Nvidia’s Huang Urges AI Firms to Hold Back Systems That Aren’t Ready

Jensen Huang offered AI companies a simple release standard at a summit convened by King Charles III in Scotland: test systems properly, and keep them from the public if they are not ready. The Nvidia chief executive paired that warning with a notably bullish forecast, saying he expects Nvidia to sell twice as many chips next year as it sells this year.

A release gate, not a call to stop building

Huang said AI companies themselves are responsible for developing technology safely and testing it properly. His proposed line is not a general brake on AI development. It is a judgment about whether a particular system is ready to be released: “If it’s not ready, just hold it back.”

That puts the decisive responsibility with the companies making the systems. Huang also argued that governments need not build entirely new AI-specific rulebooks, saying existing laws in sectors including healthcare and transportation could be adapted as AI is introduced there.

The sales forecast behind the caution

The context for Huang’s release advice was an expectation of continued infrastructure spending. He said Nvidia expects to sell twice as many chips next year, attributing the projected increase to countries investing in localized AI infrastructure and cloud data centers. It is a management forecast, rather than a reported sales result.

The two messages sit together uneasily. Huang’s standard asks companies to delay systems that fail their readiness test, while his sales outlook assumes that the physical buildout supporting AI will keep expanding. He presented safety as a condition for deployment, not as a reason to ease the push for more computing capacity.

Power demand is part of the same argument

Huang acknowledged the heavy energy demands of AI data centers, but argued that demand could prompt investment in new power sources. He named fusion, fission, wind and hydro as areas that could benefit from that investment. That is an expectation about how the buildout could reshape energy spending, not a guarantee that it will do so.

Huang’s remarks leave a practical question at the center of the industry’s safety debate: when a system is not ready, the company building it is the party he says should hold it back. The effectiveness of that approach depends on whether corporate release decisions can command trust while AI infrastructure and demand continue to grow.