Potential Sciences, a London-based startup founded by former Graphcore researcher Hatem Helal, has stepped out of stealth with an eight-figure seed round backed by the UK government's Sovereign AI fund and Stripe chief executive Patrick Collison. The company, based in Cambridge, is rebuilding quantum chemistry software to run on modern GPUs, aiming to generate the high-quality molecular data that AI drug-discovery models need.
The round was announced on November 5. Twin Path Ventures is a key investor alongside Sovereign AI, with Blue Wire Capital, Lightfield Equity, Nauta Capital, Odyssey, Portfolio Ventures and Rule30 also taking part, and Collison joining as an angel investor. The company did not disclose the exact size of the round, describing it only as eight figures in capital and compute.
Physics data, not bigger models
The pitch is that drug-discovery AI is held back less by model design than by training data. Existing chemical datasets are sparse, noisy and expensive to extend, because quantum chemistry — the mathematics that predicts how molecules behave — was written for CPUs and runs slowly. Potential Sciences says rebuilding it for GPUs cuts the cost of those calculations by roughly 100 times.
Helal previously built a billion-example quantum chemistry dataset at Graphcore. The company argues that far cheaper physics calculations let it generate millions of highly accurate molecular calculations for AI models to learn from, instead of relying on slow lab experiments or rough approximations.
State cash plus state compute
Sovereign AI's backing goes beyond money. The fund is also giving the startup access to national supercomputing through the UK's AI Research Resource, with up to one million GPU hours available per portfolio company. The fund launched in April 2026 as a 500 million pound vehicle that the technology secretary said would operate like a venture capital firm "with the muscle of the state behind it".
The investment fits a wider push. The UK chancellor's October budget added 60 million pounds to the Sovereign AI fund and pledged up to 200 million pounds in AI model credits from OpenAI and Anthropic. The fund has also backed Oxile Computing, Ineffable Intelligence and Callosum.
Why it matters
If the 100-times cost claim holds up, physics-generated data could become a new fuel for AI drug discovery — a field where several well-funded labs have stumbled on data quality rather than computing power. The real test will be whether the generated data produces better molecules in the lab, not just cheaper calculations on a screen.
Potential Sciences is advised by Andrew Fitzgibbon, formerly of Graphcore and Microsoft Research, and James Hawkins, co-founder and co-chief executive of PostHog. Alongside Cambridge, it has researchers in Bristol and Tubingen.
