Jensen Huang on cutting off a dangerous AI lab: "of course"
In London with Emily Chang, the Nvidia chief argues that AI safety and AI capability are one conversation, not two.
Why this story? The video opened at a pace of 58,552 views a day, 12 times a normal first day on this channel (4,923).
Nvidia CEO Huang Says He Would Not Sell Chips to Irresponsible AI Customers
Things we can learn from the video
Press to jump
- 0:07Huang defines sovereign AI as controlling your own data, culture and knowledge, like owning water or power infrastructure.
- 1:31Asked about cracks in the AI boom, he says AI's capabilities will never go backward.
- 3:06On OpenAI and Anthropic going public: generative AI became profitable this past year, and their revenues are real.
- 4:16He says the computing cost of a token is falling 30 to 40 times per year.
- 4:43Against calls to slow down, he wants guardrails and monitoring accelerated alongside capability.
- 5:53Would Nvidia stop selling to a dangerous lab? Yes, if it knew for certain an application caused harm.
Yes, with a condition. Asked by Emily Chang whether Nvidia would stop selling chips to a lab building AI too dangerously, Jensen Huang says the company would cut off sales if it knew for certain that an application was causing harm. "It's a hypothetical, but of course," he adds. It is a promise about what Nvidia would do, not news that it has dropped a customer.
That exchange closes a short interview recorded in London, on the same day Bloomberg Tech's daily show reported Huang's plans for another $2 billion in European deals. It has passed 30,000 views in its first nine hours, far beyond the channel's usual full first day for a regular video. The comments moved even faster. A typical first day here brings a handful, and this one drew dozens within hours. Several viewers doubt that a chipmaker would ever turn money away, and some wonder whether banned buyers would simply find chips another way. Others like the idea of chips as a real consequence for unsafe work, and a few just want the full interview.
Chang starts with a puzzle: how can a country build "sovereign AI" on American chips? Huang's answer is that sovereignty means controlling your own data, culture and institutional knowledge, and that "nobody could deny you service." He compares AI to water and electricity.
She then asks about cracks in the boom: blocked data center projects, delayed listings, and what happens when OpenAI and Anthropic go public. Huang calls them "the two fastest growing technology companies in history," says their revenues are real, and points to costs falling fast: "The cost of a token is reducing by 30, 40 times per year."
To those who want development slowed, he offers a different remedy. Guardrails and monitoring should speed up alongside the technology. "I think it's in fact a mistake to separate AI capability and AI safety as if it's two conversations," he says, which is what makes the final question land: it asks him where that one conversation would end a sale.
Published on Trari
Where next?
What viewers say
Where's the full interview?
@stevethemacguySee it on YouTubeIf a company is building unsafe AI, cutting off its access to the chips powering it is real accountability. AI safety can’t just be about talking responsibly..it has to come with consequences when companies cross the line.
@YolandaCole-v2p7cSee it on YouTubeExcellent
@BilichaGhebremuseSee it on YouTube
Public comments from the video's page on YouTube
- Channel
- Bloomberg Tech
- Subscribers
- 742,000 ▲ 1,000 this week
- Channel views, last 7 days
- 784,524
- Channel views, pace per day
- 60,592 (on a normal day: 94,432)
- Video published
- October 14, 2026
- Video views so far
- 31,308 (less than a day online)
- Video views, pace per day
- 58,552
The numbers
A short interview that drew far more viewers, and far more comments, on its first day than this channel's videos usually do.
💡 Test your knowledge
Which piece of clothing has become Jensen Huang's signature on stage?
Huang is famous for wearing a black leather jacket at keynotes and public appearances.



