The AI designing catalysts aims to accelerate clean industry
Alexandre Duval, co-founder of Entalpic, applies AI to the discovery of materials and catalysts. Objective: to help chemistry and energy reduce their emissions and research costs.
Alexandre Duval, a researcher turned entrepreneur, puts artificial intelligence at the service of discovering new materials, a less visible field than chatbots but crucial for the climate. Co-founder and scientific director of Entalpic, he aims to accelerate the development of catalysts capable of making industry less energy- and carbon-intensive. This topic matters because these materials directly impact chemistry, fertilizers, and hydrogen, three sectors difficult to decarbonize. At this stage, the promise is not to replace laboratories but to prevent them from too often working blindly.
When AI leaves the screen for matter
Alexandre Duval's journey illustrates a branch of AI that moves away from consumer uses. After studying at the University of Warwick and then the MSc Artificial Intelligence at CentraleSupélec, he discovered at the Mila institute, founded by Yoshua Bengio, the uses of deep learning applied to chemistry and materials design, alongside researchers like David Rolnick. In this field, models do not produce text or images: they represent atomic structures, predict their properties, and propose new candidates to test.
This approach changes the research method. Rather than multiplying costly trials, AI is used to explore a vast chemical space, select the most promising paths, and bring together quantum simulations, scientific literature, and experimental validation in a single loop. The sought-after gain is simple: to move faster towards useful molecules without spending more than necessary.
Entalpic relies on physical models and targeted trials
In March 2024, Alexandre Duval co-founded Entalpic with Victor Schmidt and Mathieu Galtier, where he became Chief Science Officer. The startup develops a platform that combines predictive and generative models, atomic simulations, and laboratory validations. Its goal is to reduce the time needed to discover materials useful for energy, chemistry, and more broadly, hard-to-decarbonize industries.
In 2024, Duval also co-authored PhAST, for Physics-Aware, Scalable, and Task-Specific, a graph neural network method designed to accelerate catalyst research by integrating physical knowledge. A graph neural network, recall, processes relationships between atoms as a set of links rather than a simple list of data. This technical detail matters because it helps the model better align with chemical reality.
What this means for industry and climate
Entalpic's bet is as much industrial as environmental. Catalysts play a central role in many reactions, from fertilizer production to hydrogen, and their improvement could reduce energy consumption as well as associated emissions. According to the article, better identifying these materials could help significantly reduce CO2 emissions from the most emitting industrial processes, without waiting for a unique breakthrough from a providential major laboratory.
This approach also marks a strategic shift of AI, which moves from screens to matter. A model's performance is no longer measured only by its ability to respond quickly but by its capacity to bring forth a truly useful molecule. It is less spectacular than an assistant writing an email but much more concrete for industry. And often, this is where the future is decided.
Key points
- Alexandre Duval co-founded Entalpic in March 2024.
- PhAST was co-authored in 2024 with a physical approach.
- Catalysts target chemistry, fertilizers, and hydrogen.
- Entalpic combines models, simulations, and experimental validation.
- AI is used to explore a vast chemical space.
- Yoshua Bengio founded the Mila institute, [to be verified].