Berkeley Lab's 13 New AI Projects Map What Federal 'AI for Science' Money Is Buying

The U.S. Department of Energy on Wednesday named Lawrence Berkeley National Laboratory to lead 13 new AI projects and to partner on more than 30 others, the first publicly visible awards under a federal program the DOE calls the Genesis Mission. Read as a portfolio, the 13 projects are the first public map of what "AI for science" is now being asked to do in U.S. federal strategy: not bigger chatbots, but AI-controlled laboratories, instrument-coupled forecasting for the grid, and AI-driven discovery for materials, fusion, and quantum hardware.

The Genesis Mission is a Department of Energy initiative that pairs AI with supercomputing, quantum systems, and scientific instruments to speed discovery in energy and the physical sciences. Energy Secretary Chris Wright announced the first round of project funding on July 22. Berkeley Lab is the single largest lead in this initial batch. The lab's own Genesis Mission page counts 13 lead projects and roughly 30 partner roles, and the spread is deliberate: critical minerals, materials, manufacturing, fusion, and energy research all sit inside the same federal envelope.

Three projects illustrate the shape. ASPIRE puts AI on top of DOE precipitation data to forecast U.S. hydropower output and grid conditions. A separate project funds AI-driven autonomous laboratories for materials discovery, with the lab and its partners describing AI agents that run experiments, not just suggest them. A third effort, on AI-Powered Discovery of Quantum and Optoelectronic Materials, applies the same toolkit to the components behind quantum hardware and advanced semiconductors. Three more Berkeley Lab-led project labels appear on the lab's program page: MOAT (a multi-office particle accelerator team), OPAL (an orchestrated platform for autonomous laboratories), and SYNAPS-I (synergistic neutron and photon science with AI). They reinforce the same pattern: AI that touches instruments, not just models.

"This research builds on decades of work at Berkeley Lab in high-performance computing, large-dataset analysis, and AI modeling," Berkeley Lab director Kathy Yelick said in the announcement. "The goal is to help scientists reach answers faster and develop better AI tools for science." The line is also positioning: the Genesis Mission is a federal claim that the next decade of AI for science will be built on top of the national labs' supercomputers and instruments, not on a generic foundation model trained on web text.

Every Berkeley Lab project in this first round is Phase 1, which the DOE announcement defines as identifying promising research pathways, designing workflows that integrate AI with scientific investigation, and evaluating whether AI accelerates discovery, improves prediction, enhances experimentation, or generates new insights. None of the 13 projects is producing outcomes yet. The Phase 1 framing is also a federal hedge: the program wants to learn which of these workflows actually pay off before the next dollar goes out.

The bigger read is a redefinition. "AI for science" in the U.S. has, for the last three years, mostly meant foundation models for chemistry, biology, and weather, plus a steady drumbeat of benchmark results. The first Genesis Mission portfolio is something different. It treats AI as a control system for laboratories and instruments, and it routes that work through the Department of Energy's national labs, where the supercomputers, particle accelerators, and light sources already sit. Foundation models for chemistry, biology, and weather are still funded. The first money named is going to projects that put AI in the lab loop.

For Berkeley Lab, the 13 leads are also a strategy document. They sit across the lab's computational, materials, and physical-sciences divisions, and they are paired with another 30-plus partner roles on projects led by other national labs and universities. The portfolio shows where the lab expects AI to do useful work in the next funding cycle: grid and water, materials and manufacturing, fusion and quantum. The next visible step is the Phase 1 evaluations, which the DOE will use to decide which workflows get scaled.