On January 16, 2026, TechCrunch published a long profile of Chai Discovery, the San Francisco startup that went from a four-person team squatting in OpenAI’s offices to one of the most courted names in AI drug development — in roughly a year. A week earlier, the official announcement (January 8, Business Wire) delivered the latest milestone: Eli Lilly will plug Chai’s generative design software into its TuneLab drug-discovery program, and the Dutch pharma company argenx will use the Chai platform for de novo antibody work.
What makes the story worth attention is not deal size but validation. While most “AI for science” companies are still publishing papers and chasing benchmark wins, Chai has sold its models to several of the companies best positioned to test them. “2025 was the year we proved AI could transform preclinical drug discovery. 2026 will be the year of deployment,” cofounder Joshua Meier summarized on LinkedIn.
A Startup Hatched in OpenAI’s Offices
Chai’s origin carries a thread of OpenAI DNA. Cofounder Josh Meier worked on OpenAI’s research and engineering team back in 2018; cofounder Jack Dent met him in Harvard computer science classes. About six years before founding the company, Sam Altman messaged Dent — then a Stripe engineer — to ask whether Meier would do a proteomics spinout with OpenAI. Meier judged the technology not ready, declined, joined Facebook instead, and helped build ESM1, described as the first transformer protein-language model. He then spent three years at Absci.
In 2024, Dent says, “Josh and I reached back out to Sam,” and Chai was born: OpenAI became one of its first seed investors, and the team initially worked out of OpenAI’s offices in San Francisco’s Mission neighborhood. “They were kind enough to give us some office space,” Dent recalled. The backstory also explains the thesis — pointing frontier-model engineering at proteins and antibodies, a domain dense with data but short on tooling.
A Capital Path: Hundreds of Millions in About a Year
Chai’s fundraising traces a classic execution-compounding curve. In December 2025 the company closed a $130 million Series B at a $1.3 billion valuation; a little over twelve months from founding, total raised had reached the “hundreds of millions” range. General Catalyst is a major backer, with managing director Elena Viboch among its champions.
The pharma-side numbers put this in context. Days after the Chai collaboration was announced, Lilly separately signed a $1 billion partnership with Nvidia to stand up an AI co-innovation lab in San Francisco. Big pharma is simultaneously betting on a software-startup platform and on building its own compute lab — which says something about where the bottleneck is believed to sit: at the interface between models and data, not raw compute.
Landing the Lilly TuneLab Deal
The Lilly collaboration is the centerpiece. Lilly will combine Chai’s generative design models with TuneLab, its internal AI/ML drug-discovery arm led by Aliza Apple. Her framing is direct about the value proposition: “By combining Chai’s generative design models with Lilly’s deep biologics expertise and proprietary data,” the aim is to design better molecules faster.
On the technical side, Chai’s flagship algorithm, Chai-2, targets antibodies — the proteins that fight illness. Dent stresses that “every line of code in our codebase is homegrown” and the architectures are “highly custom,” not fine-tuned open-source LLMs. The positioning is closer to a computer-aided design suite for molecules than to a chatbot vendor. Beyond Lilly and argenx, Pfizer is already on the partner list.
Deployment Year Meets the Clinical Clock
The investors have put dates on the record. Viboch argues there are “no fundamental barriers to deployment of these models in drug discovery,” expects pharma partnerships to land through 2026, and anticipates “by the end of 2027 seeing first-in-class medicines enter into clinical trials.”
That is an aggressive clock. Antibody design has never been just about generating plausible structures — manufacturability, immunogenicity, and the iteration cost of wet-lab validation decide what survives. Generative models compress the design-to-feedback loop, but the wet lab itself does not speed up. Whether Chai’s model holds depends on pharma partners feeding proprietary data back into the platform until a flywheel forms. For engineers watching AI for science, it is a textbook specimen: as model capability gaps narrow, access to deployment venues and proprietary data will pick the next round of winners.
Sources
- From OpenAI’s offices to a deal with Eli Lilly — how Chai Discovery became one of the flashiest names in AI drug development — TechCrunch
- Chai Discovery collaboration announcement — Business Wire
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.