Google DeepMind is putting 40 million dollars in AI tokens and cloud credits into the Genesis Mission, the Department of Energy program to double the pace of American scientific discovery over ten years. National laboratories will get access to AlphaEvolve, AlphaFold 3, AlphaGenome, WeatherNext, and Gemini for Government. The same day, OpenAI published its own commitment with the DOE to bring frontier AI to national laboratories.

It’s the third signal in a week putting compute front and center. As we reported on July 22, OpenAI confirmed Project Camellia with 3.2 gigawatts in Georgia and AMD signed a 5 billion dollar deal with Anthropic for MI450 GPUs. Those with frontier models are moving massive resources toward whoever does research.

Why it matters to you. If you work in a field that produces lots of data, access to tools like AlphaFold 3 or AlphaEvolve is shifting from luxury to infrastructure. A researcher at the Pacific Northwest National Laboratory is already using AlphaEvolve to explore mathematical systems too complex for the human mind. At the Rocky Mountain National Laboratory, Gemini cut microscope calibration time from 90 minutes to 13.

Free tokens have expiration dates and access programs are selective. The signal is that laboratories are becoming the first enterprise customers for frontier AI, and the skills to use it well will soon become market requirements.

In detail

The Genesis Mission launched in December as a White House initiative: use AI to double the pace of American scientific discovery within a decade. Google had already moved with an early access program for the DOE’s 17 national laboratories. The July 22 announcement at the Genesis Mission Summit 2026 raises the stakes: 40 million dollars in tokens and cloud credits, plus access to a portfolio of science-specific tools.

Researchers get access to AlphaEvolve, a Gemini-based agent that designs algorithms and explores mathematical spaces too vast for the human mind. Then AlphaFold 3 to predict protein structure and interactions, AlphaGenome to understand how DNA variations affect biology and disease, WeatherNext for weather forecasting, and AlphaEarth Foundations to map the planet. Added to this are workstations and Gemini for Government tokens for tens of thousands of laboratory users, from research to administration, for one year.

The most concrete case comes from the Pacific Northwest National Laboratory. Senior scientist Henry Kvinge uses AlphaEvolve to map combinatorial systems connecting geometry and algebra. “We can automate exploration of countless angles,” he says. The discoveries are already guiding his group’s future research programs.

At the Rocky Mountain National Laboratory, Steven Spurgeon’s team has integrated Gemini into laboratory tools to build an autonomous experimentation system. Microscope calibration went from over 90 minutes to 13, and manual steps to focus from 50 to 2. The system observes, reasons, and decides in real time, opening parts of the materials design space that manual operation couldn’t reach.

The bigger picture. The same day, OpenAI published its own commitment with the DOE and national laboratories. It adds to Project Camellia, the 3.2 gigawatt data center in Georgia, and the AMD-Anthropic 5 billion dollar deal for MI450 GPUs. Read together, these signals say the competition front is shifting from benchmarks to real use cases. AI labs are looking for partners who consume large amounts of compute on hard problems, and government laboratories combine stable budgets, complex problems, and steady demand.

Limitations. The 40 million in tokens are in-kind credits, not direct funds. Access is reserved for Genesis Mission program winners and national laboratory staff: an outside university researcher can’t sign up. Duration is one year for Gemini for Government, and it’s unclear what happens after. Specialized tools like AlphaEvolve and AlphaGenome are in experimental phase: researchers describe them as “still in testing.” The bottleneck is the competence to frame them well, more than access itself.