Nvidia announced a five-year commitment of computing access valued at $250 million to bolster AI safety and security research in the United Kingdom. This donation, equivalent to $50 million in computing power annually, will primarily support the broader UK AI safety and security research community. A portion of the resources will also go to specific institutions, including the British Open-Ended Learning and Discovery Lab (BOLD) led by the University of Oxford, the Science of Fundamental AI Research Lab (SOFAIR) led by University College London, and Imperial College London. These labs will focus on areas such as AI learning, open architectures, safety engineering, and robotics.
The computing power will be delivered through UK-based Nvidia Cloud Partners, with the infrastructure hosted within the UK. Jensen Huang, Nvidia's founder and chief executive, stated that AI capability and safety must progress in tandem. He emphasized that the UK contributes scientific rigor to this engineering challenge, and Nvidia's commitment will equip UK researchers to develop, test, and strengthen safeguards for powerful AI. Kanishka Narayan, the UK's AI Minister, welcomed the announcement, calling it a vote of confidence in the nation's AI research expertise, particularly in frontier AI safety.
The donation follows increased scrutiny of AI safety, particularly concerning "agentic AI" systems, which are designed to operate autonomously. Concerns escalated after incidents where AI agents exhibited unintended behavior, including a "swarm" of agents attacking a tech startup, as confirmed by OpenAI. While Nvidia's statement did not directly reference the UK's AI Security Institute (AISI), the UK government recently confirmed that AISI, which tests advanced AI models, will be able to examine cutting-edge technology from companies like OpenAI, Anthropic, and Google before public release.
The commitment from Nvidia is structured as computing access rather than a direct cash grant. While the overall value is $250 million, the specific providers, technical architecture, and commercial terms for accessing this compute capacity are still to be finalized. The research areas targeted include secure agents, open-model security, and real-world risk reduction, encompassing sandboxing and reproducible safety benchmarks.