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

Good morning. Welcome to the 43rd Annual JPMorgan Healthcare Conference. My name is Michelle [indiscernible]. I'm an associate in the JPMorgan Healthcare Investment Banking team. It is my great pleasure to introduce our next presenting company, Ginkgo Bioworks. With that, please welcome Co-Founder and CEO, Jason Kelly.

Thanks, Michelle. Hi, everybody. Welcome to the last day of the JPMorgan conference. You've almost made it. I'm Jason Kelly, I'm the Co-Founder, CEO here at Ginkgo. I just want to start by saying, I'll be making forward-looking statements. I encourage you to read this before my talk and think about it. Okay. So Ginkgo's mission is to make biology easier to engineer. There's been a lot of change at the company in 2024. This doesn't change. So this is why we exist as an organization. I think there's lots of ways to get at this, lots of different business models and technologies that can serve this mission. So I'll tell you a little bit today about what we've changed in the last year as we pursue this. Three big topics for today. First, I want to talk about how Ginkgo continues to grow our business in biopharma. So this is the fastest area of business for us to grow in, in general, in this market, that's really tightened up on biotech. We're seeing a lot less business in the industrial and ag sector. So this is really important for Ginkgo. Second, we did a big restructuring in 2024. So I want to talk about the progress we've made there. And then finally, alongside that sort of tactical changes in the business, we've also changed our strategy a little bit and in particular, expanding into direct sales of our platform via life sciences, tools and services channels. So I'll talk about that. All right. So first, pharma. So I'm really excited about this. We had an amazing year expanding and adding new logos to our pharma partnerships. To give you a sense, we have about 130 ongoing R&D partnerships at Ginkgo, mostly with mid- to large biotech companies. So about 90 of those are outside of pharma. So companies like Bayer in agriculture, Sumitomo Chemical and so on in the industrial space. But now more than 40 in the biopharma sector, including Merck, Novo Nordisk. You can see Marcus Schindler, the Head of R&D, CSO for Novo Nordisk on stage at Ginkgo's event in Boston last year, Ginkgo Ferment. Biogen, Pfizer, Boehringer Ingelheim and also places like the Gates Foundation, all customers now for us on the biopharma side. To give you an example of some progress in one of those partnerships, just in November, we announced meeting one of the middle technical milestones for our relationship with Merck. We have 2 big projects with them. This is from one of them. We hit that technical goal. We got a $9 million cash milestone payment and with that complete, we got to actually expand into Stage 2 of this partnership. So this is pretty typical for these types of we call these solutions deals where we have a multiyear partnership with a large biopharma, we're meant to deliver technical results. If we hit them, we get paid additional sort of success fees and things like that along the way. And then in the future, God willing, if the products go to market, we get royalties and milestones and things like that, also on the commercialization, okay? But this is really a research milestone along the way. Novo Nordisk is another thing I wanted to highlight. We started working with Novo back in June 2022 to work on improving expression systems for some of their pharmaceutical products. We've now expanded that relationship 4x since. And so one of the things to keep in mind with these like R&D partnerships are kind of like co-development, right? They're really close relationship. It's a lot of trust building. And so as we grow in that relationship with a biopharma customer, it is actually, I think, easier for us to keep adding if we show them results along the way. So I'm really happy to see things like that Merck milestone. That means it's easier to add more deals with them in the future. It's really a trusted relationship. And you see that in some of the comments. I won't belabor this but great quotes from our customers publicly about the work we're delivering for them at the very high end of R&D partnerships. Okay. I want to now talk for a few slides about our restructuring in 2024. This was really my main focus last year. And so the big goal here was quite simply to get down our OpEx, our cash OpEx. So we started the year in quarter 1 with an annualized about $500 million of cash operating expenses. Our goal was to get that down $100 million in '24, additional $100 million in '25. We actually are ahead of that. This is a slide from my Q3 earnings call. We are ahead of that $100 million goal down to $375 million cash OpEx in Q3 annualized. The way we're doing that, we did about a 30%, 35% layoff. So that's a big change midyear for the team. And then alongside that, we've also consolidated our laboratory space. So we closed 2 sites in Europe. We closed our site in New York City, our offices in Cambridge and we basically consolidated down to our large laboratory facility in the Boston area to do the majority of our lab work. We have about 200,000 square feet of space there. And then we also have sites in Emeryville that are specialized for building robotics, which I'll talk about in a minute and our site up in Davis, where we do our agricultural work for customers like Bayer. We have greenhouses and things like that. So those sites will keep along with the big site in Boston. But we do have a bunch of extra space in Boston. I mentioned this on my last earnings call for the people tuning in online. If you're interested in 60% lab, 40% office, beautiful lab space in the Seaport in Cambridge, I am happy to be your landlord. So give me a call. Okay. We're also -- I think the team has done an absolutely amazing job. So when we announced that we were going to be doing the RIF. We updated our guide for the year on revenue. I was very happy the November 12 earnings call to say that we were reiterating that. So you can see on our services side and cell engineering, that $120 million to $140 million in revenue. Additionally, we had a onetime, just so you know, a noncash revenue release of $45 million. So we did update our revenue target for that but that's noncash. We did want to highlight that. And then finally, biosecurity, $50 million plus as well. So that level of change in the team while still being able to hit the targets we said we would, I felt very good about that in November and still do. Okay. All right. So I want to expand -- talk last about sort of how we're now making strategic changes alongside those tactical changes. All right? So the point of the tactical changes is real simple. I want to get my cash operating expense down. I want to get my cash revenue up. I should mention, we ended Q3 with $616 million in cash and cash equivalents in the bank and no bank debt. Okay. So I actually have a very strong balance sheet, particularly in this biotech market so long as I don't spend it off too fast. So that was the real big tactical change we made in '24, was just to squeeze that down. So we get our cash burn to where I want it to be in '25 and we don't go through this money and I don't have to fund raise if I don't want to. And I think I feel very, very good about our ability to do that and get ourselves in a position where we're always raising money from a position of strength so that we're not unnecessarily diluting ourselves as shareholders, as employees of the company or our investors. So that was the -- that's really was the focus on the OpEx. But it was a good opportunity for looking at the strategic direction of Ginkgo as well. And we did make a pretty big change there that I think is -- I would love for folks to understand this better because I think you're going to see a lot of progress on this, this year for Ginkgo. So the way we've sold our platform and just so you understand what's in the Ginkgo platform, we make lab automation directly, so we design and build hardware. We build custom software, custom [indiscernible] system, custom data informatics systems that are handling the huge amount of data that's coming out of our labs. We have custom intellectual property around biological code base. We have scientific workflows we've developed over the last decade. That big stack of things, I'll call the Ginkgo platform and we sold it via the business model of those 130 R&D partnerships I mentioned at the beginning, where I basically try to get to know the Head of R&D or someone senior at a biopharma or bio ag company and I say, listen, this technology is amazing. You can't get it anywhere else. You should really -- we should help you codevelop your products, okay? And they say, "Great, here's my business development team. Talk to them about a partnership." And thus begins a 6-month to 18-month dance where we meet people here at JPMorgan and so on and work out IP sharing rights, royalty rights, milestones, the types of things you would see in a partnership between, say, like a CRISPR Therapeutics and a Vertex or something like a small biotech working with a large pharma, those types of deals, that's roughly like what those 130 deals look like for us. It takes a lot of time, right? And so the thing I'm most excited about is we're now taking that same platform and selling it in a new way, additionally, through traditional tools and services. So think more like a CRO or like an equipment vendor where you sell directly to the scientists, like your customer, not to their boss's, boss's boss. Okay? Does that make sense? And just to help you get in your head because I've spent a lot of time thinking about this. And I think for folks in the room that are life science tools folks, I think this is at least my way of thinking about it. So that big, dotted line in the middle, on the left-hand side of that, our offerings where you get a piece of the value of the customer's product. And then to the right, you are basically selling to the customer's research budget. In other words, you don't get any of the value of your drug or anything like that. You're just being paid fees or capital equipment purchases, all right? And the Y-axis is the technical risk and the customization level that you are doing for that customer. So in other words, are you taking on the technical risk as the vendor and how customized is it to what the customer wants? So the extreme form of this is a small biotech, developing an asset, hoping to license it to Vertex, right? That is extremely custom and they're taking a huge amount of risk. They're spending all the money themselves. And if it doesn't work out, they get nothing. All right? But if you come down a little, you get my solutions business, those research solutions, where a company like us or a company like, for example, Adimab in the antibody space, they share the risk with the customer. The customer pays some fees but then the service provider ends up getting royalties and milestones on the back end. Does that make sense? All right. So that's what we've got. I like it. But over here on the right, there's 3 new products we're launching and have launched this year. The first data points is our CRO service effectively where we generate large data assets for customers, typically building AI models. So I'll talk about that in a second. The second is, we are building our own AI models. I'm not going to spend too much time on this today but we have them up on our website, buy an API where you can pay for them with a credit card effectively. And then finally, we're starting to sell that in-house automation we built directly with an equipment vendor service model, okay? And so I'll talk about data points and automation now in a little more detail. But again, just to remind you, still doing solutions. These are larger deals. They have tons of upside with the drugs. We've had billions of dollars in biobucks and milestones specified through those projects and they're great because they get to that head of R&D but they are slow. Still going to do them but adding in now near-term fees, faster sales cycle, tools and services sales of that same platform at Ginkgo. And what's great is I was able to launch data points and automation in a matter of months because we didn't have to build anything new. We were just selling the platform we already had in a new way. Okay. So let me talk about data points. This one I am very, very excited about and I had about 40 conversations at this JPMorgan about this particular product. Okay. So basic idea here is to generate large data assets for customers. They own the IP. There's no royalties. There's no milestones, they just pay us a fee and we leverage our automation to generate that data. Okay. So why are people asking for this? So I think you saw a real leadership over the last 3 or 4 years from folks like Aviv Regev at Genentech, the team at Recursion, basically showing that you could generate big data assets and train models that would be relevant in the area of drug discovery. Okay? And so now lots of people have paid attention to that and they're asking, okay, what does that mean? How do I internalize that? You probably, if you've been at this meeting been hearing about foundation models and [indiscernible] models and virtual cell models and all these sort of AI terms in the biotech space. So here's how I get it in my head, all right, because I do spend a lot of time talking with these sort of R&D leaders in the industry. So on the left-hand side here, I want to just talk about how we discover drugs today and how we advance the science of biotechnology and sort of like disease biology. And the basic way you do it is a scientist reads a bunch of papers, gets a sense of sort of the limits of knowledge, say, in a therapeutic area and they say, okay, I have a hypothesis that I want to test to ask a question to push this field forward. And then they say, great to ask and answer that hypothesis, this is the precise experiment that I want to do. I want these reagents. I want this exact piece of equipment and I'm going to do it myself because I don't trust anybody else and I'm going to do it at a lab bench, and there's a whole tools and services ecosystem that supports the ability to ask and answer that hypothesis, collect the data back, the scientist then gets to interpret it, think about it and do that process again. And that's why, for example, like the Thermo or NEB catalog is this thick, all right? And so that -- those scientists have the ability to go get whatever exact tool they need for the job. It's why when you look around a lab, there's 400 pieces of equipment from 250 different equipment vendors. It's because for that -- they want that very specific assay to ask and answer their hypothesis. So the first point is, I don't think that's going away. Actually, the that works kind of great. It's amazing, the breakthroughs we've made in understanding biology. It's amazing the drugs that we develop with that system. So for all the scientists that are hearing about AI, I don't think it's going to just swallow and replace the [ bench ] or replace them. However, I do think you have to look at -- in particular, I think what we saw with, for example, AlphaFold from Google DeepMind, now 4 years ago, where people have been approaching it with that kind of classic approach on the left in the area of protein structure design. And there was just this breakthrough where it turned out, if you applied a big enough data set and enough compute and a big enough neural net, it was actually a better way to move the field forward and have scientific discoveries. It was a new way, a new different way to either understand therapeutic areas better or to develop drug assets. And that's what's got people excited. Okay? It's not that it's a replacement. It's just different, all right? And so I do think, like to give you some examples of the differences, you are no longer using that bench to do your 1 perfect experiment. The point of the lab is to generate as much data as you can per dollar, okay? The lab is really what we would call a data foundry. All right? That is the point. The point is to make a lot of data. Second, the data can't all be different, all right? If you want to train the model, the data needs to be comparable. The PDB is a beautiful thing because it's 200,000 structures that are all basically against the same standard. That's why that works so well. So as you're thinking about building up big data assets at your biopharma company or biotech company, the protocol you choose to generate all that data needs to be pretty similar and standardized against the same controls if you want to use it for a large training set for a model. So not infinite variability, much more specified, okay? And then finally, the amount of data you generate means it doesn't land in an ELN, an Electronic Lab Notebook, that's like it can handle small amounts of data, it means it's got to go through sort of modern data informatic pipelines, okay? Because you're just moving literally terabytes of data. Okay? And so that is a different game on the right when it comes to data, when it comes to the fact that it's not being done by hand, it's being done with automation to generate those large data sets and it's being done in a standard way. Does that make sense? Okay. All right. So let me tell you about what that means for us. So first 2 products we've launched in data points. This is where we say, "All right, maybe you'll someday build your own lab. We'll talk to that in a minute. But right now, I've already got a lab that can run the right-hand side here, okay, that can generate large data sets in a standard way for you, okay? So what might you use it for?" So the first offering we have is what we call functional genomics. So this is used in the area of target discovery. All right? So you're a pharma company, you want to know more about a certain cell type that's relevant to your disease. We've worked in more than 100 human cell types at Ginkgo now, lung, cervical, kidney, skin, skeletal, aorta, so on and so forth, all these different cell types. And you say, all right, I've got that cell type but I now want to perturb it so that I can better understand the biology. I'm going to give you 2 ways you can perturb it. You can perturb it with a chemical compound library. So you could send us 30,000 chemical compounds and we'll grow those cells of interest to you on plates, we'll hit them with the compounds, all right? Or you could tell us, "I want to make these 5,000 CRISPR knockouts." And we'll go design the guides. We'll get the CRISPR in, we'll do the knockout and we'll make those 5,000 cell lines with the CRISPR perturbations. Now that you've got those plates with the perturbed cells, we're going to run assay outputs on that, either transcriptomics, so we can do a drug seek, so a high content transcriptomic output or we can do imaging. We can do cell painting or brightfield imaging, high content imaging on that. And what's really cool is, if you're an ML person, people are really liking this. We've released now our 2 datasets and the [ GDPX21 ] is actually really excellent. It's about 1 terabyte of data. It's 4 different cell types, human melanocyte cells, aortic smooth muscle cells, dermal fibroblasts and skeletal muscle myoblasts. We hit them with 85 different compounds, 6 concentrations, so you get sort of a dose look there and 4 replicates. And then we do 2 million reads of sequencing depth on transcriptomics and you -- give you that data dump. Okay? And it's all ML ready, your AI scientists can play around with it and you can get a sense for what it would be like to generate a similar dataset for your exact cell type of interest. All right? And we're happy to onboard yourselves that doesn't -- we've had nearly basically 100% success rate, bringing on customer cell types, so we're happy to do that, okay? So QR code is right here. Please download away. People really like this. The other thing, in addition to target discovery that we're seeing people build models for is to develop their drug assets. Okay? So asset development, asset optimization, drug optimization. And the big area, at least the first one that we're pushing is in antibody developability, all right? And so the idea here is you could send Ginkgo, say 1,000 sequences, 5,000 sequences of antibodies that have been designed by, say, your model or that you think are interesting to explore, we synthesize the DNA, we express the antibody. And then we run a battery of 10 different sort of industrially relevant, developability assays on this tighter target, binding, aggregation, heparin binding, thermostability, hydrophobicity, colloidal stability. These are very difficult assays to run at high throughput. Okay? And so this is where the Ginkgo sort of secret sauce of being able to integrate complex biochemistry onto liquid handling automation and so on and bring these costs down allow us to do this at huge scale. So this is something you can get at small scale from a place like WuXi at sort of great expense but to be able to get large amounts of data in this area, I think we're the best product on the market by far. So this is something that, again, if you want to now have an antibody model that's not just trained to bind but rather has the data that also helps you have the antibodies be designed to be developable, in other words, to be ultimately good drug candidates on the other end. This is the kind of data we think would feed that sort of model very well. And I think -- I didn't mention it but we have a -- I think -- yes, top 25 biopharma company on this one, a couple of tech bio companies and top 10 biopharmas already signed up for the transcriptomic one as well. Okay. And I want to just give 2 examples of projects. You have a sense of like, at least for the transcriptomics sort of time lines. So we have like example one, data on an oncology target ID, 2000 chemical perturbations, 5 different cancer cell lines, you get a drug seek readout and then you can get that in about 10 to 11 weeks. If you're -- example 2, data for an Alzheimer's target, this case, we're doing CRISPR knockouts, times 1,000 genes, 3 cell types and then giving you back that drug seek and also high content imaging data back, that's 9 to 10 weeks. Okay? So again, this is something we're running all the time. The antibody stuff, I think, is even faster. And so we're very happy to sign customers up for this. And just from a sales cycle business standpoint, we can get -- we've gotten proof-of-concept deals done here in like 30 days or less, which back for our like more traditional solutions business that we've been running before and are still selling, I will highlight, is just unheard of. So it's really exciting for us to be able to bring online these deals so fast. The other thing we've been adding recently because we're seeing a lot of requests for folks building models on the RNA side is stability, translation, protein expression, immune response for your RNA constructs. And so we can do 20,000 designs in an assay here over a 3-month period as an example on the stability. And so if you're building RNA models, please do call us, I think. Don't bother building out the lab, we're very happy to do this. Like I said, it's fee-for-service. And if you're a tech person, this is a business model very similar to a company like Scale AI in the tech AI industry. So this is like who OpenAI paid to create data for their models, fee-for-service basis, just generate a lot of data. Okay. So that's so much as I want to say about data points. I'm really excited about that. I think it's a great direction for Ginkgo. It's a proven business model in the tool space but we're offering a differentiated product there. We're not just trying to offer a CRO service that someone else offers. Okay. Ginkgo automation. So I'm very excited about this one as well. So we acquired a company called Zymergen, now I think, about 3 years ago. And the reason -- well, not the only reason but probably the biggest reason we acquired them was for this automation technology that they had been developing, now over an 8-year period. And so I think there's probably, I don't know, call it, close to $0.5 billion invested into this equipment and software stack here. And the idea is to create modular integrated robotics in the lab automation. And this is the kind of units that you need to understand if you want to understand what we're doing with our system. So this is a reconfigurable automation cart. We call it a RAC. And these are going to launch -- there's like the Society for Laboratory Automation meeting in San Diego at the end of this month, it will be like kind of officially launched. But this is the newest generation of these that we just did in the last few months -- in this last year. And so that cart has a single piece of equipment, that orange centrifuge, for example, in this picture, a robotic arm and a piece of MagneMotion track. And what MagneMotion track is, is basically like a maglev track that allows you to move plates along a rail. And what's cool about this is, I can take that 1 cart and 1 piece of equipment and connect it to many other carts that also have pieces of equipment in them just through standard connection interfaces, I think like LEGOs a little bit, all right? And this has a few advantages, right? For starters, it means if you change your mind about the piece of equipment you want in your integrated automation setup, you just pull out a cart and pop in a new cart. Okay? That sort of unheard of, again, in the world of if you've ever been familiar with like work cell automation systems. It's like an arm in the middle of like 10 pieces of equipment and it's reaching everywhere and it bangs in the stuff and you have to like really be careful. It's very finicky. Your ability to change what's in there, to like take out a piece of equipment and swap a new one out is a several month automation engineering and software project. Whereas here, you can make that change in a few hours. In fact, we sort of proved that last night. So Recursion had a pretty fun event last night at the San Francisco Mint and so you can see people having cocktails next to our 5 RAC automation setup that we put together during that day. So over a period of a few hours during the day and that was operating. Okay? Plates were moving around, the arms are picking them up. They're putting them on the equipment, being able to stand up an integrated automation system in a few hours. Again, if you're not in the industry, that's a real breakthrough. And so this system, what I really want to achieve with it is I want to, instead of having an automation work cell, which is an integrated automation system designed to do 1 thing, say, a diagnostic assay or a high throughput screen, I want you to invest in an automation core, all right? So this is like a big room full of lots of these RACs and it can be -- it's flexible, okay? Your scientists are ultimately submitting many different automation protocols all to the same infrastructure. And that's in part because of the hardware. We can stick together new equipment as you need it in your core in the future, it's future proof. But it's also because of the software. So we have cloud-based software where we have an amazing scheduler where that system you saw in Boston, where I had those 25-ish RACs all put together, we've had up to 12 workflows submitted by our scientists at the same time, run on that system. And so they can be interleaved. So as you submit your, hey, I want to run this thing and you add your plates to the carousel, the software checks could it fit in with the other things that are running. And if it can, then you can add your workflow to the system and run it at the same time, all right? So this helps you amortize the cost of a large integrated automation system across many workflows instead of just one as is done with traditional lab automation. And I was at -- also at that Recursion event last night, Tony Kim from BlackRock, he's a tech investor at BlackRock, was in the panel and he's like I really would love to see horizontal platforms in biotech. You guys just like to make therapeutics, right? But really in the tech industry, all the money has been in the horizontal platforms. And I was like, okay, well, I can talk about business models for horizontal platforms and biotech for about 5 hours. But the point of this system, what we're trying to achieve with it, is to establish a horizontal platform for the physical infrastructure associated with laboratory automation in biotech. In other words, this system should work for everything, okay, because the RAC is a standard object. They're all the same. What's inside it can be all kinds of different equipment. And then all that equipment, we connect to our software. So you have 1 software that lets you basically push samples among any piece of equipment in your lab in a high throughput manner by betting on our infrastructure as a horizontal standard. And that's what we're trying to make happen. I think we're the first ones to really do that in a serious way in laboratory automation. So I'm really, really, really excited about this and we're excited to launch it later this month at SLAS. Okay. I'm going to end on this. I will say biotech is a little beat up at the moment. There's a lot of sadness about biotechnology. Meanwhile, out in the wider world, I would say everyone is getting pretty excited about landing rockets and self-driving cars and defense tech. I mean I think even crypto has attention again. It's an absurdity. And so like I do want to remind us in the biotech space that biotechnology is very captivating to the general public, okay? It is as captivating a technology as space or autonomous driving or cryptocurrency. And like -- just to highlight a few things like, for example, there is a company called Colossal Biosciences that yesterday, raised $200 million at a $10.2 billion valuation. They are trying to -- I think basically the business model is to make Jurassic Park. They're trying to bring back woolly mammoths. And so I think on this crowd, would be a lot of scoffing and that sort of thing. But like it's a highlight of consumer attention to biotech, right? Like that is a captivating story. The whole world would be excited to see a woolly mammoth come back. That's the thing you can do with biotech that you don't get that same captivation from, say, robotics. At the Golden Globes, Nikki Glaser gave the intro. And she said, good evening, welcome to the 82nd Golden Globes, Ozempic's biggest night, okay? So these GLP-1 drugs, I think the right way to think about them is they're really a consumer product, right? These are long-term, like longevity, health span products. They are not just treating a small set of patients. They're ultimately going to go to hundreds of millions, maybe billions of people, right? People want that. Customers want that. They're excited to see biotechnology come to them as a consumer product. I think we should lean into that. I will also direct you to this Netflix documentary that came out a few days ago. This fellow, Bryan Johnson. You probably read about them. He's trying to live forever, spending like $2 million, measuring all his biomarkers. But there's like an amazing part of -- and by the way, I'm a friend of Bryan, I like Bryan. But there's an amazing part of this documentary where he goes to Roatan, which is an island off the coast of Honduras to a company called Minicircle to get a gene therapy for a follistatin expression, for like muscle health effectively. Not approved in the U.S. That's why you're going to a special economic zone off the coast of Honduras, right? But that's -- and it's not just him, there's a lot of people doing this now. So this is -- there's an interesting thing where I think with the new administration and just how people are approaching this stuff, we could think about how do you do this sort of thing safely. I think consumers are going to want to adopt biotech as a consumer product. What does it look like to get those types of things approved? How do we make sure this is happening in the United States in Cambridge, Massachusetts or California, not Roatan, in an unsafe way. This is really an opportunity. And so I think we're coming up on a year where the general public is going to be hungry and excited about new technology. I think biology is the most captivating of new technologies. So I think it's going to be a good 2025 for all of us. So that's it for me. We should grow the world we want to see. And thank you very much for your time and I appreciate questions.

Thank you, Jason. I would like to open up the floor to any other questions that you may have. Also would like to remind the audience that there are 2 options for you to ask a question, 1 by raising your hand, and we'll hand you a microphone. Or second is submitting a question through the portal online. Questions from the audience? Okay. There is a question that came in online for you, Jason. What are the typical types of customers for your new data point service? And are you seeing interest from large biopharma or tech bio start-ups or tech companies?

Yes. So it's actually been interesting to me. I sort of assumed it would be like out of the gate, all the AI modeling companies that you've heard about that are sort of like AI experts building better versions of some of the public models and things like that. But I think a lot of those companies view that they're so good at the modeling that they can make a better version of the model trained on the same public data that everybody else has. So it's actually been less from them. Large biopharma, a lot. Yes. So most major biopharma and midsized biopharma companies are thinking about AI. They're trying to understand it for their area of interest and there's a sense that they're going to need the right kind of datasets because they've gone and looked at their internal data and realized it wasn't done in a standard way. It wasn't really comparable. You can't just throw it all together. Their AI teams have finally come back to them and sort of like read them the riot act on that. And so there's a sense of needing to generate new datasets there. Interestingly, we are also seeing it from the tech companies. So there's like a weird -- I just -- yes, it will be really interesting to see. I mean, I think there's like a natural intuition among the tech companies about the sort of 2 things driving it. One is, I think the sense that like DNA is a coded language and that's sort of the game they play in. And then second, I think for a lot of the really AI folks like the Dennises of the world at Google and so on. Part of the reason they got into AI was to drive science. And like biotech, we're really the big spend on science, if you like, look across out of all of science. And so if you were going to try to do it, that's where you'd do it. So I think for those 2 reasons, you're going to see it from the tech companies, too.

Perfect.

Which is the biopharma. We shouldn't -- for the companies in the industry that live here and just do biotech all day, I think they should lean in to make sure that it's not something that just gets grabbed up. Yes.

Questions from the audience? Okay. Perhaps one last question that I have for you. And what are you worried about as you launch the automation business? This is Ginkgo's first foray into selling equipment. How do you know it's going to go well?