Google's $40M Genesis Mission deal puts DeepMind protein-folding, algorithm-discovery, and Gemini AI into all 17 U.S. energy labs

Google committed $40 million of AI tokens and cloud credits to the U.S. Department of Energy's 17 national laboratories at the DOE Genesis Mission Summit 2026. The package gives the labs a year of access to five named DeepMind science tools, including AlphaFold 3 for protein structures, AlphaEvolve for algorithm discovery, AlphaGenome for DNA regulation, WeatherNext for weather forecasting, and AlphaEarth Foundations for Earth-systems mapping, and seats of Gemini for Government for tens of thousands of users. The $40 million is in-kind, paid as AI tokens and cloud credits rather than as a cash grant, and it runs under the White House's Genesis Mission.

Berkeley Lab announced on July 22, 2026 that it will lead 13 new Genesis Mission AI projects, the first concrete rollout of which scientific workflows the named Google tools are being wired into. The lab says the projects target fusion plasma simulation, the search of new materials, and the analysis of exabytes of data from advanced experimental facilities, the same problem categories the Google announcement names. Same-day corroboration from a national lab is the first independent evidence that the program is operational rather than aspirational.

The deal puts choice in a few private hands. A small number of AI labs now decide which scientific problems at the national labs get frontier models and which do not. Google DeepMind's earlier commitment gave all 17 labs early access to AI-for-science tools, and Google positions Gemini for Government as the AI backbone for DOE. That position loses force if OpenAI, Anthropic, Meta, or another frontier lab wins an equivalent or larger Genesis deal. This is the first major named-tool, multi-lab access deal under the program. More will follow if other frontier labs land equivalent commitments.

The DOE has not published how many tokens or GPU-hours the $40 million actually buys. Without that figure, the dollar amount is a notional commitment, not a measured runway. Who picks the Genesis Mission awardees, and on what criteria? How long does the access last across 17 labs? What changes about who has frontier AI for publicly funded science when a small number of private labs become the de facto model-and-compute layer? The labs that take these tools to fusion plasma simulation or materials search will be the first place those numbers get tested.

Berkeley Lab's 13 projects are scheduled to start within the current fiscal year, and the next set of Genesis awardees and tools will be named at upcoming DOE program milestones. The $40 million buys a year of access. The 13-project deployment list is where the program first shows up in a working lab.