Good morning. Welcome to the 44th annual J.P. Morgan Healthcare Conference. My name is Yue Huang. I'm an associate in the J.P. Morgan Healthcare Investment Banking Team. It is my great pleasure to introduce our next presenting company, Ginkgo Bioworks. With that, please welcome co-founder and CEO, Jason Kelly.

All right. Excited to be here today. I will say I think I've been having the most fun I've had at a J.P. Morgan in the last 10 years over the last few days. We had our robotics set up the Marriott Marquis lobby and had people coming by and learning about autonomous labs. And I get to stand there and kind of talk to people about that. And I think it's what I'm going to focus most of the talk today on is explaining that to you all because we're seeing a ton of interest. And I think it's a great fit for Ginkgo's mission of making biology easier to engineer. Our view is the technological basis for that in the future is going to rest on this idea of an autonomous lab. We also were really lucky on the commercial side to win a contract recently under the new Genesis mission. This is an executive order from President Trump to bring AI into science. And so we won a $47 million contract with the Department of Energy to build out a large 97 robot autonomous lab up at Pacific Northwest National Lab in Washington. And this is actually just back in December. We opened the first 18 robots, so a smaller version of that big lab, and ribbon cut that with the Secretary of Energy, Secretary Wright there. We did a press conference at the end of December. So starting to see commercial pickup on this idea. But today, I was going to kind of do a deep dive a little bit, because I think this is really the direction Ginkgo is going to build on for the next five or 10 years. So first, I'm going to talk about why autonomous labs are going to be transformative for biotechnology and pharmaceuticals and even science broadly. Second, what is the technological underpinnings? What makes an autonomous lab different than lab automation, which is something that's been around for a long time in the tools industry? And then finally, how are we going to market with this at Ginkgo? There's really two ways. We're going to be directly selling it like we did to the Department of Energy. We'll build a lab for you. But we also have a large autonomous lab of our own in Boston that we're selling services on top of. So I'll talk about that as well. Let's dive in. So I'm a little bit of an IBM history nerd. I really like the 1950s, 60s, 70s era of IBM because I think they really were the ones who brought sort of automation into computation. They laid the groundwork for what became the entire tech industry we see today back in that period. And so this is an advertisement, 1951. It says this is an IBM electronic calculator, and it will do the work of 150 engineers. All right. And I really like that. I like this because if you don't recognize it, what all those engineers are holding is a slide rule, which is a manual computation device. OK, and I think one of the things I hear when I talk to scientists about automating laboratories, there's some concern about, oh, it's going to take scientists jobs and so on. And so one of the things I like to point out is this device replaced 150 manual computing engineers. We have many more engineers and information technology today than we had in 1950, in fact, because of the automation of computation. And that's because by automating it, we increased the return on investment dramatically for what was in those engineers' heads because they could deploy with much greater leverage after computation was automated. All right, does that make sense? And so I want to talk a little bit about a little of that history in computers and some other industries and then what it means for labs. So it's not just enough to automate. So I'll point out – this is how I kind of frame this in my mind. On the y-axis there, you have the amount of automation. So a low amount of automation is that slide rule or that paper notebook. Okay, it's totally manual. A high amount of automation up at the top there is that electronic calculator from IBM. All right, and then on the x-axis, we have another thing that's important, which is the flexibility of what the person can do with the system. So with a notepad and that slide rule, you can do all kinds of different types of computation. But the electronic calculator from IBM actually just did at the time division, subtraction, addition. It was a much more limited set of things. So it was automated but not flexible. All right. And so the true breakthrough actually came 10, 15 years later from IBM with mainframe computers. And what ultimately broke it into all of our lives is the personal computer I show there. The computer gave you both automation and flexibility. And that was because it ran on code. You could put different code in and make it do different things. Does that make sense? And so it's really that top right corner, the ability to get both automation and flexibility that built the entire tech industry. So I think this is happening again in another domain right now, which is transportation. So if we rewind the clock, you know, 100 years or whatever, we have this beautiful San Francisco trolley cars. We have subways in New York and Boston and so on. This is a high amount of automation. You get in that subway. You don't have to do anything. It takes you around. low amount of flexibility. You better be wanting to go to one of the stops on that subway train because it doesn't take you anywhere else. Bottom right, a car. We have a huge amount of flexibility, but it's totally manual. You're driving it, but you can drive it wherever you want. So for the last 80 years, 100 years, since we invented these things, we've had nothing in the top right corner. We had nothing automated that would take you anywhere you want. If you've been walking the streets of JPM over the last week, I'm sure you saw a lot of these. We just broke through on this. So we have what we call autonomous cars, and these are automated and flexible. You get in the back seat, you don't have to do anything, and you tell it where you want to go, and it takes you anywhere you want. And so my view is that thing is like the computer. We're going to rebuild the entire transportation industry on the back of autonomy because it is giving you both automation and flexibility at the same time, just like the computer did back in 1960. Does that make sense? Okay. So where are we in science and laboratory work in the underlying engine that creates all the drugs and the pharmaceutical industry and so on? So it turns out we actually do, we have automation in labs. We've had it for 30 or 40 years. It's what you see up there in the top left. We call these automation work cells. You can buy them from Thermo Fisher, Hi-Res Bio, Biocero, there's a number of specialist companies in the space. And what they are used for is things like high throughput screening. I have 100,000 chemical compounds, and I want to run them against a certain cell assay and find a needle in a haystack. We use them in diagnostic labs. I'm getting in thousands of blood samples, and I want to run the same panel of assays on each sample over and over and over again. So no flexibility, high amount of automation. Down in the lower right, a lot of flexibility in the manual lab bench, which amazingly, if you walk into the labs of Merck and Pfizer and so on, you will see lab benches that look quite a lot like lab benches looked 100 years ago in scientific settings. And that's because it gives scientists flexibility. You can pick up a pipette, you can do an experiment you just read about in a paper last week. All right. And I'm going to talk a little bit in a minute about what that looks like and what we can do about it. But what I want to point out is what we are aiming to build here at Ginkgo is the top right corner. We want to give you an autonomous lab, like an autonomous car. We want to give you a lab which is totally automated, but doesn't just do one thing, like the subway, but rather like a Waymo can take you wherever you want as a scientist. You can put in your order for an experiment, and it will do it that day, even if you've never asked for that experiment before. All right, so that's what we're trying to achieve. Does that make sense? And again, our hope is if we can establish this, then like the computer or like the Waymo, we're able to rebuild this industry around automation with flexibility, which is very, very empowering. And so that's why I want to do it. The key technical question is how do you do it? How do you get up to that top right corner of high autonomy and high flexibility? So what does it look like?

All right.

So an autonomous lab has to allow you to do the same work that you would do in a traditional lab setting, but without the manual work. All right. So let me just lay out that a little bit on like what's a traditional lab like. Also, I've entered my influencer era on LinkedIn, just so you know, in case you've seen me. So a traditional laboratory looks like this. Right. And what you have is let's say you're opening up a lab to do mammalian cell engineering, CAR T work or something. You're going to have 15 or 20 pieces of lab equipment. You're going to open a door into this lab. There's 15 to 20 pieces of lab equipment in there, all sitting on benches. There's a number of benches, not just one. There's maybe five or eight that scientists can stand at that have pipettes above them and access to a bunch of different reagents. And what the scientists are going to do when they walk into that lab every day is they're going to do, I think, three major activities. One, they're going to get reagents, and they're going to use pipettes, and they're going to manually manipulate liquids and other reagents to make reaction samples. Number two, they're going to move those samples among those 10 or 15 devices in the lab. So they're going to take the sample over to the PCR machine. They're going to take it over the HPLC. They're going to take it to a centrifuge. When it's done with one machine, they might take it to a different machine for the next step. All right. And then third, every time they put it in a machine, they're going to input the settings on that machine? What RPM do I want the centrifuge to spin at? What's my process in the thermocycler? They're going to use their scientific knowledge to set the settings on the machine. All right. That's basically what they're doing. Okay. They're doing that in many different combinations, but it's those activities that are getting done. Okay. So what if we want to replace that lab with an autonomous lab? What do we have to have? Okay. First, we need reliable liquid handling. So we need to replace those pipettes with robots that can do the liquid handling and pipetting. It turns out we're actually pretty lucky in this regard. We've had liquid handling robotics being built by companies like Tecan and Hamilton and new entrants like Opentrons over the last 30 years. They're actually quite good. In fact, more reliable, you know, don't tell people, than the scientists at the bench at doing liquid handling. All right. They're not particularly easy to use. They're pretty difficult to program. So we'll talk about that in a sec. But they do the physical work well second i need to be able to transport material to from one device to another device across all 15 of those devices i had three i need parameterized control i need to be able to set the settings on the devices four i cannot just have three of the 15 devices i need all 15. i need the whole lab in one setup because i don't know ahead of time which sample needs you know a sample might need to go from equipment a to equipment g and i didn't know that ahead of time. They all have to be on there, just like they're all in the lab. Number five, this one's tricky. You need to support parallel work. So when you go into a traditional lab, the first scientist doesn't go in in the morning and then lock the door behind them and say, no other scientists can use the lab but me today. When I'm done, I'll open the door and the next guy can come in. No. A number of scientists all use the lab at the same time. They talk to each other about the availability of equipment. I'm going to use this machine for the next two hours. Okay, great. I'll use it after you they have multiple liquid handling stations with pipettes so they can do that in parallel so you need to be able to support a bunch of people all using the same 15 pieces of equipment and then finally it needs to be as easy to use as that lab so you do not it could not be the case that you need to be a software developer in order to use the autonomous lab if you didn't need to be a software developer and you don't need to be to use a biological traditional lab Does that make sense? So if we want scientists to use this, they need to be able to interface with more like a human language interface and so forth, like we're seeing with these AI models and no code coding agents, rather than having to write code themselves. Does that make sense? Okay, so that's the list. If you knock that off, then I think Autonomous Lab would basically be able to do the same thing as every lab you see at Merck, at Pfizer, in diagnostics companies, in every academic research facility, every lab in the world, okay? It would be able to do that same work, except it would do it 24-7, and it would do it with higher reliability, and it would do it with improved efficiency, okay? And so that is very hard. Like, this is actually a very hard technical challenge, both at the software and the hardware level. So I'll talk about what we've done and our approach to trying to solve that problem with our hardware and software. And it's a journey. I'll tell you where we are in it today. So this is the key component you need to understand. This is what Ginkgo's invented. We call it a reconfigurable automation cart, a rack. It's basically a standard envelope that wraps around one of those 15 pieces of lab equipment in that lab I mentioned. So any piece of third-party equipment that you can buy from any of these million different life science tools vendors, I can wrap in one of these rack envelopes. And once I've done that, I can do a few things.

There's a six-axis robotic arm here, and there's a piece of magnetic magnimotion trap, which is a transport system.

I'll show you a video in a second that can deliver a sample to that arm, and then the arm can put a sample onto, in this case, that's a centrifuge, that red box. And then underneath, there's electronic box hardware that lets us connect to the centrifuge. We can power cycle it. We can do anything to it. We can set its settings. We can do all the things a scientist could do to it at the lab bench, but all with software. And importantly, these are structured. They're standard. There's different sizes for different pieces of equipment, but the height is all the same. so you can Lego block them together and put together three or 15, like in the example lab I gave you, or 97, like we're doing for the Department of Energy, into one big setup. All right. So what does it look like? All right. So I think it helps a little bit to see it. I'll talk in a minute about the software, but what you're going to see here is a workflow, in other words, like a protocol being submitted into our system. And this is our system in Boston where we have 40 rack cards. So you're seeing a sample. And one of the constraints we have on the system is the samples are passed in SBS format, which is 96 or 384 or 1586 well format plates. That's a standard across the industry. Many types of equipment accept that format of plate. And you see it just got delivered to a centrifuge. And then here comes one getting delivered to an echo liquid handler. This is an acoustic liquid handler from LabSite. So this is one of the devices that does the liquid handling. Then it's been passed over now to another device that does liquid handling. This is an Agilent Bravo. And you'll see in this case, it does liquid stamping. So it's going to pick up a whole bunch of tips and it's going to stamp liquid. So one point to make is that first device is from LabSyx owned by Danaher now. The second device is from Agilent. And the beautiful thing from the standpoint of a scientist using our system is they don't need to use the third-party software from either Agilent or Danaher. They're able to use our software exclusively to make their protocol, and our software does the translation down to the third-party software on all these different devices. Okay, and that's really critical because if you're programming all these different things, you're going to lose your mind. And so this is now the sample has gone off to be put on a shaker, and then it's going to end up going on to a thermocycler in order to do the final reaction and do the assay. And a few other things are great about this I'll highlight. We have cameras in all the carts, so you're getting a record of everything that happened. If there's any kind of error, you can see it. You get data coming off these things about everything that happens. So when you're doing liquid handling at the lab bench as a scientist, there is no record. There's no record if you made a mistake, if you pipetted to a wrong well or anything like that. Every step, every action taken on an autonomous lab is recorded electronically. So we have an entire record of everything that happened on the system as well. So there's a number of things that are just sort of extra benefits in addition to the reduction in manual work to zero. Okay. So then the next thing I wanted to show is that parallelization. So that was just one run happening in that video. This is a screen capture from our system in Boston yesterday. Each row on this is a different piece of equipment on our autonomous lab in Boston. All right. And what you're seeing across is time. And each color is a different protocol operating on the system. And the reason that you need to worry about this, it's like I said back in the lab earlier. Scientist one comes into a traditional lab and they say, I'm going to use the PCR machine. They start a PCR reaction. Scientist two comes in and says, hey, I need to use the PCR machine. When are you done with it? Oh, I'm done with it in three hours. Well, I'll start an incubation that goes for four hours and then the PCR machine will be ready. and so I'll put my sample on. That's what you're seeing here, done computationally. When a new scientist shows up, they submit their workflow, their protocol into the autonomous lab software here at Ginkgo. And it says, can I fit your protocol into the system today? Will the PCR machine be available when you need it? Will the centrifuge be available when you need it? Can I rejigger things across all the protocols so that you can drop right in? And this is critical for a couple of reasons. Most important, it means I can have 17 different protocols running, submitted by lots of scientists. And I just want to point out, there is no automation system in laboratory automation in the world today, other than ours that does this, okay, at this kind of scale. And so first off, it just means you can do parallelization. Number two, let me tell you what more often happens than not with that busy piece of equipment in the lab. Oh, you're already on it today? That's okay. I'll wait till tomorrow. So the utilization rate for equipment in traditional lab settings is like sub 30%, right? These things are just sitting around never being used. And so, and again, that's because the men, you know, the manual mental scheduling is not very good. Okay. Among a team of scientists working independently, but we can just do it algorithmically So we can make our scheduling almost perfect and try to get the utilization rate up dramatically on all this capital equipment, which is quite expensive in these labs. All right. You can start small. So you might have a lab where there's only three or four pieces of equipment. That's what I had set up at the Marriott Marquis all week. We had four pieces of equipment with the with the system running. Right. Our lab set up in Boston is 40. And that one you see on the right is a schematic of the system that we'll be building. the new system we'll be building for Pacific Northwest National Lab. So you can grow these to huge size. I'm not going to talk a ton about it today, but you'll hear more from us coming up on this. Obviously, a scientist can submit a protocol to an autonomous lab. But of course, an AI scientist could also submit a protocol to an autonomous lab. And so you're seeing an enormous amount of energy from the frontier labs. Like Google has an actual, has an AI scientist, OpenAI, Anthropic, or it's been announcing more in these areas. And then you have specialist companies like Edison that are working on building an AI scientist. The one thing all these companies with all their money in data centers and these great AI models do not have is hands in the lab. And so if we want to have a lot of that, that, you know, wind in our sails from AI coming into the biotechnological space and pharmaceuticals, I think it's critical that we give these reasoning models hands in the lab. And this is how you do it. A hundred percent. Okay. It needs to be end to end. There needs to be not people in the middle. It needs to be an autonomous lab. And so again, more to come on that from Ginkgo, but this is definitely one of the value props. Okay. So what are those value props for a customer buying one of our autonomous labs? The first, massive overhead cost savings on your labs. Again, my point here is not to replace your work cell that does your high throughput screening. I think I can sell some systems that way. It's a good way to get my foot in the door, it's fine. There's a $400 to $500 million a year work cell market. That's nice for me. It's a way to get started. What I want to do is have you close your labs. I want every lab in the United States of America to close and be replaced with autonomous labs where scientists are able to do their work via computers. And I think in the US, especially in terms of having competitive labor costs on doing science compared to China, this is why you're seeing this push around the Genesis mission, AI for science in the US, I think we need this improvement in research productivity just to play ball at all. Second, this point on research productivity for AI models in particular, they need very large data sets, much larger data sets than we generate traditionally at the lab bench by hand. So yes, sometimes you could use a work cell for that, you know, if you know exactly what you want. But my view is people are going to want different things over time, depending on what models they're building, and they're not going to want to make a one-off work cell to generate every data set. You're going to be much better off sitting on top of a large autonomous lab that can do different things. And then finally, that point I just made, these AI scientists are going to need hands in the lab, and this is how it's going to happen. So we're hearing that from some of the pharma companies. You're hearing it described as lab in the loop. We're seeing some demand there. And then the first two, I think ultimately every lab in America can see those benefits. Okay. So how are we going to go to market with those is the last thing I want to talk about. The first thing I'll say is when I look at the life science tools industry, like the big companies, Thermo and Danaher, they're really predicated on a paradigm that the way we do science, the way we do laboratory work is by hand at the lab bench. They sell you liquid handling, you know, by hand tools. They sell you reagents in the form of kits that have little tubes with little edges so that you can open them with your hands. They sell benchtop equipment with things like touchscreens so a human can touch it. There's a whole edifice of tools that are predicated on the idea that the way we do this work is manual. If we are successful at building autonomous labs, I think the entire tool stack in life sciences and sciences broadly needs to be rebuilt. And so we would do that at Ginkgo. Okay, so how do we do that in the near term? Two ways. First, we will build autonomous labs at customer sites. So like we're doing the Department of Energy, we're talking diagnostics companies, pharmaceutical companies, we're competing for work cell deals. We're getting in and getting started selling these systems directly. And then the second, which I'm gonna spend a minute talking about, is we have built a very large frontier autonomous lab, largest one in the world today at our site in Boston that has 40 of our racks and by May or June should have 100 racks in that system. And that's the one I just showed you the data from a minute ago. We offer two services on top of that system in Boston. The first we call Solutions. And if you follow the Ginkgo story, this is something we've done for a long time where Ginkgo scientists use our technology, our autonomous lab to deliver research outcomes for customers. And then the last year and a half, we launched a new service called Data Points. This is much more akin, oh, sorry, in solutions, we get royalties and milestones and biobucks and that sort of thing. In Data Points, our customer scientists order lab experiments from our lab in Boston. Okay. And we offer that through today, a menu of services, and I'll show you them in a second. But I think in the future, My hope is if this gets good enough, they could really use it like a cloud lab, which is a model that's been tried a few times, not successfully, where people could just order whatever experiment they want. But today in data points, we have a menu that companies order from. So I'll talk about how that's going. I will make the point that us running, this is a picture of our lab in Boston, and you should come visit, running this large autonomous lab ourselves gives us a couple advantages. One, we can showcase the art of the possible for customers. So I can be the first one to show that you can put 10 scientists submitting 20 protocols a day on an autonomous system without it catching on fire. I can do it first so that the head of R&D at Merck knows it already works. Second, we develop and improve the technology for our robotics and software much faster than traditional robotics vendors and tools companies that don't do science on top of their hardware. So my people are breaking the system every day. So then that means my software teams can be updating and getting bug fixes constantly, not waiting around for that, you know, a customer to try something and having a much more gapped feedback loop. Does that make sense? Dog fooding, so to speak. Okay. So on the solution side, we've done over 250 partnerships. If you've seen me talk before, I've showed this. Pharma, industrial biotech, agriculture, I'll say in the last year and a half, we tightened up dramatically where we were selling solutions. So we're largely selling it only in therapeutics now and a small amount in agriculture. OK, that's a big part of the way we got costs out of the business. And we have not stopped doing solutions deals. You can see a number of new deals, both with existing customers and new customers in 2025 and a lot of work for government R&D projects as well. So pretty decent year in terms of adding new solutions deals. Data points is the CRO service I mentioned where we generate the data sets customers ask us to make. We have three service lines in that area, one omics. So we do a functional genomics assay. So you can tell us, here's my favorite cell line, make these 5,000 CRISPR edits, and then do drug seek and send me back the transcriptomic data in a nice Amazon bucket, all clean for my ML team to do AI work. So this is the kind of stuff you can't get from Wooshie today. It is a sort of AI-minded, large data, ML-driven CRO. We also do it for antibody developability, and we recently launched AdMe. And so I've been really excited to see there's a lot of good energy around ChaiBio's deal with Lilly. And you're starting to see a business model of maybe large pharma will pay a bunch of money to just to license interesting AI models. Well, what makes those AI models interesting at those companies is they're trained on proprietary data. Like there was a little era where it was like who has the smartest AI people, but it's very quickly moving to who had what data to train a model for whatever it might be. Antibody binding, developability, small molecule, who knows, right? Anything they want to work on. I think the model of companies that get the next big deal with Lilly will do it based on making large proprietary data sets. And we would be very happy to be the sort of scale AI. This is the company that did a lot of data gen for OpenAI at the beginning and other tech companies. We'd happily generate data for anybody making AI models so that they don't need to build their own labs and they can stay focused, be computational teams, hire a lot of good computer scientists and so on, and just outsource the data gen to our sort of cloud autonomous lab in Boston. And so that's really what we're aiming to do with this. And we release data sets publicly. We've released the largest antibody developability data set that's available publicly. Same with the DrugSeq and so on, ADME. And so we'll just keep doing that. And I think this group, the team at Ginkgo, I'll put in a plug, they've done an amazing job sort of becoming like a community builder in this emerging AI for biospace. So we ran an AntibioDevelopability competition, DataPoint Summit, like a big meeting. We have the Virtual Cell Pharma Ecology Initiative, and then these data drops are really popular. So I think we've done a nice job. Oh, and I should mention, first thing I should have said, yeah, this guy got launched a little over a year ago, And now we're working with 10 of the top 20 pharma customers and generating data for them, in addition to some of the, like, more startup AI bio folks. And you can see the ones that are – some of the ones that let us talk about it down below. Not everybody does. The other thing that Solutions and Data Points do is they provide a revenue base for us while we're building up the direct sales autonomous lab business and that equipment and software and ultimately even reagents that will sell to those customers that buy our rack carts. You can see this is just from my last earnings call. We reiterated our total revenue for the year, 167 to 187, and services are a substantial portion of that. I will also mention we have greatly reduced our cash burn over the last year and a half. This was like a lot of pain at Ginkgo. We brought our cash burn down 73% while still ending last quarter, ending Q3 with $462 million in cash and no bank debt. So I really like our position in terms of having our good, solid runway and burn under control as we move into this leadership role in autonomous labs. So this is that lab in Boston. Again, I encourage folks to come visit. We're down in the seaport if you want to come see it. But GIGO is the right company to lead in this transition to autonomy in scientific work at the lab. We are well capitalized. We have a reduced cash burns. We have a long runway. We have extensive practical experience. So we spent the last 10 years trying every way under the sun to automate and scale lab work so we know where the bodies are buried. And we incorporate a lot of that directly into both the design of the rack carts, but more importantly, into the software that helps when a scientist pushes a protocol onto the system. There's a lot of checks that include the tacit knowledge Ginkgo has built up over the years. And then finally, we're mission driven. We're not giving up on this. This is, you know, we believe that biotechnology has not had its IBM moment. We are still very much living in the manual era. I think if we can overcome that, it will be some of the most important work we can do. This is the ad I would like us to be thinking about, you know, 50 or 60 years from now that we were able to unlock the programming of DNA like we were able to unlock the programming of computer code through automation and removing manual work. And so if you're a scientist listening in today, please put down your pipettes and join us in the Autonomous Lab. All right, let's grow the world we want to see. that's my email up there if you need anything and i'm happy to take some questions if folks have them thank you thank you jason uh a question from the floor yeah yeah so the question is what's the longest uh piece of dna that we could synthesize in the automated workflow or yeah the um so the the way and we talked about this a little bit at the beginning but i think the easy way to think about it is the system is not designed to do like one activity, right? Whether that's DNA synthesis or protein expression or, or, you know, high throughput screening or anything like that. It's really designed to replace a lab. And so if you had a protocol that was great for DNA synthesis and it used say an echo, a bunch of thermocyclers and so on, we would be able to replicate that protocol on the system. And so we do happen to do a chunk of that at Ginkgo. We've built very, I don't know, probably, I think using our old Gen 9 technology, we've probably built things that are 50 or 100 KB. But that's not the point, right? The point is that whatever idea you had for a protocol, you should be able to do. And I want to move away from Ginkgo has some special proprietary protocol that's excellent, but rather that I provide the infrastructure for other people to develop great protocols. Does that make sense?

Yeah. Yeah. Yeah. Yeah.

So I think an important, yeah, there's an important point there, which is, um, you But in a data center, for example, in those computers, it's all information moving around. But with a lab, there is atoms moving around. And so one of the things that needs to happen is we need to be loading atoms into the system. And so today, that means I have a team who are basically preparing reagents in sort of reagent plates that are going into the system. And those are then able to be part of reactions that are ordered custom by scientists. I think what will happen over time, this is back to my point about needing to rebuild the entire stack of sort of tools in an autonomous lab world, is we should have, you know, a cart that has every reagent you might need. That's the sort of part of this. And as you order, it just gets drawn out of there and it's like a usage based pricing. Right. And places already do this. Like, you can get a freezer from NEB or Thermo. It's just your people pulling out the stuff and getting charged by NEB and Thermo.

I think you'd have a similar idea with these systems. yeah yeah so follow up on our conversation yesterday perhaps um software integration so for this to work uh you will need to hardwire into a lot of third-party equipment that was not designed to it doesn't have an api of sort and in fact some of the vendors i'll try to not call out names are quite aggressively trying to make it very difficult to integrate their systems by others because they want to charge you for the software yeah so how are you overcoming that and what what's the what are the pros because you need an api standard for that yeah this has long

been desired uh there's very few standards across benchtop equipment um and again i think this is because the paradigm is it's just a single device sitting alone interacted with and the human is the glue between everything and so nothing has to talk to each other and so there's never been an api standard. We're actually even lucky we have the SPS standard on the plates. Chemistry doesn't even have that. And so that I think is, you're 100% correct. It would be very obvious that we should have API standards for third-party equipment. I think things like that will come eventually. In the meantime, and there's open source efforts like PyLab Robot and things that are working on this. In the meantime, we basically, as customers ask for equipment, we do the work of bringing it on. Sometimes it's easier than other times. Sometimes there's an API, sometimes there's not sometimes we write our own drivers directly into the equipment like we do what we have to do if it's really impossible then ideally there's a different for many of these functions there's a second company that has the same piece of equipment basically um and we can swap but we've had pretty good luck um integrating things and that that's a one to two month process to bring on and then you only do it once and then that equipment is on so and we do it all the time I think we're, I don't know, nearly 100 now.

I have a question. So how do you see the autonomous labs being adapted by biopharma companies? Or in other words, what would be drivers for adoption?

Yeah, that's a key question. So right now, the market today for automation, like the top left corner, the work cells, is people buying high throughput screening systems or diagnostic labs and so on. And so that is bought every day of the week. Not every day of the week, but their RFP is coming out all the time. And again, let's call it a $400 to $500 million a year market served by three incumbent vendors. And I think we can play in that market. So one of the ways we get in is we just sort of, it's harder, you're not an incumbent, but we kind of muscle in and we say, hey, look, our systems are more general. They can turn into an autonomous lab someday, even if you only need this thing to do one thing. Let's get started. And then the second area is basically at all of these pharma companies, there's someone in charge of AI. And so that person is thinking about the labs differently than the historical drug discovery person is thinking about the labs. And they're, depending on the company, have some level of mandate to generate data in-house in a new way. And that, I think, are more direct autonomous lab sales. But it's, you know, it's going to take some time, I would say. And that's why I'm excited about our services is because in the meantime, I grow this giant lab in Boston. And that also is a great way to show people the art of the possible in the pharma industry. And the other one is, I'll say, is in academia. So academia is another place where we're sort of doing this thing, which I think is a bit pathological, which is we are underpaying grad students and postdocs to be low-cost manual labor in the United States. And then they graduate and there's not actually that many places that can afford to do manual labor at the cost it is to actually employ an adult full time in the United States of America. And so you have this like crazy drop off and basically only pharma can support it. And it's pretty messy. If instead, all of our scientists were running on autonomous labs, then I think you might end up with science and R&D much more broadly distributed across industries, because you wouldn't need to carry the cost of manual scientific labor, if that makes sense. Sorry, you're going to say something? yeah yeah there there's yeah the physical transition that happened in semiconductors and electronics i think is a good history right like yeah yes there was an era there was an era with vacuum tubes and a lot of manual work around electronics and we squeeze it all out.

I guess I will ask just one last question. There have been a lot of companies like Emerald Cloud Lab and Surtayos. They tried and struggled with the Cloud Lab business model. What do you think it takes to open up that go-to-market in biotech?

Yeah, so this is something I've thought a lot about. So we have this big lab in Boston. I want to make it available. You had folks try these Cloud Labs previously. And I think the issue, the basic, the main problem is like a business model problem. If you go into a therapeutics company and you say, Hey, I've got this lab, it can do whatever you want. Why don't you order for me? You won't have to build a lab. It's great. It's like what you do with your data centers at Amazon. Like, don't you want to not have all this capital infrastructure in house? They say, great, sure. I would like this protocol. And you're like, okay, cool. Let's do it. And their question, first question is, have you done it before? And your answer is like, well, I'm a cloud lab. Like, no, you know, you're asking me for your unique protocol. Of course, I haven't done exactly that protocol before. That's the whole point of this thing. And they're like, well, if you haven't done it before, I don't trust that you'll be able to do it. Okay. And so that was the, that's why that first generation didn't really work out. And so the way you get around that is you offer a menu of things that you have done before, that's traditional CRO, but you do it on top of flexible lab. And then you kind of, you boil the frog. You start to say, Hey, you know, you do get this from us. Do you want this slight variation? Did you know? And then eventually I think people come around but uh it'll it'll be a journey the other way you'll do it i'll make one last point is if they have an autonomous lab in-house they've gotten familiar ordering from it i can be overflow uh and so i'm hopeful that works too so all right i think we're at time but thank you so much jason for the presentation yeah thanks everybody yeah