The Pulse
Lila Screens 2,942 Catalysts and Finds Palladium-Based Leads
Lila Sciences says its AI-directed lab screened 2,942 oxide catalysts across 53 systems and 26 elements for acidic oxygen evolution. The work identified six palladium-based material families, including a candidate that a research preprint s
AI.info Team ·
Results on a computer screen pointed Lila Sciences researchers toward a catalyst they thought would fail. The material contained palladium, a metal long discounted for the acidic oxygen-evolution reaction used in water electrolysis. Lila says its AI-directed laboratory screened 2,942 oxide catalysts across 53 material systems and 26 elements before identifying six palladium-based material families with promising combinations of activity and stability.
The company published its account on September 25, 2026, a day after the research team posted a preprint describing the work. The result is a laboratory discovery, not a commercial hydrogen-production technology: durability and performance at industrial scale still need testing.
Palladium returned to the candidate list
Making hydrogen by splitting water requires two reactions. At the anode, oxygen evolution takes place in a strongly acidic environment that can corrode many materials. Commercial systems rely heavily on iridium oxide, while ruthenium is another option; both metals are scarce, and the supply of iridium is especially limited.
Lila’s researchers had built a closed-loop workflow that combined AI-guided material selection with synthesis, characterization and electrochemical testing. Results from each round informed the next set of proposed compositions. The company says its lab could synthesize 96 catalysts in parallel, then test them for both activity and stability.
During the screening, the system proposed palladium compositions modified with small amounts of other elements. “It was not an obvious pick for any OER scientist,” Ken Jenewein says in Lila’s account. Rather than treating the prediction as a result on its own, the researchers made and tested the materials.
Six material families, but durability varies
The research preprint reports that the platform evaluated 2,942 catalysts and identified palladium-based oxides including InMnPdOx and NiTaPdOx. Its long-term tests distinguish the candidates: InMnPdOx kept its overpotential below 0.5 volts for 1,000 hours, while NiTaPdOx crossed that threshold at about 470 hours. The preprint says PdOx crossed it at about 200 hours.
Those measurements matter because a catalyst must combine useful reaction performance with resistance to degradation. Lila says its program found six palladium-based material families on or near the activity-stability frontier. In its announcement, Rafael Gómez-Bombarelli described the unexpected result: “If we had done this without a lab to confirm the finding, we would have chalked it up to a hallucination. But we did the experiment. It is the real thing.”
The system still needs human oversight
Lila reports that the screening pipeline ran 17 times faster than a standard laboratory and saved more than 90% of human time per sample. Those are company-reported comparisons. The preprint characterizes the platform as more than 90% automated, while Lila’s account says people still transferred samples between instruments and checked proposed materials for safety.
The work also does not establish that palladium catalysts can replace iridium or ruthenium in commercial electrolyzers. Lila says it is continuing long-term durability tests in a form closer to industrial use. The preprint’s strongest reported result remains specific: InMnPdOx stayed below the stated overpotential threshold for 1,000 hours in acidic testing, not proof of operation at commercial scale.