Huang says he would halt chips to a customer developing unsafe AI

In an interview with Bloomberg’s Emily Chang in London, Huang said he would halt chip supplies to a customer developing unsafe AI software.[1] He described cutting off an irresponsible AI developer as a hypothetical prospect, while saying he would do it.[1] Huang also said there would be a conversation before supplies were stopped.[1] The distinction matters for organizations assessing Huang’s remarks about supplier responsibility: he expressed willingness to act in that hypothetical case, and said there would be a conversation before supplies stopped.[1]

Huang said Nvidia does not run its customers’ companies and places limits on how much it judges customers’ use of its technology.[1] He also said he did not believe the current level of concern about AI was warranted.[1] Huang described the prospect of cutting off an irresponsible AI developer as hypothetical.[1]

Nvidia announced a five-year compute donation valued at $250 million for U.K. agentic safety and security research, providing $50 million in compute access annually.[3] The company said most of the donated compute would support the wider U.K. AI safety and security research community, with a portion allocated to BOLD, SOFAIR and Imperial College London.[3] Nvidia described BOLD as developing approaches for AI systems to learn, adapt and operate alongside people, and SOFAIR as exploring open-source AI architectures and more accessible development and deployment.[3] Imperial’s work would include safety engineering and robotics, Nvidia said.[3]

Nvidia said the initiative’s infrastructure would be hosted in the U.K., while provider selection, technical architecture and commercial terms remained to be agreed.[3] The company said the research programs would focus on secure agents, open-model security and real-world threat mitigations.[3] It also said the commitment builds on collaboration with the U.K. research community, including work on the University of Bristol’s Isambard-AI supercomputer.[3] For prospective research partners, the announced amount and annual compute access establish the scale of the commitment; the unresolved arrangements mean the announcement does not yet provide the operational details needed to plan around specific providers or terms.[3]