TL;DR

The University of Oxford has signed a non-binding memorandum with the chipmaker AMD covering the BOLD lab — British Open-ended Learning and Discovery — under which AMD would supply as much as £5m in computing resources. The lab is led from Oxford and draws researchers from UCL and Imperial College London. Nothing in it is binding, and the language throughout is conditional.

What the lab is for

BOLD’s remit is deliberately awkward for the current orthodoxy. The named research directions include training methods that dispense with backpropagation altogether, discovery systems built from several agents working together, and physical AI that adapts to its environment — in other words, questions the frontier labs are largely not asking because the existing recipe still scales.

Jakob Foerster, an associate professor in Oxford’s engineering science department, framed it as pursuing a more open and efficient route through academic and industrial collaboration, with wider access to computing as part of the point. For AMD, Thomas Zacharia, who leads its public sector and partnership work, described the aim as helping researchers experiment faster and contributing to Britain’s standing in AI research.

How the money arrives

The £5m is not a cheque. It would come as a mix of infrastructure AMD runs itself, capacity bought from cloud providers, public sector arrangements and donated hardware, released in phases — exploratory work first, then concentration on whatever early experiments suggest is worth pursuing. Alongside it sits software work on AMD’s open ROCm stack, with the lab committing to publish its benchmarks, evaluation tools and code rather than hold them.

Looking forward

For UK readers the significant word is “intends”. Two things are true at once: this genuinely adds research compute to three of the country’s better AI departments, and it costs AMD very little against the up to £2bn it has already said it will put into UK research and infrastructure, or its Cambridge work on the Zenith machine and on a Sovereign AI Innovation Lab there. Vendor-supplied capacity is how most British academic AI research now reaches hardware, given that publicly funded machines arrive in £36m increments. That solves an immediate problem and creates a slower one about who sets the research agenda.