Ginkgo Bioworks Holdings, Inc. (DNA) Earnings Call Transcript & Summary

Hi, everyone. Thank you so much for coming to Ginkgo for Ginkgo Bioworks 2023 Investor Day. For those of you that don't know me, my name is Megan LeDuc, I am the Manager of Investor Day and of Investor Relations in general. So if you have any questions, feel free to e-mail me. Before we get started, I just want to give you a few reminders that one, these presentations today will contain forward-looking statements that involve certain risks and uncertainties. And if you have any questions or want to learn more about those risks and uncertainties, please see our SEC filings. And number two, for any of the folks online that are watching, if you have any questions today, please e-mail those to investors at ginkgobioworks.com, and we will try to get those answered throughout the day. Without further ado, please let me hand it over to our CEO and Co-Founder, Jason Kelly.

All right. Thanks, Megan. And excited to have you all here today. Thank you. Okay. you got that Okay. All right. So before we get started, I did want to -- we're going to get into the agenda and the plan for today. I do want to start by talking about our subsidiary, Zymergen. There's news this morning. They filed for Chapter 11. So I want to give a little bit of context on that since I have a bunch of in the room, and I'm sure there'll be a bunch of questions about that later. I will say that we'll be able to talk about this at the Q&A, so you get to ask questions. But first off, you should review our 8-K. We put as much as information as we could in there. At the time of the Zymergen acquisition, we were aware of the liabilities at Zymergen and that if they couldn't be resolved. That Zymergen would potentially consider a bankruptcy. So that was known to us. Zymergen has been operating as a distinct legal entity since we did the acquisition and has had arm's length agreements with Ginkgo throughout that time. Just as a reminder, we ended last quarter with $1.1 billion in the bank in cash and cash-like securities. So, Ginkgo is in a strong position, this is really about Zymergen the subsidiary. Previously, Ginkgo and Zymergen entered into a nonexclusive license agreement to the intellectual property at Zymergen. So Ginkgo can continue Business as usual. So the technology we have been using from Zymergen we still have access to that. So not really a disruption there in terms of our business operations. And then finally, just to help the bankruptcy process along, Ginkgo has submitted a bid to purchase the bulk of Zymergen's assets, transfer employees and assume some liabilities, including one of the leases to the real estate out there. And so I do want to just give a little bit of that context since I figured people would have questions about it before we got into things. And again, feel free we can talk more about it at Q&A or if you want to go at the breaks or things like that. Okay. All right. Okay. With no further ado, welcome the Ginkgo Investor Day. I want to highlight what we're hoping you can do today is get a picture of a wider set of the bench here at Ginkgo. You hear a lot from me. on earnings calls. I will obviously be here today. I'm very happy to talk to all of you, but you're going to get to hear from Barry Canton, our CTO, Patrick Boyle, our head of code-based, Anna Marie, who, as of today, as her expanded role as our Head of AI. So you're going to get to hear on a number of our technical fronts. Jen Wipf who heads up our Cell Engineering, Matt in biosecurity. And then in the breakout sessions, you can hear from some of our leaders in certain market areas. So Kevin Madden on industrial biotechnology, Patrick in agricultural biotechnology. And then Mark, our CFO; and Jason Berndt, who heads up our operations will be in a breakout around ops and finances as well, okay? And so what I would encourage you all to do, and I have told the team to do this to leave time at the end of each of their talks to take questions. And so really, what I'm hoping is that we can really dig in, you guys should ask deep questions at the point of the Investor Day from my standpoint as compared to an earnings call is there's a lot more opportunity for a 2-way conversation here. So please do take advantage of that and for the folks that are joining us on the webcast. I know there's a bunch please do submit questions, and we're happy to work those into the discussion today. And then finally, we're very lucky to be joined by Board Chairman, Shyam Sankar, CTO of Palantir. We've been doing an AI fireside chat with Barry and Shyam, Dmitriy, who heads up our AI on the technical side and Anna Marie in the afternoon. Okay. All right. So Ginkgo's mission is to make biology easier to engineer. I start a lot of our talks internally at the company with this almost all of them as well as our earnings talks. This is a key idea. So this is what I think distinguishes us from most other biotech companies in the industry that might have a mission like to cure a certain disease. Or to develop more efficient crops for growers in the Midwest, right, Or animal-free meat. Or like there is outside new fragrances produced with biotechnology. Those are the missions of Ginkgo's customers. Ginkgo does not have a product pipeline. Our mission is to make all of those customers more likely to deliver on their products in biotechnology in their markets. And we have been committed to that for 15 years. I do think this is what makes us unique. It's why our business model looks the way it does. It leads to a lot of the questions and confusion about Ginkgo because it is kind of unique in the biotech sector, that is very product focused, but this is where it comes from, and why we have that sort of platform services business model. This is my favorite slide at Ginkgo Bioworks, all right? So this is not a subset, not all of the customers that have come on to our platform in most of them in more recent years. What I want to highlight is, Ginkgo is trying to get people to do R&D differently in the biotech industry. All right? And so if I were a product company, and I wanted to show that my platform was relevant in say, R&A therapeutics I would just start developing my own drug in R&A therapeutics. And the only person I would have convinced that I was any good at doing R&A therapeutics was myself and my investors who are betting on me to develop that drug. That's not how it works for Ginkgo. If I want to convince you that my platform is relevant to a certain type of biotech R&D, I have to go get a customer which means I have to convince typically the head of R&D of a company that what I have and what you'll get to see on the tour today is either additive or better than what they have in-house for doing that kind of work. If I can't convince them that, the logo does not go on the screen, all right? And so that sets a higher bar for us to validate to all of you that our platform is relevant, and so what I'm really excited about is -- you can see it in so many different areas, right? So you can go start an industrial biotech, or some of our earliest customers like Robertet and Givaudan, the flavor and fragrance industry. Now, places like Sumitomo for large-scale chemicals, in agriculture, we work with Corteva, Syngenta, Bayer big 3 at biotech companies, all customers of Ginkgo. And then you're going to hear a bunch about this today, but our fastest-growing area right now is in the biopharma space. And so you're going to get to hear from Kevin and Patrick on industrial and agricultural biotech and Jen Wipf, who heads up our commercial on biopharma later today. okay? And so I do want to highlight one of the deals. I've highlighted that Pfizer deal. It's a deal we just signed last week. I think this is something investors should be paying a lot of attention to because this deal is special in a couple of ways. First off, obviously, it's a great big biopharma like I just said a minute ago, they have large existing in-house infrastructure. So they had to decide to use Ginkgo instead of do this in-house. But importantly, here's a comment from Will, Head of Biomedicine design at Pfizer, Will Summers. It says, access to Ginkgo's proprietary platform will help enable Pfizer to search for novel and exciting R&A constructs with improved stability, right? The progress into that biology, we have the potential to create new R&A treatments. So the key here is this is not about manufacturing therapeutics but not R&D to support that, like, for example, our deal with Novo Nordisk, our deal -- historical deal with Biogen, our first deal with Merck. We're all in the area of optimizing R&D, doing R&D to support manufacturing. This is R&D to support drug discovery, all right? And so why is that exciting? That is the biggest R&D market in the biopharma industry. And if you think about the markets, the TAMs that Ginkgo goes after, 1 of them is the research budget of our customers. So the money they are spending on their labs, on their equipment on their kits and employees to do research. And then the other is the product revenue that we get a royalty on, okay? In biopharma drug discovery, those are the two -- that is the biggest R&D budget market and the biggest product market in biotechnology. So this is a very exciting area for Ginkgo going to be getting into. Obviously, this is a deal in R&A. We are hopeful we could get similar deals in gene editors in protein therapeutics in cell therapy. And we're working and you can talk with Jen and ask questions, we are working to add more discovery deals in those areas as well. Does that make sense? You might ask why didn't you start in this market if it is the biggest market for biotech R&D and for products? And the answer is, 8 years ago, when I was doing the first deals in fragrances, there's no way I could have got a deal with Pfizer. Because our platform, as you're going to hear about today on the technology side improves with scale. So every year, as we build this place bigger and every time you see me add a new customer, my infrastructure is improving. And so it has taken us the last 8, 10 years of scaling this place to be good enough that this comment is you would hear from a R&D leader at a major biopharma company doing drug discovery, okay? And I will just point out the platform you see here is used for all of those projects. So we are nailing now the biggest market for biotech R&D on a platform that can do any type of cell engineering R&D. That generality is why I think Ginkgo's a big deal, it's what makes it special. Does that make sense? Okay. I'm not going to belabor this too much, but you are going to get on the tour, you're going to hear from Barry, I encourage you to ask Barry questions about our foundry. This is our oldest asset that we started developing first at the company. It is a lot of high throughput automation to do the kind of work I did back when I was at the lab bench at MIT, working by hand on engineering cells and creating research data about how genetic designs perform. Here at Ginkgo, you're going to get to walk through about 100,000 square feet of facility here that is automating that lab work so that we generate more research data per dollar than a Pfizer could in-house with their research teams. That data then goes into our code base, which Patrick Boyle heads up, and we keep it. So one of the things Ginkgo does uniquely is when we work out these deals with our customers, we retain the reuse rights to the data for future projects. All right? And that is, again, I think something that's special in the industry because if you just think of a single product company, they have access to only their data. They don't have access to the data of a bunch of other companies in the industry. But with our code base, that's what we're accumulating. So you can push on Patrick. What do we have access to? What are the things that are really reusable, what can cross over customer to customer when we do this work. And so we've been building that as the scale of the automation went up, we started to generate organized and reuse that code base. And then Anna Marie is going to be on the tour as well. The thing we're doing with the data, the most recent activity is we announced a partnership with Google Cloud about a month ago, where we're going to be training AI models, particularly in the protein design space on all this data, okay? And that provides a number of things, but we are hopeful, even better predictive ability in terms of what biological experiments you want to run, what type of genetic designs are going to be most effective for a customer project. The key thing at Ginkgo, all of these things get better with scale. Okay? The bigger the foundry, the more data you get per dollar, just like a semiconductor fab or an auto plant like it has a physical scale economic. The more data I have from previous projects, when I do a new program, the more likely I can take something off the shelf, and it's relevant from my code base. So like you want to do the next R&A project because I've done a lot of work on the Pfizer project, I will have something that's useful to you. And then finally, the scale of compute we do and the amount of data that we can pump into a single foundation model also has a major scale effect, and you're seeing that in the natural language models today. That's why you have basically 3 that are coming to dominate the natural language processing space when it comes to foundation models because there is a scale effect in AI models. That's the kind of technology Ginkgo likes to bet on. General purpose technology that gets better with scale. We're just not doing it in the tech arena. We are doing it in the biotech arena. Okay? But these all get better every time we sign up a new customer. And so again, I already highlighted that you're going to get to hear from all these leaders. The only other point I would make is Jason Berndt, who heads up our operations. He's really focusing on, can I use the foundry and code base particularly in the areas of industrial biotechnology to drive massive efficiencies and make that market into a place -- almost where like we can when we just get like fees from customers that we are covering the cost of doing that work, that would be ideally where we'd like to get to with that type of project. And that allows us to kind of have as much time as we want to get to royalties. So I think you'll see that type of flywheel spin up first in some of our older areas like microbial engineering, where we have more built out scale. Does that make sense? So talk to Jason about that. Okay. Finally, you're going to get to hear from Matt McKnight, who heads up our biosecurity business. The key idea, I think, to take away on biosecurity and this is work that came out of what we did in COVID around infectious disease monitoring. So things like radar stations looking for viruses is that this is moving from a public health market to something that looks a lot more like a national security technology market, okay? Because when we go out and interact with our government and foreign governments around this type of technology, we're often interacting with the national security groups who are trying to understand is something coming to disrupt my country, I need to know about that ahead of time. That's a different kind of solution space than what we've seen in public health, which is often responding to an emergency. Public health is a little more like FEMA, like I'm showing up after the problem, okay? The national security people about preparedness. They're like, "I want to know the missile is coming before it hits us. okay? That's their job. And so I think that's a real shift in how people have been approaching infectious disease. I think we're building a leading brand in this space. We're again fortunate to have Shyam. He's obviously done a lot of work at Palantir and building out. I think you are seeing a wave around technology right now, companies like Palantir and Anderol leading there. We see an opportunity to lead in biosecurity similarly in defense. And so you're going to hear about that from Matt. Okay. So since I've told the other leaders at the company at the ends of their talk, they should leave time to take questions. I'm very happy to take a few questions just right now. And then we're going to go out and take a walk around for some tours and then come back and do some more talks that folks on the live stream will be able to join. So happy take a couple of questions, if anybody has any?

[indiscernible]

Okay. So just you get a sense of the layout of Ginkgo. This was a building built for World War I. And so it's about 100 years old. Each one of these little squares between the wall behind you and that one over there is an 18,000 square foot box all right? The Army built this building, so they're all just stamped out, okay? There are 8 stories of these and there are 12 across the building. So if you're a biologist, this is a 96-well plate, okay? And you are currently in A4, Barry? yes. A4. Okay. You were sitting in A4. We have filled, I think, 7 of these wells along the top of the building. You're going to get to walk through them. And I think one key point I would make before you jump on the tour with the foundry is to understand that it is an evolving infrastructure right? So don't think of it like a chip fab or a car plant where you kind of build it once, that this is the 6 nanometer production node at Intel, and we'll build a new one in 3 years. What goes on in these rooms turns over on probably a couple of year time scale. And so you'll see new technologies coming in all the time. And so kind of get to see even different epics of what we had in the foundry on the tour. So excited to have you all join us for the tour and looking forward to the day and welcome to Investor Day here at Ginkgo. Thanks.

People in the room, we have the first 2 rows they're going to follow Barry... [Break]

Hi, everyone. Welcome back. Hope you all enjoyed your tours. We're going to jump into some presentations on Cell Engineering and Biosecurity. So without further ado, Jen Wipf, our Head of Commercial for Cell Engineering.

Thank you, Megan. Hi, everyone. Great to see you today. I'm going to talk a little bit about our commercial business in the cell engineering space. So for those of you who had the benefit of going on the tour, hopefully seeing the infrastructure give you a sense of why customers might be doing work with us. But if I outline a few examples, certainly, one of those is the ability to get more data per R&D dollar so to be doing more work for the same amount of cost and what that might look like, let's say, in the pharma space is that maybe you can carry therapeutics longer in your pipeline. As opposing to having to narrow those down early on. So your likelihood of success of sort of finding something of therapeutic importance or finding something that's going to be successful in the clinic is higher. For nonpharma applications, maybe that's just the total likelihood of success of bringing something to market. Another reason is just existing data assets that we have. So that's something you can't see as you walk around the foundry, but as you can see the scale of which we're producing data. You can imagine all of the data that's behind that. And certainly, some of the reasons here that we're seeing are things that are driving success-based pricing deals, for example. So a lot of times, we have a lot of data that's already in existence that helps customers start at a place that's further along the sort of discovery or development pathway and therefore, reach the endpoint much faster. Another reason might be to launch work fast. So I think a shining example of this is Arcaea, right? So the company started in April of 2021, the next day, there was R&D work happening in the foundry. And fast forward to today, they now are launching their second product with a extinct flower fragrance that I think some of you can see here. And so really, the ability to start that work and, therefore, get to a commercial product much faster is a huge advantage. Also saving on laboratory CapEx. I've had some customers come and see the space and say, "Oh, I really should think about what I build versus what I maybe consider outsourcing. And along those same lines, sort of replacing fixed cost with variable costs. We know that a lot of companies, particularly small companies have really high R&D spend. And then as they get closer to commercialization, they're focused on driving that product to market, getting into the channels and the R&D team can sometimes not have enough work. And so the ability to kind of be able to flex that spend and the work up and down is really advantageous as well. So some of the reasons that we're seeing. And I think those reasons are applied across all of the industries that we're in, like all of these companies are struggling with those same kinds of challenges. And so remember, we are industry agnostic platform. And so I wanted to give folks a sense of what we're seeing across some of those industries. So today, we have 105 programs going on concurrently in the foundry. And the sort of mix of those, you can see here along the column. I think the notable thing is that what we're seeing in pharma and biotech is this growth of the total percentage. And I think that's actually a commentary on really how tip of the iceberg we are in this sector. Remember, we started working pharma later than the other sectors. And we're continuing to build out both capabilities and physical lab space and data and all of that. And so we're really seeing a lot of momentum picking up in the pharma space. And so you can imagine some of the momentum that's happening there, and I'll talk about it a little bit today. And we're still seeing a lot of strong growth in the other sectors as well. And so that's sort of keeping up with our ambitions to grow in those sectors. So in terms of pharma, I want to spend a little bit of time talking about what are we doing in that space. And I think it's important to remember that since we're not talking about products, our customers are doing those, we're thinking about that work in terms of modalities in terms of sort of the scientific problems that pharma companies are trying to solve. And so that spans sort of a ton of modality, cell therapy, gene therapy, biologics, RNA therapeutics I'll talk about a few of those. But importantly, it also spans from discovery to manufacturing, and I think one of the other really interesting things about the solutions that we're providing is the ability to think about manufacturing and scalability alongside discovery sort of earlier in the development pipeline. And so a lot of these -- these kind of look like they're in different sections, but really, many of them span sort of the R&D development pathway. So I wanted to highlight 2 of these that I think are interesting case studies for what we're doing in pharma. And so one of those is Merck's. So remember, we started with a collaboration that was focused on enzymes, so really biocatalysis for APIs. And again, we're really strong in this space because we've been working on enzymes for many, many years in many, many other industries. And that data is really applicable. Biology didn't decide that they were going to work on enzymes for agriculture or for pharma, biology is working on a biochemistry solution, right? And so we're able to apply that in many different places of the work. And Merck saw a lot of our strength in the enzyme space. Remember, they are very strong in the biocatalysis space as well. and saw that sort of our data and capabilities would give them a strategic advantage and could help them do more biocatalysis work than their own internal teams could do. So again, expands their pipeline. On the backs of that, as they sort of started to see us delivering in those space, we were also able to sign a second deal that was in the biologics sort of manufacturing space. So again, like as we're starting to work with these kind of blue-chip pharma companies, they're really learning more about our strengths, we're learning more about what problems they're trying to solve, and we're able to kind of find the next fit of what we can do together. And we're seeing that play out in Merck. We're seeing that play out in Novo Nordisk as well. And so we started a pilot program with them. We reached that over the summer, and now we're expanding into the next set of work with them in sort of a multiprogram biologic space. And so I think what's important about both of these deals and sort of a number of the kind of blue chip pharma companies is that now we're in a place where our tech is good enough, right? These are players who are scientifically very strong, have a long history of developing products and they're seeing us as an advantage to their own R&D teams. So that's one thing that's really important. The second is that these are large companies that are often thinking about a particular therapeutic area. And remember, we're a platform. And so we're learning also how to navigate those companies to get these really big collaborations together, and then from there finding ways to expand the work that we're doing with them. And so I think you'll find that there are a number of these other sort of logos on the screen where that same sort of enterprise sales thinking is playing out for us. So one of these kind of new enterprises that we're working with is the Pfizer deal, and so in the Pfizer space, we're working with them on RNA molecules, and remember, they're thinking about trying to get -- sell clinical problems, get products to market. And what they found with Ginkgo is that because we have the scale to kind of explore a really broad set of scientific questions, we're able to also think about not just a particular solution in the RNA space like circRNA or mRNA. We're actually able to think about the problem that they're trying to solve and then apply whatever tools or modalities might be best to advance that. And so I think that's a real strength of our platform as well. The breadth can sometimes seems like, well, let's come -- like what are you actually doing in the pharma space? But when you're trying to solve a sort of clinical problem, that breadth can be a real advantage, right? You can start to say, I don't -- not looking for a particular solution. I'm trying to get a product into market. I'm trying to discover therapeutic. I'm trying to make sure it's scalable and manufacturable. And I don't necessarily have a way to do that in mind from the start. And so again, like this is one where -- this is really very much a discovery space. So again, we're moving from a lot of the earlier deals in pharma in the manufacturing space, moving more and more into discovery. And I think also we're able to think about not just discovering RNA therapeutics, but think about the path to manufacturing those, the path to making those scalable, the path to bringing those to market really early in the process. So as you walked around the facility. For those of you who are here, you can see how even early on, we're able to grant design-build test for men cycles. Do you kind of see where that's going and use that as a part of our process. So again, we're really excited about this particular program, not just for the work that we're doing with Pfizer now but for the connections that we're making about other challenges that Pfizer has and other places that we can potentially expand. Okay. So I talked a little bit about kind of enterprise sales in the context of Merck and Novo Nordisk beyond pharma, we have a number of other enterprise customers, right, Bayer in the agriculture space, Sumitomo Chemical has been a long-standing customer that continues to do work with us and others. And so what's really interesting about this is once we have that sort of customer acquisition costs, Again, they're starting to understand our capabilities more. We're starting to understand their scientific challenges in whatever industry they're in be a part of their scientific teams and try to find other ways we can advance their R&D pipeline. Also, I think what's interesting about these deals is remember, we're signing these big kind of collaboration contracts, which are -- they're not sort of turn the crank kind of I know exactly what the work is and do you do the work and you deliver it to me. are these collaboration contracts, and once you have that first one done from a contracting perspective, from a deal execution perspective, it becomes much easier to do work from there on. And so I think one of the things we're unlocking with these customers is the ability to start work with Ginkgo really quickly. The other thing we are focused on in the sales space is really helping customers understand what it is we can do and how we can deliver that work to them. And so you'll see that in a lot of our kind of service offerings work, so in enzyme and protein services, which we now have the success-based pricing, helping customers really understand what it is that we can do and how to, again, start that work with us really rapidly. And so that is really a driver of helping to make these deals for us, execute quickly and bring people onto the platform really rapidly. So we get out of the cycle of what do you really need? What can you do? Enzyme services, success-based pricing, like what's your target, let's go. And oftentimes, we have the data to start that already. And so you'll see us helping to explain in the marketplace more and more what that looks like and how you can start work with Ginkgo more quickly. Okay. So I think if I sort of summarize a few key points, and I'll open up for questions after this, I think we continue to sign deals with really kind of blue-chip customers, right? These customers are scientifically of high value. They're hard to sign deals with getting these customers means that like our tech is good -- and is good. Like it's better than what they have. It is -- we've really advanced sort of not just the foundry, not just automation, but also the data that we have, also the ways we work with customers, also the ways we sign those deals. We continue to have a lot of success with repeat customers. So once we sign that first program, we're delivering on that. We're opening up the doors to more and more work with those customers, and we've built out a team to drive the expansion of work with these enterprise customers. Also, we're really focusing on what it is that we do and helping the marketplace understand that and helping customers come to us with what they're looking to achieve and finding ways to make that really easy to start work with us. And then we'll have a whole section on AI today, so I'll leave most of that to the team. But I would say this is a real value driver for us. Lots of people are trying to figure out what to do with the data, how AI is going to change their business. And I think we're really seen as a thought leader in that space and that's driving a lot of conversations with particularly large pharma companies, and opening up new avenues for us to explore and even build off some of the collaborations that we have. So that's roughly what I wanted to share today and I know Jason really wanted me to make sure I have time for questions. So hopefully, there are some in the room and happy to open those up for a few before I move to biosecurity Matt, Mark?

How has the success space pricing relatively early I know that the permits you talked about involving enzyme services [indiscernible]. However on earnings call you talked about being able to expand that, especially if it's kind of bump up your program [indiscernible] just outside -- and are you seeing new programs coming on beyond enzyme services?

Yes. So the question was for the audience is, how is success-based pricing going? Are we doing that in anything beyond enzymes? And how is that ramping up? And so remember, we launched success-based pricing at Ferment, so that was April of last year, and that was really focused in the enzyme space. And we did see a really interesting uptick in sort of interest in that. And we even had a few of our deals that were signed in the second quarter and to be announced in the third quarter come in basically as inbounds from that. That we were able to kind of scope and sign basically within a quarter. And those were companies that I didn't even know -- I didn't even know we're sort of in existence. They weren't necessarily on our target list. So I think that's been a really interesting thing in the market to kind of see how broad that application really is in some surprising places. And what we have seen from that beyond enzymes is that we have a real strength also in the protein space. So protein expression, in particular, and so we have been doing some success-based pricing in the protein expression space, and I would expect that to continue. Again, it's the same kind of idea, right? We have a lot of code base. We have a lot of assets that are in existence. We know that these are going to be successful. Like we've changed the sort of like ability to engineer biology in those spaces, and so we'll continue to push on that. Yes.

So what sort [indiscernible] of the interesting things you mentioned at outlets pushing more into Discovery. Discovery is something that pharma likes to pull those and adjust in general. Is that not changing as the advantages of scale and the reality [indiscernible] become more apparent to them? And is that so that how is that potato this emphasis on success-based pricing. So at some point, you more gets around that you guys are just setting and where perhaps there is a larger upfront [indiscernible] able to for people on the...

[indiscernible] Streaming?

Yes. So if I try to summarize the question, so Pfizer kind of represents some work in discovery, what am I seeing in terms of sort of interest in discovery work, why are pharma companies coming to us for discovery work? Did they sort of realize that we could do that? Or how has that played out? Is that right? Yes. So I think some of that has been both building up our internal capabilities across those modalities. So remember, we have been building up our team. We've been building up our physical space in Biowork7 to expand to some of those modalities and in discovery space. So I think customers are noticing that. Also, a lot of those customers are already doing work with us in the manufacturing space and kind of through those programs are thinking, "Oh, gosh, if I -- this could really help me if I thought about it a little bit earlier" and probably the same types of strengths are useful in the discovery space, the ability to really go broad with scale, the ability to really look into sort of the genetic design space. So yes, I think in some respects, like pharma customers are starting to realize that we have a real strength in that space and it's our job as sort of the sales team, I could go to help show that to them. So I think we'll continue to see more of that. And you said pharma companies tend to have discovery and kind of hold that close. I think also the sort of changing landscape of what is happening with data and AI and even the fact that we have a lot of data is making pharma think about how to stay competitive in discovery. So I think, in particular, our strength in AI and what we're building are going to continue to help that. Yes?

Can you guys [indiscernible] talked about 40% reduction in [indiscernible] can you talk a little bit about how that's impacting your discussion with customers [indiscernible] adding new programs.

Yes, sure. So the question was the reduction in sort of program launch or program scoping and how that's driving sales. So it's a really important point for us because some of the hurdle of getting deals signed is figuring out what is the work that we're going to do, what are we going to deliver to the customers, what are the terms on which we're going to do that. And so that is probably the biggest linchpin in the sales cycle is defining the work and making sure we're delivering something to the customer that they want. And so any decrease in doing that is really helping the deal throughput. And remember, that's growing quarter by quarter. And so we're focusing a lot on that. And so one way we can reduce that -- sort of like time to start a program is some of the sort of product-based marketing or success-based payments that we have. It's like very clear what the work is that we'll be doing. Another is thinking about sort of the work and the modalities that we're doing and helping to be really crisp on what the deliverables can be for customers. And so that is a focus of the sales team in general, is to figure out how to continue to drive that down. We're doing that in order to kind of increase the deal throughput. Yes.

Is there any other customer types that that's more interested in the success-based payment. So is it more focused like or more interest coming in from like smaller startups? Or are you seeing larger companies like Pfizer be interested in success-based payment?

Yes. So the question was like what types of customers are interested in success-based payments. I would say, to some extent, all types of customers. I would say it leans a little bit more towards smaller customers. It tends to be the case that some of the enzymes that the larger customers are wanting to do are just harder, and so they're not quite the mold of what the success-based structures look like. But they're embedded in those teams. Often there are some sort of simpler, let's say, enzyme programs that are coming from large pharma as well. But SKU is a little small.

Can you talk a little bit about a couple of years ago, sort of equity or royalty and milestone [indiscernible].

Yes. So the question is really around sort of our philosophy on downstream value and the breadth of kind of the offerings that we have for deals and downstream value. So the breakdown of ends of royalties, commercial payments, equity. So in general, we've -- again, we anchor on sort of something in downstream value because we want to have shared success and we want our incentives to be in line with the customer so that we deliver something that goes to market. So we really just anchor on that as a philosophy. And then however that works out with customer we're often in a lot of discussions with what's best for them, right? And some of them prefer sort of royalty tail. Some of them prefer a payout upon commercialization. Some people are start-up companies that have equity more at their disposal than cash. And so we continue to have a lot of open discussions with customers about a fit that works for their business.

But how has that changed numbers [indiscernible].

I don't know if we're sharing that data exactly. I would say that as you see more of these kind of blue-chip customers on the Board, like they're not doing equity, right, that's like not their preference.

[indiscernible]

It's a pretty healthy mix of both. I would say we try to manage sort of the portfolio so that we have a mix of like near-term payouts and further term payouts. And we have a broad portfolio off of which we can do that. So -- we're not leaning towards one or the other. Alex?

I'm just curious if there are maybe just bars you put in what needs to be in the downstream or upstream bucket. I think you guys believe in the technology and if you can expand access to it and generate that flywheel data generation that's awesome that will prove it out over a longer period. But just wondering, you have something in the downstream about it. Is there anything that would have to be required in the upstream bucket, like in 5 years would you run boundary below cost upfront or just to increase the accessibility? Or were there always be -- we need to at least set upstream guarantee the cost of running all these facilities?

Yes. So again, we think about our sort of commercial portfolio like we're managing across a bunch of different industries, a bunch of different project types, a bunch of different paths to commercialization. So in that sense, we're able to really leverage the breadth of the portfolio for near-term payments, for R&D services and for downstream value. So you can see already that some of what term payments being different and retaining value that we can get later on. You'll also see some sort of licensing deals for assets that we already have, where the R&D is sort of done in arrears, and we're able to sort of license some of those assets. And so you'll see us experiment with a lot of those terms, yes.

On one of your earlier slides, you had a breakdown of active programs like one, [indiscernible] ag and industrial. How do you see that evolving, especially in light of success-based program to sort of [indiscernible].

Yes. So I think if I were to try to put your -- the question is like what's the mix of industries and what's that mix going to look like in the future potentially and where the success-based enzymes and payments fall in that? So I think if I were to project what that looks like, there's a lot of momentum in biopharma space. So expect the proportion of total biopharma against the rest to be growing. But success-based payments are embedded in all of those, right? And so success-based payments are part of how we're expanding in pharma. They're part of how we're doing more work in industrial biotech, how we're doing more work in ag. And so consider those kind of in all of those spaces for a certain type of program with customers. Is that helpful? All right. Okay. I think we'll close there, and I will hand over and introduce my -- when did I lost the clicker here? Yes, sure.

So yes, maybe one quick thing I would add on the success space since it got asked a lot. The -- in the near term, it's a lot of what John was talking about like it's another tool at our disposal to get deals done with customers. The most important thing for Ginkgo is to add new logos and add repeat programs with existing customers because they drive scale, makes the platform better, makes our lives easier next quarter, okay? No question about that. The long-term thing is all biotechnology product development is considered R&D. Like we would not say that about like electronics, okay? It is not considered an R&D project to develop the next iPhone. We appreciate that it has got business risk in it, and it is an engineering project that involves a bunch of engineers laying out the spec in scope and understanding what the iPhone can do. But we don't consider it a R&D project like we consider drug development or trait discovery, things that just might not work. And so a big part of success-based pricing is to price what today we think of as biotech R&D like engineering. In other words, you don't pay for your engineering design on the iPhone, if it doesn't fricking work, right? Like -- and so that is actually the macro shift that we're like in the long arc of Ginkgo and making biology easier to engineer, we're trying to accomplish. We want these things to become predictable product development by biotechnology companies like there's predictable product development for electronics companies because that is what blows the market wide open. Okay? And so success-based pricing is a price it is a pricing scheme stepping in that direction. Everything you just toured is a technological scheme stepping in that direction. But you're starting to see us starting to align the business model in that direction. I'm really excited about that. Separately, it also gives enterprise sales team more weapons to get deals done, which is great. But that's really the longer-term nudge I would push you to that we're trying to pull off here, I think, uniquely at Ginkgo. Does that makes sense? Okay. Next up will be Matt Mcknight, our General Manager for Biosecurity, who can talk about how we're growing that business and in particular moving it towards national security.

Thanks Jen. Thanks, Jason. Cool. I'm excited to be here. We -- this is something we haven't had a chance to talk about as much publicly. So it should be a really good discussion. I'm going to endeavor to do about minutes and so please do ask questions. This is definitely a topic that there are a lot of questions about and just in how the market is developing. So it's super good to hear how people are thinking about it. I'll do like 2 halves. I'll do a little bit of like footing of how we're thinking about what is happening in biosecurity, biodefense like in the world that we are all living in. And then very specifically, going to -- like how we're thinking about product development, where we're deploying what we are deploying as a product and how we're thinking about sales. Maybe the first piece, Ginkgo, we've had 2 choices, basically. And it's actually important to identify these as 2 choices. The first choice is when you're building a cell engineering platform, you're on the forefront of biological engineering. You can choose to care deeply how that platform is deployed or you can choose not to, right? And I think very clearly, you've heard from us, we care very deeply how people use this technology. It is a transformational set of capabilities for the world. And so we've chosen both to do that in how we talk about it in the world, how we engage and everything from policy to our own decisions internally, but also with technology. So the first piece is build biosecurity technology so that people on our platform are able to build things responsibly, and that's a 25-year objective. The second one, which is a little bit less kind of obvious as a choice is you could pick that and make sure everybody that does things on the Ginkgo platform is doing things in a biosecure or a safe and secure manner with technology. Or you could also say we need to make those technologies available to the world because also, by the way, in the world, biology does not respect borders. Every country in the world is going to need to deploy these capabilities, like they deployed cybersecurity, but that's a conscious choice. And I would say that we made that conscious choice to build this business to make it available globally, not just for companies on the Ginkgo platform. And I think the very blunt like -- why to do that is if you look at what happened in cybersecurity. Cybersecurity was not a market category that any of us would have like in 1960 and like, you know what, there's going to be a huge market category of technology companies across the world building amazing capabilities in cybersecurity. And today, obviously, it's something that is hugely thought of as an important category within the computer engineering ecosystem. Obviously, to us, we believe there's going to be and why we use like the little B not on this slide, Little B Biosecurity is because we believe it will be in an industry category. There's going to be a huge ecosystem of technology companies building biosecurity capabilities in a bioengineered future. And Ginkgo can go absolutely should be the leading company building into that market. And I don't think it's 10 years. I think it's 2, 3, 5, 7 years, you're going to see this category emerging. It is kind of like the historical fact as you invent new engineering disciplines that you also invent the security components of those. So kind of -- one, we've chosen to make this a global business; two, we believe it's a big market coming up. I think the second one, Jason alluded to this, like the reality here is that the threat kind of envelope that the world is seeing from biology, now that's natural mistake and engineered, it is changing dramatically. And like if you spend time talking to people in D.C. policymakers, everything from ChatGPT plus bio, all the way to the kind of massive diffusion of capabilities, both in scale and kind of distribution and power, what nation states are doing behind the scenes with investing in high security laboratories. The threat envelope of bio is being identified as clearly not just -- and certainly, there's going to be benefits to public health, but not just an episodic pandemic boom and bust, we need to be prepared kind of and is being seen by the places where, like the annual budgets are spent to do preparedness. i.e., DoD national security type communities, not just in the U.S. but everywhere. This is becoming a massive category of concern. And you can see in the documents that have come out. But I think that the key point to think about there is that when in every point of history, when you've had a new threat category turn into a national security concern, whether that was nuclear weapons through to cyber the persistent budgets, we spend over $1 trillion a year in the United States of America on DoD plus the Intel community. The persistent budgets for preparedness change the market dramatically. And this is my favorite. So this is the National Biodefense strategy, and you can just go back in the last 18 months, the number of policy documents that have a decidedly different tone on how to think about Biodefense is pretty substantial. And so this last line, this is pretty -- this is actually new. And like the vision for how DoD thinks about this. It used to be like, oh, these are the 7 bad things and only these bad things we should worry about. This is -- it comes from anywhere and we seek to create a world free from catastrophic biological incidences. Like that -- the kind of the only other statement of that is that's an acknowledgment that there's basically 2 things that -- not to overdo it, but like 2 things that can kill tens of millions of people in a very short period of time, and it's nuclear weapons and bio. And that is the national security community saying, we're not going to let that happen. Once that flows through the system, that changes the following, which is they look for technology to answer that problem because it's not just a policy problem. And so this is where this like big emerging kind of Shyam led it, Palantir doing this for complicated data problems. Andrew is doing it for robots, right? This is where the defense tech ecosystem where American technology companies need to provide -- that's just the way our system works, need to provide capabilities for these big categories of threat becomes a symbiotic relationship between government and the private sector. And it's certainly what we're thinking about on the biosecurity side, how do we build into that demand for like cutting-edge technologies in biodefense? Something as like, okay, like what is that in reality, like what does that look like from a tech stack? What is the product stack for Biodefense or Biosecurity. The cool thing is you don't actually have to like invent anything new here. Like this is a lesson that has been learned over and over again in the national security community. Like it's the defensive targeting cycle, essentially. If you are looking for -- Jason used the analogy before, if you're worried about somebody shooting a missile, what you are doing is you're constantly watching. Every day, every second, persistent pervasive monitoring. If you're worried about cyber threats, you're monitoring every 0s and 1s at all times, most of them are boring. Usually, you're not finding anything. You're persistent pervasively monitoring. You are then prioritizing and characterizing those threats into what is concerning. By the way, this is especially in by this is where AI is going to be incredibly important over time. If you're trying to shorten the time from detection to mitigation, which is how national security thinks about this. You have to be able to monitor data and figure out what's important very quickly. And then you get to what we actually like relatively better at, which is where almost all of the resources are going today, which is medical countermeasures, vaccines, therapeutics, right? We think a lot about how to develop drugs and other things to mitigate the thing that we don't know is showing up, and so the big change moment, we believe, is this is what's missing. If you want to think about it from a DoD tech stack standpoint or from a national security tech standpoint. What we are missing is persistent pervasive monitoring so that you can enable rapid response. So all 3 of these things need to be built, by the way, right, where we have -- where we see the need really and what is kind of in significant demand relative to where there's a lot of expertise is in that first piece. Also, we are not insane, if you read all those documents, like the national security community of the world like has figured this out, right? And the most recent -- this is the bot-defense posture review. I love it. They actually I thought it was very -- for the first time, very thoughtful about saying, essentially, we need to build early warning around genomic sequencing. Early warning in genomic sequencing, Why? Because that is the -- and I'll talk about in a second, that's the high fidelity data asset that you can get novel information from. And then second, the bottom one is the U.K. this summer, launched their biological security strategy super need because it's language that we've been using for 18 months, their kind of top priority is launch biothreats radar. It turns out in the physics engineering area of the 1930s, the U.K. were the first to launch radar. This is the same thing, it's literally launched biothreats radar to get early warning. Okay. So like this is my favorite analogy, just go like nerdy for 1 minute, right? So in the '50s and '60s, we invented the high altitude camera, right? So we have the ability to take pictures from high atmosphere or space of land. That was like -- that was a new technology like Snap picture have cool picture of ground, right? And what was need, Edwin Land, who is the CEO of Polaroid at the time said, "Wow, this is going to be a new domain of comprehensive intelligence that we don't have, and it's going to be useful for national security purposes." Oh, by the way, we're going to be able to see where the Soviets have nuclear weapons and whether they're spinning them up or doing something weird with them, but also it's going to be super useful for economic purposes down the road. And then they're kind of like, well, also, we just need to figure out how to get these cameras up high. That's the issue. So if you look at the history of the U2, if you go to like the National aerospace Museum and you actually look at U2, you're like it is an amazing plane, but it's like a piece of metal built solely to get a camera up really high. It is the system around the first-generation high altitude cameras so that we could take pictures over the soviet union above surface to air missile height. And then play that through, you get satellites, better system, better cameras. And then ultimately, over the course of 3 decades, you get us some really substantial industry, which is commercial companies running satellites or to mix metaphors, radars, taking pictures of things, feeding data for national security purposes. Almost the vast majority of national security imagery today is not owned by the government. It is commercial satellites feeding national security organizations and also feeding the private sector. So if you can think about that, that's a high fidelity data asset pictures turned into a really substantial business model. What I love about where we're building is this transition to product. So -- you would not have built this except for COVID, but we spent a lot of time and effort building what is essentially very neat. The operating system, the U2, if you will, does not look like an airplane, it would be cooler if it did. The operating system around what is today the genomic camera, the sequencer. So the sequencer, 20 years of genomics revolution, still super nascent. Bloom and Oxford. Everybody else has done a great job getting those 2 places. Almost every country that we visit in the world and they spent basically the last 24 months on the road globally because governments are the customer, have sequencers, but they use them for very niche purposes. It is not the high altitude version. They are using them for clinical. They are using them for the high-end diagnostics but they're not using them as a general-purpose aperture to take pictures of DNA and RNA on the planet where the biothreats will emerge from. And so what we've been able to build is essentially an end-to-end software and services system. The software that runs a radar station and stamp it around sequencers around the world and work on biosecurity, biosurveillance programs with countries that are looking to utilize their installed assets more effectively. So when we go in, we don't need to buy assets, we don't need to buy sequencers. We are building a system end-to-end, which collects data from airplanes, that's our CDC program, moves it through to a lab. It feeds genomic data off the other side into a usable form for biothreat detection. And that is the core set of assets that we're able to take and deploy. Now what we think of this as is much more like a software problem than a wet lab biology problem. So now once we have sequence data coming up, each day, we were just versioning newer and newer assays and analytics into the system. So folks have read about -- we have one program that we did with IARPA, which is the intelligence community's DARPA, where we're essentially a good old-fashioned AI version of comparing sequence data against known natural and known engineered sequences where we can determine if something has been genetically engineered or not. That's just a software package that goes on top of diversion, a radar station, and we're constantly thinking about what are the next detection algorithms that we can put on top of this network. And that gives us a lot of stickiness with our partners, countries around the world who are constantly looking for better and better detection because they all have this security mandate in their own countries. And are also interested in preventing what frankly happened over the last 3 years or what they're all seeing in the threat service coming up. We've taken a very laser-focused approach to deploy this country by country by country. Again, because for the near term, countries are the customers because they have the national security mandate that is who spends resources to secure their populations first. So we're operating 8 international airports. We've announced 11 MOUs. We're monitoring essentially just over 100 countries because we're getting flight origin and that gives us unique data that nobody else has about how pathogens are flowing around the world. I'll give you a sense of -- so you can like understand our press releases. It's essentially our enterprise sales model. What we do, as we go in and we look for countries that are interested in investing in biosecurity overall. And we sit down with them, we talk to them about what is it going to look like to deploy biosecurity capabilities in the 21st century. We start scoping that out we generally, in the way it works with countries signed an MOU to then explore things further in a program design mode. We're almost always starting with airport programs but not exclusively. And so this is an example of what we're doing in Botswana, which we've announced. We signed an MOU. We've had teams going back and forth scoping out monitoring programs, and we'll be launching that one, and then they move into operational programs. These have no money associated with them. This is where you get into paying programs and our view is that we should be the biosecurity infrastructure and partner for Allied nations around the world. And that's essentially how we're thinking about the process of expanding our footprint and expanding our biosecurity network. To kind of come back to the beginning and close. I think there's a very cool synergy that happens over time, and it's not a long period of time between biosecurity at Ginkgo and cell engineering. To build that defensive targeting cycle that we're talking about, detection to prioritization characterization to building mitigation or enabling mitigation, medical countermeasures, vaccines, therapeutics. You have to have the whole arc. You have to have the network of the whole Arc. And over time, we very much see both sides of Ginkgo's business feeding each other to kind of close the loop from biothreat detection to mitigation and response. This is certainly like how you think about detection designing and responding over time, enabling a network of partners, something we are playing for. But today, we are laser-focused on how do you build that network that we just talked about, how do you deploy the bioradar systems most effectively. I think that's all I have and happy to answer questions.

Can you give couple of examples on 6 to 9 months back [indiscernible].

Yes. I think our favorite example is just an early proof point. When BA2 Omicron came out, we detected it 42 days before it was found clinically. In -- because this is the big -- so like the big transition is not that you should replace current surveillance systems, current monitoring systems, which are mostly or most often clinically based, right? Somebody shows up, they're sick, flag it. It's that you should build a layered system. So that was a very cool example. We were able to flag that to CDC, they're able to characterize it. Understand what the flow in and out of the country was. In that case, it's a public health use case, right? It is let me make nursing homes aware that a new highly transmissible variant is coming. This was a very cool example of an end to end where we were still serving nursing homes kind of during the outbreak mode, and we were able to notify nursing homes in our network with kind of 42-day-43-day notice. That there was this variant that was showing up on the shores. They were able to reinstitute masking policies, push vaccines, et cetera, in their own communities. So that's a public health use case, just a little bit more early warning. What we're starting to see now, we're working -- we've done pilots with the CDC flu program. So you can imagine identification and collection of flu variance from the Southern Hemisphere being able to go directly into the vaccine manufacturing process, and that's something that's been explored just in kind of a pilot mode.

[indiscernible] How much advance press and identify exactly what's [indiscernible].

So I think the 100-day like mission is like very lofty, but I also think that from a technology aspiration standpoint, we should be thinking about how we collapse that even more. And our view is the biggest gains are in how early you detect all the variants that are circulating. And then there's a separate [indiscernible] of how you model those variants, right? But I don't have a great answer for like what the art of the possible is. But kind of like in every other domain we've been able to take and shorten time lines far more dramatically than I think our aspirations are today.

One last question, what's the pushback you get from [indiscernible].

Yes, I still think it's -- something we fight through just -- this is not about COVID, right? So this is about understanding that you've got to be multipathogen detection. A lot of times, people will say, well, if it's coming through an airport -- there's delay 5, 6 days before we get sequences, like what are you going to do? And our general view is that this, the 2 things. One, you have to start with 5 or 6 days. That gives you a lot more early warning than you used to have i.e., people showing up in the clinic. My favorite analogy on this, by the way, is I note is, when we launch regular radar in the 1930s. The first use case was like flying airplanes to look for new boats. They were hitting Atlantic shipping going between England and the U.S. And we were like, I don't know what the exact date is. We're like successful 5% of the time. Nobody was like, you know what, we should stop detecting 5% of the Uboats, we should just get better, right? And so a lot of the times, you get the pushback [indiscernible] from folks who are like sequencing is too slow. It is -- it's only detecting these 3 -- when we first launched, we usually COVID-flu. So only last 2 or 3 pathogens like we want to do it all now. And we're like, okay, but you have to start laying infrastructure and then you have to build this plan that we were talking about to delay that over a number of years, so you actually have comprehensive detection. But we fight this battle of like it's not the all seeing eye today, but it will be and trying to get people activated even around that.

[indiscernible]

So. I love that second question. We have first one question too. Oh, I'm sorry. I'm sorry. Yes, I apologize. I was informed that I needed to do that, and then I failed miserably. The question was, first, like how should we measure how well this is going? And then the second was, are you still just doing wastewater? Or how do you think about other modalities like air, et cetera? First one is I think the best -- I mean, look, we're -- this is a market that's building, right? And we're watching governments change their biodefense posture like monthly. But I think the right way to think about it is that like enterprise sales pipeline, if you will, like we're able to announce MOUs and every one of those MOUs were in engagements, some fast, some slower talking to people about what their program design is. Sometimes countries are willing to announce the operational program. It's really not on us, we'd love to announce all of them. Sometimes they want to hold and like see it working. But it's really like we can show you that top of the pipeline of the MOUs and then kind of those flow through. That would be the best answer I can give you on that one. On the modality side, this is why we think about his infrastructure. And Ashish Jha, who has been a great thought partner for us, both before we went to government and in government he said this really, really well about the CDC program. who's like, look, is it wastewater, is it airplanes? What we're laying is biosurveillance infrastructure. And so it's exactly how we think about this operating system. We are sample agnostic. Like right now, it's a lot of way to answer question directly, but there's a lot of wastewater, but we're also doing in Rwanda, U.S., other places, we're also doing different types of samples, all anonymous focused on nonhuman biological data. So we have a very hard line on that. But -- and then we're starting to pilot air in airplanes, like why wouldn't we go to that? But it's really the key thing is like you got to get to sequences. It's not about yes, no detection. We're like really focused on getting to sequences.

So this year you guys talked about a $100 million plus a revenue, 50% in comparing -- as you regard to one platform is the one application of which testing -- how do you actually believe in a referring.

Yes. No. I think the big picture way to think about that is the 3 years of COVID response dollars, which essentially ended on June 30 versus long-term national security programs. That's the most detail I can give you right now, but you can look at it. June 30 was all public, it was the last day officially of kind of the federal dollars, that ended with the May 11 public health emergency being over. And so it's really how do you think about the transition to long-term national security? Always with the infrastructure in place if something else happened that we can spin it up super fast, right? Like that's the -- I can't think about what that looks like over time nor is it the foundation of the business, but there is that kind of the kind of emergency response capability that could spin up super fast.

So 50% would be PCR test [indiscernible]

I'll leave that for a later. Yes.

[indiscernible]

Yes. Last question, by the way, was, how do you disambiguate between nonrecurring and recurring revenue, plus or minus, how do you just invite between those 2 PCR testing to sequencing and bioradar-type work just to repeat the last question. So this question is, if I say back to you, like, first one is kind of why wouldn't anybody just do this? Like what's unique about what you guys are doing? Are you just comparing these to databases and finding new variants? And then the second half, say, the second half again.

[indiscernible]

Yes. Government is putting their head in the sand. Government is not wanting to know the information. Yes. So in the first half, like one of the really cool things that we've been working very hard on, and it's like the outward-facing version of what we do every day here at Ginkgo, like a massive, highly complex bioinformatics and computational bio kind of expertise. We've kind of the first time at scale, basically turned a separate version of the team, which lives inside of biosecurity and pointed that out. So we're now we work very closely inside on everything that you're going to hear about AI later, et cetera. But it's not just about can we -- can we run sequence data and compare it to variant databases. They're constantly looking for novel variants running custom computational tools against that -- against the data that's coming across for many of these countries in partnership with the smaller kind of by definition, less capable of bioinformatics teams in each country, we're essentially running remote bioinformatics for a lot of our customers. So it is very much a core kind of piece of our value proposition. When I say like versioning the operating system, that's very much what I'm talking about. And the second side is like an interesting it is clearly a dynamic that happens, like do governments like hesitate to know. I think we've got -- we're seeing that like go away pretty rapidly. It's like -- it is the previous question of like what do governments push us on like, do we want to know. We can get through that pretty quickly. And as the network has gotten big enough, enough places are able to kind of share their data. So it's not -- and we see how global air travel works, like people have realized that it's not like you can stop it from going one place to another. You actually want to know early enough. And one country is like willing to share their data because they will know it earlier. And there's still like a public good component to this that people are pursuing. We haven't -- I think where the big question marks will come in is when we -- when you get into the really serious disease and you start finding those in places you didn't expect. But generally, people realize that they should -- they really do want to know that information. But it's a question that people ask a lot.

[indiscernible]

Yes. So there's like 3 layers of this question. This kind of integrated service for biosecurity globally, the market dynamics to be really direct where we run into other companies, it's really BGI. Building strategic biosecurity biodefense relationships with countries. That's kind of who we see when we're talking about this category. That's a global market dynamic. Separately, I think you will -- and I can't talk about all of it today, you will see kind of the more technically developed economies coming online with similar versions. And I think we see a market there for sure, even being an American company. It's just a developing market. And they also have homegrown capabilities that are more substantial. Whereas in a lot of the more kind of nascent biotech markets, people are starting from scratch. And they know they're not going to develop homegrown solutions. So they're like, okay, we're going to -- we want to invest in this like we invest in CyberSecurity. We know we had to buy cybersecurity from Norton and McAfee and whoever. We know we're going to have to buy biosecurity from where -- where do I want to buy it from somewhere, where do I want to buy it from? from Ginkgo, do I want to try to buy some aluminum sequencers and hire some buy-in from cities and do it myself? Or do I want to buy it from do I want to buy from BGI, which is a solution that is available. And so that's kind of like the choice that we're seeing. And the question was, what are the market dynamics I'm like doing good afterwards, Megan.

We have one question from online from Eric Kahuna Capital Biotech for biosecurity, which I agree needs more love from investors. Do you see any addressable markets in the private sector? Or will governments be the exclusive consumer of this product line for the foreseeable future?

Yes, it's a super easy question. So yes. So for those in the room, the question was biosecurity needs a way more love from investors. I think that was the first part. But the second part was, do you see private sector customers emerging? Or will the foreseeable future be government customers only? And I think the short answer, I showed that the Edwin Land kind of how did Geo develop as a dual-purpose data assets that you could turn into dual-purpose national security governments and private sector tool. I mean, we don't have Zillow today without that, right? Or Google Maps, et cetera. 100% do we believe that the like DNA and RNA of the planet will be a high fidelity data asset that will be commercializable both by governments and private sector. It is a foundation of what we think about here from a cell engineering standpoint at Ginkgo. It can go like absolutely. And I think the nearest-term use cases in biosecurity, you would see that as like things like supply chain management. right? Like what happens when CVS stocks out of Tylenol in Arizona because they didn't know RSV was going to spike. That's insane. Like we should know that, that's happening. That's a very pretty basic, like biosurveillance and monitoring projection question, right? Like -- and you looked at the supply chain kind of management that happened via HHS and friends of ours during COVID, the infrastructure for like sharing how you should distribute medical products around the U.S. already exists. We just don't have the data feed on the front end to do it proactively. Now that's all well and good. We do think that, that's a little further out. There's a lot that needs to go into that to make it a really viable data feed. So I'd say not for the next 10 years, but like next 2, 4 years, like the 100% the major customers are governments, like that is without hesitation, and that's certainly where we're focusing our time. National Security almost always comes first before commercial applications of this stuff, if you look at like kind of the history of deployment of them. Great. Thanks, everybody.

Thanks, everyone. Thanks, Matt. We'll be taking a break for lunch. So for the people online, we will be back at 1:30 for a presentation and fireside chat on AI. We'll see you then. For the folks in the room, we will be taking a break for lunch, we will do breakout sessions... [Break]

Hi, everyone. Welcome back. I hope you enjoyed your breakout sessions and got some quality time with our execs and our great exec team. Our next panel and presentation is going to be from our Head of Corporate Development and new Head of AI, Anna Marie Wagner.

All right. Thanks, Megan, and I really appreciate everybody coming out of this. I know it's a full day, and -- but it's great to see you all, and I know the rest of the team enjoyed getting to meet you. As I have been spending more time in AI, one of the things that has been sort of remarkable to me is just thinking about the last couple of years and just how much sort of public appreciation of science has evolved in that time. Like a couple of years ago, we were all remarking that PCR was on the front page of the New York Times. And now our parents are talking about large language models and neural networks. And I think one of the things that's really appealing about AI today is that it's actually quite accessible. The transformer architectures are relatively easy to understand, there are rules that we understand about how this kind of stuff works. And you've got 26 letters that make 200,000 words or so, at least in the human language. And you can assemble those into sentences that with enough diversity that can sort of represent human understanding. Biology is also a language. You've got 4 base pairs in DNA. Those 4 base pairs combined made 64 kind of words or codons, and you can assemble those codons by making amino acids into proteins that follow some biochemical rules fold up. And then that one little thing creates all this diversity of life that we see around us. But there's one really important difference. We created human language. We understand the rules. We know what good looks like. Biology created us. And so when we think about the last couple of hundred years of biological innovation, it's really been an area of discovery, right, scientists, bread goes moldy, maybe got antibiotics now. How do bacteria defend themselves, "Oh look, we just discovered restriction enzymes and CRISPR." So all of this, like these major scientific breakthroughs have really been a process of discovering what nature created for us. But here at Ginkgo, we are trying to make biology engineerable. And to make this engineerable, we actually have to understand these rules, we have to understand what takes this into this, we need to understand how these things work, how they function. That's something we can do in human language. It's not something we really know how to do yet in biology. And so one of the things I'm most excited about as we think about AI is that it can really help us deal with the sheer complexity of this problem. It's not intractable, like there are rules that govern this stuff. We're still living in the rules of physics and chemistry and math. But it's just so complicated that we can't look at the rules and figure it all out. And AI, with enough data, with enough compute, can really help us start making those types of advances. Now if we think about human language, it's sort of in this interesting spot, we're really good at human language. Again, we wrote the rules. We know what good looks like, most of us can speak it. And so interestingly, that gives us actually a really high bar for AI. It takes a lot for us to be impressed with AI. Most of the news articles about ChatGPT are like, "Huh, isn't it funny it can't add 2 plus 2?" And we feel good about ourselves because we're still smarter than AI. We get frustrated when we try to bring an AI assistant on board and it can't really replicate our tone of voice when it's writing an e-mail for us. So it's really hard to make AI models that actually make us better. And what's been so remarkable over the last year or so is we have seen that with enough data, enough compute, these AI models are starting to do something that looks a little bit like reasoning, they're starting to connect the dots. I'd like to believe that wisdom is still the domain of humanity and not AI, but it's starting to look a little bit like that. And this is one of my favorite, examples from learning [ grammar ] as a child, but you can plug this in, and it knows that pandas are not murderous [ dinner gas ] and that they just like bamboo. It's kind of impressing, it's impressive. It's starting to impress us. I'll leave this one for you just because I still like feeling good about myself and my own wisdom. And so I tested ChatGPT on that little cartoon, and I will leave that for you, for your enjoyment when you have more free time. But we're here to talk about biology. So biology is more complex than human language. We don't understand all the rules. And by the way, we don't all speak biology today. Only a few folks that have gotten PhDs are really capable of making the types of discoveries that are advancing the field in meaningful ways. And so we're living in a very different part of this curve. And again, I would pause it that AI has the potential to be really impactful here. There's a lot of data that we sometimes dilute ourselves into thinking we have about biology. We've got sequence data, but I can assure you that none of you know what that sequence makes, even though it's only 613 letters long. I can even tell you what amino acids that make. I can tell you how it folds like plugging in an alpha fold. We still have no idea what this protein has. No one in this room does. And I can assure you, no one at Ginkgo would be able to look at this and say, "I know exactly what that protein does. What you really need is you need functional information. You need to know, where does this show up? Is it secreted? What metabolites are created when this thing is around? What does it bind to? Where else have I seen this? And what can I deduce from that complexity, from that level of information? And this is the type of information that Ginkgo is generating every single day in service of our programs. By the way, this protein is the reason we can all walk. It's the reason we have balance. It literally creates little calcium crystals that makes old rocks in your ears. It's amazing, tiny little protein. So in Ginkgo's world, we are generating this kind of labeled training data, that functional data that answers the questions of what does this thing do and why, in service of the hundreds of programs that we are working on. And I can't emphasize enough the cultural shift that's happening here. When I was in your shoes about 5 years ago and I was doing diligence on Ginkgo, in that case, deciding to leave a cushy investing job to come here, the critique that I heard about Ginkgo was Ginkgo just throws spaghetti at a wall and sees what sticks. They're just doing brute-force experimentation to figure out the results. They're not -- it's not the art of biology. And now, I was walking in to work last week and listening to a podcast, and you have the CTO of Bristol-Myers Squibb saying, you know what, these models are saying we should put a compound together. And any of our scientists who knows anything knows that that's not going to work. But we're going to do it anyway. It's still worth doing because understanding how and why that experiment fails, is going to make our model better, and that model is an important component of what is making us better as a company. And so he's highlighting this big cultural shift, and I can't emphasize enough just how much we're seeing that cultural shift across our customers. Now because Ginkgo has been generating all this data for so long, we've been thinking about how to apply that data for AI and ML programs for many years. For those of you that are new to this space, Josh Dunn, who's our Head of DNA Design at Ginkgo, read a great little kind of summary paper on machine learning across kind of biological engineering applications with some folks at Lawrence Berkeley National Labs. He also has the technical lead for our NDAR program, which for those of you who joined our Biosecurity breakout session or spent any time around Matt [ under ] know is part of our kind of biosecurity infrastructure around identifying genetic engineering in biological samples, kind of genetic tampering, you might think of it that way. And then our protein engineering team is a really remarkable use case of this technology. We do protein engineering across basically every program that we do at Ginkgo. Each of these lines, those are protein engineering programs. And each of those colors are different classes of enzymes we've had to engineer for those programs. And so the reason that we've been able to offer some of these like really game-changing kind of value propositions to customers like success-based pricing that's unheard of in our world, the reason we're able to do that is because we have now gotten so good at predicting whether or not a program is going to work and reducing the cost of doing that work in the lab through computational design through applying these tools and that we're able to do something really quite different. All right. So we talked a little bit about data. We've talked a little bit about the AI models. What I want to emphasize is that Ginkgo thinks this needs to come together. And we will be consuming this data into our models as fast as we can create it. And there are different levels of data and there are different levels of models that matter here. And what really makes the difference in kind of next-generation AI tools today is if you have a really strong foundation. So I think this tells me what proteins look like, proteins that work, this is kind of what they look like. Stick just a ton of protein data in there, and you've got a model that understands generally, right, that's a protein. That's not good enough to tell you, "I need a protein that binds to this organ and doesn't create this immunogenic response and is thermostable because this is my supply chain and, and, and." For that, you need these task-specific models. And to answer those questions, you need the kind of data that is coming out of our foundry, that labeled, functional data. And so we really view this as an -- kind of an interactive process. It's the same flywheel we've been talking about for as many years as you've known us. But AI is really a tool that allows us to take advantage of all that data that we're generating and not just the data for the successful experiments but also the data for any experiment we do that really helps us understand how and why things work or don't work. The most common question we've been getting since we announced our Google collaboration is, "All right, what models are you building first? And how much data do you need, and what is that data?" Jason mentioned at the beginning of the day that our first set of models are going to be in the protein world, so a protein foundation model and set of applications on top of that. The reason for that is it's at that nice intersection of we have a lot of data, and we have a lot of customer demand in that area. And so it's a great place to get started and start building the foundations. But we absolutely see the value and intend to be building models across, if you will, the central dogma of biology. There is a lot of value in the work we do to understanding the rest of the genome and how DNA works, not just how proteins work. And then if you want to think about the data that we have that's going into these models, yes, we also benefit from discovery. There's a lot of natural genetics that are out in the world. We have a couple of billion proprietary sequences at Ginkgo that we can combine with all the public databases to create a really rich data set to start training these models. But we also focus on creating diversity so that we can start testing new things that maybe aren't showing up or aren't showing up very often in nature and we can start understanding those new functions. We then can apply a whole set of measurements on that, right? What are the genetics? Is it being transcribed? What metabolites is it making? What proteins is it making? What are those proteins binding to? Et cetera, et cetera, et cetera. And that is the data that has been training all of these kind of applications that sit on top of the foundation model that we're building. I think by now, you're all familiar with our Google deal, but certainly happy to speak more about it. We've got the data. We've got the sort of biological insight. What we needed, though, in this new world of how do you build foundation models with lots of data is you need compute that will scale with you. And so we were looking for a partner that would give us that asset. And in the same way that we've built a scalable foundry that can work on many different programs at low cost for customers, we needed the same asset for compute. And so really happy with our partnership with Google there. But then what was so interesting about creating this relationship with Google was that they saw and asked a couple of things. One was what we've just been talking about. Ginkgo has a lot of data. And I think they're hearing over and over from their potential customers in the life sciences is, "Yes, we'd love to use AI. But we don't like -- how, we don't really have enough data to make it super useful." And so it's been really hard for them actually to go penetrate big pharma with compute as the product. And so when they look at Ginkgo, the way they think about Ginkgo is well, maybe compute isn't the product, maybe Ginkgo's model is actually the product. And what we want to do is we want to help Ginkgo build better models so that Ginkgo can then help bring the Pfizers of the world, the big pharmas of the world on to AI, onto the cloud via their models. And so that was really the basis for Google giving us funding to really accelerate the development of our models. We announced it about a month ago. It's been a really, really great relationship getting started with them, and we're sort of off to the races. So that's been fun. I'm going to turn it over to a panel in just a second. I'm really fortunate to have 3 great thinkers about -- in AI with me here. But knowing that this is an Investor Day, I did just want a nod to the how-do-you-make-money-in-AI question. Obviously, this is going to be a part of our platform. Platform improves with scale. AI improves the platform. We will deliver better value to our customers, we hope. And we will generate value in that way. But one of the reasons I'm particularly interested in spending more of my time in this space is I do think that there are new opportunities that are emerging with AI today. Again, there's this kind of cultural shift happening, especially in biopharma and especially in biosecurity, thinking through what is our AI strategy, what is our data strategy? Who do we work with, who helps us figure that out? And I want Ginkgo to be the partner that is helping all of these customers figure it out. And that could be how do we help create data together that helps answer these questions. And it could be, how do we create models that are useful for the types of problems that you're interested in? Certainly, many of these models we will keep internally. But we do plan to release models broadly. And you can also imagine models that we would be developing in close collaboration with partners as well. So more to come here. I'm very, very excited about this space. But I wanted to give just a quick introduction before launching our panel here. And so while I do introductions, I would invite our panelists to come up, and we'll get some chairs set up here. And there will be time at the end for questions, don't worry. But joining me is Shyam Sankar, who's been on Ginkgo's Board since 2015. And very happy he recently agreed to serve as our Chairman. But he's better known as the CTO of Palantir, which he joined in 2006 as the first business hire and has, as far as I can tell, led just about every function in that company at some point. And so as we've been building Ginkgo, any new function we had to take on, Shyam has been a wealth of knowledge and experience and advice over the years. We've got Barry Canton. Many of you were fortunate to tour the foundry with him earlier today. Barry is Ginkgo's Co-Founder and CTO. And just a little side note, something I've noticed is like a super power of Ginkgo's is that we have 5 founders, all of whom are still with the company after 15 years, somehow. And they all really complement each other. Barry, who is the mechanical engineer of the bunch, he's really responsible -- as he is fixing his microphone for those of you who can't see that, we'll see if he can figure it out. He actually was a mechanical engineer. That was not just a really well-timed joke. He is responsible for figuring out like how we were going to make this like messy, wet, unpredictable field of biology scalable and standardized. I think honestly, we're here talking about the potential of AI at Ginkgo because of many of the decisions that Barry has made to create the framework with which we could then generate biological data at scale. And then last but certainly not least, we've got Dmitriy Ryaboy. Dmitriy is Ginkgo's VP of AI Enablement and is responsible for our long-term technical strategy in AI. Today, he's focused on building our AI infrastructure, optimizing our model architectures and working with our scientific teams to design, train and assess our AI models. I found this out recently. So Dmitry has a long history of working at the intersection of biology and computation. He started at Lawrence Berkeley National Labs in the late '90s, meandered through the Internet boom, including building a lot of Twitter's data architecture. And then most recently, he led the digital technology organization as CTO of Zymergen. So please welcome this panel, and we'll get kicked off. All right. I feel very far away from you guys, I'm going to scooch up a little bit.

So the first question here is, I'm going to give it to Barry and Shyam. So both Palantir and Ginkgo have something called a foundry. And I think they look pretty different. But I'd be curious, where did the term come from? And what does it represent to you? I'll start with you, Shyam.

Great. Yes, they are, in fact, quite different, but I think they probably share a philosophy in common. So at Palantir, when we think of data, we see fuel, not exhaust. And that's the raw material to decisions and the decision-making process and really thinking about the flywheel effect you get from doing that right. If you visualize an institution, it's not one decision, it's kind of a chain of decisions. And anywhere you're touching and poking this pressure system, you have an ability to affect how the institution runs. And so you want this factory that allows you to make and produce those decisions as effectively as possible. And I think one of the things that's exciting with AIP and AI is taking that same philosophy from how do I turn data into decisions to how do I help customers build AI-enabled application forges to do that across their business.

Yes. I would like my answer to be data is fuel, not exhaust as well. I'll add the kind of some of the Ginkgo-specific stuff. And yes, I think a lot of what Shyam said applies to Ginkgo from -- at a philosophy level and at an abstract level. But to give you a little bit of a Ginkgo-specific context, obviously, we are a platform company. Our platform is about transforming how R&D is done anywhere that biology is used. And the physical facility, the lab is at the center of how R&D is done. And we wanted to transform and continue to want to transform the vision for how that works. And a part of that was to kind of establish a break and use a different name, a different term and lay out a different vision that kind of makes a break from the conventional notion of what a lab is. And for us, the foundry [ biology ] to a semiconductor foundry was the perfect term for us to settle on at a number of levels. First of all, as you heard from Jason this morning, the separation of the physical activity of the R&D from the design activity is something -- is a separation that we wanted to create, and that wasn't really present at all in the life sciences when we started building all of this. Once you have that separation, now you can start to have a lot of design activities centralized on a shared general purpose, a physical platform for doing the work, the foundries that we toured this morning and you get all kinds of operational efficiencies from doing that. Further, once you centralize it, you can start to automate it, you can increase the operationalization. It becomes capital intensive. It becomes all of the things that semiconductor fabs are, and we see a lot of analogies there that we want to -- that we are continuing to push after. So yes, great term from our perspective for what we're trying to do.

Appreciate it. All right, Dmitriy, when you were at Lawrence Berkeley in the late '90s and early 2000s, again, I found this out recently, it's been fun; people you work with most closely, you don't know them until you introduce them for a panel. You developed something called the VISTA Genome Browser 2.0. What was that? And could you predict then what we would be doing at the intersection of computation and biology 20, 25 years later?

The Internet remembers. Yes, I then took like more than a decade off of biotech and went into consumer Internet. So I came back. I came back. So it was the second version of the VISTA Genome Browser, obviously. The VISTA Genome Browser was -- this was happening when -- so I was an undergrad and I just happened to get a [ program ] and job. I was looking for pay my tuition at Eddy Rubin's lab in [ LBL ]. And that was right around the time when the human genome project was just about to finish, and then I was there as it published, both of them published, and a lot of other sequences started coming online. So you got human, you got monkey, you get pig, and then you start getting microbes and all kinds of things. So there's this explosion of data that was just coming at a very different pace than the field had been used to. And that was before NGS came along, next-generation sequencing. And people are doing a lot of very interesting comparative genomics, compare the human genome to the mouse genome, find areas of the genome that are very similar and get multiple genome alignment, and that tells you where the genes are, right? Like that was a problem, figure out what's conserved between -- through evolution. And so it turned out that the actual Genome browser was like a Java applet, if you remember those. JavaScript wasn't quite right yet for providing the results of these alignments to scientists across the U.S., who could kind of navigate them and look at them. But behind that was the more interesting part to me, which was how do you actually get those alignments because now we were at a place where the data was way too big for us to do it the way people were used to doing it. So we had to get into distributed computing. We literally built a rack, like a wire rack, not -- if you've seen server racks, they don't look like wire racks anymore. It was a literal rack like you've seen around here, with [ beige ] boxes on it, all wired together in a literal broom closet. We had to keep the door open, so it wouldn't get too hot. And so we built all that up and wrote up a bunch of software to have these things [ talk to each other ], so they could run these alignments. And so that was kind of my introduction to distributed computing and to what became known as big data. And then that actually set me on the path of doing big data for internet companies that have a lot of data and eventually for returning here.

So could you predict then what we're doing now?

Right. The second part of the question. I think because it was -- I think it was the seeds of what we're doing now. It was right at the point where things let into data is much bigger, it doesn't fit on a single computer. And all these problems that were before kind of more pure science problem, lab problems became, "Oh my God, like we really have to do a lot of computational analysis of this data," and that's how we're going to move the field forward. So it was a seed maybe in acorn. And now, we're like in oak territory. And with AI, we're heading into sequoias and beyond. But there were glimpses.

All right. Shyam, so the last 12 months have really, I think, like rocked the world of a lot of corporations that are all kind of struggling to figure out, "okay, like what do we do about AI?" Didn't think AI was relevant for them before. But suddenly, they're feeling the basis of competition sort of change under their feet. I would guess Palantir sometimes sees this coming before the rest of the world does. And so I'm curious how you think about product development in a world where your customers might not necessarily know what they need and what they want. Do you approach it like, "Hey, we know best, and we're going to develop the product that's right?" Or do you find that it's still a very kind of collaborative and consultative project?

That's like the hardest question because I think the answer is that you have to find a way of doing both. You have to both meet the customer where they are today and what they understand. But if you haven't already developed meaningful conviction in what you think they're going to need 2 years, 5 years from now, then it's not actually going to work. And the way that we've squared this -- what I'm quite excited about with this -- and part of this, I think, is just that people now have an expectation that software is supposed to work fast. If you just take that as a rebaselining of the amount of energy people are putting into this, what we've really seen work now is getting people with their hands on their keyboard. This is not the sort of problem or technology that you can admire and think your way through. Like you have to actually experience it and iterate with it. So like getting multiple customers in a room for a week to actually build something where they're going to exit that boot camp with something they can put in production, has been so efficacious because I can scream until I'm blue in the face that like chat is a limiting paradigm and not how you should be thinking about applying this in the enterprise, that [ LLMs ] are statistics, not calculus. Any of these like deductive frames and it's just like, "Okay, conceptually may be interesting, maybe." But really, it's like, "Oh, I just build something that saved my users 50% of their time in a day." Like they get it. And so I think squaring those two things is the art of this.

Barry, maybe turning to you. We talked just in the introduction a little bit, just at the highest level, the difference between foundation models and task-specific applications. And so as we think about building product at Ginkgo either for our own internal use, for our scientists or more broadly, where are we spending our time and how are we thinking about the technology that our customers and our scientists are going to need?

Well, I think the answer is that we have to work on both of those things. So as has always been true in the history of Ginkgo, we have to think about the platform. And then as Shyam said, we have to think about meeting the customers where they are today and helping solve the problems that they have today. And so the way I think about the foundation models that we're building is analogous to the foundry that we've been building and the code base that we've been accumulating, it's a general-purpose asset that gets better with scale. And the more broadly useful it is, the better it will get. So we are absolutely using the data that we already have access to, both public data but also the proprietary data that we have, to train foundation models that we hope will be broadly useful across markets and projects. Second, on the task-specific model side, the -- our partners have very specific needs. They need a particular protein to be more active or more soluble or expressed at a higher level or they need a promoter that has greater tissue specificity. Whatever it might be, these are very specific problems. And to solve those, it is not sufficient to have a foundation model, you need to have a task-specific model that can address those particular questions. To be able to build a task-specific model, you need relevant data, you need data with the right kinds of labels, tissue specificity labels for promoter sequences, for example, stability data for proteins. If you think about our foundry, what we've been building here is a way to generate labeled data sets for very particular problems that are commercially relevant, and we've been working on doing that now for 15 years. So we have the engine and we have the capability to collect data to train fine-tuned models. And so we'll absolutely do that to help solve customers' problems today. The last thing I want to say is I would like to zoom out a little bit because while there's an enormous amount we can do with better modeling of protein and DNA using the language models that are emerging over the last couple of years, to some extent, the most powerful thing in biology are our cells. These are fully featured little machines that can do incredible things and that self-replicate each other and that we can't build with any other technology. I think the language model tools that we're all able to use today are going to make it easier to understand and program cells at the cellular level, but I don't think they're going to be sufficient. We're going to need better AI tools, and we're going to need to be able to integrate mechanistic modeling techniques in order to be able to model and predict at the cellular or tissue level. And that's where a lot of the true value in the future is going to be, proteins and small molecules and DNA-level work can be extremely valuable today. But I think the broader potential of this technology is going to be at the cellular and tissue level, and we're going to need to do a lot of new things on the modeling side and on the data collection side to be able to enable those.

Sounds like a lot of complicated, messy data. Dmitriy, you have spent the last 25 years or so working on lots of complicated, messy data. How do you think about building the infrastructure for Ginkgo that can handle that complexity, that diversity? What allows us to start thinking about those bigger questions that Barry is outlining?

Yes. you have to be thoughtful about the foundation. I don't mean the foundation model, but the foundation of the foundation model. Because there's a lot of engineering that goes into sort of enabling those things. That's why my title is in the AI enablement, that's the tricky problem. So organizing the data, making sure the data is captured in the right ways, the data is relevant, you can actually interpret it later, you can look at your models and find out where the data that went into those models came from, you can figure out when your results are off, and you can build a feedback loop. And that's all kind of abstracted from what exactly is the data, is the data from the LCMS? Is it the data from our [ HDS ] screens? And then, how do you organize that information so that it can provide appropriate input to a model that sort of fundamentally, you don't want to overspecify to the individual types of input. So there's a lot of mess in there. Fortunately, to some extent, this new paradigm for how we build AI models lets you get away with a fair amount of mess, right, like human language is messy, images are messy, and yet we're able to extract meaning out of them. So -- and the models really shine in sort of one -- a domain that's very expressive. You can express pretty much anything with language, that's what we do. Images are very expressive, right? So that's a very good kind of problem for these models, where it's something that's very expressive and very complex, right? But there is an internal structure to it. It's not random. And they're able to elucidate that structure internally. And that's the kind of data we're dealing with in biology. And that's why, fundamentally, we think this is going to work.

So I think we biologists like to think we're special and feeling like data is all that matters, and it's the big, hard problem. Shyam, do you run into the data problem in your world? And if so, what does that look like?

Yes. I think one of the exciting things about the current approach with LLM, in particular, in the enterprise, is that there's all sorts of data that nobody used to even think was economical to capture that you now can. And part of this is actually elucidated by trying to solve problems. So if you are trying to use something like retrieval augmented generation to service very high-end equipment, to automate and build a copilot for maintenance, the first thing you're going to go to is like the maintenance manuals, except the reality is those 10,000-page PDF documents are wildly out of date. No one maintains them. And you kind of have to mark-to-market as soon as you try using them that it doesn't actually have the source of truth. But you know what does have the source of truth, the slack rooms, the [ Jira ] tickets, the audio recordings of the video conference calls you're using in your incident response to debug these things. That is otherwise historically treated as ephemeral and useless data that is unstructured and irrelevant, but it's actually completely trivial to structure now, and it is actually the most relevant data. In fact, all the canonical historical sources are known to be inaccurate. And so if you kind of string this together where it's not just how do I solve the end part of this, but what are the new sorts of data that actually are much higher fidelity but historically harder to capture, you get a lot of value. And I suspect there would be analogy there to the sorts of data you're able to capture to the foundry.

Do you agree, Barry?

Yes. Yes, I think that is absolutely true for us. A lot of the insights, a lot of the interpretation have been treated as being a femoral, and now they can be integrated with the kind of the harder, more structured data, and that's exciting. Actually, one of the things that Dmitriy is working on how to surface and structure that information to make it easier to get at.

All right. So shifting gears a little bit. Shyam, this is still for you. So you spend a lot of time working with government customers. We obviously have a biosecurity business also working with that type of customer. How do you see them thinking about AI? Are they viewing it more as a threat or an opportunity? And how do we build that trust?

I think folks in government, thinking is the right word. They're doing a lot of thinking about AI and maybe not as much acting on it. I think that's created a lot of opportunity for companies that are incorporating this into their products, so just taking ground, manifesting the facts on the ground by rolling out their products that incorporate these things. But it speaks to the underlying reality that AI is this experiential technology. You're not going to be able to think your way through it. And the advantage is going to accrue to the people who are experimenting most rapidly with it. So I don't want to paint the government in broad brush strokes. So there are pockets of deep innovation where people are going fast. But by and large, there's a lot of admiring how exquisitely interesting the problem is and not enough hands on keyboard.

Yes. Well, I think at least when I hear folks talking about the dangers of AI, biology is often one of the things that they bring up, right? Like what happens when we can start messing with biology, it's sort of a touchy subject. So Barry, I'd love to get your thoughts on just these concerns that these times hear about the risk of moving too fast in AI, especially as it relates to biology. And so how do you think about reconciling the huge opportunities? As I think about biology, it's probably the only technology that can deal with hunger and health and climate in a real scalable way but with the potential for misuse or risk.

Yes. I think why we started Ginkgo and why all of us are still here is because of that potential to solve global scale challenges with biology. So I would say that in the biological domain, we have 50 years of history now of dealing with really kind of monumental breakthroughs that have raised questions of biosafety and security. And so actually, the community has spent a lot of time wrestling with those questions and thinking about them going all the way back to the discovery of recombinant DNA in the '70s, the publication of the human genome, the discovery of CRISPR. We have been -- deployment of engineered crops. These are questions that society and experts have worked through. Are the systems perfect? Are the regulations perfect? Obviously not. Would we like things to move faster, obviously. But we do have a kind of a multilayered protection system in place, everything from the level of DNA synthesis screening to labs being able to operate through the controls around how engineered biology is deployed. So I think we have a lot going in our favor. I think it's really gratifying to hear every time I hear from Matt about the progress on the bio security side because I think that's like filling a huge gap. Also on the biology side, we spent a lot of time wrestling with what data it makes sense to have in the public domain versus to be kept private, both from a commercial perspective, but also from a from a bio safety perspective. So I see AI as being an accelerant here to what we're able to do. But I think that we've been working on a lot of these questions for a long time with other scientific breakthroughs. And I think that we will be able to use a lot of that in this case, too.

One of the things I think is sort of most interesting as we think about this problem is some of the same things that we think about purely for commercial reasons, right, how do we protect our IP, how do we protect our data, how do we protect our insights? So the same types of questions I think you need to wrestle with when you're also thinking about kind of the security implications. Biosecurity, it's still how do you regulate access, how do you regulate data, et cetera. And so it feels like we have a sort of a unique place in the world just by virtue of the way that we've built the business to try to wrestle this with a lot of these hard challenges.

Yes. And I mean I think the way the commercial incentives will line up is that it's fast-moving commercial entities, working with proprietary data and data generation platforms that are going to make the most progress. I think efforts to build large federated public databases. They're going to run into all kinds of misaligned incentives and structural problems, and we'll just move very slowly, and the speed and the progress will come from commercial entities operating in a kind of an agile and focused way. The other point I forgot to mention before is with biology, unlike, say, software, the -- even if we said, "Hey, we're going to slow down technology. We're going to try to somehow use social, cultural and political approaches to slow down the rate of technology change, evolution is out there running billions of experiments every day finding...

Sort of humbling, isn't it? What's the foundry throughput of mother nature?