Chai Discovery's $400M Round Buys 7 Résumés. The Real Pool Is Alumni.

On July 14, 2026, Chai Discovery closed a $400M Series C at a $3.8B valuation, led by Index with Kleiner, Sequoia, Dimension, and OpenAI returning as a co-investor. The company has roughly 30 employees. If you are trying to hire against that round, the math you should be staring at is not the valuation. It is the addressable talent pool, and it is almost comically small.

The pool Chai is buying is 7 people deep

Chai Discovery's $400M Series C targets a global talent pool of about 7 people who literally match its ideal profile, and a broader buildable pool of roughly 200. That is not a marketing number. That is what shows up when you filter for the actual work.

In Refolk's index of professional profiles, only 7 profiles worldwide currently hold a Research Scientist, ML Scientist, or AI Research Scientist title and list "Protein Structure Prediction" as a skill. Broaden to "Computational Biology + Deep Learning" with no title constraint and the count balloons to about 11,330, but the top employers in that broader slice are Intel, Haut.AI, and Xbiome. Those are adjacent shops, not frontier-model biology labs. The true builder pool for what Chai is doing sits in the low hundreds.

7

The company's own job spec confirms the profile it is chasing: "a highly experienced AI Research Scientist to play a crucial role in the development of our next-generation AI x biology foundation models," with a bar of publishing at top conferences or serving as staff-plus at a frontier lab. That description carves the world down to a few dozen people, most of whom already know each other by first name.

Chai has $2M of dry powder per addressable candidate

Divide Chai's $400M raise by the ~200 credibly buildable AI-x-biology researchers on the planet and you get roughly $2M of dry powder per person Chai could realistically hire. That is a stunning number and it is also completely useless if you are competing against them.

Here is the comparable dataset the Chai round writes onto the sourcing wall:

$20M raised per current employee is the useful frame. Chai does not have a hiring budget problem. It has a candidate existence problem, and so does every founder trying to build a competing lab. Isomorphic Labs, Alphabet's drug-discovery subsidiary, raised $600M in 2025. Recursion and Exscientia merged in a $688M deal. All of them are fishing in the same 200-person pond, and none of them can print people.

OpenAI is on the cap table and that is the whole moat

Chai's structural advantage in this fight is not the model. It is that OpenAI is a returning investor, which turns OpenAI's biology-adjacent alumni into a warm-referral channel Chai's competitors cannot replicate.

The returning investor list on the Series C reads like a talent map: Thrive, OpenAI, Oak HC/FT, Menlo, General Catalyst, Glade Brook, Avenir, Lachy Groom, Yosemite. OpenAI is the one that matters for recruiting. Sam Altman has repeatedly named scientific discovery as AI's most important application, and OpenAI's cap table exposure to Chai means every OpenAI researcher curious about biology gets pointed to exactly one landing pad. Isomorphic has DeepMind. Chai has OpenAI. Everyone else is fighting for scraps.

The founding profile explains why this works. Joshua Meier, Chai's CEO, is the archetype: OpenAI, then Meta FAIR where he co-led ESM1, the first transformer protein-language model, then chief AI officer at Absci, then Chai. His co-founder Matthew McPartlon was tech lead for de novo antibody design modeling at Absci. Jacques Boitreaud came through a similar route. This is not a random slice of ML talent. It is a specific two-hop path: Meta FAIR generative biology, then Absci, then Chai.

Chai's cap table is a referral engine. Its competitors have to cold-source the same 200 people from scratch.

Absci, not OpenAI, is the highest-yield sourcing list in the industry

The single most valuable list of names for anyone hiring in this space is not OpenAI's biology alumni. It is Absci's mid-tenure AI staff, because three of four Chai founders routed through there.

Absci is the Vancouver, WA-based antibody-AI company that has become the de facto training ground for the Chai cohort. Meier ran AI there. McPartlon led de novo antibody design modeling there. If you are a founder or recruiter trying to build a competing team, the sourcing sequence looks like this:

1. Absci alumni and current staff in AI research and modeling roles. Mid-tenure, past initial vest, before the next liquidity event.
2. Meta FAIR generative biology alumni with authorship on ESM1, ESM2, or ESMFold papers.
3. DeepMind AlphaFold alumni who did not follow Hassabis to Isomorphic.
4. OpenAI researchers who have publicly posted about biology, protein folding, or scientific discovery use cases.
5. Pharma internal AI teams at Pfizer, Eli Lilly, and Novartis, all of which now have Chai partnerships and are the next feeder pool as internal AI groups get frustrated with slow enterprise cycles.

Job-board keyword filters will not find these people. Their titles do not say "protein foundation model engineer." They say "Research Scientist" and "Member of Technical Staff" and, in Meier's case, "Chief AI Officer." The pool is defined by a two-hop career pattern, not a skill string, which is the exact gap Refolk closes: you describe the profile in plain English (something like "ML researchers who worked on ESM at Meta FAIR and then went to Absci or another antibody AI startup") and get a ranked shortlist, not a keyword dump.

The wet-lab distinction makes Chai's hiring harder, not easier

Chai runs a pure AI model with no wet lab, which sounds like a simpler company to staff and is actually the opposite. It puts Chai in a bidding war with Anthropic, OpenAI, and DeepMind for pure ML researchers, not with Recursion or Genentech for bench scientists.

Recursion, Genentech, and Isomorphic all own or contract significant wet-lab capacity. That means they hire computational biologists, structural biologists, and ML engineers who are comfortable working next to biologists holding pipettes. Chai does not need those people. It needs frontier ML researchers who happen to care about proteins. That is a much smaller, much more expensive pool, and the competitive set is Anthropic and OpenAI, both of which pay staff comp that until recently was unimaginable in biotech.

The good news for Chai is that the 50% hit rate reported for Chai-2 and its Eli Lilly deal (a mid-eight-figure annual access fee) give it a technical story that can compete with generic LLM work for the "I want to do meaningful science" cohort. The bad news is that FAIR and DeepMind can now say the same thing, and OpenAI has been increasingly public about scientific discovery as a north star. Mission is no longer a differentiator on its own.

The "no approved drug yet" ceiling changes retention math

Despite roughly $20B poured into generative AI drug discovery, zero AI-discovered drugs have been approved, which means Chai's researchers will start optimizing for equity liquidity over mission sometime in the next 12 months. That reshapes both retention and poaching windows.

Here is the mechanism. When a scientific mission has not yet produced a shipped result, and the researcher's stock is already worth real money on paper, the psychological anchor shifts from "am I doing the most important work" to "when can I actually sell." A $3.8B valuation means Chai has the paper to do outsized signing and retention grants, but so does Isomorphic at its $600M raise and Xaira and every other well-funded competitor. Expect the next 12 months to feature signing bonuses that dwarf FAANG staff comp for the top of this pool.

For recruiters, that means two things:

• Poach windows tighten around vesting cliffs at Chai, Isomorphic, and Absci. The 12-to-18-month zone after a large funding round is when researchers reassess. Map those dates.
• Second-tier feeder pools become more important. As primary pool members lock into golden handcuffs, the next hire will come from pharma internal AI teams or from academic labs at MIT, Stanford, and ETH Zurich. Those people do not show up in keyword searches. They show up in citation networks and PhD advisor graphs.

7.2x

How to source this pool without a $400M round

You source it by mapping alumni graphs and second-degree lab networks in plain English, not by filtering titles or skills. Chai's competitors will lose this fight if they use conventional recruiting tools, because conventional tools do not describe careers, they describe résumés.

The playbook that actually works:

• Start from named labs, not skills. Meta FAIR generative biology, DeepMind AlphaFold, Absci AI, OpenAI, Baker Lab at UW, Ovchinnikov Lab at Harvard. Name the labs, then find current and former members.
• Use career-path descriptions as your query. "Went from FAIR to a startup," "was at Absci in 2023 and 2024," "co-authored on ESM2." These are the queries that surface the real pool. Refolk handles this natively: you ask in plain English and it resolves the career path across GitHub, LinkedIn, and the open web, which is the whole reason keyword sourcing dies on contact with this segment.
• Track publication authorship dates. New co-author additions to protein-model papers are the earliest signal that a lab has hired someone worth calling. Faster than LinkedIn updates by three to six months.
• Watch the pharma partnership announcements. Chai has Pfizer, Eli Lilly, and Novartis deals. Every partnership creates a small internal team on the pharma side who will be interview-ready in 12 to 18 months when the honeymoon ends.
• Treat Chai Discovery jobs postings as a decoder ring. The exact language Chai uses in its specs ("next-generation AI x biology foundation models") tells you what to search for in other candidates' bios. Reverse the pitch.

The strategic point is that AI drug discovery hiring is now an alumni-graph problem, not a job-board problem. Whoever maps the graph fastest wins. Refolk exists specifically for this shape of search: describe the person you want in a sentence, get the ranked list of people who match, and skip the six weeks of Boolean-string archaeology in the middle.

FAQ

How many people can actually build an AI x biology foundation model?

Realistically, about 200 worldwide, and only around 7 currently list the exact combination of a Research Scientist title and Protein Structure Prediction as a skill in Refolk's index. The buildable pool sits in the low hundreds because the work requires deep ML research chops plus meaningful protein or biology background, which historically only converges in a handful of labs: Meta FAIR generative biology, DeepMind AlphaFold, the Baker Lab, Absci, and now Chai itself.

Why is OpenAI investing in Chai Discovery instead of building its own biology lab?

OpenAI has been a returning investor in Chai since earlier rounds, and the arrangement functions as both a strategic bet and a talent alliance. Building a competitive biology foundation-model group internally would require OpenAI to reallocate senior ML talent away from its core work, whereas backing Chai gives it exposure to scientific discovery outcomes and, in effect, a preferred landing pad for OpenAI researchers who want to work on biology without leaving the extended network.

Where do most Chai Discovery hires come from?

The dominant path is Meta FAIR generative biology, then Absci, then Chai. Three of four founding-team members route through Absci, including CEO Joshua Meier (who co-led ESM1 at FAIR before becoming Absci's chief AI officer) and Matthew McPartlon (Absci's de novo antibody design lead). Recruiters chasing OpenAI biology alumni should also be watching Absci's AI staff, because that is where the actual pipeline sits.

What is the best sourcing approach for protein foundation model talent?

Skip skill filters and map career paths instead. The pool is defined by two-hop patterns like "FAIR to Absci" or "DeepMind to Isomorphic," not by title strings. Use plain-English career descriptions, monitor paper co-authorship for new hires at target labs, and treat pharma internal AI teams at Pfizer, Eli Lilly, and Novartis as the next feeder pool when their enterprise cycles frustrate frontier-minded researchers.