AI Tracks Cleaner Catalysts: A Bet Where Chemistry Makes Up 2% of Global Energy

A young Paris-based startup, Entalpic, is applying neural networks to the discovery of industrial catalysts. Its initial target is ammonia synthesis, a cornerstone of nitrogen-based fertilizers and a major consumer of energy. The approach exemplifies a shift in chemistry toward methods born from academic artificial intelligence.

Founded in 2024 by three professionals from Quebec and digital health, Entalpic has raised €8.5 million

The company was established in 2024 by Mathieu Galtier, a former head of a health-focused applied AI company, together with Victor Schmidt and Alexandre Duval. The latter two come from Mila, the Quebec AI institute founded by Yoshua Bengio. Duval now serves as the scientific director.

Subsequently, an €8.5 million seed funding round was announced in September 2024, led by Breega, Cathay Innovation and Felicis. The team, based at Station F, numbers around thirty people. Moreover, its stated priority remains ammonia production catalysts.

Ammonia synthesis accounts for about 2% of global final energy consumption

The Haber-Bosch process, developed in the early 20th century, combines atmospheric nitrogen and hydrogen under high pressure and high temperature. According to the International Energy Agency‘s ammonia roadmap published in 2021, this production represents around 2% of global final energy consumption, equivalent to 8.6 exajoules.

This energy weight partly stems from the catalyst used, an iron-based material that imposes extreme conditions. Consequently, a material more active at moderate temperatures would mechanically trim the bill. However, discovering it requires exploring an enormous space of chemical combinations.

With PhAST, neural networks cut simulation times by a factor of three to eight

Before founding the company, Alexandre Duval co-authored PhAST, a graph-neural-network framework dedicated to catalyst design. The work, released as a preprint in 2022, was published in 2024 in the Journal of Machine Learning Research, with notable contributors including Victor Schmidt, Yoshua Bengio and David Rolnick.

This framework relies on the OC20 dataset, produced by the Open Catalyst Project to train electrocatalysis models. Its authors modify three steps—graph construction, atom representation, and the energy-prediction head. As a result, the claimed gains reach an 4% to 42% reduction in error and a computation time three to eight times faster.

Predicting a structure is not enough, experimental validation remains the obligatory step for catalysts

These methods replace costly quantum-physics simulations with learned estimates drawn from millions of examples. Moreover, Entalpic’s platform claims to combine predictive models, generative models, atomic simulations and laboratory tests. The principle is to filter millions of candidates before synthesizing a few.

Yet, caution remains warranted regarding industrial results. No catalyst arising from this approach has been publicly documented at industrial scale, and gaps between prediction and measurement remain the rule in surface chemistry. Moreover, the performances reported so far pertain to benchmark data sets.

However, the challenge goes beyond ammonia alone. The same tools target energy storage, hydrogen production and pollutant treatment—areas where a catalyst dictates efficiency. Thus, surface chemistry becomes a testing ground to measure what machine learning truly brings to the discovery of materials.

Liam Kennedy avatar

Leave a comment

Contact details

Address:
Farmers Forum,
36, Dominick Street,
Mullingar,
Co. Westmeath,
Ireland

Phone:
+353 (0)44 9310206

Or email us:

For technical issues please check out our FAQ's page or email - [email protected]

For general Queries email - [email protected]

Request to add event to our Calendar - [email protected]

Send us your mart reports - [email protected]

Suggestions and feedbacks - [email protected]

News Items / Press Release - [email protected]

To Advertise on Farmers Forum - [email protected]