Chai Discovery becomes pharma AI tool standard after model quality leap

In January 2025, four major AI × pharma deals were announced at the JPM Pharma conference in San Francisco, a week-long annual event that brings together pharma deal-makers. Chai Discovery, an OpenAI-backed startup founded two years prior and valued at $4B, found itself at the center of this shift, despite being young and new to the space.

The historical bottleneck in pharma AI was structural. Companies developing AI tools for drug discovery faced a fundamental challenge: pharma buyers demanded proof that the tools actually worked. Generating that proof required identifying good drug targets with clinical validation—expensive and time-consuming. If you had that proof, it was easier to raise capital or license the tool for a specific target than to convince many companies to adopt a tool on the promise it would work broadly across their portfolios. As a result, most AI × pharma startups ended up building their own drug pipelines to prove their tools' value, rather than licensing the tools themselves. This path persisted until the tools became genuinely trustworthy.

The breakthrough was technical: structural AI models—tools that predict how proteins fold—evolved into binding models that predict binding affinity, measuring how strongly a molecule binds to a target protein. This shift from structure to binding changed what was possible. Matt McPartlon, Chai's cofounder, described the evolution as a series of unlocks: "Can I beat a mouse, and then can I do what mice can't do? And then how many levels of interaction can you just keep building on top of that?" The body explains the practical impact: designing an antibody that triggers a very specific molecular cascade takes years of trial-and-error in the lab; designing bi-specific antibodies (binding two different proteins) is similarly difficult. Good design tools can now unlock these capabilities directly.

Chai's strategy has been to invest heavily in close partnerships with pharma counterparts, learning from them what researchers actually needed rather than building tools in isolation based on theoretical value. Neil Patil, Chai's product lead, emphasized this: "We get to really learn about what is the stuff that would be helpful in research. So rather than doing research in a vacuum, based on what would hypothetically be cool, we're able to do informed research based on what our partners have just been organically asking us for help with." This customer intimacy translated into better product design—a molecule editor modeled on CAD or graphics design software, not a chatbot. Chai's observation is that in engineering-driven problems, the best product tends to win, and that good technology is necessary but not sufficient.

The payoff has been rapid. Since June, Chai announced three major deals: partnerships with Eli Lilly, Novartis, and argenx, plus an expansion of their Eli Lilly program. The January JPM announcements, combined with these signings, signal that pharma has crossed a threshold: the tools are now trusted enough to become standard infrastructure in drug design workflows. This unlocks a different kind of value for pharma companies: more and better candidate molecules reach the lab and animal trials faster, screening for toxicity and delivery improves, and molecules that reach clinical trials are more likely to succeed. For Chai and the broader AI × pharma ecosystem, it means the "tools deal"—licensing AI software to pharma—is now viable as a path to scale, not just as a stepping stone to building drugs in-house.