DNA - NYSE

[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

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.