For more than half a century, scientists have struggled with one of the most persistent bottlenecks in green hydrogen production: a heavy reliance on some of the rarest and most expensive metals on Earth. Now, an AI-directed laboratory at US startup Lila Sciences has uncovered a ground-breaking alternative in just three months.
By proposing, synthesizing, and screening 2,942 distinct catalyst compositions, the company’s autonomous AI platform identified six high-performing material families. The discovery challenges long-held consensus in electrochemistry and offers a vivid demonstration of how artificial intelligence can accelerate the clean energy transition.
The Scarce Resource Stifling Scale
Producing green hydrogen via water electrolysis relies on driven chemical reactions. The most demanding half-reaction occurs at the anode under strongly acidic conditions—a highly corrosive environment that rapidly degrades most metals.
Today, commercial electrolyzers rely almost entirely on iridium and ruthenium to survive these conditions. Iridium is primarily mined as a byproduct of platinum extraction, with global production limited to just a few tonnes per year. This severe supply constraint has long threatened to cap the global expansion of green hydrogen infrastructure.
Challenging Decades of Scientific Consensus
When Lila’s AI model began suggesting compositions based on palladium, the startup’s own research leads initially assumed the system had made an error. In electrochemistry, pure palladium had long been written off as a dead end for acidic oxygen evolution reactions due to poor stability.
However, the AI’s prediction proved correct. By adding precise ratios of complementary elements to palladium, the platform created a modified material that resisted acid corrosion while delivering catalytic performance comparable to industry-standard precious metals.
“My former group at Caltech explored catalysts for this reaction for 12 years, and if a student had suggested trying this combination of elements, I would have advised them to try a more promising direction,” said John Gregoire, Chief Autonomous Science Officer at Lila Sciences. “This combination defies all traditional wisdom.”
During physical lab testing, lead candidates successfully demonstrated stable performance across more than 1,000 hours of continuous acid exposure. Because palladium is far more abundant and widely traded than iridium, the discovery could substantially lower raw-material costs for electrolyzer manufacturers.
17 Times Faster Than Manual Research
The breakthrough was made possible by Lila’s “AI Science Factory”—a closed-loop platform where an AI reasoning model formulates chemical hypotheses, directs physical robotic equipment to synthesize candidates, analyzes testing data, and iterates autonomously.
The automated pipeline screened potential catalysts 17 times faster than a conventional laboratory, reducing required active human research time by over 90 percent per sample.
The team emphasizes that the novel catalysts are a materials-science proof of concept rather than a finished commercial product. The next phase of development will focus on testing the materials inside industrial-scale electrolyzer cells and expanding full robotic automation across the laboratory workflow.