Nvidia commits $250 million in computing access to UK AI safety research

Nvidia is committing computing access it values at $250 million over five years to UK research on making AI agents safer and more secure. Announced October 14, 2026, the donation is intended to provide $50 million annually, with most going to the wider research community rather than only the three institutions named in the announcement.

The commitment is for computing resources, not a $250 million cash grant. Nvidia says it will fund access through its UK-based cloud partners, with the infrastructure hosted in the country. That makes access to computing the means of support; researchers are not being promised unrestricted funds to spend on other needs.

Broad access, alongside three named research programs

A portion of the donation will support the British Open-Ended Learning and Discovery Lab, known as BOLD; the Science of Fundamental AI Research Lab, or SOFAIR; and Imperial College London. Their stated research roles span how AI learns, how models are built and how systems act in the physical world.

• BOLD, led by the University of Oxford, is developing approaches that help AI systems learn, adapt and operate alongside people. Its founding director, Jakob Foerster, emphasizes building governance and human-AI improvement into that work from the start.
• SOFAIR, led by UCL, is exploring new open-source AI architectures and more accessible ways to develop and deploy AI. Its part of the program therefore reaches beyond testing safeguards to the design of AI systems themselves.
• Imperial’s supported work will include safety engineering and robotics. Alessandra Russo, director of its Frontier AI Research Lab, described the challenge as making physical AI trustworthy for the communities and industries that will depend on it.

We are experiencing a critical moment where AI safety as an afterthought is no longer appropriate for the open-ended, evolving world.

Jakob Foerster, founding director of BOLD

From secure agents to repeatable safety tests

Nvidia identifies three research priorities: secure agents, open-model security and ways to reduce real-world threats. The planned work includes secure sandboxes—isolated environments for running systems—along with reproducible research and benchmarks for assessing safety. These are intended research activities, not safety results delivered by the donation.

Some research will use Nvidia’s OpenShell secure runtime software and its Sentry reference design. The company says findings are intended to inform safeguards across the wider computing ecosystem. The program thus combines work using Nvidia’s own safety tools with an ambition for the results to be useful beyond those tools.

UK hosting is specified; the delivery setup is not

Nvidia has specified a domestic route for supplying the computing: UK-based Nvidia Cloud Partners operating infrastructure hosted in the UK. The company presents that arrangement as support for both research capability and the country’s computing ecosystem, rather than research access supplied through infrastructure located elsewhere.

But provider selection, technical architecture and commercial terms are still to be agreed. The announcement therefore establishes the value, duration and research purpose of Nvidia’s commitment without settling those implementation details. The annual computing pledge is defined; the providers and arrangements that will turn it into usable research access remain unfinished.