Biohub, DOE, NIH, Google DeepMind, Isomorphic Labs and Meta expand biological data collaboration
On 7 October 2026, Biohub, the U.S. Department of Energy and the National Institutes of Health announced a $1.8 billion coordinated commitment to generate AI-ready biological datasets, with DOE investing over $500 million over five years, NIH coordinating datasets from more than $500 million in prior federal investment, and Google DeepMind, Isomorphic Labs and Meta collectively contributing $300 million to the Virtual Biology Initiative.
Where the rake sits
No commercial rake is visible from any seat in this announcement. The public funders (Biohub, DOE, NIH) are paying to generate the substrate and publish it; the three technology companies are co-funding generation of inputs they intend to train on, which keeps the capability-stack climbing — and whatever rake it yields — inside their own model businesses rather than inside the data-generating consortium. The split between open substrate and proprietary models downstream is the live design question, but no party is paying another for data here.
What happened
- Total commitment announced 7 October 2026 is $1.8 billion in funding, data, computation and measurement technology.
- DOE commits more than $500 million over five years through the Genesis Mission, drawing on exascale supercomputing, the Joint Genome Institute and the Environmental Molecular Sciences Laboratory.
- NIH contributes datasets, repositories and knowledge bases built on more than $500 million in prior federal investment, coordinated via its Bio Genesis Mission and NLM/NCBI infrastructure.
- Google DeepMind, Isomorphic Labs and Meta collectively invest $300 million in the Virtual Biology Initiative, announced April 2026.
- Biohub's founding $500 million splits into $400 million for new measurement technologies (cryo-ET, large-scale microscopy, perturbation tools) and $100 million for research outside Biohub.
- Allen Institute, Broad Institute, Gladstone, Human Cell Atlas, Human Protein Atlas and Wellcome Sanger join as scientific partners; NVIDIA will support the initiative to leverage accelerated computing infrastructure.
Who is involved
Nonprofit biomedical research network funded by the Chan Zuckerberg Initiative; generates and standardises foundational biological datasets (imaging, cell biology, measurement tech) as a public-good research producer.
Founded 2016 with a $600M endowment; CZI committed a further $800M-$1B over 10 years in 2021 to extend operations through 2031.
US federal department; here acting as a funder and operator of national-lab compute and measurement infrastructure contributing to AI-ready biological datasets.
Cabinet department; n/d for this specific commitment beyond the collaboration's aggregate $1.8B figure.
US federal biomedical research agency; co-funder and data-source partner for the collaboration's shared biological datasets.
US federal agency, FY budget ~$47B (publicly reported); n/d specifically for this commitment.
Alphabet's AI research division; contributing models, compute and funding (part of the $300M technology-company commitment) and an obvious consumer of AI-ready biological data for foundation models.
Division of Alphabet Inc. (NASDAQ: GOOGL); Alphabet FY2024 revenue approximately $350B (parent-level).
The reading
Per the Newswise release, Biohub, DOE and NIH are generating and standardising the datasets while the three technology companies (Google DeepMind, Isomorphic Labs, Meta) contribute compute, measurement tech and $300M of the ~$1.8B total; no single party is clearly selling to another.
The substrate's value is to foundation-model builders training on biology (predict/treat disease), but the announcement does not name any paying buyer whose decision the data would feed, so the enrichability call is not grounded in a buyer here.
Not settled by the public facts: the release describes making datasets 'accessible' but does not specify access terms, exclusivity, or what, if anything, remains proprietary to the technology-company contributors.
Not determinable; the parties have chosen a public-good configuration where capture is not the aim, so the under-capture question does not apply in the usual commercial sense.
Why it matters
The design is the tell: $1.5 billion of public and philanthropic money generates an open substrate, and $300 million from DeepMind, Isomorphic Labs and Meta buys a seat at the specification table for the inputs their proprietary models will train on. The capability stack climbs inside Alphabet and Meta; the datasets stay open at the base. Biohub, DOE and NIH are funding the floor on which someone else's rake will sit.
The argument this deal tests: The Unsexy yet Fundamental Part of AI Projects: Data
Related deals
- Anthropic's $400M acquisition of Coefficient Bio for domain-specific biotech data underscores substrate dimensions2026-04-01
- OneMedNet signs $11.5M+ agreement to supply regulatory-grade real-world imaging data for foundational AI model training2026-07-28
- Merck and Mayo Clinic announce R&D collaboration for AI-enabled drug discovery and precision medicine2026-02-18
- MITRE and The Weather Company announce collaboration using MITRE's Weather 1K AI training data2026-04-23
Sources
- newswise.com — primary
- en.wikipedia.org — party background
- chanzuckerberg.com — party background
Announced 2026-10-07