Chai Discovery Just Tripled to $3.8B. The Hire-Ready Pool Is 19.

On July 14, 2026, Chai Discovery closed a $400M Series C at a $3.8B valuation, tripling the $1.3B mark it set seven months earlier. The company lists 11 open roles on Ashby. If you are the recruiter or founder trying to fill them, the relevant question is not "how do we compete with Index Ventures money" but "where do the people who can actually build zero-shot generative models over protein structure spend their weekdays." The answer is almost never LinkedIn.

Why Chai's raise is a sourcing story, not a funding story

Chai's valuation math is downstream of a model math problem that only a couple of hundred people on earth can solve. Chai-2, released in 2025, was the first zero-shot generative platform for fully de novo antibody design to hit double-digit experimental success rates, roughly 20% hit rates versus the ~0.1% of prior computational methods. That is a 200x improvement, and it is the ratio investors are underwriting when they triple a valuation in seven months.

The funding stack tells the same story:

• 2024: ~$30M seed
• August 2025: $70M Series A
• December 2025: $130M Series B at $1.3B (General Catalyst, Oak HC/FT)
• July 14, 2026: $400M Series C at $3.8B (Index Ventures, with Kleiner Perkins, Sequoia, Dimension, and OpenAI participating)

New backers on the C include Bain Capital Ventures, Battery Ventures, Baillie Gifford, BDT & MSD, Sapphire Ventures, and Avra Capital. Named pharma customers are Eli Lilly and Pfizer, with Novartis as a collaborator on AI-driven antibody discovery. That customer list buys Chai time. It does not buy Chai people.

200x

The talent pool has two tiers, and most recruiters source the wrong one

There are roughly 376 people globally whose public profiles mention "AlphaFold," but only about 19 who are actually building generative protein design systems. Sourcing the first pool and hoping to filter down is the standard failure mode.

In Refolk's index of professional profiles, 376 people globally reference "AlphaFold" in a headline or experience. The top employers in that broader pool skew toward Google and pharma bioinformatics shops like Jubilant Biosys, with a long tail across Vietnam, France, Bengaluru, and the SF Bay Area. Most of those profiles are downstream users: bioinformaticians running inference on public checkpoints, pharma associates, computational biology grad students who ran AlphaFold once for a paper. They are not the people Chai, Isomorphic, or Cradle need.

The tighter query, public mentions of "protein design generative," returns 19 profiles worldwide. That is the hire-ready cohort. Top employers in the 19 are GenBio AI (3 profiles), Cradle, Isomorphic Labs, the Institute for Protein Design at UW, Archon Biosciences, and Incyte.

The 20:1 ratio is the number to internalize. For every one person who has actually shipped a generative model over protein structure, there are twenty who can talk about AlphaFold in an interview. LinkedIn keyword filters cannot distinguish between them, which is the exact gap Refolk closes: you describe the person in plain English ("built a generative model for de novo protein or antibody design, has a first-author paper or a public repo") and get a ranked shortlist that does not include the 357 bioinformaticians.

Over half the cohort is still in academia

About 53% of the hire-ready pool sits in postdoc, PhD, or PI seats, which means the correct sourcing motion is arXiv scraping and lab-website crawling, not recruiter InMail. Looking at the 19 profiles, the top titles are Research Scientist, Postdoctoral Researcher, Research Assistant, PhD Student, Principal Investigator (Computational Antibody Discovery), and AI Research Scientist.

That distribution changes everything about how you approach these people:

1. They do not respond to recruiter templates. They respond to technical questions about their last paper.
2. Their "job change" signal is a completed postdoc or a defended thesis, not a LinkedIn "Open to Work" toggle.
3. They are reachable through advisors and co-authors, not through Talent Solutions.
4. Their venue is NeurIPS MLSB workshop, RosettaCon, and ICML generative biology tracks, not tech job fairs.
5. They publish on arXiv q-bio and bioRxiv before they update any profile.

The practical implication for Chai and its competitors: the sourcing team needs someone who reads MLSB accepted-papers lists the day they drop, cross-references first authors against RFdiffusion and Chroma GitHub contributors, and can start a conversation with a citation instead of a comp band.

Their job-change signal is a defended thesis, not a LinkedIn Open to Work toggle. </pull> ## Paris and Toronto beat SF in the tight cohort The geographic concentration of hire-ready talent runs Paris (3) and Canada (3) ahead of Seattle, London, and Copenhagen. San Francisco is not the center of gravity here. This undercuts the default Bay-Area hiring pitch. The lab clusters that actually train and retain this cohort include: - **Paris**: Prescient Design (Genentech, Kyunghyun Cho's group), Institut Pasteur, Owkin adjacent groups - **Toronto and Montreal**: Mila, University of Toronto computational biology, Vector Institute - **Seattle**: Institute for Protein Design at UW (David Baker's lab, source of RFdiffusion) - **London**: Isomorphic Labs, direct Chai competitor - **Amsterdam and Zurich**: Cradle - **Copenhagen**: Novo Nordisk Foundation adjacent groups Chai is headquartered in San Francisco. Every hire out of Paris or Toronto is a relocation conversation, and relocation friction is a bigger blocker than compensation for scientists with tenure-track optionality. The teams winning this cohort are opening satellite offices, not raising base salaries. Isomorphic's London presence is a structural advantage against Chai for European candidates that no US comp package fully neutralizes.

refolk prompt: Find people who have first-authored a de novo protein or antibody generative model paper on arXiv or bioRxiv since 2023, currently in postdoc or PhD roles in Europe or Canada. note: You get a ranked list of the ~15 to 25 researchers who fit, with lab affiliations, advisors, and links to their papers and GitHub contributions, not 376 bioinformatics profiles. slug: ajt1wkhv6r

## The competitor set is also the training ground

Three of the top ten employers in the hire-ready cohort (Cradle, Isomorphic Labs, GenBio AI) are simultaneously Chai's competitors for pharma deals, which means the fastest path to hiring is poaching from someone Chai is trying to out-bid at Pfizer.

That creates a specific dynamic:

- **Expect counter-offers, not passive candidates.** These companies know exactly who their key researchers are and what it would cost to replace them.
- **The "why leave" story matters more than comp.** Candidates moving between Isomorphic, Cradle, GenBio, and Chai are choosing between similar cap tables and similar problem statements. The differentiator is model access, wet-lab loop speed, and who they get to work with.
- **GenBio AI is under-covered relative to its density.** Three of nineteen profiles in the tight cohort work there. Most recruiters have not indexed it as a primary source; the ones who have are winning uncontested outreach windows.
- **The Baker Lab pipeline is public.** UW's Institute for Protein Design published RFdiffusion, and its alumni network is the single largest training ground for de novo protein design talent. Track defenses, not job posts.

For a founder or head of talent working through this list, the useful move is to stop treating the competitor set as off-limits. Refolk can build the poach list directly: ask for "ML researchers at Isomorphic Labs, Cradle, or GenBio AI who have co-authored on generative protein design and joined in the last 24 months," and you get names ordered by how likely they are to have vested.

## Chai's founders are themselves a sourcing map

The Chai founding team's prior collaborators are a pre-built shortlist that most recruiters never think to pull. Chai was founded in 2024 by Joshua Meier, Jack Dent, Matthew McPartlon, and Jacques Boitreaud, with backgrounds spanning OpenAI, Meta FAIR, Stripe, and academic molecular design.

Two founder threads are particularly rich:

- **Joshua Meier** was on the Meta FAIR ESM team. ESM (Evolutionary Scale Modeling) is one of the two foundational protein language model lines, and its co-author list is a direct sourcing map. Anyone who published on ESM-1b, ESM-2, or ESMFold and is not currently at Chai is a warm intro away.
- **Matthew McPartlon** worked in structural ML before Chai. His co-author graph on structure prediction and antibody modeling papers overlaps with the same 19-person cohort in Refolk's tight query.

The founder-as-sourcing-signal move works for every company on this list. Isomorphic's team traces back to DeepMind's AlphaFold group. Cradle's team traces through EPFL and Adaptyv. GenBio AI's team is publishing under known advisors. If you know the paper trail, you know the shortlist.

19

What this means if you have 11 roles to fill

If you are hiring for Chai, Isomorphic, Iambic, Cradle, or Generate Biomedicines right now, treat the sourcing problem as a graph problem, not a keyword problem. Each of Chai's 11 open roles represents roughly 9% of near-term headcount growth capacity, so every miss is expensive and every unnecessary bake-off with a competitor for the same candidate compounds.

The concrete moves:

1. Stop filtering on "AlphaFold" as a LinkedIn keyword. You will surface 376 profiles and interview the wrong 20.
2. Build the arXiv layer first. MLSB, ICML generative biology, NeurIPS LMRL, and bioRxiv q-bio are the first-mover indexes.
3. Cross-reference GitHub. RFdiffusion, Chroma, Boltz, ESM, and OpenFold contributor graphs are public.
4. Map advisors, not just candidates. Baker (UW), Cho (Prescient), Correia (EPFL), Ovchinnikov (Harvard) each anchor a dozen-plus alumni.
5. Open a satellite or accept remote. Paris and Toronto are the two cities where the cohort is denser than SF.
6. Move on defense dates and grant cycles. The signal is academic, not corporate.

The 141-person or 150-person framing is directionally right for the ceiling. The floor, the people who have actually built and shipped generative protein design work, is the 19 in Refolk's index. Everything else is a training pipeline: real, worth watching, but not ready this quarter.

FAQ

How big is the AlphaFold-native talent pool globally?

The broad awareness pool is 376 profiles worldwide that publicly reference AlphaFold, according to Refolk's index of professional profiles. The hire-ready pool, defined as people who have actually shipped generative protein design work, is 19. External framings that cite "the AlphaFold-native pool is 141 or 150" are directionally right for the ceiling but overstate the number of researchers who could step into a role at Chai Discovery, Isomorphic Labs, or Cradle tomorrow.

Why does LinkedIn keyword sourcing fail for AI drug discovery hires?

LinkedIn surfaces the 20x-larger downstream user pool: bioinformaticians who ran AlphaFold once, pharma associates, and computational biology grad students. It cannot distinguish between someone who used a public checkpoint for a paper and someone who trained a generative model over protein structure. More than half the qualified cohort is also still in academic seats (postdoc, PhD, PI), and those people do not maintain current LinkedIn profiles or respond to templated recruiter InMail.

Where do Chai Discovery's competitors hire from?

Cradle, Isomorphic Labs, and GenBio AI show up as top employers in the same 19-person hire-ready cohort, which means the four companies are drawing from an overlapping pool and poaching directly from one another. The academic feeders are the Institute for Protein Design at UW (Baker Lab, source of RFdiffusion), Prescient Design at Genentech (Kyunghyun Cho), Mila in Montreal, and Institut Pasteur in Paris.

How should a small team compete with Chai's $400M for the same candidates?

Compete on problem access and wet-lab loop speed, not on cash comp, because the hire-ready cohort has similar offers from four well-funded companies. Open a satellite office in Paris or Toronto to remove the relocation blocker that hurts SF-based competitors. Source through advisor networks and arXiv co-author graphs rather than recruiters, and move on defense dates rather than job postings. Chai's own founders came out of the ESM team at Meta FAIR; the equivalent map for any founding team is already public, if you know where to read it.