Antibodies by Algorithm: Chai Discovery Hits $3.8 Billion as AI-Designed Drugs Reach Lilly, Pfizer and Novartis

A two-year-old startup just tripled its price

On July 14, 2026, Chai Discovery said it had raised a $400 million Series C at a $3.8 billion valuation. The number is arresting on its own, but the trajectory is the real story: according to the reporting, the company closed a $130 million Series B at a $1.3 billion valuation in December 2025. That means Chai roughly tripled its valuation in about seven months, and it has now raised something on the order of $630 million total since it was founded in 2024.

The round was led by Index Ventures, with a syndicate that reads like a roll call of the venture industry — Kleiner Perkins, Sequoia Capital, Dimension, Thrive Capital, Menlo Ventures, General Catalyst, and, notably, OpenAI among the existing backers, joined by new investors including Bain Capital Ventures, Battery Ventures, Baillie Gifford, Sapphire Ventures, and BDT & MSD. OpenAI's presence on a drug-design cap table is its own small signal: the frontier-model world and the biology world are increasingly funded by the same pockets.

Chai is led by CEO Joshua Meier, previously of Meta AI Research and the AI-antibody firm Absci, alongside co-founders drawn from Absci and the French AI-drug startup Aqemia. In other words, this is not a lab experiment that stumbled into a business — it is a team that has been building protein models professionally for years.

What Chai actually builds

Chai makes foundation models for molecules, aimed squarely at antibodies. Antibodies are among the most valuable and most stubborn products in biology: they are large, floppy proteins, and designing one that latches tightly and specifically onto a chosen disease target has traditionally meant screening enormous physical libraries and hoping something sticks.

Chai's pitch is that you can generate candidates computationally instead. Its 2025 model, Chai-2, added a generative layer — described in the coverage as a full-atom diffusion architecture — that can design complete antibody sequences and structures from scratch, conditioned only on a target protein and a specified binding site. The current generation, Chai-3, is presented as a step change over that: the company says it substantially cuts the failure rate of designed antibodies and improves binding affinity and multi-specific engineering.

The framing that matters here is "de novo" — designing a binder from scratch rather than optimizing an existing one. If that works reliably, it compresses the earliest and most failure-prone stretch of drug discovery from months of wet-lab screening into a compute run.

The benchmark claims, read carefully

This is where an editor has to slow down, because the numbers are the load-bearing part of the hype. Per AllSci's account, Chai-2 achieved "16–20% experimental hit rates in fully de novo antibody design," against rates "below 1% for prior computational methods." SiliconANGLE's write-up frames the newer model differently, describing Chai-3 as roughly doubling success rates and citing hit rates in the 35–40% range for molecular targets.

Those two framings don't line up neatly — one is about Chai-2's de novo hit rate, the other about Chai-3's improvement — and neither is something WelclAI can independently verify. Treat them as company-reported figures repeated in trade press, not as peer-reviewed results. What is fair to say is the direction: the reported hit rates are one to two orders of magnitude above the sub-1% baseline attributed to older computational approaches. Even discounted heavily for optimism, that is the kind of gap that would change how discovery teams allocate their lab time.

Why Big Pharma is signing

The most persuasive evidence isn't the benchmark — it's the customer list. Three of the largest names in the industry have signed on in quick succession, and the timeline is tight enough to look like momentum rather than coincidence:

• Eli Lilly, in January 2026, in an agreement to accelerate biologics discovery.
• Pfizer, in June 2026, described as a license to Chai-3 plus a model trained on Pfizer's proprietary data.
• Novartis, dated July 13, 2026 — the day before the raise was announced — as an antibody-discovery collaboration.

That sequencing is worth dwelling on. Announcing a Novartis deal one day and a $400 million round the next is a textbook way to convert commercial validation into valuation. But the deals are also the substance: large pharmaceutical companies are conservative buyers with deep internal AI teams of their own, and getting three of them to license outside models — in some cases trained on their own confidential data — is a stronger endorsement than any leaderboard. CEO Joshua Meier's summary, as quoted in the coverage, is that "AI drug discovery has moved from promise to deployment."

The part that is not proven yet

Deployment is not the same as a drug. The sober counterweight, acknowledged in the reporting itself, is that no AI-discovered drug has yet been approved anywhere. The industry has reportedly poured on the order of $20 billion into generative AI for drug discovery, and there are said to be 173-plus AI-originated programs in clinical development — but "in clinical development" is a long, expensive corridor with a high failure rate at every door.

Designing a molecule that binds its target in a dish is the beginning of the story, not the end. A candidate still has to survive animal studies, then Phase 1 safety, then Phase 2 and 3 efficacy trials that take years and routinely kill programs that looked perfect on paper. AI can sharpen the first step; it does not exempt anyone from the rest. A designed antibody that hits its target beautifully can still be toxic, unstable, immunogenic, or simply ineffective in a living body.

So the honest read of Chai's valuation is that it prices in a future the clinic has not yet confirmed. The $3.8 billion is a bet that faster, cheaper generation of viable-looking candidates will eventually translate into more approved medicines — a plausible thesis, but an unproven one.

Hype versus real

Strip the round down and two things are simultaneously true. The hype is real in the sense that money and marquee customers are moving decisively: a two-year-old company with three Big Pharma deals and a who's-who investor base is not vaporware. And the caution is real in the sense that the ultimate product — an approved therapy that owes its existence to these models — does not yet exist, and the benchmark figures circulating this week are company claims rather than independently validated results.

What makes Chai a meaningful marker isn't that it "solved" drug discovery. It's that the center of gravity has visibly shifted from demos to contracts. When Lilly, Pfizer, and Novartis are paying for model access — and one of them is letting a startup train on its proprietary data — the question has quietly changed from "does generative biology work?" to "who owns the models the incumbents rent?"

The takeaway

Chai Discovery's $400 million round is best understood not as a single funding event but as a status update on an entire field. The reported hit-rate numbers should be held at arm's length until independent data appears, and the absence of any approved AI-designed drug remains the industry's honest asterisk. But the commercial signal is hard to wave away: three of the world's largest drugmakers have chosen to buy rather than build, and investors just tripled a startup's valuation in seven months on the strength of that choice. The wet lab will have the final word — as it always does — but for now, AI antibody design has moved from the "interesting research" column into the "pharma is writing checks" column. That, more than any benchmark, is what $3.8 billion is really pricing.