yes. So think of it this way, right. Give you an oil drilling example, right? So you have an exploration drill that's drilling, you have seismic data. Today, you have to plan exactly the drill profile and what some -- what the drill operator is going to do. And so the oil companies have built these very sophisticated AI models that run in a module container typically out on the drilling side that are collecting real-time data from the drill and then feeding back instructions into to the drill master. That's an existing example.

You can go to a trading, a financial trading company. And their whole thing is speeding latency. They want their systems operating on their local network, not on a wide area of connection where there's latency because 25 millisecond delay in a response means they lose the profit on a trade. And so there are -- whether you're looking at defense, which is going to be a huge growing sector when it comes to AI, just look at the amount of AI that's needed to operate in any third of war today.

Look at health care, look at manufacturing and production, look at the movie television industry, the single largest consumer of tokens in AI are video illustrations and audio generation. Those are the systems that consume that these diffusion models are the single largest consumer of tokens.

And so cost per token is very critical to them because if you're going to generate a 5-, 10-, 15-minute clip of video, it takes multiple factors in magnitude, more tokens than asking OpenAI where your feed your lunch today.

And so I think, again, the marketplace gets all hyped up about these big contracts, but they really need to look at who's actually going to use this stuff? What are they going to use it for? What can they afford to pay for it? Will the pricing trends be over time and to use the worn out win rest technology, if you're in our business, you want to be skating to where the puck is going to be, right? You don't want to be chasing the puck. And I think there are a lot of people announcing deals out there getting on the bandwagon to pump their stock when they need to look at what's this industry going to look like in 5 years.

That actually is a -- question, Fred. So thinking about announcements and catalysts like what should we look for from MARA to know that like this strategy is taking form and we can start to frame an economic story or accretive story around some of these initiatives. Like what are the milestones and announcements we should be looking for from you guys?

So here's what I think you should look for. 4 years ago, I made a presentation at a conference where I said that Bitcoin miners are either going to be energy companies or be owned by energy companies. I think what you should look for is when large energy companies start signing partnership agreements with companies like us to monetize their energy assets at large scale. That will tell you that if that happens to be of that they have chosen us to do with because they feel we are the best option for them to maximize the value of the electrons that they produce. That's 1 step.

The next step is as you start seeing customers using more and more inference AI. And you see us reporting a greater and greater mix of inference AI in our data centers. And the real metric you should look for is what is our profit per megawatt hour that we talk about. It's not a GAAP measure. So it's not going to be reported that way, but you can think of it as an operational KPI where the profit we can generate from every megawatt hour of energy that we consume or produce, is a data point that our investors will be able to see and that will directly correlate to our profitability and ability to cash-generating business.

Okay. Fair. And just to make sure and I apologize to so many questions. Are you looking to sign colocation clients or deals for this site in West Texas? Or is this something that you're thinking about putting your own machines in...

It gives us I'm not going to answer the question directly because I think our competitors spend more than enough time listening to what I say and then emulating it. So I'm just going to say it this way. It gives us maximum optionality to decide what we want to do with whom.

Got It. Okay. I want to bring it up because you didn't mention signing a colocation as like a milestone.

No, you see, if I can operate inference AI and make money on it without signing a colocation facility. That will give you a little bit more insight into what the business model might actually be. Because think about the best thing about our Bitcoin mining business is we don't have a customer. What's the hardest thing all these colocation deals have? Does that have to go into a customer?

Yes. Okay. Okay. People say change my opinion when the facts change, and that's a pretty -- it seems like a pretty major shift for MARA like I said, you guys bought GPUs, I guess, in the last 3 months and you start to run them, like what in your mind has changed that has changed your opinion or has your opinion changed and strategically, it seems like the company is kind of pivoting. Talk to me about that. Like what have you learned or gleaned in the last couple of months or quarters it is shift.

Yes, Reggie, I wish I could tell you that I had a lightning bolt strike me and I came to Tiffany. But this is -- we're executing the strategy we decided to execute over a year ago. It's just we have decided not to go totally open to market and tell people what we're doing because it just gives our competitors insight into what we do and they can emulate it. and prefer to control the timing on how we talk about what we're doing.

But I've been talking for the longest time about inference at the edge and that's where we would make our market in the marketplace. And we are -- we've talked a long time about owning power and the desire to run our business based on controlling energy assets that were fully virtually integrated. And we're doing that.

There's no change in strategy. There's no pivot. It's just we have been purposely operating more like a startup in the sense that we have really wanted to make sure that we had everything in place so that as the market becomes aware of what we're doing, they just start seeing kind of announcement after announcement after announcement that just gives them more and more confidence in that we're executing on the vision that we set out a year ago.

Yes. No, I'd say from where I sit and I think about all the pieces you guys have, there are a lot of pieces, and I'm not starting to figure out how to put it all together, but it seems like you guys have a lot of ways to kind of win here. I guess we just have to kind of sit back and wait for those announcements as they kind of come through. I know we've kind of spent a lot of time on this. I hope it wasn't a wasted time for people.

Maybe you could shift gears a little bit and talk about your like sovereign and foreign government initiatives and things that are going on there. Like one of the questions I have you think about this, what do you think gives you guys a right to win in the sovereign have you kind of load management space versus competitors? Who's been competing with you?

I think -- so there are a couple of ways to look at it. Most of our competitors enter a marketplace by partnering with somebody or contracting for power. They don't bother talking to the government because they're afraid that if they do, they may not be allowed to do what they want to do. And that's the case in a lot of places in the Middle East.

We, on the other hand, chose to do it the other way. So in UAE, where we've been operating now for a couple of years, we chose to directly go and work with the sovereign. So we partnered with ADQ and IHC and operate a joint venture together with them where we balance the grid in UAE. It's one of the most advanced liquid immersion technology sites in the Middle East. The only one that's bigger than that is the liquid merchant site we operate in Granbury.

And so that has given us the reputation of being somebody who works well with government entities follows the rules and is focused on being a good to grid citizen in balancing the grid. So when we talk to people in other countries such as in France, such as in the U.K., such as in Kenya, in Saudi Arabia in other places. We are welcomed with open arms because we are focused on how can we make your grid more efficient and more effective? How can we make sure that every electron generators generate maximum value?

And we are here to be a good grid citizen, and we are here to operate such that your grid becomes more stable, and it becomes easier for you to bring on new types of loads and see them AI data centers or whatever. And the challenge, the way most people see it, that takes time, right? I have been crossing the Atlantic very frequently. But I have been having meetings in at the top levels of government, and we have a lot of support. We certainly have seen lot of support on the European side because there are certain dynamics in Europe that create very large opportunities for us.

And so same thing exists in Saudi Arabia, for example, in other places. And we think that we -- it's worth our effort to spend the time and takes the time to do this carefully and prudently and well thought out so that we're able to execute successfully and have long-term success in the countries. Because if we're friends with the government, then we have the advantage that as they look to expand what they're doing, if we are a good partner, they will come to us and say, hey, we want to do more with you. And that's the type of relationship we want to have with our partners across industry, be it governments, vendors or end customers.

And you haven't talked about Bitcoin mining at all, no running show. Just an update there. Love to hear about the stuff that's happening in the wind farm and some of your flared gas initiatives. Maybe talk a bit about Auradine. And then, I guess, your plans for growing hash rate here and how you think about that in the context of kind of where cash price is and why it makes us to continue to grow your hash rate at these levels.

Yes. So maybe I'll look at this in kind of a somewhat reverse order. So there is more hash rate coming online every day from lots of players. There are very well capitalized to the companies who are not public who are -- have access to huge amounts of capital who have a stated goal of becoming the largest bitcoin miner in the world. And if we don't grow our cash rate, we will have an ever-decreasing amount of the global hash rate and produce ever decreasing amounts of Bitcoin. And we think that it's our duty to continue to grow hash rate, not just in the United States but globally to support the security and diversity of the Bitcoin blockchain and the Bitcoin network because we don't want it to be dominated by any small handful of players. And so we believe it's our duty to continue to grow cash rate.

So how do we do that economically, we do it with low-cost power, which we can control. which ties to the MPLX deal, it ties to what we're doing with our wind farm Texas. It ties to what we're doing with flare gas. We have doubled by the end of this year, we'll have doubled our flare gas capacity, and we're going to continue to grow that. The wind farm is fully built out from a data center perspective, and that's running. And we're going to continue to look at opportunities to require more energy at low cost that we can then allocate between Bitcoin mining or AI. You have to kind of think of us as we are going to own lots of our electronics, and we're going to put those electrons to best possibly use.

In regards to Auradine, Auradine's more recent hydro model, which competes very well with the Bitcoin and other vendors models is doing well. We're deploying -- in our fleet. We're not putting exclusively Auradine at this point. There are still different machines, different characteristics that are really good for different environments, and we have a lot of different environments. And so we're continuing to deploy a mix of systems.

But over time, it would be logical to feel that we're going to add more and more Auradine to our fleet. Their systems offer them very unique capabilities, especially around load balancing that in a model such as the one that is beginning to gain steam in Texas, where the utility wants to regulate your curtailment and shut you off and turn you on.

That requires special capabilities in the miners, and that's something that exists in the Auradine systems. And so as more and more utilities start looking for those capabilities amongst miners who are on grid, I think they will continue to gain some market share there.

Other than that, they have spun out some very interesting AI-related businesses, one or escape, which is around securing large language models, which recently had a lot of positive reviews at the RSA show earlier this year and then also scaleup, which is a start-up around ultra-high-speed cluster interconnect switch technology. So that has been a great investment for us, and we continue to look for investments like that where we can acquire or build technologies that can become part of our solutions over time.

And I guess last one for me. You kind of talked about it earlier, but obviously, a lot of market cap, a lot of value created in the Bitcoin mining space amongst the publicly traded guys. I'd argue that you guys haven't received or gotten your share of that. Like what do you think the market is missing. And hopefully, we'll come to appreciate in the near term or medium term.

Yes. I mean I think the key for us is the floor on the valuation of our stock is essentially the value of our Bitcoin Holdings. And people don't put a lot of value on the Bitcoin mining infrastructure or the Bitcoin mining business per se. And I think as our business continues to evolve, especially with the energy generation story and as AI becomes a bigger piece of this and we generate more profit per megawatt hour consumed we'll start getting more attention from people. And I think you'll start seeing people realizing really the benefit of what we're doing in our model, and we'll get more credit for that.

Reggie, just to add to that, the power capacity that we have secured through these transactions that puts us at the forefront. And here's where the actual value flows with Bitcoin mining option value between AI-ready assets, our operational flexibility with integrated power. That's what's going to drive value for our stockholders from a long-term perspective.

Thank you, guys. Congrats in the quarter.

Thank you.

Thanks, Reggie. We appreciate it. Most of the questions that we received from our retail shareholders have been answered. We're obviously running -- short on time, but thanks, everyone, for joining us today. If you have any questions that were not answered during today's call, please feel free to contact our Investor Relations team at [email protected]. Thanks very much, and enjoy the rest of the day.