TL;DR
On Thursday 5 November Sovereign AI, the state’s venture fund, invested in Potential Sciences, a UK company building AI for drug and chemical design, in a funding round worth eight figures. The package also gives the company time on the UK’s national supercomputers. The startup’s pitch is that a model grounded in physics, not past lab results, can make the underlying calculations about 100 times cheaper.
What the company does
Most AI used to design molecules learns from records of earlier experiments, and that data is slow and expensive to produce. Potential Sciences instead trains on the basic laws of physics. Founder and chief executive Hatem Helal says the team rewrote quantum chemistry methods to run on GPUs, so that high-accuracy results become affordable at the volume AI needs. “When your code is 100 times more efficient, a seed round buys what used to take a dedicated datacenter,” he said.
The government frames the target as diseases such as cancer and Parkinson’s, and as rarer conditions that are currently too costly to pursue. That is the case for the approach, not evidence it works: the release reports no results from the model.
Who else is in
Sovereign AI invested alongside venture firms including Twin Path Ventures, Nauta Capital, Odyssey and Lightfield Equity. Angels include Stripe co-founder Patrick Collison and Chris Gibson, who co-founded Recursion. As is usual for the fund, the size of its own stake was not disclosed; its typical cheques are around £1m to £10m.
The compute is drawn from the national AI Research Resource, whose machines include Isambard-AI in Bristol. AI minister Kanishka Narayan and science minister Chris McDonald both welcomed the deal, as did Lord Vallance, chair of the PM’s AI taskforce.
Why it matters
The deal follows last week’s Budget, which added £140 million for Sovereign AI and widened its remit to physical and industrial AI. It is another name on a portfolio we have tracked since the summer. In our view, pairing cash with supercomputer time is the fund’s most distinctive lever. For a startup whose product is cheaper computation, access to Isambard-AI may matter as much as the money.
Looking Forward
The test is whether physics-derived data produces candidate molecules that hold up in the lab. Until the company publishes results, UK life sciences firms should treat the 100-fold figure as the founder’s claim.