Chai Discovery Just Hit $3.8B. The AlphaFold-Fluent Senior Pool Is Zero.

On July 14, 2026, Chai Discovery closed a $400M Series C at a $3.8B valuation, nearly triple its December mark, with Pfizer, Eli Lilly, and Novartis already running its molecular-design models in production. The round is explicitly earmarked for R&D expansion. That expansion collides with a talent pool so thin that a strict keyword search in Refolk's index returns zero senior profiles.

If you are a founder or head of talent about to copy Chai's JDs onto LinkedIn and hit boost, stop. This market does not work that way.

The pool everyone is fighting over is 540 people, and the senior slice is zero

In Refolk's index of professional profiles, only 540 US-based people list AlphaFold or protein structure prediction as a skill. When you add Senior/Director/VP seniority plus a "transformer protein structure" keyword filter, the count collapses to zero. That is the honest picture of the AlphaFold talent pool in mid-2026.

The 540 is not a shortlist. It is the total addressable universe before you filter for seniority, transformer-architecture fluency, US work authorization, non-compete cleanliness, or willingness to leave a tenured academic position. Most of the 540 are wet-lab structural biologists who run AlphaFold as inference, not ML researchers who can modify a diffusion head or a graph neural network attention block.

540

That gap between "can run AlphaFold" and "can build the next AlphaFold" is the entire recruiting problem. Chai, Isomorphic Labs, Recursion, and Genesis Therapeutics are not competing for 540 people. They are competing for the low double-digits who sit inside that 540 and also carry deep transformer, diffusion, and GNN expertise.

Why the ex-Isomorphic funnel is mathematically tiny

Ex-Isomorphic Labs is the most obvious backfill for anyone building on the AlphaFold lineage, and it is also the smallest funnel in the market. Isomorphic Labs has 116 total employees company-wide, spanning London and Lausanne, and only a fraction of those are ML researchers rather than ops, legal, BD, or wet-lab staff.

Do the arithmetic:

• 116 total headcount is a hard ceiling
• ML research and engineering is realistically 40 to 60 of those
• Attrition is low at a company launched specifically to advance the Nobel-winning AlphaFold system
• Of the leavers, some go back to DeepMind proper, some to academia, some to Google Health

That leaves an ex-Iso pool that is globally 30 to 50 people, split across geographies and often bound by non-competes. Any sourcing plan that assumes meaningful ex-Isomorphic volume is fantasy. You will hit the same six names as every other recruiter in the space, and those six names already have four inbound offers.

Isomorphic is also actively recruiting ML engineers and research scientists at ICML 2026 in Seoul with scheduled one-on-one candidate meetings. If Chai's recruiters are not on that same conference floor, the pool never hits the open market.

Where the real backfill actually lives

The realistic backfill for AI drug discovery hiring is dispersed across mid-tier biotech and academia, not concentrated at the FAANG-of-bio. Refolk's index shows the top US employers absorbing this skill set are Alamar Biosciences, Leash Laboratories, Sanford Burnham Prebys, University of Pennsylvania, Alector, and Kite Pharma. None are household-name AI-bio labs.

That distribution changes the shape of the sourcing campaign entirely:

1. Target labs, not companies. A single PI at UPenn or Sanford Burnham can be worth more than a whole team at a mid-tier biotech. Map the PIs, not the logos.
2. Chase the second-author, not the first. First authors on major structural-biology-plus-ML papers have ten offers. Second and third authors have deep skills and half the attention.
3. Read the acknowledgments. The engineer who "helped with model implementation" is often the exact transformer-plus-protein-structure hybrid you need.
4. Watch the compute footprints. GitHub commits to ESM, OpenFold, RoseTTAFold forks, and Boltz repos surface names that never appear in a boolean LinkedIn search.
5. Follow the postdocs. A 3rd-year computational-biology postdoc with two NeurIPS papers is the highest-signal profile in the market and rarely surfaces via inbound.

This is the exact gap Refolk closes. You describe the person in plain English ("computational biology postdoc, published on diffusion models for protein design, US-based, not currently at Isomorphic or DeepMind") and get a ranked shortlist across GitHub, LinkedIn, and the open web. Boolean strings do not resolve this problem because the vocabulary is inconsistent across paper acknowledgments, GitHub bios, and lab pages.

The comparable numbers, in one table

Here is the full comparison from Refolk's index and public sources, side by side.

The US:UK ratio is the row most people miss. The US has roughly 5x the UK's population and roughly 10x its biotech capital, yet the AlphaFold-skilled talent pool is only 4x larger. On a per-capita basis, London is denser. That is not a coincidence; it is the direct residue of DeepMind and AlphaFold having originated there.

London is the second office nobody is opening

London is the most under-priced sourcing geography for biotech AI recruiting in 2026, and Chai does not yet have a real presence there. The UK has 136 AlphaFold-skilled profiles in Refolk's index against the US's 540, but the density is higher and the compensation expectations are 30 to 50 percent lower for equivalent seniority.

The London ecosystem includes:

• InstaDeep (recently acquired by BioNTech, top UK employer of AlphaFold-skilled talent in Refolk's index)
• Isomorphic Labs London office (the primary poach target, if you can get past the retention packages)
• DeepMind proper, where the original AlphaFold work happened
• Sanger Institute in Cambridge, still one of the deepest structural biology benches in the world
• Exscientia alumni, now scattered post-Recursion merger

The US has 5x the population and 10x the biotech capital. On AlphaFold talent it only has 4x the profiles. London is punching above its weight.

Chai's competition is not blind to this. Isomorphic already operates in London and Lausanne. Recursion has a Toronto presence. If Chai builds a Bay Area only recruiting funnel, it is fishing in a pond that has already been fished by four competitors with better name recognition in structural biology.

Why "post the JD" fails on protein structure ML engineer roles

Posting a protein structure ML engineer role and waiting for inbound is the single worst strategy for this hiring market in 2026. The people who can actually do the job are (a) not looking, (b) not on LinkedIn Recruiter's normal filters, and (c) evaluated by their peers on paper output, not job titles.

Three mechanisms drive this:

1. Title inconsistency. The exact same person is a "Research Scientist" at DeepMind, an "ML Engineer" at Isomorphic, a "Computational Biologist" at Genesis, and a "Postdoctoral Fellow" at UPenn. Boolean searches by title miss 60 to 80 percent of the pool.
2. Skill-tag decay. Most senior researchers have not updated their LinkedIn skills since 2022. Their AlphaFold work shows up in a 2024 Nature paper, not in a skills chip.
3. Signal lives on GitHub and arXiv. The strongest candidates have three commits to a Boltz fork and a first-author arXiv preprint. Neither surfaces in a recruiter search.

Recruiters using Refolk describe the person they want ("second author on a diffusion-for-protein-design paper in the last 18 months, currently at a US academic lab, has GitHub activity in the last 90 days") and let the system reconcile GitHub, LinkedIn, Google Scholar, and lab websites into a single ranked list. That is the workflow that actually returns names in this market.

The realistic sourcing plan for Chai and its competitors

The realistic sourcing plan for any company chasing Chai Discovery jobs candidates has five parts, and none of them are "post more roles."

1. Build a 50-target lab map, not a 5-company target list

Refolk's index shows the pool is scattered across Alamar, Leash Labs, Sanford Burnham, UPenn, Alector, Kite Pharma, plus a long tail of academic labs. Map 50 PIs and their current postdocs and PhDs. That is your true universe.

2. Send recruiters to ICML, NeurIPS, and RECOMB

Isomorphic is already booking one-on-one candidate meetings at ICML 2026 in Seoul. If your recruiters are not on the conference floor, you are competing on second-best channels for the pool that Isomorphic is meeting in person.

3. Open a London sourcing arm

Not a full office. A single senior recruiter based in London, with a mandate to work InstaDeep, DeepMind alumni, the Sanger Institute, and Isomorphic's London bench. UK-based hires can work US hours or relocate; either way, the density is better.

4. Use plain-English search over boolean

The pool is impossible to define with keywords. Titles vary, skill tags are stale, and the strongest signal lives in paper acknowledgments and GitHub commits. This is exactly where Refolk earns its keep: describe the person in one sentence and get back a ranked shortlist that spans GitHub, LinkedIn, and the open web.

5. Move on offers in days, not weeks

The AlphaFold talent pool is so small that every candidate has multiple concurrent conversations. Offer cycles longer than 10 days lose. Chai has $400M in fresh capital and can move fast; if the process drags, the money does not matter.

FAQ

How many AI drug discovery ML researchers exist in the US?

Refolk's index shows 540 US-based profiles listing AlphaFold or protein structure prediction skills across all seniorities. The senior slice, filtered by Senior/Director/VP title plus transformer-plus-protein-structure keywords, returns zero. The usable pool that Chai, Isomorphic, Recursion, and Genesis are actually fighting over sits in the low double-digits.

Is ex-Isomorphic Labs a realistic backfill for Chai Discovery?

Only in tiny volumes. Isomorphic Labs has 116 total employees company-wide, of which perhaps 40 to 60 are ML research and engineering. The global ex-Iso pool is realistically 30 to 50 people. Any sourcing plan that assumes meaningful ex-Iso volume is fantasy, and the same six names are already saturated with inbound.

Where should I actually source protein structure ML engineers?

Academic labs and mid-tier biotech, not the household-name AI-bio companies. Refolk's index flags Alamar Biosciences, Leash Laboratories, Sanford Burnham Prebys, University of Pennsylvania, Alector, and Kite Pharma as the top US employers absorbing this skill set. Add London (InstaDeep, DeepMind, Sanger Institute) for a disproportionately dense per-capita pool.

Why doesn't boolean LinkedIn search work for AlphaFold talent pool sourcing?

Because titles are inconsistent, skill tags are stale, and the strongest signal lives on GitHub and arXiv, not LinkedIn. The same person is a "Research Scientist" at one lab and a "Computational Biologist" at another. Plain-English search across GitHub, LinkedIn, and the open web returns candidates that boolean strings systematically miss.