The Chan Zuckerberg Biohub is teaming up with Meta, Google DeepMind, Isomorphic Labs and the U.S. government on an $1.8 billion effort to generate the kind of massive biological datasets that artificial intelligence models need to actually predict how cells behave, rather than just describe what scientists have already observed. The project, dubbed the Virtual Biology Initiative, aims to measure how cells respond to changes across far more conditions than researchers have ever systematically studied.
According to The Lufkin Daily News, which ran the Reuters report from reporter Krystal Hu, the initiative has already drawn in Meta Platforms, Alphabet, and the federal government as partners building open datasets for AI biological research. Meta, Google DeepMind and Isomorphic Labs jointly invested $300 million, while Biohub itself put in $500 million back in April. Biohub co-founders Mark Zuckerberg and Priscilla Chan established the nonprofit as a philanthropic venture, and Chan framed the stakes bluntly, saying biology has been just sort of a clever discovery-based science until this point.
The U.S. Department of Energy is committing more than $500 million over five years toward laboratory measurement, modeling and computation, according to the Reuters report carried by The Lufkin Daily News. That federal money flows through the Genesis Mission, a program that taps exascale supercomputers, X-ray and neutron scattering, cryo-electron microscopy, and autonomous self-running laboratories spread across 17 national laboratories, as detailed by TNW. The National Institutes of Health will separately coordinate datasets and repositories built with more than $500 million in earlier federal funding, which Biohub is now standardizing for AI training.
A Data Gap Measured in Trillions of Cells
The core problem the initiative is trying to solve is scale. Existing single-cell datasets used to train AI models max out at hundreds of millions to nearly a billion cells, but Biohub scientists say accurate predictive models will ultimately require datasets covering billions, and eventually trillions, of cells, a gap Quartz notes is staggering given that human tissues contain an estimated 37 trillion cells. The project intends to lean on tools like spatial transcriptomics, which maps molecular activity inside intact tissue, and cell screens that record how cells respond to environmental changes.
Biohub launched the groundwork for this push in April when it unveiled its Virtual Biology Initiative with its own $500 million, earmarking $400 million for internal cellular measurement and imaging technology and $100 million for external research grants, according to The Rundown AI. That earlier commitment built on Biohub's track record in the space; in May, the organization released a world model of protein biology built on Evolutionary Scale Models that can predict protein folding and interactions using an open atlas mapping more than a billion structures, per the Chan Zuckerberg Initiative.
Open Data for Taxpayers, a Head Start for Investors
The arrangement comes with a built-in tension over who gets access first. Biohub says it will release datasets publicly, but commercial funders will receive a head start on datasets they fund, with the organization imposing embargo periods before public release. Government-funded work, by contrast, will carry no restrictions on data access from the start. The embargo for Google DeepMind, Isomorphic Labs and Meta specifically runs one year, a mechanism TNW describes as typical of public-private research coalitions trying to attract corporate investment while still serving open-science goals.
Biohub plans to approach pharmaceutical companies and philanthropies for additional funding as the project scales, and Nvidia is already supplying supercomputing hardware and software while Renaissance Philanthropy leads broader fundraising efforts, per the same TNW report. The academic bench is deep too, with partners including the Allen Institute, Broad Institute, Gladstone Institutes, Wellcome Sanger Institute, and the Human Cell Atlas and Human Protein Atlas consortia all contributing to the effort.
Racing the Clock on Drug Development
The payoff Biohub and its partners are chasing is speed. The group plans to compress work that would normally take decades into roughly five years, aiming to produce a first dataset within about a year and accurate predictive models within five. That timeline is aimed squarely at an industry bottleneck: traditional pharmaceutical development takes an average of 10 to 15 years and costs $2.6 billion per approved drug, according to PhRMA, with fewer than 12% of candidate molecules that enter Phase 1 trials ultimately winning FDA approval.
Biohub is not alone in betting that physical lab data, not just computation, is the missing ingredient for AI in biology. Anthropic expanded its own biology efforts with a wet lab after acquiring startup Coefficient Bio, building a physical robotic lab in the Bay Area to test whether its Claude AI model can directly instruct automated equipment to run experiments, according to explainx.ai. The OpenAI Foundation has taken a similar path, starting a grant program exceeding $125 million for biological and medical datasets, which included a $40 million award to UNC Lineberger Comprehensive Cancer Center, per Venture Atlas.
The push fits a pattern Biohub has followed before in the Bay Area. The organization, which co-founded the $95M Seattle synthetic biology lab with the Allen Institute and University of Washington in December 2023, already runs research facilities at Mission Bay in San Francisco. Whether the newly formed coalition can actually deliver trillion-cell datasets on a five-year clock remains to be seen, but the money and the computing muscle behind the attempt are now firmly in place.
