Biohub teams up with Google DeepMind and the US government to build a universal virtual cell

Biology has a simulation problem. Engineers can model a jet engine down to the bolt before building it, but scientists still mostly learn how a cell reacts to a drug by testing it, one experiment at a time.

The Chan Zuckerberg Biohub wants to change that. On October 7, 2026, it expanded its Virtual Biology Initiative to a total of $1.8 billion, bringing in the US government and several of the largest names in tech to help build AI models of living cells.

The goal is a “virtual cell”: software that can predict how a real cell behaves when something perturbs it, such as a drug, a mutation, or a disease.

Who is putting money in

Biohub first committed $500M to the effort in April 2026, then broadened the initiative with outside partners.

The US Department of Energy is set to invest more than $500 million over five years to provide technology support.

The National Institutes of Health is coordinating work to standardize AI-ready datasets. Its prior funding for that standardization effort exceeds $500 million.

On the private side, Meta, Google DeepMind, and Isomorphic Labs contributed $300 million to the initiative.

The full roster of partners includes Biohub, Tahoe Therapeutics, Arc Institute, NVIDIA, the DOE, the NIH, Meta, Google DeepMind, and Isomorphic Labs.

The data comes first

The focus is on large multimodal biological datasets. That means combining several types of measurements, including single-cell omics and imaging data, so a model sees a cell from multiple angles at once.

A key milestone arrived in January 2026. Biohub, working with Tahoe Therapeutics and Arc Institute, produced a dataset covering more than 120 million single cells and 225,000 perturbation interactions.

That January dataset is the largest created under the initiative. It is more than four times richer than the earlier Tahoe-100M dataset, which had previously set the bar.

Open science, on a deadline

Biohub says all data and models produced through the initiative will be openly accessible to researchers worldwide.

The timeline is ambitious. The first dataset is expected to become available in approximately one year from October 2026, and accurate predictive models are targeted within five years of the announcement.

The initiative also aims to compress drug development timelines, from decades down to five years.

Biohub is part of the broader philanthropic network founded by Mark Zuckerberg and Priscilla Chan. With Meta also writing a check, the Zuckerberg name effectively appears on both sides of the funding ledger.