Nvidia Boss Says AI Won't End the World and Should Be Built Fast
Nvidia CEO Jensen Huang said there is a “0% chance” that artificial intelligence will destroy humanity.
He spoke in an interview with CBS News reporter Jo Ling Kent, published as a longer video on Sept. 20, 2026. They talked directly about fears over what AI can do and predictions of disaster. CBS News
Huang rejected those predictions directly. He told CBS News that claims AI could destroy humanity in a few years are irresponsible and not based on science. In another report on the same interview, he was quoted as saying: "Scaring people is unnecessary. It is irresponsible." The Verge
He said the same about calls to slow down. Huang said calls from leaders like Anthropic's Dario Amodei and OpenAI's Sam Altman to slow AI work are "not grounded in science." He rejected those calls and said AI should be built "as fast as we can," according to CBS News reports published Sept. 18 and Sept. 20. CBS News
Huang also said no extra oversight is needed. He argued there is no need for new AI rules, laws, or guidelines.
On claims about human extinction specifically, Huang called those warnings "doomsday narratives." He told CBS News there is a "0% chance" that 2030 will be the end of the world. In the video interview, he said he completely disagrees that AI will destroy the world by the end of the decade. CBS News
The broader context here starts with the wording. Zero does not mean small risk. It means no risk. Much of the safety debate has been about tests of what AI can do, controls on how it is released, and safety fixes added after training. Huang is asking a different question: whether warnings to slow down meet the normal standard of proof before holding back a technology.
Looking at what this means in practice for people who build and use AI, it comes down to speed and who decides. A slowdown would mean more checks, tighter reviews, and possible outside limits on building and releasing systems. Full speed means developers keep responsibility under current laws, and learn mainly by testing with real users, a bit like learning how a car drives on real roads rather than only in the lab.
In my view, both paths have familiar tradeoffs. Moving fast finds problems sooner, but users feel those problems first. Moving slowly allows more careful testing, but it puts risk decisions in fewer hands and slows learning from everyday use.
Worth flagging is how Huang talks about fear. He treats public alarm not as normal confusion but as irresponsible behavior. There is a difference between clearly measuring a danger and spreading fear about it. Mixing the two makes it harder to judge risk well, and that leads to poorer product and policy choices.
In my view, confident forecasts about 2030 are not like lab measurements. They are guesses about surprising new behavior as systems get bigger and more widely used. Readers should treat them as guesses and focus on what can be checked: how systems act, how they might be misused, what safety controls exist, and what incidents actually happen.
In practical terms for teams planning hiring and building, what is concrete is Huang's preference. Build without artificial delay. Add no new AI-specific rules. Treat disaster forecasts as stories, not findings. That choice is worth weighing on its own, apart from whether zero percent turns out to be right.
