AI-Directed Laboratory Discovers Iridium-Free Palladium Catalyst for Green Hydrogen Production

On September 24, 2026, a research team led by Ken J. Jenewein submitted a preprint detailing the results of a three-month autonomous laboratory campaign that synthesized and tested 2,942 distinct metal oxide catalysts. The system, operated by Cambridge-based startup Lila Sciences, identified six palladium-based material families capable of driving the acidic oxygen evolution reaction—the chemical bottleneck in green hydrogen production. The lead candidate, an indium-manganese-palladium oxide (InMnPdOx), maintained its performance for over 1,000 hours in sulfuric acid.[1][2][3][4]

Producing hydrogen from water requires two simultaneous half-reactions: the hydrogen evolution reaction at the cathode and the oxygen evolution reaction (OER) at the anode. The anode reaction is notoriously difficult because it operates at high electrical potentials in strongly acidic environments that dissolve almost all known metals.[1][4]

In today's commercial proton-exchange-membrane water electrolyzers, the only catalyst that reliably survives these conditions while remaining active is iridium oxide, sometimes alloyed with ruthenium. Both are among the rarest elements on Earth. "Catalyzing acidic oxygen evolution... relies almost entirely on iridium or ruthenium, drawn from concentrated supply chains that constrain gigawatt-scale deployment," the authors note in their paper. Nearly all global iridium is produced as a byproduct of platinum mining, yielding only a few tonnes per year.[1][3]

To find an alternative, Lila Sciences built what it calls an AI Science Factory (AISF). Starting in late 2024, the team designed a closed-loop platform that integrates combinatorial sputter synthesis, high-throughput electrochemical screening, and machine-learning models. The system operates with greater than 90 percent automation, requiring human intervention primarily to transfer samples between testing stations.[3][4][5]

During the screening of 53 material systems and 26 elements, the AI model began proposing compositions based on palladium. "It was not an obvious pick for any OER scientist," Jenewein stated, noting that while palladium is a workhorse metal in other industries, it had long been considered a dead end for acidic OER due to its tendency to corrode.[1][4]

The model did not select pure palladium, but rather modified compositions containing small additions of other elements. The researchers found that these additive elements promote a specific nanostructure that maintains catalytic activity while stabilizing the palladium against the harsh acid.[1][3]

In long-term validation testing in 1 molar sulfuric acid at 10 milliamps per square centimeter, pure palladium oxide exceeded a 0.5-volt overpotential after approximately 200 hours, indicating degradation. A nickel-tantalum-palladium mixture (NiTaPdOx) survived for roughly 470 hours. The lead candidate, InMnPdOx, demonstrated a dramatic increase in stability, retaining an overpotential below 0.5 volts for the full 1,000-hour test.[1][2][3]

The speed of the discovery highlights a shift in materials science methodology. Lila Sciences reported that its pipeline screened catalysts 17 times faster than a standard laboratory, processing up to 240 samples per week. "If we had done this without a lab to confirm the finding, we would have chalked it up to a hallucination," said Rafael Gómez-Bombarelli, a senior scientist on the project. "But we did the experiment. It is the real thing."[4]

While palladium is itself a precious metal, it is significantly more abundant than iridium and trades in a much larger, more established global market. Substituting palladium for iridium could ease the immediate supply constraints facing the electrolyzer industry.[3][4][5]

The researchers caution that the current result is a laboratory-scale materials discovery, not a commercial product. Any new OER catalyst must eventually be manufactured at an industrial scale and proven under the intense, fluctuating operating conditions of commercial hydrogen plants. Long-term durability testing of the InMnPdOx compound is ongoing.[1][4]