Ethics and safety·October 18, 2026, 20:33

Nvidia CEO: We would stop sales if AI demonstrably harms people

AI-generated from the sources listed, with automated source checks.

Jensen Huang says Nvidia would "of course" stop supplying a customer if it were certain the technology was harming people. He stressed, however, that the question was hypothetical and that a conversation would come first.

Nvidia CEO Jensen Huang would stop supplying chips to an AI lab if it became clear the technology was being used to harm people. He said so in an interview with Bloomberg's Emily Chang in London.

Huang was asked whether Nvidia would stop selling to an AI lab if he believed it was doing something too dangerous. He called the question hypothetical, but said Nvidia would of course do so.

Conversation first

Huang stressed that Nvidia would first have a conversation with the customer. Nvidia does not run other companies, and there are limits to how well Nvidia can judge how others use its technology. He called the technology broadly applicable and said that imposing Nvidia's values on other companies, institutions or countries is largely out of bounds.

But if it were certain that a use case was harming people, Nvidia would stop, he said. Nvidia has not announced criteria or a set procedure for when a customer could be cut off.

Safety and capability go hand in hand

Huang believes it is a mistake to treat AI capability and AI safety as two separate conversations. He wants to speed up the development of "open shell," technology for containing AI systems and monitoring them, including by having AI monitor AI.

He also believes the industry's narrative that AI is out of control is wrong. According to him, the young AI companies have a responsibility that they are growing into.

Sovereignty and money

Huang rejected the idea that AI built on American chips cannot be sovereign. According to him, sovereignty is about controlling your own data and experience, about no one being able to deny you service, and about having the ability to act. He calls the computer merely an instrument.

He also said that the price of a token, a small unit of text that AI models count in, is falling 30-40 times a year. He described Anthropic and OpenAI as the fastest-growing technology companies ever with real revenue and profit. He expects their IPOs to give them plenty of money to grow. These are Huang's own statements, and they have not been independently verified.

Emily Chang noted that more than 100 data center projects in the US have been blocked or delayed this year, and that several AI IPOs have been postponed or dropped. She also pointed out that Nvidia is a major investor in, and a customer of, both Anthropic and OpenAI.