DOE funds a SLAC-led project to let AI choose catalyst experiments

The U.S. Department of Energy is funding a project that would let an AI agent choose which catalyst experiments to run next. Announced October 8, 2026, the Genesis Mission award supports a SLAC-led collaboration with Caltech and Lawrence Berkeley National Laboratory, linking AI reasoning to physical measurements in the search for better catalysts—materials that speed chemical reactions.

A slow search through millions of possibilities

The work builds on the Liquid Sunlight Alliance, a DOE-funded research hub led by Caltech’s Harry Atwater. Its researchers seek catalysts that help turn water, carbon dioxide and nitrogen into liquid fuels and useful chemicals.

To understand those materials, alliance scientists have used X-ray beamlines at SLAC’s Stanford Synchrotron Radiation Lightsource, or SSRL. The measurements reveal how a catalyst’s physical structure changes as it becomes active. But the work is slow: millions of compositions are possible, small chemical changes can sharply alter performance, and only a handful have been examined this way.

The new award gives the team short-term seed funding to explore feasibility. It is one of six new Phase I selections in DOE’s second Genesis funding round. The same announcement includes 12 Phase II awards totaling $159 million; that figure covers the Phase II awards, not the catalyst project.

From competing ideas to the next measurement

The proposed system will compare competing explanations of how catalysts work. An AI agent will identify the evidence needed to support or disprove those explanations, then select the next experiments. The goal is to use AI and automated experimentation to speed testing and identify the best catalysts for particular reactions, including liquid-fuel production.

At SSRL, the system will combine real-time X-ray measurements with electrochemistry, computer models and published knowledge. Each result will feed back into the reasoning process and guide the next measurement. That feedback is what makes the proposed laboratory system a closed loop.

The AI weighs the scientific evidence, decides which measurement would tell us the most, and that decision becomes a real experiment at SSRL.

Dimosthenis Sokaras, project principal investigator, in an SLAC press release quoted by Caltech

The first hurdle is understanding, not volume

Caltech researchers plan to prepare libraries containing hundreds to thousands of catalyst compositions for the X-ray beam. Caltech’s Joel Haber says the team will measure catalyst performance and structure at the same time, allowing researchers to connect structural changes with how well a material works.

Those data will be combined with electrochemical measurements from Caltech laboratories and information from computational chemists to continually update the AI model. Phase I will test whether the approach yields scientific understanding with fewer experiments than standard methods. That is the feasibility question the funded work will investigate, not a result announced with the award.