Every large molecule model in drug discovery is trained on the same scarce resource: experimental results. Potential Sciences argues the bottleneck is no longer the model but the data, and that physics can supply it.
The London startup emerged from stealth on November 5 with an eight-figure seed round, anchored by the UK government’s Sovereign AI fund. Other backers include Nauta Capital, Twin Path Ventures and Blue Wire Capital, plus angel checks from Stripe co-founder Patrick Collison and Recursion co-founder Chris Gibson.
Rebuilding quantum chemistry for GPUs
The company’s premise is that simulating molecular behaviour with quantum chemistry produces gold-standard accuracy but has always been too expensive to run at scale. Founder and CEO Hatem Helal says his team rewrote those calculations to run on modern GPUs, making the work roughly 100 times cheaper per result.
If it holds, generated data could train drug-design models on chemistry the experimental record never captured – rare disease targets nobody has funded a wet lab to explore.
What the government is buying
Sovereign AI is contributing equity and access to the AI Research Resource, including the Isambard-AI supercomputer in Bristol. AI Minister Kanishka Narayan framed the bet around Britain’s strength in fundamental science. Investment partner Josephine Kant, who first met Helal in Cambridge, called the ambition a frontier lab for chemistry.
A week earlier the Chancellor added 140M pounds to Sovereign AI. The same week, a Nature paper showed AI agents running entire biology experiments – a reminder that the data problem cuts both ways.
No product has been announced, and the efficiency claim rests on company benchmarks. What the round buys is compute, staff and time to prove the simulations match reality.
