Analysis
Chan Zuckerberg Biohub is joining Google DeepMind, Meta, Isomorphic Labs, the U.S. Department of Energy and the National Institutes of Health on a roughly $1.8 billion effort to build what Biohub calls a "universal virtual cell" -- an AI model detailed enough to let scientists run biology experiments digitally before committing money to a physical lab, according to Axios and Dealroom. "We are at the beginning of a new scientific paradigm with artificial intelligence," Alex Rives, Biohub's head of science, said of the effort.
Who's actually paying for what
The figures, as reported, don't cleanly add to $1.8 billion -- itself a sign of how sprawling the coalition is:
“- NIH -- coordinating datasets built on more than $500 million in earlier federal funding; unclear how much is new money versus already committed.”
- Google DeepMind, Meta and Isomorphic Labs -- $300 million combined.
- Department of Energy -- more than $500 million over five years through its Genesis Mission, plus access to exascale supercomputing, X-ray and neutron scattering facilities, and autonomous labs.
- NIH -- coordinating datasets built on more than $500 million in earlier federal funding; unclear how much is new money versus already committed.
- Chan Zuckerberg Biohub -- $400 million for measurement tools like cryo-electron tomography, separate from the $500 million it pledged to its broader Virtual Biology Initiative in April.
Rives, who joined Biohub as head of science when it acquired his AI research lab EvolutionaryScale, says current datasets cover hundreds of millions of cells -- an accurate model, he says, will need billions, eventually trillions.
The AlphaFold precedent, and why this is harder
The closest comparable success is Google DeepMind's AlphaFold, which cracked protein structure prediction and reshaped structural biology research practically overnight. A whole living cell is a much bigger problem than a single folded protein: cells involve constantly shifting interactions between proteins, RNA, metabolites and organelles, and most of the data needed to train such a model simply doesn't exist yet -- it has to be painstakingly measured. Other groups are chasing pieces of the same goal, including the Arc Institute's own virtual-cell research efforts and AI-drug-discovery shops like Recursion Pharmaceuticals and Insilico Medicine, though none has assembled a coalition -- or a budget -- at this scale.
Dealroom notes the $1.8 billion ranks in the top 1% of comparable U.S. health grant rounds over the past four years, without naming the comparison set.
What the headline misses: a one-year head start for commercial funders before data goes public means Google, Meta and Isomorphic Labs effectively get first use of taxpayer- and philanthropy-funded research before competitors do, complicating the "open science" framing. And Rives's own billions-of-cells target is an admission that even hundreds of millions of cells -- what's already been measured -- isn't close to enough.
Partners say they're aiming to compress what would normally be decades of biology research into five years, with the first usable dataset expected in about a year. That timeline, more than the headline dollar figure, is what will tell VCs and biotech LPs whether this becomes a real tool or another ambitious research program that slips.
