Chai Discovery raises $400M at $3.8B valuation to bring AI-designed drug molecules to Eli Lilly, Novartis, and Pfizer

We'll talk to you soon.

Let me tell you about Railway. Railway is the all-in-one intelligent cloud provider. Use your favorite agent to deploy web app servers, databases, and more. While Railway automatically takes care of scaling, monitoring, and security. Our next guest is Jack Dent from Chai Discovery, and he's here with us now in the TBPN. Welcome to the show, Jack. How are you?

Welcome.

Thank you. Great to be here.

Thank you so much. Uh, could you quickly introduce yourself and the company and then I want to go through the news, but I'll let you start with an introduction.

Sure. So, I'm Jack. I'm one of our co-founders here at Chai. Uh, at Chai, we're building what we like to call the computer a design suite for molecules. So, think like Photoshop or Blender or Solid Works except instead of designing a car, you're designing molecules. These are, you know, small collections of atoms which are often used for things like medicines and that actually go inside our bodies and do things like like cure us. And so we uh operate at all levels of the stack. We're both a uh you know frontier research lab, also a product company, also a bit of a science company and tie all of those things together to put that uh that piece of software in the hands of some of the largest pharmaceutical companies in the world. So specifically narrow to pharma, but I imagine that you will have applications in material sciences all all over the place. But uh is is pharma just the the best landing zone or is that such a deep market that you'll stay there in perpetuity?

Yeah, I think pharma is right now where we're really focused. uh you know our models uh they work at the level of atoms and molecules and that um you also require different categories of models when you're trying to do different tasks. So it's not like a specialized mo model that will design an antibbody will necessarily like out of the box work for material science as well. That's still some additional work. Um but but farmers also, you know, this is a multi- trillion dollar industry. It's actually larger than the chip industries in in in revenue in and of itself. Um and so is it's this massive market um that obviously uh does huge amounts for human health and there's just so much runway to to build into that. Uh so what actually is different about the models that you're developing? Do you have the same sort of massive capex around pre-training runs and scale is all you need and you need to marshall all this compute. Does it have a different shape? Is there a different data puzzle that you're solving? uh how uh how important is your feedback loop with your customers? Uh how does your lab look different than uh sort of a a lab that's focused on coding for example?

Yeah, it's a great question. It's really all of the above. So when I say we're a research company, that means that we actually train our own models from scratch. These aren't like LLMs that we take off the shelf and fine-tune some open source architecture. We're starting at, you know, PyTorch and then below that down even at the kernel level sometimes and uh creating models which actually look quite unlike uh the sorts of models that you build elsewhere. It's sort of like how if you're training a video diffusion model that's a pretty different architecture to what you train for language. These these just different model categories require fundamentally different avenues of research. You know, one piece of intuition here is in the same way that video models sort of think in 2D and they put pixels on a grid and so uh you know, auto reggressive token prediction with a language model maybe isn't the best fit. You know, our models think in uh in 3D really they put individual atoms in 3D space to construct these molecules. And so that means we really need to have a command of the the entire stack. And then the data as well, this is not natural language data. uh it's very domain specific data. It's these vast troves of uh protein sequencing data sets and structure data sets where u people have used essentially really powerful microscopes to go and look atom by atom um at the structures that that that um that exist in nature and then we also do a lot of data generation in house as well. We we throw that off as well. So um and that's just at the the research layer. Then there's a whole you know product layer on top of that as well. It's almost like you know we're training Claude as we're also building you know cursor or cognition and so we need to do those things hand in hand. So there's this also this this feedback loop with the the scientists um and our um at both at the farmer companies but our users many of whom sit inside the company and dog food and try and push the frontier of these models. Um so it it requires yes to be just A+ across both science AI research and and product engineering as well. Please,

have you seen any scientists get uh quote unquote one shot in the way that certain vibe coders have over the last where they're just like staying up all night open with chai discovery running at all times?

Yeah. Because I mean it's such an interesting challenge for that that the labs have faced where you're trying to balance like making products that are enjoyable to use, effective. Um uh in a weird way, you know, you are optimizing for engagement. if you're optimizing for revenue uh just because someone's more engaged, they're they're using more tokens. But, uh I'm I'm curious if you've seen any signs of that because so far the labs that have experience the most extreme product market fit have gotten to the point where there's this idea of like oneshotting and then you have to like you know align the model uh better and and and uh improve that.

Yeah, totally. And actually, you know, maybe it's worth just providing some historical context here on just how quickly the research has moved in this field.

You know, a year a year and a half ago, basically none of this stuff worked. You know, it was or the success rates were extremely low, you know,.1%. And over the course of 2025, these models went from being these research curiosities in that they were maybe interesting things to study in some academic labs, but they weren't really being used in real drug discovery workflows. Um, in uh July, June, July of last year, we put out a paper called Chi 2 where um, you know, the title of that paper was zero shots antibbody design in a 24 well plate. literally getting at this exact point where you could start to design these molecules without needing to fine-tune them on a lot of data. You could really just prompt them in the way that you'd prompt an LLM with an input, you know, a target, maybe some some disease you're going after, a target that's implicated in that disease. And it would, you know, design a number of candidates uh that would uh that would help do that essentially. Um and so you that created some really uh you know interesting uh situations where you know we would go into you know a farmer company and uh we'd be presenting some data and there's one that that comes to mind in particular where somebody pointed to one of our slides and says oh my god you solved that one. Did you choose that on purpose? And we're like no we didn't we didn't choose that one on purpose. Why? Um and they said uh you know I spent five years of my life working on that that target trying to trying to find a bind it to it. you know the this technology has really moved beyond I think AI and drug discovery has been in the realm of promise for a really long time where it's been there's been a lot of uh hype and attention around it but in 2025 we really like crossed the you know uh crossed the Rubicon in a way uh and now these technologies are are you know be actively being deployed in companies like Eli Liy Novatist FISA um you know these are some of the uh most uh scientific highest tech companies in the world and they're putting them into real drug programs, you know, making them part of their core discovery engine and I think that's that's what might get missed here by you know so much has changed in the last year on the technology side and it's created this wave of um of new applications and what what's amazing is the scientists themselves are the most happy about it. They don't love the fact that they fail 90% of the time or that their work

or they could dedicate half of their career to one disease and and make you know modest progress or maybe their entire career. Do you think do you are do you think there will be something uh the music industry is funny right now because you know every artist is using AI or not every artist but many of them uh both casual artists non-professionals you know amateurs but also professional musicians are using it but no one wants to talk about it I expect that uh pharma will be much happier to talk about it but is that the right intuition? Yeah. Uh I think that's exactly right. I think 2026 is the year where these technologies are going uh from being fringe technologies which people are adopting on the side to actually technologies which you know entire discovery programs are are built around and so the uh you know pharma companies are not necessarily uh the loudest you know they care a lot about their IP and uh they uh you know there's obviously some um some alpha in figuring out some you know an advantage in getting a a drug to market faster than than someone else. Um so I think that will be be lagging and then I think there will also be some time uh for some of these uh you do denovo drugs to get into clinical trials and humans and there'll be validation there still. But what's important is you know that adoption curve is is um is well underway now. It's kicking off. You know it is the most sophisticated um companies and you know some of the stories we're hearing already are just remarkable. on the on the business model side, you're you're very full stack already. You're building the, you know, the harness, the model, and then the research that goes into the model. Uh, at some point, I imagine you'll have a customer that develops a new drug using Chai, and that drug, you know, goes on to generate, you know, tens of billions or hundreds of billions of dollars of revenue. To me, that's like what success

blockbuster

a Blockbuster drug created with Chai. Uh depending on how you charge for that, it's possible that you end up capturing, you know,

200 bucks a month.

Yeah. Yeah. Just some some like very, you know, minimal amount of value, which is good. Every company, you know, want to capture less value than than they create. Um, but at some point would you consider um or or have you considered a model where you're a true partner uh to these companies and and you're doing sort of joint joint ventures where you're basically giving them access to the product in exchange for long-term upside.

Yeah. You know, that traditionally has been the model. If you look at how biotech deals tend to work, a lot of them includes things like royalties and upside sharing in the drugs. You know, this is really just all partner-driven for us. You know, what does you know, we live to serve our partners really and what do they they want to do and what they want to do right now is a lot of them want to take our technology and deploy it you know fairly broadly across the portfolio and um that uh that means that uh you know we get uh rewarded financially for that. We also can reinvest some of the money they're paying us into building even better models for them. So that that way of doing business is working well for us right now. There may come a point where you know some pharma companies turn around and say hey we'd love to do a joint venture or you know go 5050 on this thing with you and you know I think uh certainly open to that but really it's it's partner driven for us. Um and uh we we we are we're trying to you know what we see ourselves as as this you know core um platform layer which enables their you know their discovery engines and you know that if you look at the top farmer companies they might spend 5 10 15 billion a year in R&D and so

even that that's not even you know on a uh what they're spending you know on uh on royalties or anything that's just that their R&D cost. So um it's a massive market to be building into. In fact, you know, it's even larger uh than some of the the the money that's spent on, you know, semis R&D every year. Farmer is one of the most enduring and durable categories of applied R&D spend in the world.

That's a good point. Uh well, you have some new partners on the financial side. Tell us about the round. I want to hit the gong.

How much did you raise?

Yeah.

So we raised 400.

Who came in? Everybody everybody came in.

So we got uh yeah we we raised $400 million at a $3.8 billion valuation. We're um super lucky to be partnered uh with Index, Kleiner, Parkkins, Sequoia, and Dimension um who are the sort of the largest checks in in in the round. And you know, it's an it's an all allstar cast. Uh some people we've known for

very long time. uh Nina from Index, Ilia from Kleiner, Pat Grady from Sequoia, and then Zav mainly, but really the whole dimension team, you know, they're all um yeah, we're uh it's rare to get this caliber of investors on the cap table and especially rare in one round. So, we feel um we feel incredibly fortunate.

Well, congratulations and thank you so much for coming on the show with us. This was fantastic exchange. Want to change the world?