Zuckerberg and Chan's Biohub launches AI model to speed drug discovery

Biohub has unveiled an open artificial-intelligence system that it says can do more than predict protein shape: it can help design new protein binders that work in laboratory experiments. The release, announced in Redwood City, California, bundled three systems, ESMC, ESMFold2 and ESM Atlas, into what Biohub calls a world model of protein biology.

The scale is enormous. Biohub says ESMC was trained on about 2.8 billion protein sequences, while ESM Atlas maps 6.8 billion proteins and 1.1 billion predicted structures. The organization says the tools can map proteins across the tree of life, predict how they fold and function, and support the design of proteins intended for therapeutic discovery, including cancer and immune targets.

The promise is clear, but the question is narrower and harder: can this kind of AI move from impressive modeling to usable drug development? Biohub says the system has already been validated in immune disease and cancer cases, a claim that matters because protein biology has long been full of systems that look powerful in demonstrations but prove harder to use in real laboratories. Biohub’s leadership says the platform is meant to make biology more computable and to compress years of protein research into hours or days.

Access is part of the pitch. Biohub says the May 27 release is available to researchers everywhere, placing the model in the open-science camp rather than behind a closed commercial wall. That could matter for universities, biotech firms and medical labs that need shared tools and datasets if they are to test whether the model’s predictions hold up outside Biohub’s own environment. Nature has described the ESMFold2 atlas as greatly expanding the known protein universe, but the practical standard will be whether outside scientists can reproduce and extend those results.

The new release also sits inside a larger philanthropic wager by Mark Zuckerberg and Priscilla Chan. Biohub says the couple has pledged to give away 99% of their Meta shares over their lifetime, with Biohub as the main vehicle. On April 29, 2026, Biohub committed $500 million over five years to its Virtual Biology Initiative, including $100 million for coordinated global data generation and $400 million for technologies to measure, image and engineer biology. Biohub first signaled its frontier AI and biology push on November 6, 2025, and introduced ESM3 on June 24, 2024 as a generative model that simulates 500 million years of evolution.

That makes the new protein model more than a software release. It is a test of whether philanthropy-backed science can build infrastructure that changes timelines in drug discovery, or whether the field is still waiting for AI to prove it can reliably turn biological prediction into medical progress.