Biohub, DOE, NIH, DeepMind, Isomorphic and Meta commit $1.8B to open data for virtual cell models

Why it matters

【Key points】

On October 7, the nonprofit Biohub, the U.S. Department of Energy (DOE) and the National Institutes of Health (NIH) announced an expansion of the Virtual Biology Initiative, committing $1.8 billion in funding, data, compute and measurement technology. The goal is open, standardized, AI-ready cell data so models can predict how cells respond to interventions such as drugs or genetic perturbations, the so-called "virtual cell."

Google DeepMind, Isomorphic Labs and Meta are collectively investing $300 million. DOE plans more than $500 million over five years, drawing on national-lab supercomputing, cryo-electron tomography and autonomous labs. NIH contributes datasets built from more than $500 million in prior federal investment, which Biohub will standardize. Biohub committed $500 million in April. The Allen Institute, Broad Institute and others are also involved.

【Verifiable】

The primary source is Biohub's press release (October 7, 13:00 UTC); Pushmeet Kohli, who leads AI for Science at Google DeepMind, shared it on X. Axios reports that commercial partners get one year of exclusive access to the data they fund before public release; the press release itself does not state this. Related X discussion reached about 1,800 posts in 24 hours, steady at 50 to 140 per hour.

【Our read】

The signal is that AI labs are now paying for data, not just compute. AlphaFold worked because decades of public protein-structure data existed; cells have no comparable foundation, so bigger models alone just guess from sparse data. $300 million from three companies is modest, but it leverages roughly $1.5 billion in public and nonprofit resources, effectively handing the costliest data production to public institutions.

Watch how open "open" really is. If the one-year exclusivity holds, the funders use the data first, and academic and Chinese teams start a year behind. Competition in biology AI may shift from models to who gets intervention data first.