Ginkgo Bioworks Aktie

I'm Daniel Marshall, Senior Manager of Communications and ownership. I'm joined by Jason Kelly, our Co-Founder and CEO; and Steve Cohen, our CFO. Thanks, as always, for joining us. We're looking forward to updating you on our progress. As a reminder, during the presentation today, we will be making forward-looking statements, which involve risks and uncertainties. Please refer to our filings with the SEC to learn more about these risks and uncertainties, including our most recent 10-K.

Today, in addition to updating you on the quarter results, we're going to make the argument that autonomous labs are an imperative for American Science. We're also going to provide insight into how we are going to scale the capabilities of Nebula, our autonomous lab in Boston, and share an update on how we are getting autonomous labs like enhance the next generation of scientists. As usual, we'll end with the Q&A session, and I'll take questions from analysts, investors and the public can submit those questions to us in advance via X, assenting results or e-mail, [email protected].

All right, over to you, Jason.

Thanks, Daniel. We always start with our mission here, which is to make biology easier to engineer at Ginkgo. In 2026, our goals remain the same. We want to focus and invest to win in this new category of autonomous labs. We want to focus Ginkgo's efforts really on the technology side, largely into autonomous labs, and we're going to invest to extend our lead there.

Second, we want to demonstrate the capabilities of an autonomous lab by using our big system here in Boston, Nebula, which I'll talk about today that we, in the last quarter, expanded that substantially so that we can sort of move the majority of our work on to that system over the course of the year and into the future. That's a great chance to both improve the economics of our services and also demonstrate to other potential buyers of autonomous labs, just what you can do with a system like this. And so I want to talk a bit about that today as well.

And then finally, we want to book new sales of autonomous labs, in biopharma, National Labs and as I mentioned today, research universities, which we're very excited about. We have made a lot of headway, as you know, and we've been talking about for a couple of years now on improving our cash burn. You can see that in the second half of this year, we intend to improve on that burn even further than we did in the first half of the year, and that is really work we've been doing in the first half of the year sort of paying off and bearing fruit.

So really excited. This gives us this plus our $302 million in cash and cash equivalents as well as we have an additional $87 million that we've set aside for restricted cash for various customers and certain operating activities puts us in a really nice spot going into the second half of this year and the future to really have the capital we need to continue this growth into autonomous labs.

So with that, I'm going to pass it over to Steve in order to dig into the financials then you'll hear from me again in the strategic session. Thank you.

Thanks, Jason. Before I walk through our financials, I want to remind everyone that following the previously announced transaction that closed on April 3, the divestiture of biosecurity is classified as discontinued operations within our financial status. Accordingly, we have and we will retrospectively recast all prior periods presented to conform to this presentation.

The formal biosecurity results are now reported as loss from discontinued operations below loss from continuing operations. All of our financial commentary I will provide today relates exclusively to continuing operations where we now operate as a single segment.

With that, I'll now discuss our Q2 results. Revenue was $20 million in the second quarter of 2026, down 48% compared to the second quarter of 2025. For the first 6 months of 2026 revenue was $40 million, down 49% compared to the same period last year. As previously disclosed, revenue in the first 6 months of 2025 included $7.5 million in noncash revenue relating to the mutual termination of the BioMed agreement.

Excluding this, revenue for the first 6 months of 2026 was down approximately 42% from the prior year period. It is important to note that our net loss includes a number of noncash and other nonrecurring items that are detailed more fully in our financial statements. Because of these noncash and other nonrecurring items, we believe adjusted EBITDA is a more indicative measure of our profitability.

A full reconciliation between adjusted EBITDA and GAAP net loss from continuing operations can be found in the appendix. In the second quarter of 2026, R&D expense decreased 4% from $31 million in the second quarter of 2025 to $30 million in the second quarter of 2026. G&A expense decreased 26% from $16 million in the second quarter of 2025 to $12 million in the second quarter of 2022. These decreases were primarily driven by our restructuring efforts, which was substantially concluded at the end of 2025. Net loss from continuing operations was $57 million in the second quarter of 2026 compared to a loss of $53 million in the prior year period.

Moving further down the page, you'll note that adjusted EBITDA in the second quarter of 2026 was negative $36 million compared to negative $25 million in the second quarter of 2025. It is important to note that adjusted EBITDA includes the carrying cost of excess leading space, which you can see was $14 million from the second quarter of 2026, up from $12 million in the prior year period. This cost represents the base rent and other charges relating to lease which we are not occupying net of sublease income.

This is a cash operating cost that is not related to driving revenue right now and can be potentially mitigated through cell leasing. And finally, Cash burn in the second quarter of 2026 was $45 million compared to $38 million in the second quarter of 2025.

For the first 6 months of 2026, cash burn was $93 million down from $96 million in the same period last year, a 3% decrease. As previously reported, we paid Google Cloud $14 million in the first quarter of this year relating to the 2025 amended commitment, which increased our cash burn for the period. Resetting the commitment reduced our future minimum commitments by more than $100 million compared with the original terms and extended the commitment term from 3 to 6 years. Excluding this payment, cash burn reflects a significant decrease in the first half of 2026 compared to the first half of 2025, which was a direct result of the restructuring.

During the second quarter, we raised $17 million through our at-the-market equity program. Consistent with our methodology, these related proceeds are excluded from cash burn for all periods presented. Now turning to guidance. As we discussed earlier this year, 2026 is about continuing to be cost efficient, while investing in our AI robotics and software to bring autonomous labs to our bioscience customers, including the build-out of our Frontier autonomous lab in Boston.

We have turned the page from focusing on restructuring actions to focus this year, not only on cost efficiency, but on investing in what we see as our opportunities while continuing to provide our customers the advanced services they have come to expect. For these reasons, we believe cash burn best reflects our continuing services and tools and further investments in autonomous lots. In terms of outlook for the full year, we are reaffirming our overall cash burn guidance for 2026, totaling $125 million to $150 million.

This range reflects a firm balance amongst cost efficiency, continuing services and tools and further investments we are making.

In conclusion, we are pleased with the continued improvements in cash burn efficiency and our business pursuits for 2026.

And with that, I'll hand it back over to you, Jason.

Thanks, Steve. As I said, Ginkgo's missions to make biology easier to engineer. We going to have 3 strategic topics today to dig in on first. There's been a lot of activity in sort of U.S. science, a new report coming out of Office of Science and Technology Policy, I'm going to touch on. Autonomous labs are becoming a real imperative for the U.S. to stay competitive in science and in particularly in biotechnology versus China. So I'm going to speak to that. Second, Nebula, our large autonomous lab here in Boston, is the largest in the world. It's growing rapidly. I want to showcase what we've been doing with it. And then finally, we are using that lab and all our infrastructure here at Ginkgo to offer up competing services to offshore CROs that are quite economically competitive for customers, and I want to highlight one of those in particular.

All right. So let's dig in on the autonomous labs. There's been a lot of news in the last in particular, an article coming out in Stat magazine that highlighted featured Ginkgo quite heavily about this question within the biotech industry of, should we be offshoring our work to China for the discovery of drugs? And is that a concern in a world where there's sort of increasing geopolitical tensions between the 2 countries.

Ginkgo is featured around how our automation could be a counterweight to lower-cost labor in China. But this is a hot topic. And the reason it is, is highlighted in that Wall Street Journal article, you've seen the number of newly acquired drug assets. In other words, drugs bought from start-up biotech companies, go from almost none coming from Chinese startups about 5 years ago to last year, it was 48% in the first quarter of this year, it was more than 50%. And then that's obviously borne out in our sort of jobs ecosystem and our technology ecosystem. This is a post in Reddit in the biotech Forum. I'm extremely frustrated bench scientists having no luck finding work in 6 months after layoff. I did get an interesting suggestion. One biopharma start-up CEO told me, he hasn't hired for any bench work in the States, outsources at all to China. He said, have you considered working in China. And this person says, "Is that a good idea considering I only speak English. I don't think that's a great idea. I don't think our scientists should be moving to China and search of biotech jobs. I think the U.S. needs to become competitive with China and the way we're going to do that is we're going to automate the laboratory work at the lab bench.

And you're seeing a lot of energy around this. So there's an absolutely great report out of the office and Science and Technology Policy from Director Mike Crazio there, highlighting the new strategy for science in the United States. This is partially under the umbrella of the [indiscernible], which I'll talk about, to bring AI into science, but also highlights NSF's new program to spend $400 million on a national network of cloud laboratories. And if you look in the document, you'll see the section on autonomous experimentation loop autonomous laboratories can collapse discovery time lines by orders of magnitude and enable science at a truly industrial scale, focused investments in robotics and automated laboratories, leveraging industry demand and federal R&D to ensure our scientific equipment industrial base is built on the world's best hardware and software and leads the charge in the coming scientific revolution.

This is awesome. It's really great to see, I call it to action like this out of the OSTP it's exactly what they should be doing. If you see here on this next slide, Ginkgo has been building the first autonomous lab for a national lab here in the U.S. I had the chance to ribbon cut the first 13 of our racks at Pacific Northwest National Lab, with the Secretary of Energy Secretary right in December.

On the right-hand side, you can actually see all the expanded 97 rack system that we'll be building a schematic of that that this is going to be expanded into coming up under the Genesis mission. So really excited to be a part of that. But -- very excited to announce just yesterday that we have been selected to build autonomous labs for MIT, Caltech, Maryland and Northwestern, Caltech Maryland and Northwestern as part of this NSF program and MIT through a separate grant.

This is really exciting because we're getting autonomous labs in the hands of graduate students, people with my sort of training so that they're learning how to do science on top of robotics rather than how I was taught which was sort of slaving away at a lab bench doing experiments by hand. We have to think about the practice of how we do this work alongside the underlying technology of robotics that they've developed together.

So I think this program is super important. I think it's a big part of how the U.S. day is competitive. I'm quite proud we're part of it. So I wanted to, again, I'm going to highlight a few slides I showed last time, but I think it's important point to make. When I say autonomous lab, what do we even mean by that? And I'll draw an analogy to the transportation industry. So on the y-axis of this chart is sort of the amount of automation of a given transportation technology.

And on the x-axis, is the flexibility of a request from a user of that automation that the technology will allow. So low amount of flexibility, high amount of automation top left, that's a subway. That's our red line tea here in Boston. It's totally automated. You sit down. It takes you away but you better want to go to 1 of the stops on the subway. It's not going to pull up in front of your house. low amount of automation, a high amount of flexibility is a car. You put your hands on the wheel, your foot on the pedals and you can go straight to your house with a grocery store whoever you want, it's highly flexible, but you have a human in the loop to manage the variability and that's the transportation system for the last 100 years unless you've been in [indiscernible] which is what we call an autonomous car. You'll notice we don't call it an automated car, because automated, it sounds like automated door or something. It's just doing the same thing over and over again, but an autonomous car magically goes wherever you ask it to go without a human in the loop. And here's the kicker.

If you look at miles traveled in the United States, subways versus cars and trucks, it's 99% cars and trucks because you need the flexibility. It's not like we don't know about railroads and tracks. It's that people need to go where they need to go in their lives. And so that's why this is such a disruptive thing coming with Waymo's is they're going to go after the 99%. It's going to automate the overwhelming majority of the transportation ecosystem, which is what subways never got to.

Here's what it looks like in the lab, low amount of flexibility, high amount of automation. We actually have our subways. They're called work cells. And they're used for things like high throughput screening and pharma companies or for running a diagnostic tests in a clinical lab, where you've got the same experiment being run over and over again. And they're wonderful because they're fully automated, you can walk away, you can run them 24/7, you don't need a person in the middle, but they are not flexible. You cannot read the new experiment in the paper and then have it running on your work sell tomorrow. Low amount of automation, a high amount of flexibility. This is that car, right? It can do whatever you want, but you have to have a human in the loop, this is the lab bench and the manual laboratory, right? And again, much like cars and transportation.

The bench is 95% plus of the $60 billion to $80 billion a year that pharma companies spend on research, not clinical trials, but their research labs and the $40 billion a year like the NIH spends on doing research laboratory work. And so all that money is going towards the benches and almost none of it today is going to robotics because not because we don't know about robots, but because the robotic systems so far have not been flexible enough to do science and to do drug discovery. That's what we're trying to build. I think go. We're trying to make our version of a Waymo, that top right corner, it should have the automation of the work cells, you should be able to walk away and run at 24/7, but the flexibility of the lab bench. A much bigger prize than the work sell price, but a much harder technical challenge. The ROI for an autonomous lab is quite clear. If you compare it to our manual labs at Ginkgo, which we have Plenti, you can see some very obvious differences for starters, you cram the same amount of equipment that you would have spread out around a manual lab with humans moving through it. to about 1/3 of the space.

So it's much smaller. And then additionally, our lab is running 24/7. So Nebula, the autonomous labs running 24 -- if you haven't done the math on a week, work week for a lab technician is like 40 hours, and there's 168 hours in a week. So you're getting a fourfold increase in the hours that, that big sunk cost laboratory is being used. Finally, repeatability, traceability, electronic records are all just right inside of an autonomous lab without even having to work for it. AI-driven science is going to need these things. I think it's going to be hard to connect that into the manual infrastructure. The way that we're going to do that is through robotic labs, it's intrinsic to those systems.

All right. I get asked a lot, and I started this off with a bench scientist worried about a job that our autonomous lab is going to exacerbate the problem of scientists having a hard time getting jobs in the United States. I don't think so. This is our advertisement from IBM back in 1952. I love this ad. It says, hey, here's the IBM mechanical calculator -- it's actually predated the computer. It can do the work of 150 extra engineers. And there, they are these engineers with their slide rules, right? And this was the era before computation had been automated. And you might have said, "Oh, well, this machine over here will, of course, replace these 150 gentlemen with their slide rules and that is not at all what happened. In fact, we had enormous explosion in the number of engineering jobs. And the reason was the actual limiter on the market size for computation was the fact that we were doing it manually.

Once we automated computation, it turned out there was a vastly bigger market for computation than we thought there was. And that what was really valuable what was in those engineers heads their knowledge of practice and computation, their knowledge of the problems you want to solve with computation. And once you can get a much better ROI on that through the automation of computation with computers, that field exploded. That's really what I see as the opportunity for us in biotechnology. We're being limited by our manual labs. Our scientists jobs are limited by the manual labs and manual science jobs, in particular, are being offshore as fast as possible. The way to stop that is with laboratory automation.

Okay. Let's talk about an actual existing autonomous lab that we have here in Boston. I love this video. So Nebula is our -- the name of our autonomous lab here in the Seaport. We have now 105 racks on it, it really is huge and awesome to see in person. If you remember how this works. We have a track system that's moving samples from device to device on the system and then the arms pick up the samples, put it onto that particular device and then that device does whatever particular lab protocol, step in the lab protocol is asked for by the scientists that submitted the job.

One thing I'll highlight is we actually roughly doubled the size of the system. We added 50 new racks. Basically over a 3-week period, right? We had built the racks in advanced manufacturing, but just to put them in connect up all the hardware, do like a cycle of debugging on things that broke on the software when we expanded to be that big. And we had it up and running doing experiments about 3 weeks later. That in the world of like subway work cell automation is just crazy, Building a new automation system with 50 new devices on it and having it up and running over 3 weeks is just not a thing that happens. So I do think we are really benefiting from the fact that we productized through our rack carts, what has until now been a custom process of integrating devices in an autonomous lab. We now, like I said, have 105 racks, this is running day and night.

p'll just point out like an average-ish day would be 30 unique protocols coming from scientists more than 100. If you start -- if you count copies of protocols running across those 100 devices, I don't think there's anything else like this running in the world today where new experiments are submitted by scientists, not automation engineers, but scientists every day onto the system, and the system just handles that variability and manages it. This is that like Waymo phenomenon, like being able to handle the variability at scale is pretty crazy. And it's not like we don't have bugs, we don't have issues to work through, we do. But just even being able to do that is pretty nuts at this point. And it's running 24/7. It's a picture of our scheduler, the colors or different protocols, x-axis is time, why access is all the different racks on the system, and you can see how we have to sort of jigsaw puzzle in different protocols. And so if you submitted a new job to the system, it would check to see, is the device you need available in the times that you needed and could you fit your particular set of protocols into this jigsaw puzzle? If so, you would get to go in. And so a lot of the work we're doing is on improving the schedule and improving robustness of the system and all kinds of really interesting stuff. But it's very much engineering work to continue to drive up the variability that scientists can put on the system as well as increase the total number of catcalls we can run at any given time.

So really exciting engineering work. You should come to take a tour of Nebula. We've got a lot of people come through now, many hundreds of people in the first half of this year. There's lots of really fun videos on Instagram and TikTok and everywhere else. It's a need system to see in person. We do tours 3 days a week. Anyone is welcome to sign up for it. Please do. We really love to have people come by and see it. If you're sort of a pharma company or even an academic scientist or someone who has a particular protocol that you really would get value from automating but you've never automated it before. If we have the same equipment that you use in your manual lab, we're happy to try your protocol in Nebula, would just have 1 of our scientists submit it as their protocol that day, and we would see how well it would work. And so you can kind of do this sort of try before you buy on integrated automation. And that's again not a thing that's that happens with the subways. They're sort of built, you test them with water and then you ship it over and cross your fingers and the customer kind of hopes that what the vendor showed works with clear with water runs, ends up playing out in practice with biological runs once they get it in-house, and it's their job to debug it if not. We're able to bring that sort of debugging work earlier in the process.

So if that's of interest to you as a buyer of automation, we're finding people really like that. Okay. Lastly, so we are using our autonomous lab. One way we do business as you could buy it. But the other way we do business is we run our labs as a service, as a CRO, contract research organization. And increasingly, we are -- we've always done that for sort of like very high-end specialized services that can go most notably, our solutions business. We have these large projects with like Bayer or Novo Nordisk, where we're doing like multiyear research projects using our infrastructure. That's not what I'm going to talk to you about today. I'm going to talk to you today about going straight at the traditional CRO work that pharma companies have been offshoring to scientists in China, companies like Wuxi for over the last 20 to 25 years. Once you have a lab that doesn't have people in it, we really think we can compete on a cost basis very well with those offshore CROs. This is not unique to bio. There's a company we really I really like. It's called Send cut and where you can -- I don't know if anybody has done this, but you can order sort of custom sheet metal fabrication. And this is, again, back to that graph I drew of like throughput and or automation level and variability. This is custom sheet metal fabrication, which means we basically offshore it.

Because it was a labor-intensive custom process to cut this in a particular way that a customer would want them to cut it to -- and so we lost this industry over the last 50 years. It's really exciting to see this coming back via Send cuts end. And that's through a mix of some automation but also through really smart software to turn customer requests into smart geometries and how they're doing it and basically use technology to bring costs back in line with what you would have got by offshoring the old generation of approaches to lower-cost labor overseas. I think this is how the U.S. is going to bring back the world of Adams. Right? Like we should not just be a country that only does information technology and services, we should also be able to build things and in order to do that, we need to rethink the way that we work with Adams. And that's the only way I think you bring Adams back versus lower-cost manual labor. We're coming after that when it comes to these CROs, so these contract research organizations, most notably WuXi has really been sort of the centerpiece of offshoring, starting with chemistry but then increasingly biotech CRO services over the last 20 to 30 years.

We launched a ServiceNow about 6 weeks ago called ADME 1, ADME stands for absorption, distribution, metabolism and excretion -- this is sort of a standard panel of in this case, 5 Tier 1 assays that are run on small molecules, so chemical drug candidates to see how good they are on the sort of -- not drug property specific to your disease, but just these general drug properties about like how your body processes the small molecule. And to give you a sense, you can buy these. These are very standard assays.

You can get them from Western CRO vendors for $2,000 to $5,000 for the panel or from Chinese CRO vendors for $1,000 to $2,500 for the panel. [indiscernible] and get them from Ginkgo data points for $199. And that's not just the assay, we've also partnered up with inducted Bio, intangible Scientific to handle both a PK projection as well as compound management for your small molecules. So you're getting sort of the whole kit and caboodle here for close to 10 price. We've done a lot of work to validate these assays, I'll just flip through a few slides, but you can also go see check this out on our website, both internal as well as very importantly, we've compared to external vendors.

So we had the same sample go get tested by this [indiscernible] panel at external vendors and compared it to what we were seeing with our robotic automated approaches to doing [indiscernible] We've seen really great results. I'll just flip through a few of these on kinetic solubility. On the left, you can see that how we rank so like perimen coefficient, like how well do we put our the molecules in the same order, our industry peer lid on this particular assay and then as well as this bidding low, medium, high, and we have good agreement there for Kinetic solubility, also for permeability, again, same set of assays for microsomal stability in human microsomes same set of assays P450 inhibition and plasma protein binding. And we do have done this also for a very popular small molecule library called Lowpack 320 different compounds. We went ahead and tested all those across 3 of our Tier 1 assays and put that data set up on the web.

So you can download that and then you can use that to compare to the literature. So this is up online. It means other people have been able to go download and check it out. There's a company called Inflex AI that did a bunch of work with this data set and they published the platform is technically clean and talked about our replicates in assay controls and so on.

So we really encourage folks to check it out themselves. We think we stand up very well to Wuxi in terms of technical capability and throughput and we kick their butt on price. So I don't know why you can use us. What's coming soon. And this is another thing which she does well, which is chemical synthesis, so being able to build the molecules in addition to test the molecules, [indiscernible] is about testing. So we bring online plate-based chemistry. We already actually do a lot of chemical purification historically, I can go because all our work in natural products, really just bringing that into an automated environment. And then finally, we want to have in atmospheres, in other words, like anaerobic chambers to do chemistry in here, we're fortunate because of the first system we delivered to Pacific Northwest National Lab with our racks in it that I mentioned earlier with the sector of energy, that was actually an anaerobic system. And so we've already had a lot of experience getting our robots into an anaerobic environment. And so we're going to be doing that with pointing it towards doing chemistry. And so if you wanted to sort of data test that with us, give me a call if you're interested in sort of the chemistry half of things.

This is a natural complement to the biological assays we've developed a Ginkgo over the years. A lot of times in drug discovery, you're either making a chemical or you're making a protein drug, but depending on the disease you're going into, they're both funneling into similar set of biological assays about either that disease area or what it might or whatnot. And we already have a lot of those assays running at high throughput on our automation adding chemistry is a really natural match for us, and it's a bigger fraction of the CRO business today in China.

If you want to learn more about [indiscernible] you can go to data points to can go by, there's a banner at the top and you can check out our [indiscernible] service.

Okay. I want to add, just as a reminder, you can buy an autonomous lab from us. So if you really like this or even like the types of assays we're doing, many customers might want to run their ADME internally, right? Maybe you want to build a service, whatever it might be. We're happy to sell an autonomous lab to anyone that wants to use it to offer whatever types of products and services they want to develop. Or if you want to get experience trying 1 out, please try our lab services and do consider reshoring your work if you're concerned about this offshoring trend that we want to keep adding more and more of the services you're currently getting from offshore CROs to our offerings in data points and Ginkgo Cloud Lab.

Okay. Let's go to the world we want to see my e-mail is up there. Always happy to get e-mails for folks if you have more questions, and happy to do Q&A.

Thanks, Jason. As usual, I'll start with the question from the public and remind the analysts on the line that if you'd like to ask a question, please raise your hands on zoom, and I'll call on you and open up your line. Thanks, everyone. All right. Let's get started. So just a reminder, I'm going to start with some questions that were sent in beforehand. [Operator Instructions] So we're going to start with 2 questions from Brendan from TD. The first question is, what can you confirm in terms of revenues for the Rac/autonomouslab segment and the AI data points? How should we think about order funnel, backlog, revenue recognition for both moving forward? Jason, I think you might be muted by accident.

So as a reminder, we're not doing revenue guidance this year, so forward-looking, we don't have. We also aren't currently breaking out revenue we're bringing in to date. We do have pretty different rev rec for automation versus data points and our other services as well. So Steve, you look for sharing a little bit on just how we approach that.

Sure. Give a little insight. So from the large government deal, we did have a preliminary contract with them. And from that standpoint, there's some small amounts of revenue. But the larger deal that everyone's talking about is that revenue will come about when we deliver and complete the install. And right now, we're really in the planning coordination phase with that. So that will be at a point in time -- with regards to data point, standpoint is very much like the solutions business where we recognize revenue over time. Reminder, smaller projects than we've seen in the past, good growth level. We're very, very happy with what we're seeing from growth in that, but it's spread out over multiple quarters from that standpoint. So a reminder, most of those projects take anywhere from 3 to 9 months, maybe it's a little bit longer against smaller deals compared to what we used to, but it will spread out. And so some of that's reflected in the numbers for Q2 for sure.

Yes. So I buy that back. the revenue on the data points business looks similar to what you would have seen before. But all these automation deals, including like the new academic deals we just signed with these 4 universities, those really are -- for the hardware part of it, it's recognition on delivery. I will point out, we also have like an ongoing services and SaaS revenue for those. So once they're deployed, that would come in more regularly. But you have to wait for deployment for that to show up and you have to wait for the deployment for the revenue rec to show up even if we get cash earlier.

Exactly.

So Brennan's second question was, how should we think about the cadence of revenues to be recognized as part of the MSL project at [indiscernible], which is [indiscernible]

Yes. That's a big national ad project, Steve is just mentioning. So I think we covered that.

Let's move on to X. So our first question is from Vision. And this question is for data points and Cloud Lab solutions, what does the customer repeat order rate and what is the average follow-on order value as a percentage of the initial order value.

Yes. So we're not -- again, we're not breaking it out in that much detail. But what I will say is the way we typically end up having these deals happen is we'll get an initial proof-of-concept deal and then a much larger expanded deal if people are happy with it and then some amount of regular recurring work. It isn't as much probably like 2 categories like the ADME work that I'm really excited about these new -- I think what do we say, 16 customers, a lot in the first 6 weeks is very exciting. And in like new -- some of her new logos for Ginkgo, which is great. But ADME is something that pharma companies that are sort of just ordering off a conveyor belt a little bit as they're developing new molecules all the time. And that's why it's been sort of like a foundation of part of [indiscernible] CRO business. The work we're doing on data points where we say like generating data for an AI model, that might come in like campaigns, where we're making a whole bunch of data -- we do a proof of concept. We do some amount of data gen. Maybe the customer says, "Hey, I actually want more data, for further model training, we generate more. And then maybe they are like, okay, in the next model, I want to trade on something else or it gets into some sort of pattern where they're actually using it a little closer like ADME where they're designing constructs on the regular and they want more and more data of that sort.

But it could be a little more campaign if it's for an AI project versus some of these traditional CRO services, which are like on and on and on. So I am pretty excited to get into I like both those areas. I mean the AI stuff is really taking off recently in general, but I'm also pretty excited to go after the traditional CRO because it's just a reliable source of demand. But we got to prove ourselves were new in that area, but I do like our odds there. It looks real good.

All right. So we have 2 questions. There's another question is also about like revenue recognition from x, but I wonder if you can kind of bundle that with another question that we got, which is about the announcement that we made today about the NSF announcements, where 4 new autonomous labs are going to be built at universities across the country. So I'll sort of ask both of these and 1 question. How did the recent autonomous -- sorry, the recent announcement regarding autonomous labs and universities across the U.S. to impact your outlook for other new academic labs? Is this just a product of the NSF investment? Or do you see this becoming more of a trend across the world? And how will revenue work with all that stuff, too.

Yes, I get pick to the sort of demand and then Steve, you want to chat on the rev rec. So the so what I'm excited about on these is, I think this is the beginning of showcasing that the academic research infrastructure, which, by the way, NIH alone spends $40 billion a year out to our academic medical and academic research institute in doing biological research. NSF spends on top of that, [indiscernible] on top of that.

So there's actually a good amount of money that flows through this community. It's sort of a attempt at a paradigm shift for that group that at least some chunk of that work. And what's pretty interesting is we have like really great partners in this. And so if you look at the group at Caltech, they're focusing on a cloud lab autonomous lab that does like basically chemical structure data generation from chemicals or originating in the natural world. If you look at the group at Northwestern, it's protein engineering. And if you look at the group at MIT, it's for education uses like training people on these things.

So really like -- it's pretty cool to see -- in Maryland, it's biomanufacturing. So those are like 4 disparate areas of biology research, but they're all running on the same underlying autonomous lab platform underneath. That's what I'm most excited to demonstrate is what we've been saying all along is this is an alternative to the lab bench. And across all those different labs doing very different things at academic research universities. They've all got lab benches. They often have 60% or 70%, the same equipment and then maybe 30% or 40%, that's a little bit specialized in their area, but it's not an infinite list of equipment. And the proposal is there should be a giant automation, autonomus LabCorp in every biology department, and you could kind of close most of the labs down. And that would be much less expensive. You'd have way more output from the graduate students. It would feel a little more like buying time on a data center. And I think that's a I don't know. We'll see. And so I think depending on how this first batch of NSF labs go, I think you will see a good amount of homo among other research institutes that don't have these if it goes well. And then that should, I think, lead to just immediate demand or new grants, which you heard from Director Cratis, OSDP, there's a push in this area.

But even without directed funding to buy them, Remember, the universities they have these overhead, they're spending to maintain all these labs. So you could also say, well, hey, listen, if I can offset a bunch of my lab spending by adopting an autonomous lab, there may be money within the university for that or donors that want to see it go in this direction. There's a lot of ways for universities to get money for, I think, interesting projects like this.

So I'm actually kind of bullish that it won't just be associated with new grants for robots, but I also think there will be new grants for robots. Maybe last but not least, I do think it also trains a set of -- you're sort of also starting to train the next generation of scientists with this approach to doing science, which I think is particularly important.

So really excited about this program. I think it would be great for us. Steve, do you comment has more to say on the [indiscernible]

Yes. So bridging up what we just spoke about a few minutes ago about revenue and the like, I should clarify. Our legacy has been services where we get paid to work over time, and that's still true, as we mentioned, with data points. Now with regards to sales, the big block is when we deliver the equipment, install. But that also comes with services, and I'm not going to get into the details of these contracts or the others. But we do get paid services will there be custom work. We absolutely support services after the install and for which we have a long tail of revenue coming from that.

So we look at it -- you have to think about that business model as equipment and support. And support could come in the front end. This would definitely come on the back end on maintenance support and access and the like. So that's sort of the model, but not getting into specifics there's a twist on different contracts for what piece is what? But that's what you should think about equipment delivery, that's when we recognize the bulk of revenue, might be services upfront, absolutely services after the [indiscernible]

And that's inclusive of software licensing as well on the back end.

All right. I think that's all we got. Just a reminder to everyone, you don't have to wait for earnings to ask us questions. You can send us e-mails at investors.ginkgobioworks.com and respond. I hope everyone is having a great evening, and we'll see you next quarter.

Good evening. I'm Daniel Marshall, Senior Manager of Communications and ownership at Ginkgo. I'm joined by Jason Kelly, our Co-Founder and CEO; and Steve Coen, our CFO. Thanks, as always, for joining us. We're looking forward to updating you on our progress.

As a reminder, during the presentation today, we will be making forward-looking statements, which involve risks and uncertainties. Please refer to our filings with the SEC to learn more about these risks and uncertainties, including our most recent 10-K. Today, in addition to updating you on the quarter results, we're going to provide insight into how and why we see autonomous labs like Nebula, our autonomous lab, replacing the laminate, which is where nearly all of biological science is done today. As usual, we'll end with the Q&A session, and I'll take questions from analysts, investors and the public. You can submit those questions to us in advance via X. #GinkgoResults or e-mail, investors at ginkgobioworks.com. All right. Over to you, Jason.

Thanks, Daniel. We always start with this. Ginkgo's mission is to make biology easier to engineer. And I mentioned this at the last earnings call, but in 2026, our focus will be on investing to win the category of autonomous labs. And I'm really excited, even since we just spoke a few months ago, this category has really been growing in attention, new companies in Silicon Valley pursuing this, a lot of interest from the AI Frontier labs about the application of AI models in science via autonomous labs. Government talking more about this. So I do think we're on to the right track with this focus for the company.

The 2 big ways I'm going to be pursuing that goal in 2026, the first is to take our services in solutions, in data points and cloud lab and run the top of our autonomous lab here in Boston that we call Nebula. That's a chance to prove out the capabilities of our system with real-world activities.

And then the second big area of activity will be getting early adopters of autonomous labs out in the world to buy our systems like we've done already with Pacific Northwest National Labs that I talked about last time. So excited to pursue both of those, and you're going to hear more about it from industry strategic section.

We also -- in the last quarter, we were able to close on a deal I talked about extensively last time, which is the spin-off of our biosecurity unit into a new company called Perimeter. I want to say congratulations to the team at biosecurity at Ginkgo and pulling that off. and a lot of great new investors coming into that focus really firmly in the area of defense tech and building sort of a biosecurity prime. Ginkgo as a shareholder in that company. We're super excited to see it succeed. And I think this is really nice, as I talked about last time, opportunity, both for Ginkgo to keep our focus on the autonomous labs and for the team at perimeter to grow under their own brand with a new set of defense tech-focused investors.

Our focus over the last couple of years was very much on getting these numbers where they are today, bringing down our cash burn in the company. We guided towards this, and Steve will touch on that in his section. But again, happy to have a very strong cash position, $373 million with no bank debt as of Q1 2026. And so you'll hear a little bit more from Steve on this. But this sets us up very nicely. We're well capitalized to pursue this area of autonomous labs. We have these based service businesses to build on top of and the lead in developing the technology and you put all that together. And I think we're by far the best bet in this sector.

All right. I'm going to pass it on to Steve to dig into the financials.

Thanks, Jason. Before I walk through our financials, I want to take a moment to frame an important change in how we are presenting our results beginning in Q1 2026. As we announced in February, we entered into a definitive agreement to sell our biosecurity business, which was previously reported as a separate segment. Further, as Jason noted, we closed that transaction on April 3.

The biosecurity transferred assets met the criteria under U.S. accounting to be classified as held for sale and the financial results reported as discontinued operations as of March 31, 2026. This is the first quarter in which biosecurity is reflected as discontinued operations within our financial statements.

And to close with the accounting rules, we have and will retrospectively recast all prior periods presented to conform to this presentation. That means the revenue, operating expenses and cash flows previously attributed to the biosecurity business are removed from each line item of our continuing operations and cash flows as the prior period information is presented, including for Q1 of last year.

The formal biosecurity results are now reported as a single net line loss from discontinued operations, below loss from continuing operations. To be clear, all of the financial commentary I will provide today relates exclusively to continuing operations. We will not be discussing the biosecurity business further in our prepared remarks. On April 7, 2026, for your information, we filed a current report on Form 8-K that includes pro forma financial information for fiscal year's 2023, 2024 and 2025 on a continuing operations basis. Following the biosecurity divestiture, we now operate as a single segment.

So with that, I'll now discuss our Q1 results. Revenue was $19 million in the first quarter of 2026, down 49% compared to the first quarter of 2025. As previously disclosed, revenue in the first quarter of included $7.5 million in noncash revenue relating to the mutual termination of the biomed agreement. Excluding this, revenue in the first quarter of 2026 was down 37% from the prior year period. It is important to note that our net loss includes a number of noncash and other nonrecurring items as detailed more fully in our financial statements. Because of these noncash and other nonrecurring items, we believe adjusted EBITDA is a more indicative measure of our profitability. The full reconciliation between adjusted EBITDA and GAAP net loss from continuing operations can be found in the appendix.

In the first quarter R&D expense decreased 38% from $49 million in the first quarter of 2025 to $30 million in the first quarter of 2026. G&A expense decreased 35% from $20 million in the first quarter of 2025 to $13 million in the first quarter of 2026. These decreases were all driven by our restructuring efforts. Net loss from continuing operations was $76 million in the first quarter of 2026 compared to a loss of $83 million in the prior year period. The reduction in loss year-over-year was due to our restructuring is.

Moving further down the page, you'll note that adjusted EBITDA in the first quarter of 2026 was negative $42 million, which was down from negative $44 million in the first quarter of 2025. Since we are now only operating in a single segment, we all present a single measure of adjusted EBITDA, and it is important to note that adjusted EBITDA includes the carrying cost of excess lease space, which you can see was $16 million in the first quarter of 2026.

Previously, this cost would not have been included in the formal presentation of segment adjusted EBITDA. This cost represents the base rent and other charges relating to lease space which we are not occupying net of sublease income. This is a cash operating cost that is not related to driving revenue right now and can be potentially mitigated through subleasing.

And finally, cash burn in the first quarter of 2026 was $48 million, down from $58 million in the first quarter of 2025, a 17% decrease. As previously reported, in October 2025, we amended and reset the annual commitments with Google Cloud for $14 million. Resetting the commitment reduced our future minimum commitments by more than $100 million compared with the original terms and extended the commitment term from 3 to 6 -- years. We paid this $14 million in Q1 of 2026, which is reflected in our cash burn for the quarter. Excluding the payment, to Google Cloud, cash burn reflects a significant decrease in the fourth quarter of 2026 compared to the first quarter of 2025, which was a direct result of the restructuring.

Now turning to guidance. As we discussed in February, 2026 is about continuing to be cost efficient, while investing in our AI robotics and software to bring autonomous labs to our bioscience customers. including the build-out of our Frontier Autonomous Lab in Boston. We have turned the page on our pure focus on restructuring actions to focus this year not only on cost efficiency, but on investing in what we see as our opportunities while continuing to provide our customers the advanced services that they have come to expand. For these reasons, we believe cash burn best reflects our continuing services and tools and further investments in autonomous labs.

In terms of outlook for the full year, we are reaffirming our overall cash burn guidance for 2026, totaling $125 million to $150 million. This range reflects a firm balance amongst cost efficiency, continuing services and tools and further investments we are making.

In conclusion, we are pleased with our continued improvements in cash and efficiency and our business pursuits for 2026.

And with that, I'll hand it back over to you, Jason.

Thanks, Steve. So I'm going to dive in on the strategic section. I'm excited to go into this today. Our mission is to make biology easier to engineer. And the way we're really aiming to solve that problem, we believe the bottleneck fundamentally is the laboratory work associated with bioengineering. And so I'm going to dig deep today and talk about why autonomous labs will be replacing the lab bench. I want to highlight some of what we're doing with Nebula, our system because we have some news this month in terms of expanding that system. And then finally, the services that we put on top of Nebula are cloud lab data points and solutions. These are sort of like, we call it, like our Starlink, right? If you think about SpaceX, 70% of the launches last year were actually Starlink, their own internal product, in the coming year, the ability for us to scale up on autonomous lab and showcase that you can make money on services without having people in the middle of the lab doing those laboratory services, I think, is a real highlight and will help drive sales of our systems into the world. So I'm going to talk about all 3, and let's dive in.

Okay. So I gave this analogy. I'm going to do it again because I think this is for new folks listening on the call, it's worth understanding what we say when we say autonomous lab as distinct from traditional lab automation. So I'm going to give an analogy from the transportation industry. on the y-axis here. We have the amount of automation for a certain type of transport. And on the x-axis, the request flexibility. In other words, the users asking the transportation system to do something different or not. And so for a low request flexibility and a high level of automation, that's your subway, right? It's the red line here in Boston, you sit down in the subway, and it takes you away. You don't have to do anything, this high level of automation, totally automated transport, but it is very inflexible. You have to want to go to one of the stops on the red line.

Low amount of automation, high amount of flexibility. That's a car, right? You get your hands on the wheel, put on the pedals and it'll take a ride, take you ride to your house or to the grocery store anywhere you want to go. And that's roughly what the transportation system has looked like for the last 100 years.

Let's go the next slide. You've been to California in the last 4 or 5 years or now L.A. or Austin or soon in Boston and you sat in the back seat of a Limo, and it is amazed. I think it is like sitting on a subway, you don't have to do anything, but it will take you right to your house. It has the flexibility of a car. And so it's those 2 things together that sort of flexibility plus a high level of automation that mean we actually give it a new name, we don't call it an automated car, we call it an autonomous car. Because up until now, you've needed a human being with our brain in the loop in order to manage that amount of flexibility into the system.

All right. So if you look at the next slide, you'll see the mile traveled buy cars and trucks versus subways and trains in the United States, it's more than 99% in cars and trucks. And that's not because we don't know about subways. It's that we need that flexibility to do our day-to-day lives, that it's required for the transportation that humans need.

All right. So now let's go into the lab bench and into the lab. So low amount of flexibility, high amount of automation. We have actually automation in the lab. It's called a work cell. That's a 3D schematic of a work cell we have here at Ginkgo. Companies like high resin, thermo and biosera make these. And they're like a subway, right? They're great. They're fully automated. They'll do an experiment without a scientist in the loop, but it better be the experiment that you ordered yesterday. It's not going to do a new experiment for you today. low amount of automation, high amount of flexibility.

We have those 2 in the lab. It's called the lab bench. And you, as a scientist are basically the human glue, connecting all of these different devices in the lab together to do whatever protocol you want to do. And so again, here's the kicker. If you look at research budgets between laboratory work cells, which are used in things in pharma companies like hydro screening or combinatorial chemistry or things like that versus at the lab bench it's about 95% plus at the bench. And again, it's not like we don't know about work cells. It's that scientists need flexibility to explore all the different hypotheses they have for discovering a new drug or developing a new crop trait or whatever type of biotechnology they're doing. So this is what we're trying to build at Ginkgo. We're trying to get up to that top right corner and make a Waymo. We're trying to make an autonomous lab that has the flexibility of the bench of scientists can order whatever experiment they want but the automation of a work cell.

In other words, they don't have to be there to do each and every step and move the samples among the different equipment and program the equipment. They can just hit go and have that protocol run end-to-end for them via the automation, but whatever they want to order that day. That's the target.

And look, if you go to the next slide, the value prop here, I think, is very clear for getting rid of the lab bench, massive overhead cost savings. We've heard a lot about overhead costs at academic research labs and things like that. That's really paying for ultimately, millions of square feet of laboratory space, it's 50 million square feet of laboratory space just in the Boston area. So you can dramatically reduce that. You can increase the research productivity of your human scientists as we need more data from AI, and we'll talk about that in a minute. And then we can enable AI scientists to run these lab in the loop experiments. We just announced a project with OpenAI a few months ago, where GPT5 ran our lab. That's that kind of lab in the loop. Experiments that we're seeing increasingly also in the biopharma industry.

But to put a point on it, like a typical large pharma, biopharma biotech spends $1 billion to $3 billion a year on research, not clinical trials, but spending on 1 million-plus square feet of lab benches. And if you look at their spending within that on automation, it's well below $100 million, usually much less. And frankly, I think those numbers should flip. I think really, the majority of the capital should be going towards automated laboratory work rather than manual benches. And the reason for that is, I think, a relatively straightforward calculation.

On the next slide, you can see comparison between a traditional manual lab and an autonomous lab. It's about a threefold space improvement. When you take all these -- this equipment that's often very spread out in a normal human-operated lab because people need to get around it and safety reasons and all these things. But in an autonomous lab, you can jam all that equipment right next to each other about as tight as you can make it for the arms to work and things like that. So it's about a threefold space reduction. And then -- the manual labs really are just run 40 hours a week, right? I mean, it's when people come in the lab and humans are there, and they got to be there, and that's when you can get people to work. There's almost never multiple shifts in these types of sort of high-end research labs. It's really a 40-hour work week. And our system here in Nebula is running 168 hours a week. So 24.7, and that's a fourfold in sort of hours available for utilization of your laboratory.

So I think a real clear, threefold, fourfold on 2 different axis. It is a clear value driver if you go to the next slide. I think there's little question the ROI is there. I think the big question with autonomous labs is a technical one. It's how do you get that high level of automation and the high level of flexibility, that top right corner without a human being in the loop, right? It's the same question of the way, how do you get it to navigate all these different environments and different roads without a human in the loop, if you can do it, it's an obvious win. If we can do this in the lab, it's an obvious win.

All right. So let me talk through a little bit the design constraints that we focused on at Ginkgo. On the next slide, you can see a work cell, one of those subways, the way it's designed is it's designed against a particular protocol. So if you're a biopharma company and you want to build a high-throughput screening work cell and you co-call a traditional automation vendor, they're going to -- first question they're going to ask you is, tell me about your protocol and tell me what throughput you need to run out. How many samples do you want to get through every week?

Because they're building you a subway line. It's going to be built to do that protocol for you. But if you're a new facility head who is opening a lab in Central Square -- in Central Square here in Cambridge, and you are building it for a scientific lead. You don't ask that scientific lead hey, what's the protocol you're going to run in this lab that 30 of your scientists are going to use to do work, you say, what kind of science you're doing. And very specifically, what equipment do you want me to install in that lab so that your scientists can be productive over the next 3 to 5 years as they use the lab? And so it is oriented around the equipment rather than the protocol. And that's a sort of a subtle point, but it has a huge amount of consequences when it comes to how you design the hardware and software that responds to this challenge. And so if you go to the next slide, you can see our hardware solution here is what we call our rack carts, our reconfigurable automation cards. And we have -- it's basically a robot wrapped around each laboratory device. So we have control over that environment. There's a HEPA filter on the top, which is important for a lot of biological work where we have opportunities for contamination and things like that.

We have a 6-axis robotic arm and a piece of magnetic motion track, allows you to go to the next slide, LEGO block these together into ultimately very large setups. And we're at 50-plus right now in the lab and -- it's growing quickly. I'll show you some photos in a second. We have 13 racks will be coming online just in about a week, all in one big setup here in Boston.

And so if you go to the next slide, we can -- I want to show this is actually a video of the OpenAI protocol. So we had the project with OpenAI, where GPT5 controlled the lab, and we made a video of one of the samples just moving through the various racks that protocol. And so this just gives you a sense of like how does it work, right? So you have these tracks -- and we're able to move like our one sort of constraint on the system is that we pass things in what's called SBS format. So that little rectangle you saw there is like a 3 x 5-inch square. Rectangle and it can contain 96 or this is a [ 384-well ] plate or 1586 well plate. It can also carry consumables like tips and other things. And the arm picks up that piece of plastic wear or samples or tips or whatever it might be, and then puts it on to a particular device. So in this case, it's going -- just went through an acoustic liquid handler it's going on to a bravo, liquid handler.

You saw, again, that first thing that got put down there was actually the plastic tips that are now getting picked up by the liquid handler. And the other 2 plates or sort of sample and destination plate. So we're -- in this case, for the OpenAI project, we're picking up some synthetic DNA, and we're putting it into reaction mixes that were designed by the model at the time, right? And so again, a key feature here is any device that accepts those SBS plates, we can integrate into the racks. It takes us usually like 1 month 1.5 months if it's a new device. We've now got 80-plus devices on there. We're adding new devices all the time. If a customer asks us for one, if we want to add one to Nebula, we just bring it online.

So now that plate is going to get chicken up and put on to ultimately thermocycler or the final analytical device to run the QBC reactions and give a readout on the performance of each one of those samples back to, in this case, an AI model. That data at most of the runs on Nebulas going back to a human scientist, as opposed to an AI scientist. But we expect there will be a mix of both as science goes forward, we'll have both scientists and their agents ordering experiments on autonomous labs.

Go to the next slide. Great thing about this system is it can expand. We started Nebula, I think was about 8 racks, doing like next-generation sequencing prep for our samples here at Ginkgo and expanded ultimately now up to over 100 racks on the system.

So let's dig in a little bit. I want to go to the next section. So that was sort of the theory, like how we design the hardware in order to solve some of the challenges of the autonomous lab and what autonomy means. And now I want to dig in a little bit on Nebula specifically, because I think what's really unique about Ginkgo is we're not just a hardware company. We actually run BSL-2 labs here in Boston and do scientific partnerships with some of the largest biotech, ag biotech, industrial biotech companies in the world. And so we can actually show what it looks like to do real science on a system like this.

And so if you go to the next slide, one of the things I'm quite proud of that we've been able to show in the last quarter is over 100 protocols with more than 30 of them being unique, submitted by scientists, and I'll mention this, but these are not being submitted by automation engineers or experts in robotics. Being submitted onto our system, Nebula here in Boston, which has 50-plus lab devices all integrated together where you can send point-to-point samples from any device to any other device as requested by those scientists.

There is nothing else like Nebula in the world today, doing sort of open-ended scientists science like this at this scale with this number of unique protocols and end users. And it's proof that autonomous labs are feasible. I mean there's work to do and talk about that. Things break as we are scaling this up. That's for sure but it is evidence, in my view, that this is going to land, like we are going to be the manually operated lab.

And so if you go to the next slide, I want to walk through a few of the key things that you got to show if you're going to take out one of those laboratory floors at Takeda or Merck or Novartis or whatever or bear crop science or any of these companies that do a lot of laboratory work.

So first, you want to connect 100-plus devices in a single automation setup, all right? So it can't just be 5 or 10. A scientist expects to have access to many different devices in order to do whatever protocol they might read about in a scientific paper this week. And then I think about 100 is the right number. So we've been able to -- this week, we'll find how we're turning it on in about 5 days. 105 all our 103 racks, all in one big setup. And the reason we can do that is because of that rack productized hardware, that cart I showed you.

We just rolled in, they came in off a truck and we rolled another 50 in and those have all gone in the actual install over the last 3 or 4 weeks. So it's pretty fast to put that many new devices on an automated setup. We'll see if it works.

Second, we have run 10 now like a 30-plus unique protocols, 100-plus different 100-plus total protocols. But that's kind of -- you got to get in that, I think, 50 to 100 to maybe 200 unique protocols, all running on the autonomous lab at the same time. And we do that with our catalyst. This is our software, our scheduler that we built is a very complicated scheduling problem. It's really easy to mess this up. Biology is very sensitive to timing. Things break all the time as we keep driving the scale up here. So we're getting to do that quick cycle of debugging and improving the system, but that scheduler is really the key piece of software driving that.

And then finally, scientists, scientists, not automation engineers, and I think on a peak day on Nebula, we had 439 or so scientists submitting. So that's really exciting. I like to have that many different scientists submitting protocols on one automation system. Again, I don't know of any automation system in the world that's been able to do that before. And we're able to do that in part by leveraging AI coding tools with custom harnesses wrapped around them that basically understand how to transfer the scientist's intent in human language into code to operate the autonomous lab. And that is a big unlock. We're very thankful for what's going on with all the coding agents. That's a real help for improving the ease of use because at the end of the day, to make robots do something, you have to program them. And to walk up to a lab bench and do your work by hand, you don't.

So we have to solve this problem. We can't make it so that scientists have to become coders to do their job, and we've really just been giving a gift by these AI coding tools, again, like the codecs and the cloud codes and things like that, that can sit inside other tools that are specific to the automation to get this done.

So those are the 3 big ones, and I'm pretty happy with the progress at all. So if you go to the next slide, as I mentioned several times now, we're going from 50 to 105 racks by the end of this month. It's going to be awesome. It's a really cool system to see people should come out and visit it. If you go to the next slide, that scheduler is not trivial. So this is an example of our schedule, I think, running 17 or 20 different protocols at the same time. each color is a different protocol. Each row is a different device on the system. The x-axis is time. And so you can imagine if you want to add a new protocol to that, and you're like, okay, I need to use the device on ROI, the device on ROV and the device on row 9 and I need the device on row 3 for the first 3 minutes that I'm not willing to tolerate up to 1 hour gap then the second device for 15 minutes, then up to a 30-minute gap than the last device.

It will check, can it fit you in. And if it can fit you in, or if it can fit you in by moving a couple of other things that doesn't disrupt them in a way that breaks the protocol, it will fit you in. That's awesome, right? That's very much not how the traditional lab automation, the subways work. They're running a batch, that subway line is showing up at a certain time. You can't just jump and insert yourself in the middle, but you can with our scheduling software here.

On the next slide, the little green one, it's hard to see, but that column, third column over from the left over there has the names of all the different scientists submitting. So I'd really love this. I love that we're seeing different people submitting different orders for protocols every day. It's really exciting. And again, I think it's unique. We're also seeing a lot of energy on the U.S. government side. If you have the next slide, a lot of new policy action here, there's the genesis mission, which we're fortunate to be a part of from the White House to bring AI into the national labs. But there's a big motion right now where we're seeing an increasing amount of drug discovery work moving to China from candlesware, I was talking about earlier here in the Boston area. And that's because Chinese scientists are paid 1/3 as much and they're doing equal work to what's happening here in the U.S., like they're just as good. They're just as smart.

And so I think if we want to remain competitive, we got to think about doing our research in a fundamentally different way in the United States. I don't think we can just rest on our laurels of having the only smart scientists in the world in this area or at least versus China. And I think that era is over firmly over at this point. And so we've got to think about a new way to do it. I'm pretty heartened to see activity out of the National Science Foundation is funding $100 million for a network of cloud laboratories and autonomous labs. There's a new bill introduced by Senator Young to sort of do more of this cloud labs and autonomous labs. So hopefully, we see more here, but I'm encouraged by what we see already.

If you go to the next slide. We're obviously very fortunate. I had a chance in December to sort of ribbon cut the first our RAC robots going into Pacific Northwest National Labs and signed a new contract for $47 million, much larger autonomous lab set up nearly 100 RACs going in a new building in a couple of years at P&L. So this is really exciting, and I think sort of highlights the direction I believe our national labs will go our scientific research in the country.

If you go to the next slide, we were lucky to give ARPA H, a tour of Nebula. We have a great project with them. And the work is accelerated by having these autonomous labs available to our scientist that can go. I think this is something that makes a lot of sense for a lot of labs at the National Institute of Health, for example, or NSF-funded labs or academic research universities. They would all be accelerated if our scientific talent could get many more of their hypothesis tested than are today due to the limitation of the manual lab.

Next slide. listen, Nebula is showcasing what is possible, and that means that early adopters are getting excited about it. So we are also building autonomous labs for that left end of the chart here, the very earliest adopters, the people that are excited to try this out as a different way as an alternative to their lab benches. And so we'll keep leaning in there, building those systems as that demand comes in.

And we are seeing, if you go to the next slide, a lot of interest. So we've had 600-plus visitors in the first quarter. I'll show it at the end, we have a great -- like a little sign up. You can sign up. We do tours weekly, if any of you want to sign up or listening in, we're very happy to give you a tour. So okay, so that's Nebula and that's the dive on that, right?

Now I want to talk a little bit about our service businesses, Cloud Labs, data points and solutions, which I think of a little bit, like I said, like our Starlink, right? So last year, 70% of the launches at SpaceX were Starlink, go the next slide. That's a huge advantage for SpaceX. That means they get to be creating an asset, a moneymaking asset in the form of Starlink while also getting to test over and over again, their launch platform.

And their launch platform, ultimately, I think in their view, is the big product, right, that they can have that sort of transportation layer to space. But Today, they're 70% of the demand for that platform, right? I see a similar situation with the autonomous lab. We are able to have a big system here in Boston and basically prove out moving over our work from data points, Giga Cloud Lab solutions, even our reagents business onto that platform.

And if you go to the next slide, really excited. We got our cloud lab off the ground just in the last quarter, it's really been exciting. This is from the times in London. Do you want to run an experiment for $39 -- there's a lot to do it for you. Go check out cloud.ginkgo.bio. You can go in the estimate tab at the top type in whatever protocol you're interested in. It will look up and see do we have the equipment needed to do your protocol? And if so, it will make an estimate of what the price would be to run that protocol in a cloud lab.

And people are, I think, pretty surprised at how inexpensive it can be. And that is a reflection of where all the costs lie in doing lab work, which is in. Manual lab work done, 40 hours a week done at low equipment density, low equipment utilization in laboratories that cost a fortune to run. That then flows through, and it means all of the CRO services you order and so on are very expensive. We think we can solve that problem through automation and the cloud.ginkgo.bio or cloud lab service is really a great way to do that.

If you go to the next slide, this is what OpenAI took advantage of when we did this project where GPT5 ran the lab, and we had an awesome result back in February, we showed that after 6 rounds of design, we had improved the cost of cell-free protein synthesis by 40% over scientific state-of-the-art that opened a lot of eyes. I think people weren't really -- we didn't know ahead of time whether the models would even be able to design experiments and interpret data at this level of sophistication. So really excited about that, really excited about future work we're going to be doing to keep proving this out with AI.

It's a neat line of work. I would say it's distinct from the autonomous lab, I call this really an AI scientist. Using the autonomous lab, using a cloud lab to get its work done. But it is all a really important thing to watch if you're following kind of how AI is changing science.

On the next slide. Also excited just -- 3 new channels coming to our delivered business to our cloud lab and data point service. Amazon Biodiscovery got launched by AWS which is basically a platform to allow you to design antibodies. All 3 of these are sort of in the antibody space, benching similarly and then Tamarind Bio. These are Tamarind and Amazon are sort of ways for pharma companies to access these frontier bio models.

So if you think of things like Alpha Fold, which got the Nobel Prize for [ Denis ] at Google, those -- that was like one of the earliest protein design models. There's many more now. They're computationally intensive. They're interesting and they help drug discovery scientists come up with a design for an antibody or a protein for their drug. But then you got to test it, right?

Like we don't know if these things work in biology unless you go into the lab. And so the idea is, could you have these layers where you access the latest models and all the compute to power them. And then when you're ready to do your experiment, you hit a button and it kicks the designs to a cloud lab to do it for you and the data flows back very nicely, well packaged right to the model. and you can run that loop as many times as you want. So that's sort of what's going on with Amazon and Tamarind and then Bench link is really the leader in electronic lab notebooks. And it's a similar idea.

If you're in ELN as a scientist and you've designed this experiment, could you ultimately hit go and kick it off to a cloud and we partner there with our data point service again around antibodies. So super exciting to see these. I think this is like early indications of a way that could become a norm for how scientists do their work in their future and kind of order their laboratory experiments.

I'll just say a couple of more quick things about data points. Really excited about the progress here, working with 10 of the top biopharma companies in the world just in the first year of running it. It's a good mix of pharma and government and even tech companies and tech bio companies. We've done a nice job on the next slide of really being a community leader here. We're running competitions. There's a virtual cell pharmacology initiative where we'll actually test compounds for free. People should definitely check that out if you're in the small molecule drug discovery space. So really need opportunities and we hope to see summits and things like that. It's been good.

I think AI has applied to the design of drugs is a big area. And with data points, we're sort of operating almost like a scale AI, like creating those just big data packages to train the models.

All right. Next slide. We have had a long-standing business in solutions, more than 150 of these research partnerships over the last 10 years. It's gotten us to work with the R&D groups of some of the largest companies in pharmaceuticals, industrial biotech and agricultural biotech. And uniquely at Ginkgo, it is a huge range of different kinds of research from Microbes associated with the roots of corn and trying to engineer them to produce fertilizer to mRNA therapeutics or antibody development and pharmaceuticals to enzymes for industrial biotech, really wide range of different types of genetic engineering and biotech lab work that has happened at Ginkgo in sort of not totally automated way.

In other words, not like no people in the lab, but like semi-automated. So human interacting with a liquid handling robot and a human interacting with various benchtop devices that can take a lot of samples at once. So we were sort of like not all the way to an autonomous lab, but we're doing a lot of variable work for years in semi-automated setups.

And so if you go to the next slide, I'm most excited to move this kind of work on to Nebula. It is the hardest work to move, right? This is the stuff that really is that car, I mentioned earlier, the lab bench. It's totally variable. It's really different. It's not just doing the same experiment over and over again like you would in a traditional CRO. But if you remember my slide, it's where 95% of the spending is going at all of our customers. They spend a bit with us, but they mostly spend on huge internal research labs to do this kind of work. And so we want to replace the manual lab bench, migrating the work from our solutions business onto Nebula is a really critical demonstration. So I'm excited about the progress there. We're trying to share that publicly been yet. We bring people through if you go to the next slide, one of the best things we do is we bring people through, show them a lab, let them talk to our scientists, see how scientists are submitting new protocols every day. And this has been really exciting to bring research leaders from -- I don't know, heads of farmer ag R&D come through to visit just this year, right, to see the system.

And so if you just want to visit, there's the link, you really should come by. But I think Nebula and our services on top of it is a truly unique asset to demonstrate what we think fundamentally is a better way to do biotech R&D. And we would love to get it in at every company out there and replace their benches.

So go to the next slide. That is the world that I want to see. And so please, if you're interested, you can e-mail me at Jason at ginkgobioworks.com. Happy to follow up and happy to take your questions now. Thank you.

[Operator Instructions]. We have one to start off submitted from Brendan at TD. We got over e-mail. He has 2 questions. So the first one is, how should we think about the potential impact to revenues this year from the AWS and benching announcements? How have the launches gone thus far? And what is baked into your assumptions for the rest of 2026 for these new platforms?

Yes, I can take that one. So yes, we talked about AWS -- the other one in that same category as the Tamarind Bio partnership as well. I'm super excited about this. I mean this is the first time I've seen this sort of kind of like cloud layer talking directly to labs as a sales channel. So I'm excited to see where it goes. It is definitely new, right? So like a flood of inbound there. We are seeing some people are reaching out to us because of the channel, so that's exciting.

I'm most excited that it's starting around antibodies, right? Because that's just kind of naturally there's a number of these AI models associated with antibodies and so on and because there's a few different providers that will do these antibody services for you. But what I'm most excited about is with our cloud lab, we're not limited test an antibody binding, right? If you look already on the, I don't know, 8 or 9, 10 protocols we posted, we're posting a new one every week. It's a pretty wide variety of stuff we're doing mass spec metabolomics, all kinds of things.

And so you can come and ask for a protocol and Cloud lab, we'll add it. I'd love that to turn into a channel straight from an electronic lab notebook or whatever, where a scientist is like, this is the protocol I want, price it. You got to price back from cloud.ginkgo.bio and then you go run your experiment. I think that's a much -- that feels a lot closer to AWS and sort of like what we saw is successful with cloud compute.

Then where these are today, which is really much more just in a more narrow lane around antibodies, which I think is an exciting place to start. But I am super excited to fan that out. I think that -- then it could become really quite an interesting channel and something that scientists just don't have access to today. At the end of the day, you can't get custom stuff done. So I think that's 1 of the most excited about there.

All right. Next question from Brendan. What are you hearing on data points and the collective AI-driven offerings with Ginkgo as especially attractive for customers as biotech and pharma companies continue to roll out their own AI capabilities. In other words, what kind of demand dynamics are you seeing here? And are there any potential revenue funnel unlocks we should watch for over the coming quarters from this part of the business?

Yes. So super -- I mean we launched data points, almost 1.5 years ago now, and have 10, the top pharma companies as customers now is really exciting. I think the revenue unlock is just repeat business for those customers. And so we are starting to see that and what we saw, what sort of like pilot projects, data gen projects and then now you've got again because you are seeing people trying to build in-house models.

Now remember, like these are not reasoning models. These are not like in-house versions of Cloud or codecs or OPUS or whatever or GPT5. They are models trained on biological data. So they're much more specialized. And so I do think it makes sense actually in the field that you're going to see a lot of people having their own data sets, their own models that are sort of tuned up versions maybe of various protein models. That's not going to be uncommon at all, much more common than I think you'll see in the reasoning model and coding space because these things are very different and people have different data sets.

And so I'm so hopeful as people are building these models, we'll keep seeing the sort of repeat demand as they're like, okay, I found one. I like what I'm seeing in terms of return on data and performance of my internal model, give me more data. And so that's the revenue unlock. And the more that we see, then I think we become sort of like a default provider that's certainly what happened with scale and other places in the early days of image models and then language models when people saw, oh, I'm seeing performance increase with more data, they turn around a lot more data. That's what we're going to be watching as these protein models and other -- and it does other types of models come out to in the future. I think that's the lane for data points.

Sort of on the theme of AI, either someone who is on X who asked us a question. I think this is sort of based on our project with OpenAI. How much efficiency improvements after using GPT 5.5. Any idea for space left for improvement? Will this be a transitional factor?

Yes. So we had this project and just to remind that we announced back in February with OpenAI, our first project with them, where we had GPT it's actually not 5.5, it's 5. We started much earlier, and that was when that was the model that was out and we kind of kept the same model through the whole thing for like more scientific paper purposes. And so we were able to show over a series of 6 rounds of running the model with [ 103, 84 ] well plates designed by GPT5 per round, a 40% improvement over state-of-the-art in the scientific goal we were trying to achieve. I think there's real interesting questions, hey, how much further could you push that, like sort of what is actually diminishing returns look like in some of these scientific areas? Can the model have sort of breakthrough ideas that create really new ways of doing this, TBD.

And then as the models have gotten better, Yes, it would 5.5 be better than what we got with 5, right? I think that's all going to be exciting up to test. So we're excited to do more with OpenAI and we're planning to. And so I think this is an open terrain in terms of how good the reasoning models can be at basically experimental design and experimental analysis.

That's -- those are the 2 things it's really doing. It's like here's an experiment. I want to run, give me back the data cloud lab, to this lab, give me back what are the results of my experiments I just designed and then I'm going to analyze them and design more experiments. We'll say, I think it's real exciting to watch what's going to be capable of there. It's a new way to do science. It really is like -- and I won't belabor this too much, but I think it roughly can turn individual. Like the access to a model like that plus an autonomous lab can let individual scientists operate closer to how a principal investigator of an academic lab or a head of a drug discovery group who has lab of 8 people or a lab of 30 people and it's sort of assigning hypotheses to different people and kind of pursuing that over time. An individual could push that out. for probably close to the same cost as they are currently costing to be themselves at a lab bench in terms of their just -- their utility costs and everything else and utilization, low utilization of equipment, they could push out 5 agents on top of an autonomous lab to go pursue a bunch of experiments.

That is real exciting if that works. I think it really fundamentally changes the rate we can do science. That's why you see the Genesis mission in the U.S. investing in this sort of stuff because their goal is to 2x the output of U.S. science. The way that will do it. And our science-based industries, of which pharma is the biggest will be completely changed by this. If you can, 2 or 3x the rate, no question about it.

All right. Our next question is really a bundle of questions from DK who's writing from South Korea. And these questions are all about how the move on to Nebula, our autonomous lab has sort of changed the science that we're doing. So the questions are, how does the use of Ginkgo's automated lab affect overall costs? Are there meaningful differences in speed, for example, turnaround time for experiments? And have you observed improvements in success rates, reproducibility or scalability since moving to the autonomous lab?

Yes. So on cost, I try to touch on this a little bit in the talk. But I think like the clear ROI not just for us but for any one of our customers looking at an autonomous lab, is about a threefold reduction in space utilization compared to a manual lab and a fourfold increase in that time. In other words, like the amount of time the lab is being used to do lab work, right, from that 40 hours to 168, 24/7 week. That's really that -- those improvements is where it's going to yield the cost reduction. But that is a huge amount because those are really the 2 -- like sort of people time and space time are the 2 big things we spend money on in research. On the speed front, yes, it's interesting. An individual protocol doesn't really get shorter like than necessarily you would do it at the bench. You can imagine ways to do that in the future rebuild protocols differently.

But the first thing scientists are going to do is just take work they're doing at the bench and move it on to the autonomous lab. And in that world, it does not need to get faster in terms of like end-to-end time for the protocol. However, in practice can get faster because you can start a protocol at 4:00 p.m. in the afternoon where you never would have planned to spend the next 7 hours in the lab, kick it off and have the thing run overnight.

So in that world, you took an experiment that you would have started tomorrow at 10:00 a.m. and started at 4:00 p.m. and have the results by tomorrow at 8:00 am or 10 a.m. And so that can shave a whole day off. So I think you will see actually a massive speed up because scientists will start taking advantage of the 4x more time that they have available every week. So if they plan it right, in theory, you can see a fourfold improvement in a lot of the times, depending on how serialized your experiments need to be. So I think that's really exciting in our side, I think I really like that.

And then on the -- just sort of like improvement in like I say, I would call this like the quality of the experiments. I think reproducibility is inherently advantaged on automation. And that's mainly to do with like the audit trail. Like you kind of if an instrument errors, if a liquid handler makes mistake, these are all tracked. So you kind of know like those experiments that you don't catch at the manual lab bench, you catch if there's such a mistake on the autonomous lab.

So if you saw really, wow, that's a surprising result. You might go back, look at your experiments and say like, oh, I see what I did there, I like design this experiment in a way that was like a little silly, and that's actually what's giving me this result as opposed to assuming you did the experiment you wanted to do and that was the origin of this like amazing result you got. I think yes, that's a common thing that can happen. For no nefarious reasons for scientists at the bench. And so I think that you will see a big improvement in reproducibility. And then the other thing that got brought up there was throughput, the throughput increase is going to be the same.

I think people are surprised when they go to cloud.ginkgo.bio, which are triple to do and type in a protocol and see how much it costs. Because I'm basically pricing that protocol based on reagent use and equipment time and a markup on that. And it is not the insane costs that you have when you have a whole team doing this work at the bench, it's just not.

Like -- so if scientists really understood, just how low cost each sample could be in an experiment, and they did -- in order to do many more, they just hit a button rather than have to slave in the lab for 3 days is doing 1,000 experiments. They're going to just order those 1,000 experiments. And so I think you will see an explosion in the amount of data. And this is 100% what happened in every other field that's ever been automated, right? It's like the beginnings of the automation of computation, right? Like when we went from slide rules to automated computation and explosion in the amount of compute you use, and a massive increase in the return on investment from what people who understood how to design a computation could do. And that's what I want to do for the scientists for drug discovery leads when they have access to an autonomous lab compared to the ROI and the throughput that they can get out of manual labs. It's just going to be no comparison. So yes, I think all 3, you're going to see big gains on.

And the cool thing is we're going to keep showing this on Nebula. So we just had -- Head of R&D for today, and we went through with his team and showed all the games, and it's -- yes, it's really exciting right now.

So I think kind of related to that, which is you guys mentioned in the call, you've mentioned other places you're trying to get to 100 RACs. When do you actually expect to get there?

Yes. So it's been pretty fun. We have to put behind the scenes videos up, but we have been installing racks for the last 3 weeks here at Ginkgo -- showed up on RACs built by our team in Emeryville. And we just added the additional 50. They are all fully connected now in lives to a tour of it, it's insane. And so -- and we can run them now, like the original system is running and now the new 50 is running, and there is a connection. Between the two, and that connection is going to get turned on, I think on the 14th next week. So it is imminent. So I'm really excited to see it all come together. But we already have it up now running as 2 several loops.

So to put in 15 new pieces of equipment in 3 weeks. Again, these are just things that no one's ever done in laboratory automation. So I do think we are doing a very unique thing here at Ginkgo. That's the bet. That's certainly what I'm leaning in on the company. It's what we're investing our capital into. It's where our new customers are coming from and so if you like that idea, I think that is a really exciting time to get involved with the company in any way. But yes, we're going to be at 100 next week. 103 or 105. I got a count, yes.

All right. And if you want to follow us on that journey, you can go to X or LinkedIn, Instagram and keep watching. We'll have a lot of content coming about the unveiling of the new full system. And as always, if you have questions, you can reach out to us at investors at ginkgobioworks.com. Thanks so much, everyone, until next time.

Thanks, everybody.

[Audio Gap] joining us. We're looking forward to updating you on our progress. As a reminder, during the presentation today, we will be making forward-looking statements, which involve risks and uncertainties. Please refer to our filings with the SEC to learn more about these risks and uncertainties, including our most recent 10-K.

Today, in addition to updating you on the quarter results, we're going to provide insight into the autonomous lab, how we believe it will transform biotechnology and how we plan to commercialize autonomous labs going forward. As usual, we'll end with a Q&A session, and I'll take questions from analysts, investors and the public. You can submit those questions to us in advance via X, [indiscernible] results or through e-mail, [email protected]. All right. Over to you, Jason.

All right. Thanks, Daniel. So Q4 was really a breakout quarter for us in sort of defining and really leading in the category of autonomous labs. And so you're going to hear a lot from me about that in the future. But I want to start by saying our mission remains to make biology or engineer. But in 2026, the technological focus for the company and really the business focus is going to drill down on investing to win in this category of autonomous labs. And this is really a part of what I see as an emerging movement around robotics and AI and autonomy that's coming to a lot of sectors in the economy.

And I think we happen to be in a sweet spot in bringing that into really a high-value area around laboratory research, but there's an increasing amount of excitement about, and I intend to win that. All right. So how are we going to do it in '26? So first, we want to focus our investment in our platform into that area primarily. And I'll talk in a minute, but we mentioned in our recent earnings announcement just now that we'll be divesting our biosecurity business. That allows me to focus Ginkgo's investment and our dollars really into autonomous labs and bring in other new investors to invest alongside us into biosecurity.

So that focused in investment. Second, internal to the company, we want to demonstrate the capabilities of our large autonomous lab here in Boston. And the way we're going to do that is we're going to start to systematically decommission our lab benches, our walk-up automation, our work cells, the way that we've traditionally done our R&D services over the last 10 years and move more and more of that work on to a single large autonomous lab that's software controlled here in [indiscernible] is that serves as a demonstration to the Mercks and the Takedas and the Pfizers and all the [indiscernible] huge investments in traditional manual laboratories that it is possible to take open-ended research and run it through a large autonomous laboratory system.

And so I think that's really fundamentally the most important work we're doing this year. And then finally, I want to book sales of autonomous labs. So you're going to hear in one of our big announcements from last quarter that we did a $47 million deal with Pacific Northwest National Labs. So I want to sell Atomos Labs to National labs like that DOE deal, but I also want to sell them to biopharma. I want to sell them to research universities. And so that sort of bookings and landing new deals is the other thing we want to do in 2026 in this direction. I do want to take a minute and talk about that biosecurity divestiture.

So you might remember, over the last 5 years, we've invested a ton of energy into this space. This really came about starting during COVID. because we honestly just saw a need. COVID was sort of a global scale biological disaster, and we felt we should lean in and help where we could. The niche that we found in that moment was doing really monitoring for -- so not diagnostic testing, but rather monitoring testing in order to reopen congregate areas and in particular, reopening schools here in the United States. So I'm really proud of this is a decent sized business for us.

But really importantly, we helped open 5,000 schools nationwide. And this is one of these like really political topics. And I think what's neat about technologies, you can sometimes find a third way between and one end [ at the time it was ] which was hey, we really should be closing the schools standed for teachers, we care about spread in disease. And then on the other side, hey, this is hurting kids, and we need to open the schools, everyone should just go back and whatever it comes comes.

And there was a third way, which was, why don't we open the schools and a rein so that if an outbreak starts to happen in a school, you can send 2 or 3 hits home and stop it. And that's exactly what we want to build at a nationwide level and what's continued after COVID in our monitoring and airports that we do in partnership with the CDC, looking for viruses in the wastewater of planes and other -- both here and internationally in places like Doha, Qatar, the airport there. And so that sort of identify it, put it out, put that fire out before it spreads is something that's needed nationally and globally for the U.S. to be secure.

The other thing that's happened in that period of time, you might have noticed companies like [indiscernible] Board Chair, as the CEO at Palantir. This sort of defense tech set really exploded over the last 5 years. And so there's been increasing interest from pure-play investors in the defense space who want to see next-generation sort of bio defense primes. So again, these are companies that would be focused on serving the government and others on Biodefense needs directly. That's very exciting because it means there's lots of new capital interested in that.

But to my point earlier, where I want Ginkgo to focus very clearly in 2026 is on autonomous labs. And so 1 of the great things that happen, which we've got a lot of inbound from these types of investors. And we saw an opportunity to say, "All right, why don't we share in the upside of biosecurity by taking business unit in the company, spinning it off, taking it private, bringing in investment from some of these great investors. Ginkgo will still hold a minority position in that. So we get to get a piece of the upside of what we've built, but the investment needed to build that biosecurity prime doesn't need to come from the $430 million, as I'll mention in a second, that we had on our books at the end of the year. We can focus that into autonomous labs.

So I think this is a win-win all around. I also think bringing in these types of great folks that we have coming into the private entity or is really going to open doors with the defense sector and so on and having it be a sole-branded biodefense company, it's the right time. So I we're excited about this. I think it's -- I want to give again credit to the last [indiscernible] team at Ginkgo, who did absolutely amazing work through Ginkgo and now has a real opportunity here, I think, to build a generational business coming up in the defense sector.

Okay. Last point I want to make before I hand it to Steve. So again, I think tremendous work over the last 2 years. We sorted 2 things at the same time. We dramatically cut back spending as we saw sort of a downturn in the biotech sector and a lot of our customers pull back on outsourced large R&D projects, which is really our bread and butter here at the company on the [indiscernible] drew down on our spending and pretty substantially.

So in fiscal year '24, we were at $383 million and just last year, $171 million. So a 55% reduction in our annual cash burn. That sets us up very nicely. You're going to hear from Steve on our target for cash burn for this year, even with the investment, our focused investment in autonomous labs, and moving that investment in biosecurity into a separate private entity. We're actually able to do better than what we were spent in '25.

That, I think, for investors is important to understand where we're at from a cash position and how we've done a really nice job getting cash spending under control. as we continue to make investments and get in the right place at the right time with autonomous labs. All right. So I'm going to hand it to Steve to dive in a little more on the financials.

Thanks, Jason. I'll start with the cell engineering business. Cell engineering revenue was $26 million in the fourth quarter of 2025, down 26% compared to the fourth quarter of 2024. In the fourth quarter of 2025, we supported a total of 109 revenue-generating programs. This represents a 4% decrease year-over-year primarily attributed to ongoing program rationalization as part of our restructuring activities.

Turning to the next slide. On a full year basis, Cell Engineering revenue was $133 million in 2025 as compared to $174 million in 2024. As previously disclosed, Revenue in the first quarter of 2025 included $7.5 million of noncash revenue from our release of deferred revenue relating to the mutual termination of the biome edit agreement. In the third quarter of 2024, Cell Engineering revenue included $45 million of noncash revenue from our release of deferred revenue relating to the mutual termination of the Multi Foodworks agreement.

Excluding these impacts, cell engineering revenue was $125 million in 2025 and $129 million in 2024. We -- this decrease was primarily driven by customer program rationalization related to the restructuring, as all discussed previously. The biosecurity business generated $7 million of revenue in the fourth quarter and and $37 million of revenue in the full year 2025. It is important to note that our net loss includes a number of noncash and other nonrecurring items as detailed more fully in our financial statements.

Because of these noncash and other nonrecurring items, we believe adjusted EBITDA is a more indicative measure of our profitability. A full reconciliation between segment operating loss, adjusted EBITDA and and GAAP net loss can be found in the appendix. Cell engineering R&D expense decreased 44% from $50 million in the fourth quarter of 2024 to $28 million in the fourth quarter of 2025.

For the full year 2025, Cell Engineering R&D expense decreased 42% from $272 million in 2024 and to $159 million in 2025. As reported last quarter, the full year 2025 period, R&D expenses included a $21 million shortfall obligation related to our multiyear strategic cloud and AI partnership with Google Club. In October 2025, we amended and reset the annual commitments for future years and settled the shortfall obligation for $14 million.

Resetting the commitment reduced our future minimum commitments by more than $100 million compared to the original terms and extended the commitment term from 3 to 6 years. Cell engineering G&A expense decreased 40% from $20 million in the fourth quarter of 2024 to $12 million in the fourth quarter of 2025. For the full year, Cell Engineering R&D -- I'm sorry, G&A expense decreased 51% from $115 million in 2024 to $56 million in 2025. These decreases were all driven by our restructuring efforts. Cell Engineering segment operating loss was $17 million in the fourth quarter of 2025 compared to a loss of $38 million in the 2024 period.

For the full year 2025, Cell Engineering segment operating loss was $96 million compared to a loss of $219 million in 2024. The lower loss was directly related to our restructuring efforts, while partially impacted by the matters previously mentioned. The biosecurity segment operating loss improved 60% in the fourth quarter of 2025 compared to the 2024 period. And the Biosecurity segment operating loss improved 38% in the full year 2025 compared to 2024.

Moving further down the page, you'll note that total adjusted EBITDA in the fourth quarter of 2025 was negative $36 million, which was down from negative $57 million in the fourth quarter of 2024. Total adjusted EBITDA for the full year 2025 was negative $167 million, which was down from negative $293 million in 2024. We Again, the period-over-period declines can be attributed to our restructuring efforts, while partially impacted by the matters previously mentioned.

Turning to the next slide. We show adjusted EBITDA at the segment level to show the relative profitability of our segments. The principal differences between segment operating loss and total adjusted EBITDA relates to the carrying cost of excess lease space. which was $54 million in 2025, and this carrying cost was $15 million in Q4. The cost represents the base rent and other charges related to leased space, which we are not occupying net of sublease income. This is a cash operating cost that is not related to driving revenue right now and can be potentially mitigated through subleasing.

And finally, turning to cash burn. Cash burn in the fourth quarter of 2025 was $47 million down from $55 million in the fourth quarter of 2024, a 15% decrease. Cash burn for the full year 2025 was $171 million, down from $383 million in 2024 and a 55% decrease. Cash burn does not include the proceeds from the ATM issuances or certain cash construction restrictions. This significant decrease in cash burn was a direct result of the restructuring.

Turning to guidance. In terms of the outlook for 2026. As Jason has mentioned and will go into further, 2026 is about continuing to be cost efficient, while investing in our AI robotics and software to bring autonomous labs to our bioscience customers. including the build-out of our Frontier autonomous lab in Boston. We have turned the page on our pure focus on restructuring actions for the last 2 years to focus this year, not only on cost efficiency, but on investing in what we see as our opportunities while continuing to provide our customers the advanced services they have come to expect. We will also close our transaction for the biosecurity business as announced and disclosed.

For these reasons, in 2026, we will not be providing revenue guidance as we believe cash burn best reflects our continuing services and tools and further investments in autonomous labs. With 2026, our overall expected cash burn guidance is to be in the range of $125 million to $150 million. This range reflects a firm balance amongst cost efficiency, continuing services and tools and the further investments we are making. In conclusion, we are pleased with the continued improvements in cash burn and cost reductions in 2025, and our excited full will come in 2026. And with that, I'll hand it back over to you, Jason.

Thanks, Steve. Okay. So before I jump into my section, I want to spend a minute talking a little bit more about what Steve was talking about there at the end in terms of our guidance for the year, how we're going to be guiding on cash burn rather than revenues and sort of why we're doing that. And so this is in line with my theme for this earnings call, which is Ginkgo's focus. So one thing is we want to be focusing on investing in the right things.

And so we're -- I believe, again, it's important for our investors in Ginkgo to understand what we're doing with our cash supply, how fast that's being spent down, what we're spending it on. And again, the highlight here is we're spending it very deliberately on autonomous labs, and we're doing it in a controlled way. We're hopefully spending substantially less than we spent in the last year and our relative position there to our cash pile, it looks pretty good.

So from my standpoint, we have a solid margin of safety as we're investing to lead in this area of autonomous labs going forward. But the second thing we need to keep focus is our attention within the company. And so the majority of our revenue today does come from our R&D services. We love serving those customers. I'm hopeful we grow those services. But as I mentioned earlier, the focus of the team in 2026 is not on hitting a short-term revenue target around a service run on top of our autonomous lab to make sure we had a target or trying to predict exactly what that revenue is going to be over the next 12 months.

What I want their focus to be on is decommissioning all of the different labs here at Ginkgo and moving that work onto our autonomous lab so that we can show all of our customers that this works, that autonomous labs can be a true replacement for the humongous spending they have across their manual laboratories in both biotech and academic science. That's the main event. And I felt that, again, continuing a focus on revenue targets and things like that. It was going to take people's eye off the ball. And I also think it sort of takes away from the long-term orientation, which I think is going to be critical for Ginkgo. So that's why we made that decision. Happy to talk more about that in questions or otherwise, but just so you know where it's coming from.

Okay. So as I said, our mission is to make biology easier to engineer. We had 3 really amazing things happened last quarter, so he's going to run through them. So first, we had an announcement of a project we've been working on for the last 6 months. with open AI. This is their blog post about it, where we talked about connecting GPT 5 as sort of an AI scientist. So doing the work of a scientist designing experiments except they would submit those experiments to our autonomous lab here in Boston. The lab would conduct the work, send that data back to GPT 5 and then over the course of 6 rounds of doing that. we were able to beat state-of-the-art on a pretty complicated sort of [indiscernible] experimental scientific challenge in cell-free protein synthesis by 40%. What I think is good about this is number one, the sort of views on this ex post where they announced Kodak on the same day about equal to what they saw with [indiscernible], right?

So I think there is really a lot of excitement right now. in how reasoning models can enter the physical world, all right? I'm going to talk in a minute about that area of transportation, where like most have brought them into the physical world. But I think we really stand to be the ones to bring AI into the physical world of the lab. We are absolutely in the pole position on that. So I'll talk more about that in a second.

Second, we were -- I have the absolute privilege to do a press conference with Department of Energy Secretary Wright up at Pacific Northwest National Labs in Washington. -- when we announced in December that the first 18 robots that we were installing for PMML as part of the Genesis project. This is a new project out of the White House to bring AI into science and AI into the national labs in particular.

But alongside that ribbon-cutting and [indiscernible] to sign our system. You can see them signing it there. We also announced a new $47 million contract with the Department of Energy to build a 97 robot, 97 rack autonomous lab at that same site in P&L in the future. So really exciting, and I think this showcases that autonomous labs are of interest to the federal government, which is the other big pool of research spending. So places like the NIH is spending $40 billion a year on lab work. frankly. And that's pretty close to what you're seeing in the pharma companies as well. So those are your sort of big pools of spending. And so I think it's important to see it coming from the federal government as well as from pharma companies. Last not least, we had SLAS. This is the Society for Laboratory Automation and Screening Conference that was just at the recent it's about 5 minutes away from here, very fortunately for us, at Ginkgo.

So we hosted tours of Nebula, are now more than 50 rack autonomous lab set up here in Boston. We had 590 people come through, and it was very eye-opening to see what a difference it made for people to see a lab like this actually doing real science during the day, right? Like people coming in and just seeing what our scientists are doing with it, it was eye-opening for them. And so I think this makes it more and more clear to me that we're making the right choice with this focus in '26 on really driving the further expansion of the system. We're going to go from 50 racks to 100 racks by H1.

That's the sort of stuff I want you to be following. How quickly are we able to expand that? How quickly are we able to add more of our work onto that system because that's exactly what our pharma National Lab and university research leaders are going to be looking at to see if they want to buy a system like this.

Okay. All right. So now I'm going to do a deep dive into autonomous labs because, again, I think this is really our focus, certainly in 2016, and I think the technological foundation for the company over the next decade. So I'm going to talk about -- then is it what is an autonomous lab, why is it going to transform biotechnology? Secondly, what does it look like very specifically, like what do you need to have the lab be able to do in order to deliver biotech R&D. And then finally, how are we going to bring it to market. And the 2 ways we're doing that is one, we'll build 1 for you like we did at [indiscernible] that beautiful lab that you just saw pictures of able to run that sort of in a cloud service model through our R&D services and new services we're adding coming up.

Okay. So here's the analogy I like to give. And I talked -- I give -- I talk it [indiscernible] and I talked to this with a lot of people. I think it's a good one, all right? So I'm going to start in the transportation industry to help explain what I consider autonomy to be. So if you look at this chart on the y-axis, you have the amount of automation, all right? And then on the x-axis, you have the flexibility of the request from a user to that automation that it's willing to tolerate. So in transportation, if you have a low amount of request flexibility and a high amount of automation, that's a subway, right? You sit down the back of a subway and it just takes you away, right?

You're not having to do anything. It is fully automated, but you better want to go to 1 of the stops on that subway line because it's not going to take you to your house or the grocery store or just wherever you want to go, it's on rails, all right? So it's very inflexible. Now low automation. -- high amount of request flexibility, that's a car, right? You put your hands on the wheel, your feet on the pedals and you can drive it straight to your front door or to that grocery store, right? And that -- those 2 poles is basically what the transportation industry has looked like in the last 100 years. forward to slide, unless you've been in San Francisco over the last 4 or 5 years, and you've seen these driving around.

So this is a [indiscernible]. You sit in the backseat just like sitting on a subway seat, you do absolutely nothing. It takes you away except unlike the subway, it will take you right to your house right to that grocery store. So it has the flexibility of a car, but the automation of a subway. And that's such a surprising thing that we're giving it a new word we're calling it autonomy, all right? And I think you will see this replicated when you're seeing all this interest in like humanoid robotics and all this, like there's a huge amount of investment going into it right now.

What we're trying -- what's happening on a broad investor level is the industrial revolution was essentially the application of automation and systematization to all of the tasks that were like low flexibility required, like back to the loom, right? Everything that wasn't that required a lot of flexibility, we kept manual. And what's happening now is the AI models are getting good enough the software is getting good enough to allow automation to be applied to flexible things, and we're going to see how far we can push that. And the more you can push into flexibility, the bigger the opportunity there is for robotics. And so we're going to drive that change in labs.

Now last point, this is the kicker. If you were to look at the split between miles traveled on trains and subways versus cars and trucks in the United States anyway, it's 99% cars and trucks. -- because you need that flexibility to go places, right? It's a requirement. It's not like we didn't know about rails. They just did not tolerate the flexibility needed. All right. So let's look at the lab. -- low amount of flexibility, high amount of automation. So up where that subway was on the last slide, we do have that actually. We call it automation work cell. And you can buy this from companies like [indiscernible] and [indiscernible].

And basically, you tell them what protocol you want and they build you a work cell that will run that protocol for you. And it's great. It's totally end to end. There's not a person in the middle. It's fully automated. It's the top of that chart. But you better be asking for the same protocol that you asked it to do yesterday because it could not handle variety in the request from the scientists that are using it. drop down that automation line and go over on request flexibility, so not automated, but very, very flexible. That is the lab bench. And we've had it for 100-plus years. It lets you do whatever experiment you want and the scientists, the human scientists in the middle is what's providing the flexibility.

All right. And that's what the system has looked like. We've had work cell automation for 40-plus years now, that we've been kind of those 2 poles for the last 40 years. and much like research -- or sorry, much like transportation, I add a couple of heads of R&D as it 2 pharma companies over my house [indiscernible] dinner. And I asked the question, what's your spend between work cells and lab benches. And they said, actually, 99% on the [indiscernible].

But let's call it, more than 95% of the research budget is going to the lab bench. And it's for the same reason that 99% goes to the cars and trucks you need the flexibility to do science. And if you can tell this, if you walk around at Merck or Pfizer Takeda and you walk the hallways, you will not see robots. You will see lab bench after lab bench with [indiscernible] equipment on top of it and scientists basically being the human glue that connects all that different equipment and manages to do liquid handling by hand with pipe ads and all the things that they do. that is the overwhelming majority of research spending and pharma is doing, again, $40 billion to $60 billion of not development, but research spending every year through that platform.

All right. What are we trying to build or try to build that Waymo. What Ginkgo believes we have when it comes to our [indiscernible] and then very importantly, the software that run it runs it is an autonomous lab. It is the flexibility of the lab bench, but the automation of the work cell. And that is, we believe, fundamentally different. It's a much bigger market than the work cell market. The work sell market, again, just like the subway is very limited in terms of the amount of research dollars flowing to it. And so we want to go right at that autonomous lab market. The key technical question, the next slide is how do you get both high automation and high flexibility without having those human hands in the lab, all right?

And so that's the next thing I want to talk about, what do we actually have to pull off technically to make this a reality? What are people so impressed with when they come visit our lab here in Boston and see what we've built. All right. also, I don't know if you noticed we follow me on LinkedIn, you've seen I've become a bit of an influencer lately. So this is what it looks like if you're standing at a lab bench doing your work by hand. And the real major activities is number one. You're serving as manual liquid handler. In other words, you are moving small volumes very precisely between different liquid containers to set up an experiment with the right materials in it.

Second, you're moving samples -- in other words, that liquid you just set up in a plastic tube or whatever it might be in 2 different devices across the lab. So you are moving samples as the protocol demands across maybe 3, maybe 10 different devices depending on the complexity of the protocol you're doing. And then finally, every time that sample ends up on a device, all those devices. These are all like complicated, long tail, scientific devices. They have some settings that you need to set in order to have it do the thing you wanted to do. So you, as a scientists, are the 1 putting those settings in and you're either doing that with a touchscreen or with sort of third-party software. Okay. All right. So to replace traditional labs, an autonomous lab has to do those same things I just said, that's 1, 2 and 3, reliable liquid handling, material transport and parameterized control of the device.

But very importantly, if you think about what one of those labs at Takeda or Merck looks like, in 1 floor with a bunch of benches and maybe 20 or 30 scientists using it, you're going to have more than 50 devices around that lab that those scientists are making use of different ones, different days, different ones as part of different protocols. So you've got to be able to put at least 50 devices into 1 big setup. The other thing that those scientists are doing when they -- the first scientist gets in the lab in the morning, they do not close the door behind them, lock it and put up a sign that says, lab and use.

No one else can come in, right? It's busy, Lab is busy. But on a work cell, like one of those subway system automations that we have in the lab, that's exactly how it works. Once it's in use, you cannot interject yourself into that process and submit a new job. But in the manual lab, absolutely, 10, 20, 30 scientists are all walking around that lab, basically sharing the equipment and avoiding each other's usage of the equipment. So if I'm using something in the morning, you'll use it in the afternoon.

But other than that constraint, they have access to all that equipment and they can use it in parallel. And then finally, it's very easy to use the lab bag. You don't have the right software programs and things like that. I won't have as much time to talk about it today, but in the coming earnings call, I'll do a little bit of a deeper dive on our software. But one of the things we're really benefiting from is all this investment in coding agents, things like codecs from OpenAI and Cloud code are now allowing human language to turn into pretty complicated software. We want to turn scientific intent into work that runs on automation without scientist needing to code.

I think that is going to be very doable thankfully. And that's #6. It needs to feel like when I go in the lab every day to do my work, I don't have to sit down and right code. You shouldn't have to do that for the autonomous lab. This is really a difficult set of challenges. Work cells today do do those first 3. They deliver liquids. We have liquid handler automation, companies like Hamilton and [indiscernible] have been around for 25 years or greater, more. They're great. Second, reliable material transport can be done with arms and third, parametrize control is doable. For 5 and 6 are not delivered well by traditional lab automation today, but we do have it working at Ginkgo.

All right. The first thing to understand about how to deliver 4, 5 and 6 is that a work cell. In other words, that subway is designed around a protocol. So the first thing 1 of those companies will ask if you're going to build a they're going build an automation system for you is -- what's your protocol? Are you doing high-throughput screening, that's 1 of the most common ones, antibody developability, protein production. What is it you're doing, right? And you say, oh, I'm doing this, these are the steps. This is the equipment I need, and this is through and then they'll design a subway that delivers you to that stop. Autonomous labs are not designed around your workflow, but they're rather designed around the equipment because this is exactly what happens when you're setting up a new manual lab at Takeda or Merck.

If you're the person in charge of that lab, you're that kind of group leader, you ask your scientists, what equipment will they need to do their work over the next 5 years in that lab. They don't know for sure what protocols they're going to do, but depending on the type of work they're doing, mammalian work, bacterial or cancers, whatever, that they're going to use different types of equipment. So we oriented the design of our hardware -- our robotics hardware, not around a protocol, but around a device.

And so this -- here, you can see our rack automation cards inside each car is a device. In this case, that's a Centerfuge, a 6-axis industrial robotic arm and a piece of Magna Motion track. And that track allows you to deliver a sample between connected racks. Each one of the the [indiscernible] tracks connect to each other and you can send samples around and deliver them. And if you go to the next slide, we can show a video here of samples moving through our autonomous lab here in Boston. This is actually. Interestingly, this is one of the protocols from OpenAI, right? And so what you can see as this runs is -- we have the sample getting put on to the track. That's a 384-well plate in each well of that plate is a set of conditions that were designed by [indiscernible].

The plates travel on that Magna Motion track. And in this case, they're delivered to that center fuse, right? And so the [indiscernible] is going to spin down that sample. So just happens to be the first step in this protocol. Now it's going to one of those liquid handling devices. So this is what's called an acoustic liquid handler. It moves liquids with sound. So 1 of the things that's great about this is it's actually can handle smaller volumes at a greater precision than a scientist could do by hand. -- so we can move nano leaders of liquids around as a scientist using a pipe by hand or kind of limited to microleaders in terms of your ability to be accurate. Now we're going to be adding, in this case, DNA to each one of those wells. So the project we did with OpenAI, a piece of DNA was being added to what's called a cell-free mix of reagents. And the idea is that cell-free mix turns that piece of DNA code into a protein.

And the protein level is what we're trying to optimize with open AI we're trying to see, could you change the conditions such that you've got higher protein production than any scientists had shown before in the literature. So once that DNA got added, we now shake it up, make sure it's well mixed. And then it's going to end up onto an analytical device in order to basically run the reaction and then measure the levels of protein that are coming out of that particular -- of each well that 384 well plate.

And so to give you a sense for the OpenAI project, each time we did around with the model, we ran 100 of these 384-well plates, collected all of that data gave it back to the model and then the model was able to design the next round of experiments, okay? So that's what it looks like for a sample to move through the system. At the beginning of that video, you would have seen a quick picture of sort of the data coming in, like the particular designs from OpenAI and then the scheduler, that schedule on that 1 was just running the 1 protocol. This is what the schedule looks like when people -- when our scientists at Ginkgo have submitted 30 protocols to the system. And so what you're looking at is each row in this is a different device on the system. And then the x-axis is time.

So that orange bar in this case is like now, all right? And what is great is basically predict the future, right? We know the system knows exactly what piece of equipment is going to be used for what protocol and each protocol is a different color on this chart. What piece of equipment is going to be used for each protocol in the future. And what the scheduler does is if you showed up at [indiscernible] as a scientist and you submitted a new job into this -- into the autonomous lab, you would say, okay, I'm using the Center fuse for 5 minutes. And then I can wait up to 2 hours before I need to end up on the echo and do, you would specify with time windows, your protocol that driller will check could you fit in? And this is very [indiscernible] to 1 of those scientists walking around the manual lab, asking their bench mates, hey, when are you going to be done with the PCR machine.

How long -- is it okay if I run the HPLC overnight or something do you need to get on it, right, like having conversations about the availability of equipment, except in this case, it's all computer-controlled and computer scheduled, so we can essentially schedule it perfectly. And so as you add more protocols in, there's a complicated algorithm to handle all this. We are the only people in the world as far as I know, that are doing anything close to the scale of variable protocols on a single automated system. And that was pretty well confirmed at wide open eye is during the SLS tour when I was able to show this off to people.

Okay. So go to the next slide. This is just a different color. So each of those protocols, you can see being submitted by a different user at Ginkgo as well. So that's, I think, actually really interesting, where we have not just not just like a large number of protocols, but also a large number of unique users submitting those protocols. That's also very unique in the world. When you have those work cells, there's an automation engineer or 2 who are sort of in charge of it and everything funnels through them. In the case of our autonomous lab here in Boston, we have tens of scientists submitting protocols every day, different protocols from yesterday that are all scheduled simultaneously.

So hopefully, that next slide, that help give you a picture of how we've checked off the sort of 4, 5 and 6 on that list in terms of many pieces of equipment, all in 1 setup being run simultaneously in parallel, easy enough to use is by scientists who aren't automation engineers. Just note that system that's now 50-plus racks, again go started off as, I think, 7 or 8. It's very expandable. In fact, on the next slide, after we finished up at SLAS, we were able to bring over the rack carts that we had at the conference there was, I think, 7 or 8 and install them all in a day on the system. So the ability to really grow this system is, I think, again, unique when compared to traditional sort of subway style automation. All right. So what's the value prop to customers.

There's 3 things, I think, that the -- like a large biopharma or a national lab would get excited about. First, save overhead cost by closing your traditional apps. This is one of the things I'm most excited to do this year with our CRO, our kind of research services that we run on across all our labs at can go. As I move more and more of that work on to the autonomous lab, I can shrink the footprint of my labs, which saves me in EHS costs, saves me in rent, saves me [indiscernible] all these different things that you have to carry when you're running these labs. Second, it increases the research productivity of your researchers. So right now, a lot of their ideas are ultimately bottlenecked by the amount of time they have to spend in the lab.

We want to really open that up and get much more data per research dollar out of your scientists. And then finally, like we did with OpenAI, you can have AI scientists run what are kind of in the industry called lab and a loop experiments where the AI model is designing experiments, they're running on the autonomous lab and data is going back. And so we're seeing a lot more interest in that from pharma companies as well.

All right. Okay. So last section I want to talk about is how are we going to sell these autonomous labs, and there's sort of 2 ways we're going to do it. One, we will place a system like we did Pacific Northwest National Labs. We will place it at a customer site. We'll sell CapEx. We'll sell basically service fees, both for the software and for our maintenance of the equipment. And then the future I could see us even selling things like reagents and consumables and things like that to the users of our system that are sort of automation specific. Additionally, we have this big autonomous lab in Boston that we can offer services on top of, right?

And so what's like the overall market potential. This is back to that 1% on the subways, 99% in the cars, the overwhelming majority of research spending, that $40 billion to $60 billion in pharma, the $40 billion plus from the government and so on, that's all funneling through benches today. And that's before we also -- the other big industry we haven't talked about at all is sort of diagnostics. And I also see opportunities there as well. So all of that bench labs spending, I think, ultimately has the opportunity to funnel through our platform if our autonomous lab is able to replace the bench. The way we're going to get there is we're going to start by commercializing in 2 ways.

First, [indiscernible] for customers, second run the cloud lab. All right. So cloud lab services. Two of them are ones you've already heard about. So our solution services, this is where our Ginkgo scientists use our autonomous lab to deliver research outcome to a customer. So our deals with Novo Nordisk and [indiscernible] and Merck and Pfizer and all these people over the years where Ginkgo scientists use Ginkgo's labs to deliver research outcome.

We got a royalty, we got milestones. We can structure these in different ways. We do a lot of work with the government in R&D grants and things like that through our solutions business. Second, in data points, customer scientists use our autonomous lab, they design what they want to run on it. This is run like a traditional CRO. There's no royalties. There's no milestones. We send them a huge amount of data back usually to their ML team, and they use that to train Bio-AI models for protein design or RNA design or whatever they might be doing.

The third, and I'll have more about this in future earnings calls, but we'll be announcing this soon is our cloud lab offering. And so what we're going to have here is customer scientists, outsourcing small amounts of lab work directly to our autonomous lab. So think like a $50 order or a $200 order, where the actual experiment will get run on the cloud lab and data will go back to the scientist. And I think this is a great way for scientists who are curious about what it's like to engage with autonomous labs to sort of try it before they buy.

And there's lots of things that drive there different ways to bring it to market. You'll hear from us coming up on that. I'm really excited about it. Just to say, we're not new to the solutions business. We've done 250 partnerships in the last 10 years. We're continuing to sign these every quarter. We're doing a lot of business with the government and large pharma are really the 2 areas where we see the most of this, but also agriculture. Industrial biotech has been a lot harder since 2022, basically. But ag pharma and government still will sign up for solutions deals. The other area that's grown really well for us over the last year, and I want to give a shout out to the data points team at Ginkgo is we've been growing this business where we run our robotics to generate big data sets against the designs of customers. And this is a business that got started 1.5 years ago. We've now worked with 10 of the top I think 3 top pharma customers just in the first year we launched this thing. So people are really excited about it. It's a good fit.

We've actually released a bunch of public data sets. If you go to the Datapoints website, you can download some of the largest data sets for drug seek and for antibody developability and things like that. Next slide. I think that we've also done a really nice job being a community builder here. We've launched developability competition. We have a virtual cell pharmacology initiative where we do free data generation as part of building up a big public data set. So really, I think if you're interested in this area, if you're doing BioAI, definitely check out data points come to some of our events. The last point I'll mention about us running the CRO labs is that our scientists using our big autonomous lab in Boston is a little bit like the Waymo engineer Five years ago going through Palo Alto sitting in the driver seat with their fingers like this right next to the steering wheel, like ready to grab it if the car turns into a mailbox or something.

They are the first ones to push lab autonomy to the frontier, right? When you saw those 30 protocols running on a system, that were the first people to do that. right? And so things break. And that allows us to very quickly speed our development cycle on the autonomous lab compared to companies that really just focus on robotics or on software or things like that because we are doing lab research on our own infrastructure. We are learning really fast about what works and doesn't and very importantly, about how to onboard scientists into autonomous labs, like that is a cultural change. right? And so it involves technical tools to make that easier and faster so that they can still get their very important work done, but they can run it overnight, 1 of the things that our scientists have really enjoyed doing. If you watch like the ramp of protocols getting onto Nebula or big autonomous lab here, it spikes in the afternoon and then people whose experiments run overnight and they come in the morning to data, which spent a while it at the lab, but that sort of the dream has to show up in the morning with a coffee to a fresh data set. So I do think scientists get really excited about this as we bring the barriers down. But again, Ginkgo's team gets to be the [indiscernible] so that our customers of the autonomous lab end up being able to see what's possible and also have a lot of that debugged in advance. I'll talk a minute about our OpenAI project.

If you're sitting in the back of a Waymo, an autonomous car, you tell it where to go. Once you close the door and get out or get out and close the door, Waymo's AI takes over and it tells the autonomous car where to go. So what the autonomous car is solving is the problem of replacing the manual driving, not the directing. Same idea here. When we're solving the autonomous lab problem at Gingo, what we're solving is the manual lab work. not the directing of what lab work to do. And that could be done by scientists as is done every day at Ginkgo as you saw all those protocols on the scheduler, but you could also try to have those experiments run by AI scientists.

And so our project with GPT 5, like I was mentioning, we were doing 100 of these rated 4 well plates. We're giving that data back to the model, it was interpreting the data and then sending in new designs. We have a great archive paper on this. If you Google, if you look at the opening eye blogpost, you can find it. We learned a number of things about this. I think we did some really smart stuff with OpenAI here. I'll just give you 1 quick vignette. So the model is designing the parameters of the experiment, but we don't let it just run anything. We have what's called a [indiscernible] model, which is basically a software defined set of rules, and we have open sources, you can download it, where GPT 5 submitted designs into that, and it had to pass a series of tests for us to be willing to run it. And if it didn't, we would tell if it failed, and it would redesign until it met the test. So simple things. 384-well plate submit 384 wells. The volume of the well is this much liquid do not exceed that amount of liquid or else that's going to spill everywhere, right but more complicated things. Do your experiment in quadruplicate because we want to publish a paper about this and scientists allow to see replication. -- include a set of standard controls [indiscernible] we can fairly compare how you're doing over time. So we put those rules in. But then within the experimental wells like the rest of the plate, as long as it put the right amount of volume, they could do whatever it wanted.

And so across 500 plates in the experiment, we had only 2 that we thought were just total nonsensical designs. And one of them was a problem with our pedantic model, where designed negative volumes of certain reagents to try to squeeze more reagents in under the volume limits. Obviously, you can't do negative volumes. So we added that to the model and it learned not to do that. So Really, I think this is the first demonstration of really more open-ended experimental work, beating state-of-the-art. There's definitely really great ways to take this work in the future that we're going to continue following up on OpenAI basically used us as a cloud lab, right? They paid us to do the data generation and their model was able to send and receive commands and data back from our autonomous lab in Boston. All right. I'm going to sort of end on this next 1 or nearly [indiscernible] is the right company to bring autonomous labs to market at scale. I deeply believe this. This is now apparent to me, in particular, over the last quarter.

We have our cash burn under control. That's why we wanted to guide to that and keep the team and have investors understand how much we plan to invest in this. We have extensive practical experience, automating lab work. This is what we have been doing the last decade plus at Ginkgo. We know what is hard. We know what it takes to move benchwork onto liquid handlers to what are the little tips and tricks associated with each benchtop device when you run it at high throughput and high capacity. All of that information is getting embedded into models and our software to make this really just work magically for scientists as they move to the autonomous lab. I think we're the only ones to do it, and it is a dead mission fit making biology easier to engineer. I'm convinced that the #1 problem in that space right now is the lab work. We are just not able to try enough genetic designs to get good at genetic engineering. That's how I didn't go, that's the whole industry.

Next slide. I do want to mention because I'm sure some folks don't tuning in as scientists or potential customers and so on. A lot of times, we also [indiscernible] should I be worried about this. Obviously, part of my job is working at the lab bench generating this data. And I really like this old advertisement from 1951 IBM. And it talks about how the mechanical calculator or electronic calculator, I should say, is going to do the work of 150 extra engineers. At your company. And what I love about this is if you're not familiar with that device for the younger folks or whatever on this call, that is a slide rule.

So this is back when computation was done manually. And this device, this predates general-purpose computers. This was literally just a device that like added and subtracted and divided in basic arithmetic was going to do the work of 150 engineers, and you might say that device will replace 150 engineers. Now of course, you fast forward 70 years, and there are 100x more engineers than they were back in 1951. And that's because the return on investment on what is in the head of people who understand engineering increased dramatically on the other side of the automation of the manual work of computation. And if you go to the next slide, I very much believe that will be the case for the manual work of laboratories. What is -- it is in sanity that we take people who are PhD caliber, understand all the biology like all the ins and outs of these ridiculously complicated biological systems. They have to understand human biology, 1,800 things, and while they're added -- they have to be extremely careful laboratory technicians in order to move liquids and do this work with great fidelity to even be able to try and test and hypothesize their experiments.

We need to divide those 2 things just like computation did back in the 1950s. And if you do that, I assure you, you will get many, many more genetic engineers, many, many more scientists than we have today when our ROI is limited by the manual work at the bench, and we just got to do it and [indiscernible] going to do it. So please put down your pipes and join us if you're interested. All right. So next slide, that's my email up there. As always, feel free to email if you're excited about this stuff. I appreciate the time today and happy to take questions.

[Operator Instructions] So the first question that we have is from [indiscernible] and this is on [indiscernible] with the planned expansion of racks capacity from roughly 50 to 100 units at the Boston facility. Could you help us understand how this increased capacity is expected to translate into 2026 revenue.

[Audio Gap] our upcoming cloud lab service and then our solution service. So I can speak to sort of the repeatability of those services. The way solutions deals work, there's usually multiyear R&D deals. So there's some reproducibility like across a deal like we do a big program for ARPA age, our long-standing partnership with [indiscernible] science that's been going on for 5 or 6 years. So have some repeatability inside a contract. But each time we do a new contract that is hunting for a new project. Datapoint is a bit different. We are starting to see now, as I mentioned in the call, we're in 10 of the top -- again, I figure it's 20 or 30 pharma companies, where we have that is becoming more repeat business as we're able to basically build trust with those partners that we can serve as that outsourced data generation for their teams. The cloud on, we'll see. I mean that's a new experiment where I want to go after that smaller batch work from scientists at the bench. I think if you look at more band CROs, the folks that like build DNA, express proteins, these sorts of things. I think you do see a lot of repeat business once someone has a has confidence with the vendor. We're obviously trying to do more flexible work, and that will be a new experience. We'll see how it goes. But I'm hopeful that would also look like repeat business on the platform. So of the 3, I think solutions is the 1 where you're really off hunting each time to add new research partnerships with the other 2 are a bit more repeat business. And then just to mention it, that's all on the side of using our system in Boston, what you asked about. But of course, we're also selling our systems. So like we sell that system to a P&L or a pharma company or whomever, when our robotics go in, there's an initial spend on CapEx, but then we have a service and software license that's ongoing over time. That is also repeat more like a SaaS business. And then if you were to -- if we had like sort of specialized reagents, some high throughput things, as the system was used, that would also be a repeat business as well. But the initial CapEx would be one time. right

Our next question is from Brendan at TD. We actually had 2 questions, and so I'll start with the first one. The first 1 is how should we think about U.S. onshoring of manufacturing as a potential tailwind to Rex revenue growth? And what do you think Igowill need to do to maximize share of this trend over the next couple of years?

Yes. So on the manufacturing side, we have been seeing interest from -- on the sort of manufacturing QC, right? So again, what our systems are doing a typical manufacturing environment, you're going to be doing production in larger tanks. The racks are really about integrating laboratory benchtop equipment. Now you do have a bunch of laboratory bench top equipment in manufacturing plants, and it's used to do quality control across batches, sometimes associated with even kind of like semi diagnostics work associated with following up on patients over time in a lot of these drugs. So there's actually a decent amount of lab work tied to post clinical sort of like once the drug is on the market that does keep going in a repeated way. I think our sweet spot there would be being able to handle many different QC protocols on 1 big system, right? So again, the strength is complicated protocols, multistep -- but today, you may be doing at the lab bench. You often have folks at those research centers that are part of the kind of manufacturing team and so on, not sort of open-ended research PhD scientists. And so being able to do sort of like the latest type of assay or something that being able to deploy that out as a QC step I think the Racks open up the door for that. So we are talking to some customers about putting our automation into manufacturing sites. So I think there may be a little bit of tailwind there. We'll see.

Next question is how is Ginkgo Datapoints offering being received among customers? Are there any material tailwinds that you expect for this part of the business over the next 12 months?

Yes, really well. I think we're kind of finding the sweet spot in providing -- so like what's going -- I guess I'll just make this a quick point on the AI side, right? So there's sort of 2 halves to how AI is impacting the biotech industry. One is what I just spent a lot of time on the call talking about, which is reasoning models, and coding models, the same sort of models that everybody is using in basically any information technology space are going to impact the ability for scientists to use autonomous labs and robotics in the lab. That's all being made ease by reasoning models. Separate from that, there are bio AI models.

And the most famous 1 of these is alpha fold that Google came out with, which was a model trained on not human language, not human reasoning, but rather biological language and in particular, in that case, like amino acid sequences from proteins and the structure of those proteins. So there's a lot of work going on in that area. And in order to build those Bio AI models, you need to generate very large data sets with sort of a variety, in that case, say, of proteins and their structures, but there's many other things in functional genomics, antibody developability, other areas where pharma companies are asking us to make those big data sets for their ML teams. That side of the house has gotten tailwinds. There was a lot of -- if you're at the JPMorgan conference this year, folks like [indiscernible] announced their partnership with [indiscernible] there's a whole bunch of companies in the start-up side who are starting to partner with the large pharmas because they have great Bio models. And the reason they have great Bio-AI models is they have proprietary data. It's not just how smart they are at the modeling, it's that they've been generating these large data sets. So that's sort of opening people's eyes up and I think creating a little bit of a wave there. And we're definitely the right -- I'd say the leader in providing data sets to, again, to the large pharma, large biotechs, ML teams that are all ready to go for training and everything else. We can use -- we can do the robot half and the data cleanup hat and they can focus on the biological modeling. So it's been good. I would say Yes, I'm excited about it for this year.

All right. Next question is from at [indiscernible] The question is about basically not just the utilization of racks, but the manufacturing and deployment racks. And so the question is similar to how Tesla used the [indiscernible] to dramatically improve manufacturing efficiency, this Geco exploring any specific strategies, technologies or methods to significantly enhance the production efficiency and scalability of RACs.

Yes, this is something we are starting to think about. I mean mainly because those things take time to put in place -- we did make some good decisions. So over the last 4 years or so, I guess, since we acquired Zymergen where this technology started, we actually did do a like a generational upgrade to the hardware. -- of the racks and made them and that design change was not really about -- in fact, our old backs that we had back in the day, Zymergen or the new ones are compatible. But the system has many less components. And that was all done for design for manufacturability for exactly this reason. We make them today in San Jose, we do final assembly in our -- through a partner, and then we do final assembly and integration with the third-party equipment at our site in Emeryville, California. As we were to scale up and selling more and more of these, I think you will see us invest in like basically larger partners to repeat that manufacturing process. But even as we have it now, we can actually scale pretty decently on this. So it's something I think we want to plan ahead for, but it's not the immediate problem we have. All right. I think that that's all that we have for tonight. So of course, if anyone has any questions in general, they can always e-mail us at investors.

I can't go biox.com. Jason also put his personal e-mail up there earlier, so you can miss a him -- and yes, thank you, and have a great night and hope everyone has a great quarter. Thanks, everybody.

Good morning. Welcome to the 44th Annual JPMorgan Healthcare Conference. My name is [ Yu Huang ]. I'm an associate in the JPMorgan Healthcare Investment Banking team.

It is my great pleasure to introduce our next presenting company, Ginkgo Bioworks. With that, please welcome Co-Founder and CEO, Jason Kelly.

All right. Excited to be here today. I will say I've -- I think I've been having the most fun I've had at JPMorgan in the last 10 years. Over the last few days we had our robotic setup of the Marriott Marquis lobby and had people coming by and learning about autonomous labs and I get to stand there and kind of talk to people about that. And I think it's -- what I'm going to focus most of the talk today on is explaining that to you all because we're seeing a ton of interest. And I think it's a great fit for Ginkgo's mission of making biology easier to engineer. Our view is the technological basis for that in the future is going to rest on this idea of an autonomous lab.

We also were really lucky on the commercial side to win a contract recently under the new Genesis Mission. This is an executive order from President Trump to bring AI into science. And so we won a $47 million contract with the Department of Energy to build out a large 97 robot autonomous lab up at Pacific Northwest National Lab in Washington. And this is actually, just back in December, we opened the first 18 robots, so a smaller version of that big lab and ribbon cut that with the Secretary of Energy, Secretary right there. We had a press conference at the end of December. So, starting to see commercial pickup on this idea. But today, I was going to kind of do a deep dive a little bit because I think this is really the direction Ginkgo is going to build on for the next 5 or 10 years.

So, first, I'm going to talk about why autonomous labs are going to be transformative for biotechnology and pharmaceuticals and even science broadly. Second, what is the technological underpinnings? What makes an autonomous lab different than lab automation, which is something that's been around for a long time in the tools industry. And then finally, how are we going to market with this at Ginkgo. There's really two ways. We're going to be directly selling it like we did to the Department of Energy. We'll build a lab for you. But we also have a large autonomous lab of our own in Boston that we are selling services on top of. So I'll talk about that as well.

Okay. All right. Let's dive in. So I'm a little bit of an IBM history nerd. I really like, like the 1950s, '60s, '70s era of IBM because I think they really were the ones who brought sort of automation into computation. They laid the groundwork for what became the entire tech industry we see today back in that period.

And so this is an advertisement in 1951. It says this is an IBM electronic calculator, and it will do the work of 150 engineers, right? And I really -- I like this because if you don't recognize it, what all those engineers are holding is a slide rule, which is a manual computation device, okay? And I think one of the things I hear when I talk to scientists about automating laboratories, there's some concern about, oh, it's going to take scientist jobs and so on.

And so one of the things I'd like to point out is this device replaced 150 manual computing engineers. We have many more engineers and information technology today than we had in 1950, in fact, because of the automation of computation. And that's because by automating it, we increased the return on investment dramatically for what was in those engineers' heads because they could deploy with much greater leverage after computation was automated, right? So, that makes sense.

And so I want to talk a little bit about a little of that history in computers and some other industries and then what it means for labs. So it's not just enough to automate. So I'll point out -- this is how I kind of frame this in my mind. On the y-axis there, you have the amount of automation. So a low amount of automation is that slide rule or that paper notebook, okay? It's totally manual. A high amount of automation at the top there is that electronic calculator from IBM, all right? And then on the X-axis, we have another thing that's important, which is the flexibility of what the person can do with the system. So with a notepad and that slide rule, you can do all kinds of different types of computation. But the electronic calculator from IBM actually just did at the time, division, subtraction, addition. It was a much more limited set of things. So it was automated but not flexible, right? Does that make sense?

And so the true breakthrough actually came 10, 15 years later from IBM with mainframe computers. And what ultimately broken into all of our lives is the personal computer I show there. The computer gave you both automation and flexibility. And that was because it ran on code. You could put different code and make it different things. Does that make sense? And so it's really that top right corner, the ability to get both automation and flexibility that built the entire tech industry.

All right. So I think this is happening again in another domain right now, which is transportation. So if we rewind the clock 100 years or whatever, we have this beautiful San Francisco trolley cars. We have subways in New York and Boston and so on. This is a high amount of automation. You get in that subway. You don't have to do anything, it takes you around, low amount of flexibility. You better be wanting to go to one of the stops on that subway train because it doesn't take you anywhere else.

Bottom right, a car. We have a huge amount of flexibility, but it's totally manual. You're driving it, but you can drive it wherever you want. So, for the last 80 years, 100 years since we invented these things, we have had nothing in the top right corner. We had nothing automated that would take you anywhere you want. If you've been walking the streets of JPM over the last week, I'm sure you saw a lot of these. We just broke through on this. So we have what we call autonomous cars. And these are automated and flexible. You get in the back seat, you don't have to do anything and you tell it where you want to go, and it takes you anywhere you want. And so my view is that thing is like the computer. We're going to rebuild the entire transformation -- transportation industry on the back of autonomy because it is giving you both automation and flexibility at the same time, just like the computer did back in 1960. Does that make sense?

Okay. So where are we in science and laboratory work in the underlying engine that creates all the drugs in the pharmaceutical industry and so on. So it turns out we actually do, we have automation in labs. We've had it for 30 or 40 years. It's what you see up there in the top left. We call these automation work cells. You can buy them from Thermo Fisher, HighRes Bio, Biosero, a number of specialist companies in the space. And what they are used for is things like high throughput screening. I have 100,000 chemical compounds, and I want to run them against a certain cell assay and find a needle in a haystack. We use them in diagnostic labs. I'm getting in thousands of blood samples, and I want to run the same panel of assays on each sample over and over and over again. So, no flexibility, high amount of automation.

Down in the lower right, a lot of flexibility in the manual lab bench, which amazingly, if you walk into the labs of Merck and Pfizer and so on, you will see lab benches that look quite a lot like lab benches looked 100 years ago in scientific settings. And that's because it gives scientists flexibility. You can pick up a pipette, you can do an experiment. You just read about in the paper last week, right? And I'm going to talk a little bit in a minute about what that looks like and what we can do about it.

But what I want to point out is what we are aiming to build here at Ginkgo is the top right corner. We want to give you an autonomous lab like an autonomous car. We want to give you a lab, which is totally automated, but doesn't just do one thing like the subway, but rather like a Waymo can take you wherever you want as a scientist. You can put in your order for an experiment, and it will do it that day even if you've never asked for that experiment before, all right? So that's what we're trying to achieve. Does that make sense? And again, our hope is if we can establish this, then like the computer or like the Waymo, we're able to rebuild this industry around automation with flexibility, which is very, very empowering.

Okay. And so that's why I want to do it. The key technical question is, how do you do it? How do you get up to that top right corner of high autonomy and high flexibility. All right? So what does it look like? All right. So, right. So an autonomous lab has to allow you to do the same work that you would do in a traditional lab setting. but without the manual work, all right? So, let me just lay out then a little bit on like what's a traditional lab like. Also, I've entered my influencer era on LinkedIn, just so you know since you've seen me.

So, a traditional laboratory looks like this, right? And what you have is, let's say you're opening up a lab to do mammalian cell engineering, CAR-T work or something. You're going to have 15 or 20 pieces of lab equipment. You're going to open a door into this lab. There's 15 to 20 pieces of lab equipment in there, all sitting on benches. There's a number of benches, not just one. There's maybe five or eight that scientists can stand at that have pipettes above them and access to a bunch of different reagents.

And what the scientists are going to do when they walk into that lab every day is they're going to do, I think, three major activities. One, they're going to get reagents and they're going to use pipettes and they're going to manually manipulate liquids and other reagents to make reaction samples. Number two, they're going to move those samples among those 10 or 15 devices in the lab. So they're going to take the sample over the PCR machine. They're going to take it over the HPLC. They're going to take it to a centrifuge. When it's done with one machine, they might take it to a different machine for the next step, right? And then third, every time they put it in a machine, they're going to input the settings on that machine. What RPM do I want the centrifuge to spin at? What's my process in the thermocycler? They're going to use their scientific knowledge to set the settings on the machine, right? That's basically what they're doing, okay? They're doing that in many different combinations, but it's those activities that are getting done. Okay?

So what if we want to replace that lab with an autonomous lab, what do we have to have, okay? First, we need reliable liquid handling. So we need to replace those pipettes with robots that can do the liquid handling and pipetting. It turns out we're actually pretty lucky in this regard. We've had liquid handling robotics being built by companies like Tecan and Hamilton and new entrants like Opentrons over the last 30 years, but actually quite good. In fact, more reliable, don't tell people, than the scientists at the bench at doing liquid handling, all right? They're not particularly easy to use. They're pretty difficult to program. So we'll talk about that in a second, but they do the physical work well.

Second, I need to be able to transport material from one device to another device across all 15 of those devices. Three, I need parameterized control. I need to be able to set the settings on the devices. Fourth, I cannot just have 3 of the 15 devices. I need all 15. I need the whole lab in one setup because I don't know ahead of time which samples needs -- sample might need to go from equipment A to equipment G, and I didn't know that ahead of time. They all have to be on there just like they're all in the lab.

Number five, this one is tricky. You need to support parallel work. So when you go into a traditional lab, the first scientist doesn't go in the morning and then lock the door behind them and say, no other scientists can use the lab but me today. When I'm done, I'll open the door and the next guy can come in. No. A number of scientists all use the lab at the same time. They talk to each other about the availability of equipment. I'm going to use this machine for the next two hours. Okay, great. I'll use it after you. They have multiple liquid handling stations with pipettes so they can do that in parallel. So you need to be able to support a bunch of people all using the same 15 pieces of equipment.

And then finally, it needs to be as easy to use as that lab. So you do not -- it cannot be the case that you need to be a software developer in order to use the autonomous lab if you didn't need to be a software developer and you don't need to be to use a biological traditional lab. Does that make sense?

So, if we want scientists to use this, they need to be able to interface with more like a human language interface and so forth, like we're seeing with these AI models and no-code coding agents rather than having to write code themselves. Does that make sense? Okay.

So that's the list. If you knock that off, then I think autonomous lab will basically be able to do the same thing as every lab you see at Merck, at Pfizer and diagnostics companies in every academic research facility, every lab in the world, okay? It would be able to do that same work, except it would do it 24/7, and it would do it with higher reliability and it would do it with improved efficiency. Okay?

And so that is very hard. Like this is actually a very hard technical challenge, both at the software and the hardware level. So I'll talk about what we've done and our approach to trying to solve that problem with our hardware and software. And it's a journey. I'll tell you where we are at in it today.

So this is the key component you need to understand. This is what Ginkgo's invented. We call it a reconfigurable automation cart, a RAC. It's basically a standard envelope that wraps around one of those 15 pieces of lab equipment in that lab I mentioned. So any piece of third-party equipment that you can buy from any of these million different life science tools vendors, I can wrap in one of these RAC envelopes.

And once I've done that, I can do a few things. There's a six-axis robotic arm here. And there's a piece of magnet -- MagneMotion track, which is a transport system. I'll show you a video in a second that can deliver a sample to that arm. And then the arm can put a sample on to, in this case, that's a centrifuge, that red box. And then underneath, there's electronic box hardware that lets us connect to the centrifuge. We can power cycle it. We can do anything to it. We can set it settings. We can do all the things the scientists could do to it at the lab edge, but all with software, okay? And importantly, these are structured. They're standard. There's different sizes for different pieces of equipment, but the height is all the same. So you can LEGO block them together and put together 3 or 15 like in the example lab I gave you or 97 like we're doing for the Department of Energy into one big setup.

All right. So what does it look like, right? So I think it helps a little bit to see it. I'll talk in a minute about the software. But what you're going to see here is a workflow, in other words, like a protocol being submitted into our system. And this is our system in Boston, where we have 40 RAC carts. So you're seeing a sample. And one of the constraints we have on the system is the samples are passed in SPS format, which is 96 or 384 or 1536-well format plates. That's a standard across the industry. Many types of equipment accept that format of plate. And you see it just got delivered to a centrifuge and then here comes one getting delivered to an Echo liquid handler. This is an acoustic liquid handler from lab site. So this is one of the devices that does the liquid handling, okay?

That's been passed over now to another device that does liquid handling. This is an Agilent Bravo. And you'll see, in this case, it does liquid stamping. So it's going to pick up a whole bunch of tips and it's going to stamp liquid. So one point to make is that first device is from lab sites owned by Danaher now. The second device is from Agilent. And the beautiful thing from the standpoint of a scientist using our system is they don't need to use the third-party software from either Agilent or Danaher. They're able to use our software exclusively to make their protocol and our software does the translation down to the third-party software on all these different devices, okay?

And that's really critical because if you're programming all these different things, you're going to lose your mind. And so this is -- now the sample has gone up to be put on a shaker and then it's going to end up going on to a thermocycler in order to do the final reaction and do the assay.

And a few other things are great about this, I'll highlight. We have cameras in all the carts, so you're getting a record of everything that happened. If there's any kind of error, you can see it. You get data coming off these things about everything that happens. So when you're doing liquid handling at the lab bench as a scientist, there is no record. There's no record if you made a mistake, if you pipetted to a wrong well or anything like that. Every step, every action taken on an autonomous lab is recorded electronically. So we have an entire record of everything that happened on the system as well. So there's a number of things that are just sort of extra benefits in addition to the reduction in manual work to zero. Okay?

So then the next thing I wanted to show is that parallelization. So that was just one run happening in that video. This is a screen capture from our system in Boston yesterday. Each row on this is a different piece of equipment on our autonomous lab in Boston, right?

And what you're seeing across is time. And each color is a different protocol operating on the system. And the reason that you need to worry about this, it's like I said back in the lab earlier. Scientist one comes into a traditional and they say, I'm going to use the PCR machine, they start a PCR reaction. Scientist two comes in and says, hey, I need to use the PCR machine, when are you done with it? Oh, I'm done with it in three hours. Great. Well, I'll start an incubation that goes for four hours and then the PCR machine will be ready and so I'll put my sample on. That's what you're seeing here done computationally. When a new scientist shows up, they submit their workflow, their protocol into the autonomous lab software here at Ginkgo. And it says, can I fit your protocol into the system today? Will the PCR machine be available when you need it? Will the center fuse be available when you need it? Can I rejigger things across all the protocols so that you can drop right in.

And this is critical for a couple of reasons. Most important, it means I can have 17 different protocols running submitted by lots of scientists. And I just want to point out there is no automation system in laboratory automation in the world today other than ours that test this, okay, at this kind of scale. And so first off, it just means you can do parallelization.

Number two, what -- let me tell you what more often happens than not with that busy piece of equipment in the lab. Oh, you're already on it today. That's okay. I'll wait until tomorrow. So the utilization rate for equipment in traditional lab settings is like sub 30%, right? These things are just sitting around never being used. And so -- and again, that's because the manual mental scheduling is not very good, okay, among a team of scientists working independently. But we can just do it algorithmically. So we can make our scheduling almost perfect and try to get the utilization rate up dramatically on all this capital equipment, which is quite expensive in these labs, right?

You can start small, so you might have a lab where there's only three or four pieces of equipment. That's what I had set up at the Marriott Marquis all week. We had four pieces of equipment with the system running, right? Our lab set up in Boston is 40. And that one you see on the right is a schematic of the system that we'll be building, the new system we'll be building for Pacific Northwest National Labs. You can grow this to huge size.

I'm not going to talk a ton about it today, but you'll hear more from us coming up on this. Obviously, a scientist can submit a protocol to an autonomous lab. But of course, an AI scientist could also submit a protocol to an autonomous lab. And so you're seeing an enormous amount of energy from the frontier labs like Google has an actual -- has an AI scientist, OpenAI, Anthropic been announcing more in these areas. And then you have specialist companies like Edison that are working on building an AI scientist.

The one thing all these companies and all their money in data centers and these great AI models do not have is hands in the lab. And so if we want to have a lot of that wind in our sales from AI coming into the biotechnological space and pharmaceuticals, I think it's critical that we give these reasoning models hands in the lab, and this is how you do it. 100%, okay? It needs to be end-to-end. There needs to be not people in the middle. It needs to be an autonomous lab. And so again, more to come on that from Ginkgo, but this is definitely one of the value props. Okay?

So what are those value props for a customer buying one of our autonomous labs? The first, massive overhead cost savings on your labs. Again, my point here is not to replace your work cell that does your high throughput screening. I think I can sell some systems that way. It's a good way to get my foot in the door. It's fine. There's a $400 million to $500 million a year work cell market. That's nice for me. It's a way to get started.

What I want to do is have you close your labs. I want every lab in the United States of America to close and be replaced with autonomous labs where scientists are able to do their work via computers. And I think in the U.S., especially in terms of having competitive labor costs on doing science compared to China. This is why you're seeing this push around the Genesis Mission and AI for science in the U.S. I think we need this improvement in research productivity just to play ball at all. Second, at this point on research productivity, for AI models, in particular, they need very large data sets, much larger data sets than we generate traditionally at the lab bench by hand.

So, yes, sometimes you could use a work cell for that, if you know exactly what you want. But my view is people are going to want different things over time depending on what models they're building, and they're not going to want to make a one-off work cell to generate every data set. You're going to be much better off sitting on top of a large autonomous lab that can do different things. And then finally, that point I just made, these AI scientists are going to need hands in the lab, and this is how it's going to happen. So we're hearing that from some of the pharma companies, you're hearing it described as lab in the loop. We're seeing some demand there. And then the first 2, I think, ultimately, every lab in America can see those benefits. Okay?

So, how are we going to go to market with those? The last thing I want to talk about. The first thing I'll say is when I look at the life science tools industry, like big companies like Thermo and Danaher, they're really predicated on a paradigm that the way we do science, the way we do laboratory work is by hand at the lab bench. They sell you liquid handling by hand tools. They sell you reagents in the form of kits that have little tubes with little edges so you can open them with your hands. They sell benchtop equipment with things like touch screens, so a human can touch it. There's a whole edifice of tools that are predicated on the idea that the way we do this work is manual. If we are successful at building autonomous labs, I think the entire tool stack in life sciences and sciences broadly needs to be rebuilt. And so we would do that at Ginkgo. Okay?

So how do we do that in the near term? Two ways. First, we will build autonomous labs at customer sites. So like we're doing Department of Energy, we're talking diagnostics companies, pharmaceutical companies. We're competing for work cell deals. We're getting in and getting started selling these systems directly. And then the second, which I'm going to spend a minute talking about is we have built a very large frontier autonomous lab, largest one in the world today at our site in Boston that has 40 of our RACs and by May or June should have 100 RACs in that system, and that's the one I just showed you the data from a minute ago.

We offer two services on top of that system in Boston. The first we call Solutions. And if you follow the Ginkgo story, this is something we've done for a long time. where Ginkgo scientists use our technology, our autonomous lab to deliver research outcomes for customers. And then in the last 1.5 years, we launched a new service called Datapoints. This is much more akin -- sorry, in Solutions, we get royalties and milestones and biobus and that sort of thing.

In Datapoints, our customer scientists order lab experiments from our lab in Boston, okay? And we offer that through today a menu of services, and I'll show you them in a second. But I think in the future, my hope is that this gets good enough, they can really use it like a cloud lab, which is a model that's been tried a few times, not successfully, where people could just order whatever experiment they want. But today, in Datapoints, we have a menu that companies order from. So I'll talk about how that's going.

I will make the point that us running -- this is a picture of our lab in Boston, and you should come visit. Running on this large autonomous lab ourselves gives us a couple of advantages. One, we can showcase the art of the possible for customers. So I can be the first one to show that you can put 10 scientists submitting 20 protocols a day on an autonomous system without a catching on fire, okay?

I can do it first so that the Head of R&D at Merck knows it already works. Second, we develop and improve the technology for our robotics and software much faster than traditional robotics vendors and tools companies that don't do science on top of their hardware. So my people are breaking the system every day. So then that means my software teams can be updating and getting bug fixes constantly, not waiting around for a customer to try something and having a much more gapped feedback loop. Does that make sense? Dogfooding, so to speak. Okay?

So, on the Solution side, we've done over 250 partnerships. If you've seen me talk before, I showed this pharma, industrial biotech, agriculture, I'll say, in the last 1.5 years, we tightened up dramatically where we were selling solutions. So we're largely selling it only in therapeutics now and a small amount in agriculture, okay? That's a big part of the way we got cost out of the business.

And -- but we have not stopped doing Solutions deals. You can see a number of new deals, both with existing customers and new customers in 2025 and a lot of work for government R&D projects as well. So a pretty decent year in terms of adding new solutions deals.

Datapoints is our -- the CRO service I mentioned where we generate the data sets customers ask us to make. We have three service lines in that area, one omics. So we do a functional genomics assay. So you can tell us, here's my favorite cell line, make these 5,000 CRISPR edits and then do drug seek and send me back the transcriptomic data in a nice Amazon bucket, all clean from my ML team to do AI work. So this is the kind of stuff you can't get from WuXi today. It is a sort of AI-minded large data, ML-driven CRO, okay?

We also do it for antibody developability and we recently launched ADME. And so I've been really excited to see there's a lot of good energy around Chai Bio's deal with Lilly and like you're starting to see a business model of maybe large pharma will pay a bunch of money to just license interesting AI models. Well, what makes those AI models interesting at those companies is they're trained on proprietary data. Like there was a little era where it was like who has the smartest AI people, but it's very quickly moving to who had what data to train a model for whatever it might be, antibody binding, developability, small molecule, who knows, right? Anything they want to work on. I think the model of companies that get the next big deal with Lilly will do it based on making large proprietary data sets.

And we would be very happy to be the sort of scale AI. This is the company that did a lot of data gen for OpenAI at the beginning and other tech companies. We happily generate data for anybody making AI models so that they don't need to build their own labs, and they can stay focused, be computational teams, hire a lot of good computer scientists and so on and just outsource the data gen to our sort of cloud autonomous lab in Boston.

And so that's really what we're aiming to do with this, and we released data sets publicly. We've released the largest antibody developability data set that's available publicly, same with the DRUG-seq and so on, ADME. And so we'll just keep doing that. And I think this group team at Ginkgo, I'll put in a plug, done an amazing job sort of becoming like a community builder in this emerging AI for bio space. So we ran an antibody developability competition, data points summit like a big meeting. We have the virtual cell pharmacology initiative, and then these data drops are really popular.

So I think we've done a nice job, and I should mention, I should first thing I should have said. Yes, this got launched a little over a year ago, and now we're working with 10 of the top 20 pharma customers generating data for them in addition to some of the like more start-up AI bio folks. And you can see the ones that are -- some of the ones that let us talk about it down below. Not everybody does. Okay?

The other thing solutions and data points do is they provide a revenue base for us while we're building up the direct sales autonomous lab business and that equipment and software and ultimately, even reagents that we'll sell to those customers that buy our RAC carts. You can see this is just from our last earnings call. We reiterated our total revenue for the year, $167 million to $187 million and services are a substantial portion of that.

I will also mention we have greatly reduced our cash burn over the last 1.5 years. This is like a lot of pain at Ginkgo. We brought our cash burn down 73%, while still ending last quarter ending Q3 with $462 million in cash and no bank debt. So I really like our position in terms of having a good solid runway and burn under control as we move into this leadership role in autonomous labs.

So, this is the lab in Boston. Again, I encourage folks to come visit. We're down in the Seaport, if you want to come see it. Ginkgo is the right company to lead in this transition to autonomy in scientific work at the lab. We are well capitalized. We have a reduced cash burn. So we have a long runway. We have extensive practical experience. So we spent the last 10 years trying every way under the sun to automate and scale lab work. So we know where the bodies are buried. And we incorporate a lot of that directly into both the design of the RAC carts, but more importantly, into the software that helps when a scientist pushes a protocol onto the system, there's a lot of checks that include the test of knowledge Ginkgo has built up over the years.

And then finally, we're mission-driven. We're not giving up on this. This is -- we believe that biotechnology has not had its IBM moment. We are still very much living in the manual era. I think if we can overcome that, it will be some of the most important work we can do.

This is the ad I would like us to be thinking about 50 or 60 years from now that we were able to unlock the programming of DNA like we were able to unlock the programming of computer code through automation and removing manual work. And so if you're a scientist listening in today, please put down your pipettes and join us in the autonomous lab.

All right. Let's grow the world we want to see. That's my e-mail up there if you need anything, and I'm happy to take some questions if folks have them. Thank you.

Thank you, Jason. A question from the floor.

[indiscernible]

Yes. So the question is what's the longest piece of DNA that we could synthesize in the automated workflow. So the way -- and I talked about this a little bit at the beginning, but I think the easy way to think about it is the system is not designed to do like one activity, right, whether that's DNA synthesis or protein expression or high-throughput screening or anything like that. It's really designed to replace a lab. And so if you had a protocol that was great for DNA synthesis and use, say, an Echo, a bunch of thermocyclers and so on, we would be able to replicate that protocol on the system. And so we do happen to do a chunk of that at Ginkgo. We've built very -- I don't know, probably, I think using our old Gen9 technology, we've probably built things that are 50 or 100 kV.

But that's not the point, right? The point is that whatever idea you had for a protocol, you should be able to do. And I want to move away from Ginkgo has some special proprietary protocol that's excellent, but rather that I provide the infrastructure for other people to develop great protocols. Does that make sense?

Yes, yes. [indiscernible] Some of the smaller molecule where there could be a variety of inventory [indiscernible], which has been a traditional hurdle for automation [indiscernible].

Yes. So I think important -- there's an important point there, which is in a data center, for example, and those computers, it's all information moving around. But with a lab, there is atoms moving around. And so one of the things that needs to happen is we will be loading atoms into the system, okay? And so today, that means I have a team who are basically preparing reagents in sort of reagent plates that are going into the system, and those are then able to be part of reactions that are ordered custom by scientists.

I think what will happen over time, this is back to my point about needing to rebuild the entire stack of sort of tools in an autonomous lab world is we should have a cart that has every reagent you might need. That's sort of part of this. And as you order, it just gets drawn out of there and it's like a usage-based pricing, right? And places already do this, like you can get a freezer from NEB or Thermo. It's just your people pulling out the stuff and getting charged by NEB and Thermo. I think you'd have a similar idea with these systems.

Yes, yes. So, a follow-up on our conversation yesterday perhaps. Software integration. So, for this to work, you will need to hardwire into a lot of third-party equipment that was not designed to -- it doesn't have an API on it. And in fact, some of the vendors, I'll try not to call out names, are quite aggressively trying to make it very difficult to integrate their systems by others because they want to charge you for the software. So how are you overcoming that? And what's the -- what are the -- because you need an API standard for that.

Yes. This has long been desired. There's very few standards across benchtop equipment. And again, I think this is because the paradigm is it's just a single device sitting alone interactive and the human is the glue between everything. And so nothing has to talk to each other. And so there's never been an API standard. We're actually been lucky we have the SPS standard on the plates. Chemistry doesn't even have that. And so -- so that, I think, is -- you're 100% correct. It would be very obvious that we should have API standards for third-party equipment.

I think things like that will come eventually in the meantime, and there's open source efforts like PyLabRobot and things that are working on this. We're -- in the meantime, we basically, as customers ask for equipment, we do the work of bringing it on. Sometimes it's easier than other times. Sometimes there's an API, sometimes there's not. Sometimes we write our own drivers directly into the equipment like we do what we have to do.

If it's really impossible, then ideally, there's a different -- for many of these functions, there's a second company that has the same piece of equipment basically, and we can swap. But we've had pretty good luck integrating things. And that's a 1- to 2-month process to bring on. And you only do it once and then that equipment is on. So and we do it all the time. I think we're, I don't know, nearly 100 now.

All right. Question. So how do you see the autonomous labs being adapted by biopharma companies? Or in other words, what will be drivers for adoption?

Yes, it's a key question. So, right now, the market today for automation, like the top left corner, the work cells is people buying high-throughput screening systems or diagnostic labs and so on. And so that is bought every day of the week, not every day of the week, but there are RFPs coming out all the time. And again, it's call it, a $400 million to $500 million a year market served by three incumbent vendors. And I think we can play in that market.

So one of the ways we get in is we just sort of -- it's harder, you're not an incumbent, but we kind of muscle in and we say, hey, look, our systems are more general. They can turn into an autonomous lab someday, even if you only need this thing to do one thing, let's get started. And then the second area is basically at all these pharma companies, there's someone in charge of AI. And so that person is thinking about the labs differently than the historical drug discovery person is thinking about the labs. And there, depending on the company, have some level of mandate to generate data in-house in a new way. And that, I think, are more direct autonomous lab sales. But it's going to take some time, I would say. And that's why I'm excited about our services is because in the meantime, I grow this giant lab in Boston, and that also is a great way to show people that art of the possible in the pharma industry.

And the other one is, I'll say, is in academia. So, academia is another place where we're sort of doing this thing, which I think is a bit pathological, which is we are underpaying grad students and post docs to be low-cost manual labor in the United States and then they graduate and there's not actually that many places that can afford to do manual labor at the cost it is to actually employ an adult full time in the United States of America as you have this like crazy dropoff and basically only pharma can support it, and it's pretty messy. If instead, all of our scientists were running on autonomous labs, then I think you might end up with science and R&D much more broadly distributed across industries because you wouldn't need to carry the cost of manual scientific labor, if that makes sense. Sorry, you say something.

Comment outside of biology that, that transition has happened, you can't enter [indiscernible] because you would...

In chips, yes.

The whole thing. And so the idea [indiscernible] that only left in biology.

Yes. Yes. There's -- yes, the physical transition that happened in semiconductors and electronics, I think, is a good history, right? Like yes, there was an era. There was an era of vacuum tubes and a lot of manual work around electronics, and we squeeze it all out.

Just one last question. There have been a lot of companies like Emerald Cloud Lab and Strateos. They tried and struggled with a cloud lab business model. What do you think it takes to open up that go-to-market in biotech?

Yes. So this is something I thought a lot about. So we have this big lab in Boston. I want to make it available. You had folks try these cloud labs previously. And I think the issue -- the basic -- the main problem is like a business model problem. If you go into a therapeutics company and you say, hey, I've got this lab, I can do whatever you want. Why don't you order from me, you won't have to build a lab. It's great. It's like what you do with your data centers at Amazon, like don't you want to not have all this capital infrastructure in-house. They say, great. Sure, I would like this protocol. They're like, okay, cool, let's do it.

And their question -- first question is, have you done it before? And your answer is like, well, I'm a cloud lab, like no, you're asking me for your unique protocol. Of course, I haven't done exactly that protocol before. That's the whole point of this thing. And they're like, well, if you haven't done it before, I don't trust that you'll be able to do it, okay?

And so that was the -- that's why that first generation didn't really work out. And so the way you get around that is you offer a menu of things that you have done before, that's traditional CRO, but you do it on top of a flexible lab and then you kind of -- you boil the frog. You start to say, hey, you do get this from us. Do you want this slight variation? Did you know? And then eventually, I think people come around. But it will be a journey.

The other way you'll do it, and I'll make one last point is if they have an autonomous lab in-house, they've gotten familiar ordering from it, I can be overflow. And so I'm hopeful that works, too. So, all right.

I think we're out of time, but thank you so much, Jason, for the presentation.

Yes. Thanks, everybody. Yes.

[Audio Gap] Manager of Communications and Ownership at Ginkgo. I'm joined by Jason Kelly, our Co-Founder and CEO; and Steve Coen, our CFO. Thanks, as always, for joining us. We're looking forward to updating you on our progress. As a reminder, during the presentation today, we'll be making forward-looking statements, which involve risks and uncertainties. Please refer to our filings with the SEC to learn more about these risks and uncertainties, including our most recent 10-K.

Today, in addition to updating you on the quarter results, we're going to be providing insight into how we believe AI models will impact biotechnology, how our tools are positioned to support those impacts and how those tools are winning us new deals with customers. As usual, we'll end with a Q&A session, and I'll take questions from analysts, investors and the public. You can submit those questions to us in advance via #ginkgoresults or e-mail [email protected]. All right. Over to you, Jason.

All right. Thanks, Daniel. Ginkgo's mission is to make biology easier to engineer. We always start with that. I want to highlight the 3 big objectives for us going into 2026. And I'm going to give you a little more detail on these today. The first is to deliver the robotics and software that bring autonomous labs on-prem, in other words, at our customer sites so that they can run them themselves through our tools business.

And we really grew into that sort of tools business model last year. But this robotics and automation and AI controlling it, I think, is having a big moment right now, and I think we've got the right tool stack to bring that to customers. Second, we want to expand sort of our frontier autonomous lab here in Boston. We have the largest RAC install in the world. I want to keep it that way. We'll be continuing to expand that even as our customers build larger systems as well. And we want to use that to be able to show just the art of the possible to customers, what you can do when you have ultimately hundreds of pieces of equipment, all connected in a single robotic setup that can be controlled by AI. And so I'll show a few photos and what we're doing there coming up.

And then finally, our 2 big services, our CRO services, solutions and data points. We want to offer best-in-class services, best on the market services to customers there by leveraging that in-house robotic infrastructure. And that helps us kind of, again, demonstrate what's possible with those robotics and also offer great services to customers. So you're going to get to hear about all 3 of those things later from me.

What you're not going to hear as much about in '26, but I'm very proud of us pulling off in '25 is this chart, dramatic reduction in our quarterly cash burn over the last year, doing all that while still maintaining a strong margin of safety in our cash position. So after Q3, we have $462 million in cash and cash equivalents and no bank debt. So I think this is really, again, particularly in what's been a tough biotech market over the last few years, puts us in a very, very strong spot as a growing tools company. And so again, very proud of the team for doing that.

You're going to hear less about cost takeouts in '26 and a lot more about our investments for growth and what we're doing for customers as we expand in AI and automation. All right. With that, I'm going to pass it to Steve, but looking forward to giving you more detail in a moment.

Thanks, Jason. I'll start with the cell engineering business. Cell Engineering revenue was $29 million in the third quarter of 2025, down 61% compared to the third quarter of 2024. As previously disclosed, cell engineering revenue in the third quarter of 2024 included $45 million of noncash revenue from a release of deferred revenue relating to the mutual termination of a customer agreement with Motif FoodWorks, one of our platform ventures. Excluding this, revenue in the third quarter of 2025 was down 11% from the prior year period.

In the third quarter of 2025, we supported a total of 102 revenue-generating Cell Engineering programs. This represents a decrease of 5% in revenue-generating programs year-over-year. This decrease can be primarily attributed to the ongoing program rationalization as part of our restructuring activities. Turning to Biosecurity. Our Biosecurity business generated $9 million of revenue in the third quarter of 2025 at a segment gross margin of 19%. As a reminder, segment gross margin excludes stock-based compensation.

Turning to the next slide. It is important to note that our net loss includes a number of noncash and other nonrecurring items as detailed more fully in our financial statements. Because of these noncash and other nonrecurring items, we believe adjusted EBITDA is a more indicative measure of our profitability. A full reconciliation between segment operating loss, adjusted EBITDA and GAAP net loss can be found in the appendix.

In the third quarter of 2025, cell engineering R&D expense decreased 8% from $55 million in the third quarter of 2024 to $51 million in the third quarter of 2025. The 2025 period R&D expense included a $21 million shortfall obligation related to our multiyear strategic cloud and AI partnership with Google Cloud. In October 2025, we amended and reset the annual commitments in future years and settled this shortfall obligation for $14 million.

Cell Engineering G&A expense decreased 47% from $23 million in the third quarter of 2024 to $12 million in the third quarter of 2025. These decreases were all driven by our restructuring efforts.

Cell Engineering segment operating loss was $37 million in the third quarter of 2025 compared to a loss of $5 million in the comparable prior year period. The increased loss year-over-year was due to 2 factors. First, as previously mentioned, the third quarter 2025 expense included a $21 million shortfall related to our Google Cloud contract that was subsequently settled. Second, as previously mentioned, the third quarter of 2024 included $45 million of noncash revenue from the Motif contract termination.

Biosecurity segment operating loss improved 21% in the third quarter of 2025 compared to the prior year comparable period. Moving further down the page, you'll note that total adjusted EBITDA in the third quarter of 2025 was negative $56 million, which was down from negative $20 million in the third quarter of 2024. Again, this year-over-year decline can be attributed to the previously mentioned Google Cloud shortfall expense recorded in the third quarter of 2025 as well as the Motif related noncash revenue in the comparable prior year period.

So turning to the next slide. We show adjusted EBITDA at the segment level to show the relative profitability of our segments. The principal differences between segment operating loss and total adjusted EBITDA related to the carrying cost of excess lease space, which you can see was $14 million in the third quarter of 2025. This cost represents the base rent and other charges related to leased space, which we are not occupying, net of sublease income. This is a cash operating cost that is not related to driving revenue right now and can potentially be mitigated through subleasing.

And finally, cash burn in the third quarter of 2025 was $28 million, down from $114 million in the third quarter of 2024, a 75% decrease. Cash burn does not include the proceeds from ATM sales during the quarter. The significant decrease in cash burn was a direct result of the restructuring.

Now turning to guidance. In terms of outlook for the full year, we are reaffirming our overall revenue guidance for 2025, totaling $167 million to $187 million with Cell Engineering revenue to be $117 million to $137 million and Biosecurity revenue expected to be at least $40 million. In conclusion, we're pleased with the continued improvements in cash burn and cost reduction. In the fourth quarter, we will continue to execute against our core objectives while navigating continued uncertainty in the macro environment. And with that, I'll hand it back over to you, Jason.

Thanks, Steve. All right. So we'll start the strategic review. There's 3 topics we want to cover today. The first, I believe AI models are going to impact biotechnology fundamentally in 2 big ways, and I think Ginkgo is well positioned to sell tools into both of those. So I'm going to talk about that. Second, we are continuing to offer that Research Solutions business on top of our in-house robotics platform at Ginkgo. And we had 2 big wins in the last quarter. I want to touch on that briefly.

And then finally, we are expanding our sort of frontier autonomous lab here in Boston, big RAC set up. So I'll show you some photos and a little bit of background on what we're doing there. And please do come visit. I'll mention that when we get to that section. But if you want to come see it, yes, you're very welcome. All right.

So let's dig in on really how AI is impacting biology. Before I do that, I do want to remind, we made, again, over '25 and the second half of 2024, we made a big shift in the business where we went from just offering research solutions, which is the left-hand side of this chart here. These are these types of research partnerships, we get fees and we get downstream value share, we get royalties or milestones in the sort of ultimate end products that our customers are developing, leveraging our platform.

It's a very close partnership with the customer. There's a lot of our scientists involved as well as our robotics. We've done about 250 of those R&D partnerships over the last 8 to 10 years. That is a business we will be continuing. But in the last 1.5 years, we expanded into the tool space with our data points, automation and reagents businesses.

And so I want to spend a minute talking about how AI and what's really been coming down the pipeline, I think, offers us a nice niche and entry point into the tools market where we really have, I think, the sort of category-defining technology. So first, why is AI important right now in sort of sciences in general and bioscience in particular? So this was the America's AI Action Plan came out of the White House in the last few months.

There's one specific section I draw your attention to, which was investing in AI-enabled science. And the general idea here is to have AI reasoning models, leveraging and they highlight automated cloud-enabled labs, and that's why I'm excited to share more on what we've been building here in Boston, which I think is a great example of one of these cloud-enabled labs. That if you connect those 2 things together, you can potentially change how science is done. And the idea is the reasoning models could be thinking and the labs could be doing that lab work, and I'll talk about that more in a second.

And the reason this is important is shown here, I think we're -- particularly in the biosciences are going to be the first sort of battleground for AI-enabled science, if you look at what's happening between the U.S. and China. So there was a New York Times editorial just a few months ago saying China's biotech is cheaper and faster. I think that's largely true if you think about the traditional way we're doing biotech today, which is you basically have well-trained scientists working by hand in laboratories here in Boston, it's in the Kendall Square area here down the street, it's also in South San Francisco and California, San Diego, Research Triangle, North Carolina, a few hubs in the United States where you have sort of scientists working by hand doing biotechnology research.

For a long time -- if you go back and stay back a slide. For a long time, that was -- we had an advantage over China just in the sense that our people were better trained, and we had access to sort of like better facilities and things like that. That advantage has largely evaporated over the last 10 to 15 years. There are just as good academic institutions, just as good start-up ecosystem and so on in China, and there are more scientists trained and they're paid less, frankly. And so I don't really see where we have an advantage on physical labor anymore versus China.

And so I was really excited to see Senator Young, who's sort of heading up the National Security Commission on emerging biotechnology, put in a number of bills around this topic, NSF launched $100 million AI programmable cloud Labs initiative. And the big theory behind these things is if we're going to compete with China in biotechnology, we need to do it with robotics rather than hands at the bench. And if we don't do it, I think you're going to see what we've seen over the last 2 or 3 quarters where an increasing number of the early-stage biotech start-ups that are being acquired by large pharma or invested in by US VCs are based in China.

And so I think if we're going to turn that around, both for biotechnology and for science at large, we need to do it by investing in robotic infrastructure. And I think that's not lost on the U.S. government. And I think Ginkgo, if you go to the next slide, has exactly the right technology for that. And so I've shown these before, but these are reconfigurable automation carts, our RAC carts. And this is the first big area where I think AI is coming into biotechnology. And so this is around reasoning models. So again, I think like GPT-5 from OpenAI and so on.

These are Gemini from Google. These are these models that are able to think over a period of time, come to sort of a conclusion based on what you've asked them to do and either they can write code, they can do other things, they can kind of use browsers and tools to go off and do sort of a multistep operation and come back and bring a result to you.

I think the first big frontier here is going to be connecting those reasoning models to physical automation in the lab. And the reason this is necessary is if you think about how science gets done outside of areas like math or theoretical physics that are purely kind of people thinking about stuff, it's purely intellectual, the majority of science, experimental physics, experimental chemistry, experimental biology and so on is moved forward by lab work, right? Like we have a hypothesis. Scientist has a hypothesis about how some disease works or whatever. But the only way they really know the answer is to go off and run carefully constructed laboratory experiments.

And so if you want these models to really be AI scientists, and you're seeing FutureHouses that are had a great new model come out yesterday or now called Edison Scientific, super excited about that. Those models need to be able to do experiments. And if you go to the next slide, the way they're going to do experiments is using the technology like what we built at Ginkgo. This is our reconfigurable automation carts. Each cart has a piece of lab equipment, a robotic arm and a plate transport track, and I'm going to spend a minute later showing you this in action.

But basically, what it allows you to do is sort of LEGO block together, if you go to the next slide, 5 of these in a linear setup, 20 of these in a circular setup or here's a setup, we actually just sold one of these systems with 97 carts on it in one giant setup. And so the idea here is to be able to connect ultimately hundreds of pieces of lab equipment, LEGO block style into a huge setup where the whole thing is software controlled. And the reason it's important that it's software controlled is just like these reasoning models can write code for Python or whatever, right, for a website, they're also able to write code to run this automation and design and execute experiments and interpret data.

And so if we want to have the sort of AI-controlled science, these cloud-enabled labs, this is what they look like, and you really need a new hardware technology like what we've built with the RACs to do that. So I think we're extremely well positioned for this, and you'll see us leaning in heavily here in 2026.

The second area where we're seeing AI applied to biotechnology is in using the same kind of like math and compute that was used for the reasoning model. So large neural networks, GPUs, that whole infrastructure, except instead of training those neural nets on human language and human reasoning and code and programming, things that humans kind of read and understand and interpret, you train them on biological language. So DNA, amino acid sequences from proteins, the language of life, the language of living organisms. And you do the same type of training, the same infrastructure, but these things learn to speak biology.

And so this is a more nascent area compared to the reasoning models when it comes to AI and biotech, but I think it's also going to be extremely important. And with our Ginkgo data point service, we really want to build the community in that area. So we highlight here our antibody developability competition. This is just, I think, at the end of November, going to wrap up. So you should -- if you go to the next slide, you should check it out. You can go to datapoints.ginkgo.bio, you can sign up. We have more than 200 teams now competing in that competition.

And the idea there is build a model like the one I just mentioned, like train a model on data for the developability of antibodies. In other words, is this antibody sequence going to work well as a drug? Will it be soluble and so forth? There's other -- is it not immunogenic. That is a very valuable feature set for biopharma companies.

So if you're a bioinformatician or you're a start-up that has a great new AI model, I encourage you to compete in our competition here. We basically generate a large amount of developability data. We shared some of that with the community. We kept some of it back as like competition set and your job is to predict the held back data, and we'll rank who does the best. The other thing we're doing to help build the community is we're releasing data sets for free. Again, you go to our website there and download the sort of AI/ML-ready data sets. They're an example of the sort of data that we generate on a fee-for-service basis for customers through our data point service.

So go download those, play around. If you wanted to buy data from us, we're very happy to do that. And we're really here to build a community of folks who are trying to train AI models using biological data. And so really excited about this as a sort of a nascent area for AI applied to biology.

All right. Second thing I wanted to talk about -- so those are the 2 big buckets for AI, again, reasoning models, controlling robotics in the lab and then basically neural nets trained on biological data. And they're both involving AI, but they are different. And so Ginkgo will play there through our automation in the first one and our data points for the second one. All right.

So next category. This is now going back to that left-hand side of this chart, the business that Ginkgo sort of like primarily focused on over the last 10 years, our Research Solutions business. We are still doing these. If you are looking for sort of breakthrough research in any of the areas that could basically leverage like high-throughput biotechnology, I think Ginkgo is still a very good call. If you go to the next slide, we won a couple of great deals in the last quarter. BARDA awarded to us and our partners, $22 million around the manufacturing of monoclonal antibodies, bringing that back in the U.S., making that cheaper, particularly around producing key medical countermeasures. So I think this is both important for national security and also important for reducing the cost of manufacturing drugs, particularly biologics drugs. And you heard the administration talking about this recently on the regulatory side to try to lower the cost of biologics. This is a technical approach to dropping the cost of biologics.

If you go to the next slide, in the agricultural sector, very happy to extend our partnership. This partnership has going on for 5 years with Bayer. We're really working on engineering microbes, if you go to the next slide for the production of fertilizers. And if you remember -- this is actually, I think, a pretty amazing story. So if you think about like elementary school biology, you learned about crop rotation, right? So you would rotate in a legume like soybeans or peanuts or things like that, and they would refertilize the soil. And then you plant something like corn and corn largely takes fertilizer out of the soil. So that's sort of how we used to do it.

And then in the early 1900s, we invented the Haber-Bosch process where you take nitrogen out of the atmosphere by burning natural gas and combining the nitrogen with that and producing synthetic ammonia. And then that goes out to the tune of many billions of dollars a year and about 4% of global greenhouse gas and so on. So it's a big, big chemistry industry, and it's largely based in China. That's a huge input into things like corn farming. Well, those crops that you rotate in like soybeans and legumes, they're able to refertilize the soil because they have microbes on their roots running that Haber-Bosch process, taking nitrogen out of the air, fertilizing the crop.

So I'm really happy to see this project continuing. I think it's the kind of world-changing stuff that only biotechnology can do in the physical world. And so really excited to keep that going.

All right. Again, if you're in agriculture, industrial, biotech, biopharma, you want to try large-scale biotech on your problem. I encourage you to call us up, and we're happy to have our scientists work with yours to leverage the infrastructure here at Ginkgo to deliver that. I really like this photo. This is 2 of my co-founders, Reshma and Austin, in the lab just a few weeks ago. The reason I bring this up is Reshma and Austin had not been in the lab prior to a few months ago for like the last, I don't know, 10 or 15 years, he started the company. And the reason they're back in the lab is because what we've been doing on the automation side at Ginkgo, building out our RACs set up here in Boston has gotten sort of ridiculously exciting over the last 6 months or so.

So if you go to the next slide, I want to talk about what we're building with our frontier autonomous lab. We're getting a ton of interest in this right now, both from customers and even just internally. So we've been expanding our setup here in Boston. So you can see our RAC carts there in the photo inside of one of our kind of big foundry base here in Boston.

If you go to the next slide, we're going to have about 45 instruments -- 46 instruments on this setup, like 10 carts are getting installed right now to bring it up to 36 RACs. Ultimately, I'd like to get it in that room to about 100 RACs. You can see a photo on the left of one of the RACs going in. That's pretty exciting, right? So this is us putting a new piece of equipment on, that video is sped up, but it takes just a couple of hours really to get that device on the setup. This is because we have invested in productizing the cart hardware so that we have greatly simplified.

And if you're not in the laboratory automation business, you may not know this, but integrating equipment into laboratory setups right now is done as a custom job. You basically pay an engineering firm and they spend months making CAD designs and they build you this kind of Rube Goldberg machine device. We've taken all that and standardized it with carts, turned it into a product that you can just buy off the RAC and install in these big setups. And so we're really excited to be building this out.

The picture in the middle there that's running is actually a RAC inside of an anaerobic chamber. We built this for Pacific Northwest National Lab, PNNL. It's like, I think, 14 or 18 of our robotic arms and RAC setups inside of an anaerobic chamber where people can't go in because there's no air. And so very exciting, big setup. We're excited to see more customers bringing those in-house.

If you go to the next slide, I just want to kind of show what it looks like. So each row in that is a different piece of equipment. Those red bars are when a sample is interacting with that piece of equipment. So that's sort of like the time line of a protocol being submitted. So a plate, and in this case, this is a standard piece of labware, that little plastic rectangle you see moving on our track system is a 384-well plate. So there's 384 samples in there. It's being put on to a centrifuge in this video here. So that plate goes in and then that centrifuge is going to spin. This plate now is then after the centrifuge step being delivered to an echo liquid handler. This is an acoustic liquid handler that's able to move liquids with sound.

And what it's going to do is it's going to set up the reaction conditions on each of those 384-well plates as programmed by the software that is telling the system what to do. And importantly, again, to nerd out a little bit, each piece of equipment, this is like a Bravo liquid handler, that was an echo. Each one has its own piece of sort of proprietary third-party software that's kind of a pain to deal with, honestly. And so what we've done as part of the RAC system on the software side is we have connected into each piece of hardware with our software. So you're able to write a multistep protocol, what you're watching here, this particular protocol is protein -- cell-free protein expression.

What you're able to do is connect many different pieces of equipment in a single protocol where you're controlling in a parameterized way each piece of equipment. This is a shaker, and then it's going to go on finally to a piece of assay equipment, at thermocycler to go kind of complete this reaction. And so all of those steps are encoded in the Ginkgo software. And then the scheduler and larger system goes and talks to all the equipment in a seamless way. So your scientists aren't dealing with 18 different types of software to do an 18 equipment run. That's a really big deal, and it also means it can be connected back to reasoning models to do that type of design and experiments as well.

If you go to the next slide, we are able, like I mentioned, to set these up quickly. So this is these 10 carts that have been coming in. This is like literally from last week. And so if we've already have the equipment that's relevant, and again, we're at 45 pieces of equipment now on this setup for the protocol you want to do, if you go to the next slide, we are able to then demo it for you in pretty short order. So if your group has been thinking about just automation in general, you can try our system. If you want to see what it's like as a scientist to interact with a system through a language model, like we have human language interface now to that setup, so you can play around with that.

And then finally, if you wanted to have an AI reasoning model controlling this setup to work on a problem of interest to you, we can do that, too. And what's exciting is we do all that just on our setup here in Boston. It's very inexpensive for you. You're not buying a bunch of equipment or anything else. And you can see if it works, like try it before you buy it, right? If it works, then we're very happy to install this in your lab so that your labs could have the same sort of just very latest scale in terms of automation and AI that we're running here at Ginkgo. And I'm telling you, it is very, very exciting. It's working really well. So I do think folks should come and try it. And if you just want to come visit, please do just shoot me a note, and we're happy to do that and happy to come by.

All right. That's what I had today. Happy to answer questions about all that, but super excited. I think we've done -- the team, again, a big round of thanks for 2025. It's a very difficult year, bringing down our costs in a huge way while maintaining that sort of large margin of safety. And that's what's allowing us to really now invest for growth in the future, particularly in this area of building out basically the automation and AI tooling for biosciences. And I think that's going to be the niche that we grow into in the coming 5 to 10 years in a big way. So excited for your questions, and thanks again.

Great. Thanks, Jason. As usual, I'll start with a question from the public and remind the analysts on the line to ask a question please raise their hands on Zoom, and I'll call on you and open up your line. Thanks, everyone.

All right. Let's get started. So the first question was one that we got on Twitter from an account at [ @DavidJu ] tweets, and this question is, can you comment on the extent of Ginkgo's exposure to U.S. government business and how that has been impacted by the shutdown?

Yes, I can touch on that. So short answer on the shutdown has not had a big impact on us. So sort of the areas the grants and funding there keeps slowing during the shutdown. I would say, in general, though, we have a good amount of exposure to the government overall. So between our cloud security business and then things like the new BARDA award, you'll see us announcing some recently also ARPA-H awards. We've been doing very well, I guess, I would say, with bringing in research partnerships with the government. So overall, I think hopefully, we're even doing more in the future with some of the sort of cloud labs work and investments I hope to see from sort of government labs around automation, but the shutdown doesn't impact us.

All right. And our first question from Brendan from TD Securities. He writes, how do you see the broader development or rollout path ahead for the RAC system over the next 18 months? Are there any additional validation steps or accounts to land that you expect could really unlock this opportunity and widen the commercial funnel for this over the near term?

Yes, I can touch on that, too. So first of all, I think what's super exciting about the RACs, and again, I tried to mention this, but there's sort of like walk-up automation, companies like Hamilton and so on, where you're getting like a liquid handling deck, and that is a very productized offering. But then there's integrated automation, which basically means there's a robotic arm in the middle of a bunch of equipment. And the key there is one piece of equipment maybe does the liquid handling, but then you got to take your samples to the next piece of equipment.

And you saw in the video, the plates moving on that track and getting delivered to 6 or 7 different pieces of equipment in that single protocol. You might have protocols that interact with 15 different pieces of equipment. And a human, by and large, is doing that in 99% of the labs that are out there. There is a small niche industry around integrated automation for things like high-throughput screening, where you put an arm in the middle of 15 pieces of equipment. That is built basically application specific. In other words, it's a design of a setup just for the one thing you want to do.

Our carts are not like that. They are productized. They're coming off the line the same, and then we are just connecting them so that you have whatever equipment you want initially and then actually able to expand that equipment over time into bigger and bigger setups. So that's something you just cannot get with the traditional integrated automation. So what I'm excited about on a rollout basis is continuing to scale up our manufacturing of these carts, bring the cost down, like turn that again into more and more productized offering.

But then on the sales side, it's basically getting folks to see this distinction between application-specific work cells that they buy today and general purpose autonomous labs like what I was showing you there with our frontier lab here in Boston. It's that adoption, this idea that automation isn't the thing you build for one application and then literally decommission and throw away 3 or 4 years later, that's what happens with these systems. But something that just keeps expanding over years and then ultimately replaces hopefully, tens of thousands, hundreds of thousands of square feet of laboratory benches because we're just going to move off that system.

We have to move away from the bench as the general purpose laboratory infrastructure to the automated bench to the autonomous lab. And that's the transition that I want to drive. So if you're looking for milestones, I want internal milestones at Ginkgo. So like one of the things I want to see is 50-plus scientists internally at Ginkgo ordering simultaneously from our automation system in a single day. That's the thing I think I can get -- have happening in 2026. That's something that's never been seen with an automated lab previously. So there's internal milestones.

And then what I would love to see -- we're starting to see this on the government side, but I'd also like to see it in the private sector ideally with large biopharma, a similar -- like a purchase of a very large system with an intent for a general purpose autonomous lab. And so those are kind of my 2 big things I'd love to see in 2026. Us demonstrating just what you can do with already having one of these kind of autonomous labs and then a large biopharma leaning in and making a purchase for one.

We'll still sell opposite the work cells. That's what we're selling today into, but I would love to see someone kind of lean in on the dream of the big general purpose autonomous lab. I think it's the time for it. And we're going to prove it either way at Ginkgo. But I think our customers will be sort of adopting that mindset soon, too, that's my view. It's just so much easier to use automation with the AI stuff. And so I do think that's going to just bring the barrier down massively for this in the industry.

Cool. All right. And then Brendan had one more question, which was, as you look at the current revenue mix between cell processing, as he said, cell engineering and biosecurity and then consider your internal assumptions about the AI tools and RACs rollouts, what do you see as the ideal revenue mix for Ginkgo by 2030? What has to happen to get there by 2030?

Okay. Yes, it's interesting. I mean -- so my dream by 2030 is we're starting to put a bunch of benches to bed. And so my expectation, like if I think about the balance between -- let's leave biosecurity, I'll come back to that in a second, but between like the sort of tools business, in other words, like robotics, software on the robotics, reagents going into all that infrastructure devices, that whole ecosystem of like our tools business versus the services offerings that we offer on top of our setup like data points and solutions, that tools versus services, I would say, is like 80-20 in the tools side of the house in terms of our revenue mix in 2030.

My hope would be we are largely taking over the general purpose R&D infrastructure and being that provider of the tools into the whole industry. So that should be dominant. When it comes to biosecurity, there, it's very dependent on how things play out. It's like a very interesting time right now. So CDC is getting rebuilt. There's a great post from Matt McKnight, who heads up our biosecurity business today. I encourage folks to read about sort of what a rebuilt CDC looks like. I think fundamentally, you need persistent pervasive monitoring of viruses as foundational layer for biosecurity in the future, whether you're in an outbreak or not, just all the time. And so if that type of infrastructure gets built here in the U.S. and worldwide, then who knows? Biosecurity could be 50-50 with the rest of the business. But it does depend on whether we see that adoption of sort of monitoring technology as the core -- one of the core pillars of a biosecurity that works, a CDC that could stop the next COVID.

Cool. So we got a question for Steve. So Steve, you mentioned in October 2025, Ginkgo reset the annual commitments and its contract with Google. Can you provide a little more color on that?

Sure. When we were negotiating the Google Cloud contract, obviously, we had a shortfall to solve for in Q3. We talked about that. We reset going forward, in my view, very favorable terms for Ginkgo. We were able to reduce our go-forward commitment by over $100 million and extended out the period by 2x. So going out over 6 years over the prior 3 years. From that standpoint, I think that puts us right where we want to be.

Yes. And just a little extra color on this. We had made that investment on sort of the Google Cloud side around -- I mentioned the 2 areas of AI, the sort of reasoning model based AI and the bio model based AI. It was originally made with a mindset of that bio-based AI was going to grow quickly. And I think what we've seen in the industry is it's being adopted, but it has not grown at anywhere near the rate that the reasoning models have. And so this is more a reflection of kind of how we see the deployment of -- and like really like training needs internal to Ginkgo in the future.

It's a much more smooth ramp over a longer period of time compared to if you were seeing massive investment across bio AI models. And that just hasn't been at the rate we were expecting back then. So I'm very happy that this was cleaned up very nicely by Steve and the team and our great partners at Google have worked with us on this. So I'm really happy about where it landed.

All right. The next one is for Jason. Jason, you mentioned Future labs new announcement of its next-gen AI scientist Kosmos. Can you say more about how your experience at Ginkgo kind of informs your viewpoint on AI, not just analyzing data, but also designing experiments, et cetera?

Yes. I mean it's more folks checking this thing out. I mean, so Future labs is now called Edison Scientific, it used to be an nonprofit sort of doing the OpenAI thing becoming a for-profit. And so -- but what they're doing is they basically built up a model for -- that's read all the scientific literature, you can kind of ask it like a scientific question. It will run for several hours and then kind of come back with either kind of hypotheses or predictions or learnings or conclusions, and they were able to show this model, making several, frankly, new scientific discoveries just from reading the literature. So that's already very exciting.

I think -- and it's sort of this indicator that we're on this inevitable path where I think the logic of the models, their ability to just do complex reasoning is going to work. It already works, frankly. I think the limitation will then move to what tools can you give access to these models. And the big one we believe is important in the realm of science, like I mentioned earlier, is hands in the lab. That's just it. It's hands in the lab.

And so that type of a model with the ability to then say, well, what I actually believe I should do to really answer your question based on everything I read in the literature is run these 10 experiments or these 100 experiments, see what I learn and then run another 100 and do that a few more times, and then I'll come back to you with the answer. I mean that's what a PhD does. I mean that's what I did for 5 years at MIT, in my PhD, it's like, I got this question I'm trying to answer. I'm going to run some experiments. I'm going to look at the results. I'm going to interpret them, and I'm going to go around that loop.

And a lot of it is understanding what other people have done in the literature. I think that's what this model does from FutureHouse, Edison. And then the other half is kind of just not basic logic, but not the world's most complex analysis of what you're seeing in the lab. It's really your ability to conduct and design the experiments and then interpret the results. Just the craft of that is what keeps a lot of people out of science. And I think that can just be replaced now, I think, with programming and a robotic interface to the lab. And I don't know what that does. I mean that might blow open access to asking hard scientific questions in a wide number of areas, which would be very exciting. So we'll see. But where we want to provide the hands, that's our role in that. And we're very happy to have other places build those genius models.

So the next question is kind of a follow-up to that one actually. And so the question is, how do you see this AI plus robotics platform changing the R&D landscape sort of at large? And what has the initial feedback been from potential tools customers?

Yes. So I think like if you think commercially, how this can make a big difference, right? So what -- the way -- like take drug discovery, for example, right? You're -- you have an idea, you've read about -- again, you read the literature, you're an expert in this area, you have a hypothesis about a certain disease and how it works and you're looking for an interesting drug target around your hypothesis.

So you would sort of plan a line of experiments, you and a team of researchers will go conduct that over a period of 6 months or 1 year, 1.5 years and then try to get to an answer on your hypothesis. I think what's exciting is that for us, maybe those original hypothesis, maybe stuff like FutureHouse can just come up with those, who cares? Even if they can't, you always have a longer list of hypotheses then you have the resources to go out and test in the lab based on the number of scientists you have, fundamentally, that is the limit.

And so if instead, you could basically spider these models out and say, hey, I want you to pursue my top 100 hypotheses instead of my top 3. And for each one, again, it's not just one experiment. It's got to do some lab work, interpret the results and then plan some more lab work and keep going down that trail. You could be running that across 100 or 1,000 hypotheses in parallel as a single researcher potentially with access to robotics to go spider and then have it just come back and tell you when it gets interesting results.

And that is just -- I mean, I don't even know. That's a fundamentally different way to pursue a goal around, say, how does this disease work? It just -- fundamentally, what is limited is reasoning and experimental hands. And if we can take both those off the table, then I think all the cost just turns into like reagent costs. It's like literally the consumables you're going through, which is just crazy. Like that is not at all the cost right now. The costs now are 100% dominated by basically human time in all these areas, really. And laboratory space, just like literally square footage. And both of those could compress massively with automation plus AI. It's really exciting.

All right. That's all the questions that we have for tonight. A reminder, you can always ask questions by e-mailing us at [email protected]. And also, as Jason said earlier, if you're interested in coming by and seeing some of this equipment, reach out, and we'll make it happen.

Great. Thanks, everybody. Appreciate the questions.