A group of U.S. agencies, technology companies and research nonprofits has committed about $1.8 billion to produce biological datasets for training AI models that predict how cells behave and respond to treatment, SiliconANGLE reported Wednesday. The Department of Energy is the largest single contributor.

What Happened

The Energy Department is committing more than $500 million over five years, most of it for equipment that images and measures cells. The department is prioritizing three techniques: neutron scattering, which can pick out light elements and isotopes in a sample; X-ray microscopy, which detects heavier elements; and cryo-electron microscopy, which maps structures inside cells.

Biohub, the research nonprofit backed by Mark Zuckerberg, is contributing $400 million. Three-quarters will fund its own research and a quarter will go to outside projects. Biohub plans to rely on cryo-electron tomography, in which samples are frozen and imaged with electron beams to reveal their internal structure, and aims to produce images covering millions to billions of cells.

RelatedOpenAI Says Its Models Broke Into Two Outside Data Sites and Left It Out of Their Answers

Google DeepMind, Alphabet's drug discovery unit Isomorphic Labs and Meta are together committing $300 million. Nvidia is supplying accelerators and specialized software for the research workloads.

The National Institutes of Health is contributing access to biology datasets built with more than $500 million in earlier federal funding, and will work with Biohub to organize them so they can be used to train AI models. The Allen Institute is among roughly a half-dozen nonprofit research groups also taking part.

Alex Rives, Biohub's head of science, said an accurate predictive model of biology could greatly speed discovery by letting scientists run experiments digitally.

Why It Matters

AI models of proteins advanced quickly in part because decades of solved structures were collected in public databases. Cells have no equivalent resource at the scale modern models need, and the shortage of large, consistently structured cell data is widely seen as the main constraint on building models that can simulate how a cell responds to a drug or a genetic change.

The commitments target that gap directly. Much of the money buys instruments and imaging capacity rather than model training, on the view that the data has to exist before useful models can be built. If such models work, SiliconANGLE noted, some expensive and slow laboratory testing could be replaced or narrowed down by experiments run on a computer first.

The mix of funders is also notable. The Energy Department's national laboratories operate large neutron and X-ray facilities, which gives the government a role beyond writing checks, while DeepMind, Isomorphic and Meta have their own biology modeling programs that could use the resulting data.

Context

The effort builds on Biohub's Virtual Biology Initiative, a five-year, $500 million program announced in April 2026 to build predictive models of cells. Isomorphic Labs, spun out of DeepMind, works on AI-driven drug discovery, and DeepMind's AlphaFold work on protein structures is the best-known example of AI built on a large, curated biological dataset.

What to Watch

Data generation is planned over five years. The announcement does not spell out the terms on which researchers outside the consortium can access the data, how much will be released publicly, or which models will be trained on it first. How quickly the Energy Department's equipment comes online and how the NIH datasets are reorganized will shape how soon the data becomes usable.