Potential Sciences, a Cambridge, England, startup that uses physics simulations to train AI models for chemistry, announced a seed round on 5 November 2026. It did not disclose the amount, saying only that it had raised eight figures in capital and compute. Tech.eu reported the raise as being in the tens of millions of dollars.
The company named the UK government's Sovereign AI fund and Twin Path Ventures as its key investors. Blue Wire Capital, Lightfield Equity, Nauta Capital, Odyssey, Portfolio Ventures and rule30 also took part, along with individual backers including Stripe chief executive Patrick Collison and Recursion co-founder Chris Gibson. The valuation was not disclosed.
Besides investing, the Sovereign AI fund is supplying computing power, giving the company access to national supercomputers through the UK's AI Research Resource.
What Potential Sciences does
AI models for chemistry and drug discovery learn from data, and the company argues that good data has run short. Lab measurements are scarce, expensive and skewed towards experiments that worked, while computer simulations have tended to be either accurate or fast, but not both.
Potential Sciences says it can generate new training data from the laws of physics at the accuracy of a lab measurement. Its software runs CCSD(T), a highly accurate but costly quantum chemistry method, natively on GPUs, at what it says is 100 times lower cost than traditional implementations. That, it says, makes it possible to label millions of molecules at that accuracy rather than thousands.
It is using that data to build a large chemistry dataset and to train foundation models that predict how atoms interact. The company said it had beaten the published state of the art on reaction barrier prediction and protein-ligand binding affinity, with full results to come later.
Starting with drug discovery
The first market is small-molecule drugs. The company points out that most candidates entering clinical trials fail, often because they bind weakly to their target or bind to the wrong one. It wants chemists to be able to predict how a small change to a molecule affects binding, reactivity and side effects before it is made.
It plans to work with AI-focused drug discovery labs on lead optimisation, reaction prediction and covalent drug design, and later to move into catalysts, enzymes, materials and guidance for automated labs.
"We generate new training data from the laws of physics, at the accuracy of a laboratory measurement," the founding team wrote in the announcement.
The team and its backers
Hatem Helal, formerly an executive at chip designer Graphcore, founded the company in late 2025. Its researchers, trained at Cambridge, Caltech and CERN and previously at Graphcore, Recursion and Dayhoff Labs, work in Cambridge, Bristol and Tübingen. Advisers include Andrew Fitzgibbon and PostHog co-founder James Hawkins.
Joséphine Kant, an investment partner at Sovereign AI, told Tech.eu she first met Helal in Cambridge and called him a formidable founder. She said the company wants to become a frontier lab for chemistry. According to Tech.eu, the fund has also backed chip startup Olix and AI companies Ineffable Intelligence and Callosum.
Tech.eu described the company as London-based; the company's own announcement says it was founded in Cambridge.
Company website: potentialsciences.com