Google invests millions in Mark Zuckerberg’s efforts to create a ‘virtual cell’

Google DeepMind, Meta and the AI drug discovery startup Isomorphic Labs are putting a combined $300 million into Biohub, the biomedical research nonprofit founded in 2016 by Mark Zuckerberg and Priscilla Chan, according to Reuters as cited by The Verge. That money is one piece of a $1.8 billion effort to build AI-ready biological datasets.

The public side is just as notable. The Department of Energy plans to invest more than $500 million over five years, and the National Institutes of Health is contributing datasets, repositories and knowledge bases that came out of earlier federal investment totaling more than $500 million. The goal is a "virtual cell" that lets researchers run experiments as simulations before they ever touch a lab bench.

We think this is the right bet, and the structure is the story. Companies that compete hard on models are agreeing to fund the same data. That tells us something the hype cycle usually skips: in biology, the scarce thing is not a clever architecture, it is large, consistent, well-labeled measurements of how cells actually behave. Biohub's head of science, Alex Rives, frames it as a job that needs coordinated data generation at national and international scale. We agree, and we'd rather see that work done once, in a shared effort, than five times behind five corporate walls.

The upside is real if it works. A predictive model of a cell could let scientists screen ideas digitally, discard dead ends faster, and point scarce lab time and money at the experiments most likely to matter for preventing and managing disease. That is useful AI in the most literal sense.

Now the eyes-open part. First, a virtual cell is a goal, not a product. Nobody has shown that a simulation can stand in for wet-lab biology, and the danger if this scales carelessly is that "the model predicted it" starts substituting for evidence in grant reviews, drug development or clinical arguments. Prediction should narrow the search, not end it.

Second, governance. Public agencies are contributing hundreds of millions of dollars and years of taxpayer-funded data, and three of the world's most powerful AI companies are paying in alongside them. The reporting we have does not spell out who can access the resulting datasets, on what terms, or whether funders get privileged use. Those details decide whether this becomes a public scientific commons or a subsidized training set for a few labs. I'm watching for those terms more closely than for any early demo.

Our read: a promising, overdue investment in the unglamorous layer of AI for science. Get the openness rules right and this could be one of the more useful things big tech money does this year. Get them wrong and public data ends up underwriting private advantage.

Biohub has framed the virtual cell, a model researchers can use to run biological simulations, as central to its mission of combating disease since its founding in 2016. The new commitments combine private money from AI developers with federal funding and existing federally built data resources, which is an unusual mix for a single biology project.