Google DeepMind, Meta, and Isomorphic Labs are putting a combined $300 million into an effort to build a “virtual cell,” an AI model that can simulate how human cells behave. According to The Verge AI, which cited earlier Reuters reporting, the money goes to Biohub. That’s the nonprofit biomedical research group Mark Zuckerberg and Priscilla Chan founded in 2016. The investment is part of a larger $1.8 billion “Virtual Biology” initiative.
This is one of the largest coordinated pushes so far to turn biology into something you can model on a computer. Rival AI labs are also funding the same project, which you don’t see often.
The Quick Version
- Who’s paying: Google DeepMind, Meta, and Isomorphic Labs (DeepMind’s drug discovery spinout) are putting in $300 million together.
- Who’s leading: Biohub, the Zuckerberg-Chan nonprofit.
- The total pot: a $1.8 billion Virtual Biology initiative.
- Government backing: the US Department of Energy will invest more than $500 million over five years. The National Institutes of Health will contribute datasets, repositories, and knowledge bases built from more than $500 million in earlier federal investment.
- The goal: a “high-accuracy predictive model of the cell” that lets researchers run experiments as simulations.
What a “Virtual Cell” Actually Means
Today most biology happens at the bench. Researchers form a hypothesis, then spend weeks or months running wet-lab experiments to test it, and most of those experiments fail. That makes the process slow and expensive.
A virtual cell would flip that order. You’d ask a model how a cell responds to a drug, a gene edit, or a disease mutation and get a reliable prediction first. Only the most promising ideas would go on to physical testing. Biohub describes it as a way to “ask, predict, and answer biological questions digitally.”
“An accurate predictive model of biology could dramatically accelerate scientific discovery by enabling scientists to perform experiments digitally,” Biohub head of science Alex Rives said in the press release.
Why the Funding Is Mostly About Data
What stands out to me is how much of this initiative goes to datasets rather than model architecture. That’s the right call. AlphaFold worked because it had decades of carefully collected protein structures to train on. Nothing like that exists yet for whole-cell behavior.
Rives said as much: “It will require coordinated data generation efforts at a national and international scale, which is why these partners are coming together.”
That explains the mix of partners:
- AI labs bring modeling expertise and compute.
- The DOE brings national lab infrastructure and supercomputing.
- The NIH brings existing biomedical data that’s already been paid for.
- Biohub coordinates the effort and generates new experimental data.
No single company could build this training set by itself, so the competitors are pooling resources.
How It Fits the Bigger Picture
AI-for-biology has moved quickly. DeepMind’s AlphaFold solved protein structure prediction and won a Nobel Prize in 2024. Meta’s earlier ESM work showed that language models could learn the “grammar” of proteins. Isomorphic Labs is turning those advances into drug candidates.
Proteins are only one layer, though. A cell involves thousands of interacting molecules, signaling pathways, and gene regulation loops all running at the same time. Going from a single protein to an entire cell is a huge jump in complexity, and it’s widely seen as one of the hardest open problems in computational biology.
What This Means for You
- Biotech and pharma teams: watch for shared datasets and models. If Biohub releases them openly, as it has with past projects, they could cut early-stage research costs a lot.
- ML practitioners: biology is becoming a serious place to apply foundation models. Experience with multimodal and scientific data will be in demand.
- Investors and founders: government money alongside Big Tech money signals long-term commitment. Expect startups building on top of whatever infrastructure comes out of this.
- Everyone else: be patient. A trustworthy virtual cell will take years, not quarters.
What Comes Next
The DOE money runs on a five-year timeline, which gives a rough idea of how long serious progress will take. The question to keep asking is whether the models’ predictions hold up in real labs. If they do, drug discovery could start to look more like software development: test many ideas quickly in simulation and only build the ones that work.
The Verge AI has more details in its original report.