Nvidia will pull the plug on hardware sales to any customer caught developing dangerous artificial intelligence, according to chief executive Jensen Huang.
Speaking to Bloomberg in London, Huang framed the scenario as a hypothetical measure, conceding he would first seek talks with the offending buyer. Nvidia does not dictate how clients manage their internal operations, he stressed, and its processors remain general-purpose silicon.
Yet restricting access is not purely theoretical for the chipmaker. Nvidia established an approved-buyer registry spanning Singapore, Malaysia and Japan last year. That tighter vetting initiative eliminated more than half of its historical client base in that territory.
Selling Containment Alongside Silicon
Huang remains skeptical of the broader panic surrounding automated systems, dismissing widespread anxieties as standard growing pains. Some of the hand-wringing across the industry is self-inflicted, in his view. Instead of pausing work, he argues the correct response is accelerating safety monitoring, with software acting as the primary guardrail.
That belief mirrors a growing commercial initiative inside Nvidia. The company is assembling an internal safety group focused on vetting software agents before organizations turn them loose. The sales thesis is straightforward: mistrust delays agent adoption, and broader adoption directly fuels purchases of Nvidia accelerators.
Huang laid out similar thoughts at Dreamforce last month, maintaining that existing laws suffice and companies should charge ahead unless a specific deployment proves hazardous. That position directly clashes with Anthropic chief executive Dario Amodei, who has repeatedly called for a coordinated industry cooldown guided by shared global standards.
Other AI leaders voiced rival concerns across Europe on the same day. In Dublin, Anthropic's international lead warned that open-source models present inherent risks by their very nature. In parallel remarks, OpenAI's chief financial officer cautioned that adopting Chinese AI models introduces serious cybersecurity exposure.
Compute Pledges in the UK
Nvidia holds an overwhelming grip on the AI accelerator market, a monopoly-like footprint that leaves Huang fielding political questions about algorithmic control. His European tour carries hefty spending figures to soften those tensions.
Speaking in Paris on Tuesday, Huang previewed plans to announce $2 billion in European agreements within a week, with stops planned in London and Berlin. Part of that push includes a $250 million commitment toward agentic safety research in the UK.
That sum will arrive entirely as compute capacity rather than direct cash payments. University College London and Oxford are slated to run experiments examining automated agent safeguards and control systems once Nvidia finalizes provider selection and commercial contracts.
TBFiled by The Tech Business Desk
Big tech, startups, earnings, regulation and policy — the decisions that shape the products.
More from this desk →
Be the first to comment
Join the argument. No password, just your email or a passkey.