French AI startup Entalpic pivots from climate to chips, wins Nvidia backing

Orbital Industries also shelved its carbon-capture materials work

Entalpic, a French startup founded two years ago to use artificial intelligence for materials-science research aimed at cutting carbon-dioxide emissions, has shifted its focus to semiconductors and secured investment from Nvidia, the chip maker announced this week. Nvidia chief executive Jensen Huang unveiled the commitment during a Paris visit that included investments in other French startups. Nvidia’s announcement described Entalpic’s expertise as lying in “surfaces and interfaces for semiconductor manufacturing and next-generation electronics” — without referencing the company’s original climate mission.

The pivot illustrates the gap between AI’s theoretical promise for climate applications and the practical barriers to deploying novel materials in commercial settings, according to the Journal. Many experts, from Microsoft to McKinsey, have pointed to materials-science breakthroughs as a key way in which AI could help cut emissions, the Journal reported. Better-performing materials could improve the efficiency of batteries, solar panels and other equipment, reducing the cost of low-carbon technology, the Journal reported. Entalpic co-founder Mathieu Galtier said the company’s AI tools can screen a million potentially useful materials a day, compared with perhaps 100 tests a year in a conventional lab. The startup identified promising candidates for new battery cathodes, materials that could slow the degradation of perovskite solar cells, and catalysts that could improve ammonia production for fertilizers — a century-old industrial process. “There was no entry point,” Galtier said.

The barrier was customer adoption, not technical performance. China’s solar manufacturers are losing money, battery makers are largely focused on optimizing existing manufacturing lines, and ammonia producers are wary of trying untested chemistry in their plants, the Journal reported. Entalpic’s climate-focused customers operated on thin margins and were unwilling to take risks on unproven materials.

The AI boom itself made the chip industry a more attractive market. Chip makers have both the resources and the incentives to experiment as physical limits to chip performance push them to innovate, the Journal reported. “The semiconductor industry is more used to iterating, and they have bigger margins,” Galtier said. Entalpic is now applying atomic-scale engineering to improve the efficiency of semiconductors. Nvidia did not disclose the terms of the investment.

Entalpic is not the only AI-for-climate startup to change course. London-based Orbital Industries shelved a plan earlier this year to develop sorbents — materials used in carbon-capture devices — after market conditions shifted, according to the Journal. Orbital had envisioned its carbon-capture systems sitting inside data centers, helping tech companies reduce their carbon footprint.

Biden-era climate policies have been cut, and tech companies are now focused on getting data centers built and running efficiently, the Journal reported. “We didn’t stop because the AI wasn’t working,” Orbital chief executive Jonathan Godwin, a former researcher at Google DeepMind, told the Journal. “We stopped because the market conditions changed.”

In May, Orbital raised $50 million to develop a new product: cooling systems that use more efficient, greener alternatives to the chemical cooling agents used in some data centers. Despite being jaded by the carbon-capture effort, Godwin said Orbital would dust off that research if demand emerges. Galtier said Entalpic also plans to return to climate-tech applications once its technology has been proven.

The barriers facing novel materials are not purely financial. Godwin pointed to the blades in a jet turbine — each one a perfect crystal made of an advanced metal alloy that can withstand 3,000 degrees Fahrenheit — as an example of materials whose performance cannot be shortcut. “The only way that you know something lasts for 10 years with predictable performance is by running it for 10 years,” Godwin said.

Those structural barriers offer some industries a measure of safety from AI disruption, but they are a blow to startups — including those seeking to help the climate — that have bet on AI-driven materials science as a route to decarbonization, the Journal reported.