At the JP Morgan Healthcare Conference in January, which annually fills San Francisco with deal‑making activity, four large AI×pharma tools deals were among the headlines. Central to the conversation was Chai Discovery, an OpenAI‑backed startup that is roughly two years old and currently valued at about $4 billion.
The Science team released its first podcast interview with Chai cofounder Matt McPartlon and product lead Neil Patil, who walked through the company’s approach and recent deals.
Why tools deals are becoming material for pharma
Traditionally, AI startups that set out to build tools for pharma often ended up developing their own drug pipelines. Convincing a pharmaceutical company to adopt a tool requires evidence that the tool works — usually strong target ideas and clinical validation. If a company can offer validated targets or drug candidates, it becomes easier to raise financing or structure milestone‑heavy licensing deals (so‑called “biobucks” arrangements) than to sell a promise that a tool will work across diverse pipelines.
According to Chai, what changed recently is that the tools have become good enough for drug design teams to trust. Improvements in structural modelling and, crucially, binding‑affinity prediction mean models can now indicate how well a molecule binds to a given protein, which in turn enables design rather than just analysis.
Scaling discovery and unlocking new mechanisms
When models reliably produce better candidates from the outset, the time between idea and lab testing shrinks. More candidates can be screened faster for toxicity and delivery, and molecules pushed to the clinic are likelier to succeed. Moreover, AI design tools can make possible mechanisms that are very hard or slow to create in the lab — for example, antibodies that precisely trigger specific molecular cascades or bispecific antibodies that bind two different proteins.
In the podcast, Matt McPartlon and the interviewer (RJ) stressed that the step change in model quality is not merely an efficiency gain: it enables things that were previously infeasible.
Product focus and partner learning
Chai has invested heavily in close partnerships with pharmaceutical companies to learn what researchers actually need. Neil Patil said working alongside partners lets Chai do informed research based on real, recurring requests rather than hypothetical features. That has influenced their user experience choices: Chai’s molecule editor is intended to feel more like a CAD or graphics design tool than a conversational chatbot.
Deals and customers
Since June, Chai has announced several significant collaborations, including agreements with Novartis, argenx, and Eli Lilly, and it has expanded its program with Eli Lilly. These partnerships both validate Chai’s approach and generate direct product feedback.
Why this matters
Binding models open the door to design‑centric workflows, shifting parts of drug discovery toward an engineering mindset where rapid iteration and product quality matter. Chai’s thesis is that for engineering problems the best product tends to win — advanced technology is necessary but must be coupled with strong UX and close industry engagement to scale.
The podcast episode covers additional topics such as why protein tokens are viewed as highly valuable, how models climb levels of abstraction as they improve, and how better tools change portfolio optimization across pharma, venture capital, and research.
