Details
- NVIDIA announced a five-year donation valued at $250 million, providing $50 million in annual compute access for UK agentic AI safety and security research.
- The initiative will support the broader UK research community and selected programs at Oxford-led BOLD, UCL-led SOFAIR and Imperial College London.
- Research will address secure agents, open-model security, threat mitigation, safety benchmarks, reproducibility and robotics safety, using NVIDIA OpenShell and its Sentry reference design.
- Compute will be delivered through UK-based NVIDIA Cloud Partners, with infrastructure hosted domestically; providers, architecture and commercial terms remain undecided.
- The commitment extends NVIDIA’s UK research engagement, including support for the Isambard-AI supercomputer at the University of Bristol, and is intended to strengthen locally hosted AI-safety research capacity.
Impact
The program directs substantial infrastructure toward testing increasingly autonomous systems and could expand the UK’s safety-research pipeline while improving access to reproducible experiments and assessment benchmarks. Over the next 12–24 months, it may encourage closer integration among universities, cloud providers and government, while making compute availability a more central factor in AI-safety funding and research strategy.