Chai Raises $400 Million to Design Drugs From Scratch

Chai Discovery announced a $400 million Series C on 14 July, valuing the company at $3.8 billion. Index Ventures led, with Kleiner Perkins, Sequoia and Dimension alongside. Endpoints News noted that the valuation is roughly triple where it stood seven months ago, after a $130 million Series B in December.

The investor list is long and includes OpenAI, Thrive Capital, General Catalyst and Menlo Ventures, plus new money from Bain Capital Ventures, Baillie Gifford and Battery Ventures. Total funding is now above $600 million.

What the models actually do

Chai builds models that predict and reprogram how molecules interact with each other, aimed at the pre-clinical stage of drug discovery. That is the long, expensive stretch before anything reaches a human trial, where most candidate drugs quietly fail.

The technical claim worth paying attention to concerns Chai-2, released last year. It was the first zero-shot generative platform for fully de novo antibody design to reach double-digit experimental success rates. The jargon is doing a lot of work there, so it is worth unpacking. De novo means the antibody is designed from nothing rather than discovered by screening what already exists. Zero-shot means the model was not given examples of working antibodies for that particular target. Double-digit success means that when the designs were made and tested at the bench, more than one in ten actually bound to what they were supposed to bind to.

For a field where the equivalent hit rates have historically been fractions of a percent, that is the number that made investors move. Chai-3, the current model, reportedly improves on target success rates and binding affinity again.

The customers are the interesting part

Eli Lilly, Pfizer and Novartis are using the platform, according to Fierce Biotech. Large pharmaceutical companies are conservative buyers with their own substantial computational chemistry groups, and they do not adopt outside tooling for the novelty of it.

That said, a caution belongs here. Nothing designed by these models has yet been through a clinical trial and come out the other side approved. Binding to a target in a lab is a real result and also a long way from a drug that works in a person without harming them. The gap between those two things has swallowed a great many promising molecules.

We have written before about the distance between AI-assisted molecular design and the claim that AI designs medicines by itself. Chai's results narrow that distance. They do not close it, and the company is not claiming they do.