Entalpic

🤖 “AI” isn’t one thing and that’s exactly why it’s exciting 🔬 I just came back from Adopt AI in Paris 🇫🇷. Great energy, great people — but one thing struck me: we often say “AI” as if we all mean the same thing. In practice, it can mean wildly different things. Take chemistry alone: AI can power process engineering, supply-chain optimization, automated experiment design, data extraction, or molecular discovery. Each of these is its own universe. One of the most insightful talks came from JEAN YVES DELANNOY (Arkema), who clearly distinguished between the role of large language models and the role of AI models built for quantitative chemistry (the kind we develop at Entalpic). Maybe it’s time we stop using “AI” without context. Instead of broad “AI for health” or “AI for industry” labels at conferences, imagine dedicated tracks like “How AI designs new molecules” bringing together chemistry, pharma, batteries, defense, and more. That’s where cross-pollination becomes meaningful. Curious to hear how other sectors are structuring this, what have you seen work well? 🤔

Our team is heading to NeurIPS 2025 next week, joining the global community driving progress in AI for science and materials ⚛️ We will be present— and proud to be co-organizing — the AI4Mat workshop, where our researchers will share recent work on AI-driven materials discovery. We are excited that four of our papers were accepted this year for NeurIPS, including our paper on LeMat-GenBench, our new benchmark suite for generative models in materials science, which was built part of our open-source initiative LeMaterial. Join us for the spotlight talk on this paper as part of the workshop designed to bring together AI researchers and materials-science experts to advance AI-driven materials discovery! Entalpic will be represented on-site by Alexandre Duval, Chief Science Officer, with Senior ML Researcher Tristan Deleu, and ML Researcher Siddharth Betala We look forward to connecting with researchers, labs, and industry teams exploring how AI can accelerate scientific discovery 📍 San Diego | December 6, 2025 🔗 Workshop details & LeMat-GenBench preprint — links in the comments 👇 #NeurIPS2025 #AI4Science #MaterialsDiscovery #DeepTech #MachineLearning #Entalpic #AI4Mat

Today, Nikita Hall, Director of Lab & Material Innovation at Entalpic, will participate in the roundtable discussion, "ALD and Energy", at RAFALD, exploring the role of AI in advancing ALD innovation, from precursors to layering. Moderated by Nicolas Blasco (Air Liquide) and Lionel Santinacci (CINaM), the session will cover the critical role of Atomic Layer Deposition (ALD) in energy applications, including batteries and catalysis, with insights from leading experts, including Claire Villevielle (LEPMI laboratory Grenoble, LabEx MateriAlps - Université Grenoble Alpes). 📍 November 20th, 11:00 AM | Grenoble, France We’re excited for the exchange of ideas on how AI can drive ALD innovation and help accelerate energy transition technologies #ALD #AI4Science #MaterialsDiscovery #SustainableInnovation #DeepTech #Entalpic #EnergyInnovation

We won’t build the next generation of materials with compute alone. 🇪🇺 If Europe wants to lead in AI for materials, we need more than AI factories. We need real factories: high-throughput labs to generate experimental data, validate discoveries, and accelerate science in areas that matter: catalysis, batteries, alloys, semiconductors, and more. These labs should be designed by a consortium of experts, shared across startups and academics, plugged into AI factories, and built in the open (to let others replicate them across the continent). Similar to Max Welling vision of AI-powered “mini-CERNs”. That’s the kind of infrastructure we need: focused, ambitious, and built to empower breakthrough discoveries in material science. Europe is already behind on compute (despite the recent #RAISE announcement), let's not fall behind on experiments too ! These are slow, CAPEX-heavy builds, and we will wish we had started sooner. ⏰ This is not just about helping startups and academic research. It’s about sovereignty, competitiveness, and shaping the next industrial era.

🔋At Batteries Event 2025 in Lyon, Nikita Hall, Director of Lab & Material Innovation, presented how Entalpic applies AI-driven discovery pipelines to accelerate cathode materials innovation, from LFP to Na-ion and NFPP-based chemistries. Here are the signals shaping Europe’s battery trajectory: 1️⃣ Europe is still building momentum: Industrial players such as ACC - Automotive Cells Company, Verkor, and PowerCo continue to invest in gigafactories and reinforce Europe’s battery ecosystem. A clear signal that reindustrialisation is active, and advancing. 2️⃣ Cost and scale remain defining pressures: The EV market drives high volume and low margins. With a fraction of the funding and a compressed learning curve, Europe must compete with more mature global battery ecosystems. 3️⃣ Innovation is the lever we can pull: The energy in Lyon showed that innovation hasn’t slowed. The debate continues around performance (NMC), cost-efficiency (LFP), and emerging alternatives (Sodium Solid-State). Solid-state remains a major ambition, even as intrinsic challenges persist. A critical question for Europe, and for us at Entalpic: How do we accelerate our learning curve? AI-driven discovery offers a concrete pathway. By linking computational chemistry, high-throughput experimentation, and data-driven optimisation, Europe can shorten development cycles, derisk decisions, and scale promising chemistries faster. We’d be keen to hear how others across the battery ecosystem are approaching this challenge, what levers do you see for accelerating materials and process development in Europe? #AI4Science #BatteryMaterials #MaterialsDiscovery #DeepTech #CleanEnergy #EuropeanInnovation #EntalpicAI

The myth of the 10× developer Yes, some people code faster, think deeper, or ship cleaner. But individual brilliance doesn’t compound. Teamwork does. In my experience, the ego-filled, self-proclaimed “10×” often slow the system down. A truly multi-disciplinary team thrives on consistency, solidarity, recognition, and alignment. In AI for materials science ⚛️ , this truth is even sharper. No one can hold the full map of chemistry, computation, physics, AI, software, industry, markets, and sales in their head. I don’t believe discoveries will come from individuals anymore. This is the era of collective intelligence 👯 . Marketing will keep glorifying individual genius, the so called "legends" from big Tech, but real impact comes from teams that think as one organism.  🩷 the Entalpic team for proving this every day.

AI for Science vs Science for AI During my PhD, I worked with Yoshua Bengio on AI for Science 🧪, i.e. training machine learning models to discover better materials to improve reaction efficiency, and help mitigate climate change. Now, his focus has shifted towards Science for AI 📐. Last week at AIS24, he made a few key points that stuck with me. 1) AI progress follows its own Moore’s law. It already surpasses humans on most benchmarks, and there is no clear scientific reason for this trend to stop soon. Undesired behaviour is possible. 2) To achieve “good” goals, AI systems can take “bad” actions. And, since they learn to imitate humans, they may inherit self-preservation tendencies or create unintended sub-goals. 3) We want these large AI models to be aligned with our core values. We would thus benefit from a scientific AI that can evaluate whether an answer is harmful, dishonest or uncertain — and use that signal during training or inference to avoid reproducing harmful human behaviour. 🇪🇺 Yoshua also called for Europe to build its own frontier models, to avoid relying only on those from the US. To be sovereign, to regulate AI well, to understand its energy footprint and societal impact, we must be able to develop and measure it ourselves. My take → I am not particularly worried about current foundation AI models “taking over,” nor do I favour heavy-handed regulation. But, as with any technology that affects billions of people (cars, airplanes, food, medicines, nuclear reactors, etc.), I believe AI models should pass safety checks and audits to ensure they align with our values and behave responsibly. 🌟

We’re joining the 2025 French Tech 2030 cohort alongside 80 deeptech companies driving France’s technological and industrial sovereignty. A recognition of our mission to accelerate innovation in chemistry and materials through AI. Thank you La French Tech for the trust! #FrenchTech2030 #AI4Science #DeepTech #Entalpic