Potential Sciences, a frontier lab for chemistry that trains AI on data generated from the laws of physics, has emerged from stealth with eight figures in capital and compute. The Sovereign AI Fund and Twin Path Ventures are key investors in the round, alongside Blue Wire Capital, Lightfield Equity, Nauta Capital, Odyssey, Portfolio Ventures and Rule30. Angel investors include Patrick Collison, co-founder of Stripe, and Chris Gibson, co-founder and former CEO of Recursion.
Most AI models in chemistry learn from the same limited record of past experiments, or from simulations too crude to trust. CCSD(T), the gold-standard method of quantum chemistry, gets steeply more expensive as molecules grow, and the field's main codes were written for 1980s supercomputers with a handful of processors, so they cannot use the thousands of cores in a modern GPU. Potential Sciences has rebuilt quantum chemistry for GPUs from the equations up. Its engine runs CCSD(T) calculations at 100 times lower cost than traditional implementations, generating accurate data for millions of molecules where the field could previously afford thousands. That data trains atomistic foundation models that learn how atoms interact, so drug discovery teams can see which molecules are worth making, how strongly they will bind and whether they will hit proteins they shouldn't, before ordering a single assay.
The initial focus is small-molecule drug discovery, working with AI-first labs on lead optimisation, reaction prediction and covalent drug design. The longer-term goal is a computational oracle for chemistry across drugs, catalysts, enzymes and materials. The Sovereign AI Fund backs Potential Sciences as both investor and compute partner, giving access to national supercomputing through the AI Research Resource, including the Isambard-AI supercomputer in Bristol.
Hatem Helal founded Potential Sciences and leads a team of physicists, chemists and mathematicians who trained at Caltech, Cambridge and CERN, and built AI for science at Graphcore, Recursion and Dayhoff Labs. Between them they created QM1B, a billion-calculation quantum chemistry dataset, and won two tracks of the OGB Large-Scale Challenge at NeurIPS 2022. Based in Cambridge with researchers in Bristol and Tübingen, Potential Sciences is advised by Andrew Fitzgibbon and James Hawkins. The funding will grow the research team, scale the dataset and train the first atomistic foundation models.
AI for drug discovery has learned almost everything the existing experimental record can teach it. The next leap needs new data, and the laws of physics can supply it, as long as your software can use modern hardware. We rebuilt quantum chemistry for GPUs so that gold-standard accuracy is affordable at the scale AI needs. When your code is 100 times more efficient, a seed round buys what used to take a dedicated data centre.
Potential Sciences is a startup that’s grasping an opportunity Britain is uniquely well-placed to capitalise on: using our deep expertise in data, fundamental science and computing to potentially unleash the next great wave of medical discovery. Solving chemistry’s ‘data bottleneck’ could allow researchers to explore new treatments for diseases that are currently deemed too rare or expensive to focus on. This is AI at its best – pushing the boundaries of innovation, ultimately to improve people’s lives.
I first met Hatem in Cambridge, and he immediately struck me as a formidable founder. He combines technical excellence with low ego and unapologetic ambition. He knows the science in extraordinary detail, but he’s also deliberate about bringing in people who are exceptional in areas different from his own. Potential Sciences have the ambition to become a true frontier lab for chemistry, with the potential to fundamentally change how the field works.
Using physics to generate chemistry data only works at scale if it is both fast and cheap enough to run. Hatem and his team have the rare combination of quantum chemistry, GPU engineering and large-scale machine learning needed to make that practical. We think the next step-change in AI drug discovery comes from better data, not just better architectures, and Potential Sciences is building the engine to generate it. That’s why we’re backing them.
