Imperial and Cambridge win £700K AIchemy grant for AI-driven materials discovery

TL;DR:

• The AIchemy Frontier Fund has awarded £700,000 to Professor Aron Walsh (Imperial College London, Department of Materials) and Dr Shijing Sun (University of Cambridge) for a project titled Alignment of Generative AI for Materials Discovery via Experimental Feedback.
• The work tackles a recognised weakness in current AI materials discovery: most generative models are evaluated only in simulation, so many predicted materials are never tested in a real laboratory. The team will wire experimental results back into AI predictions to close that loop.
• The focus is optoelectronic materials, using automated synthesis and measurement to generate experimental data at scale; all outputs will be shared openly under FAIR principles to support reproducibility across the UK research community.

The grant is small in cash terms compared with frontier-model training budgets, but it sits at the practical intersection where UK AI-for-science capability has its clearest comparative advantage: world-class fundamental research, strong experimental physical-sciences infrastructure, and a national policy push to commercialise AI-for-discovery work. The “alignment via experimental feedback” framing is also notable — it imports the lab-grounded learning loop into generative materials research at exactly the moment international labs are racing on the same problem.

Why “simulation only” is the real limitation

Most published AI materials results show impressive performance on simulated benchmarks. Far fewer make the leap to “this molecule or crystal can actually be synthesised and tested at scale”. The Walsh-Sun project addresses that gap structurally rather than through point fixes, by incorporating processing conditions and atomic-scale disorder into the models themselves and feeding real experimental outcomes back into training. That puts UK research in step with the global “self-driving lab” movement — autonomous experimentation paired with closed-loop AI design — which Google DeepMind’s GNoME, Toronto’s self-driving-lab projects and US national-lab consortia are all chasing.

For UK SMEs and spinouts in materials, energy and electronics — Materials Discovery Lab, Polaron, Catagonia, Mission Zero Technologies, the cluster around the Henry Royce Institute — the Imperial-Cambridge work matters less as a single project and more as an institutional pattern. The Royal Academy of Engineering’s data-centric engineering blueprint published the same day (see [our coverage]) makes the case that UK engineering courses must embed AI fluency as a core skill; the AIchemy grants are the parallel research-side investment that turns those graduates’ work into commercialisable IP.

Looking forward

Expect the optoelectronic-materials focus to feed into UK strengths in compound semiconductors, photovoltaics and quantum-materials research; the FAIR-principles open-data commitment also matters because it lowers the entry bar for smaller research groups and SMEs to use the resulting datasets. The harder follow-on test is whether AIchemy-funded work will get to industrial partners — the Faraday Battery Institution, the Catapult network, and pharma partners — fast enough to influence UK industrial capacity rather than only academic papers.