Jensen Huang Rebuts AI 'Doomsday Theory': Safety Is an Engineering Problem, Not an Excuse for Regulatory Exemption, Urges Industry to Allocate 80% of Resources to Safety Assessment

Beating AI News Flash: NVIDIA CEO Jensen Huang, in a in-depth interview with The New York Times on the 23rd local time, delivered a strong rebuttal to recent密集 warnings about "AI going out of control" from frontier labs such as OpenAI and Anthropic. He believes AI is essentially just software and algorithms assigned an objective function, with no "will," and that safety and alignment can be fully addressed through engineering means such as isolation and sandboxing. "If you believe you can no longer control the situation, then you should not release the product until you regain control," he said bluntly, calling the logic absurd for large tech companies to claim their technology is too powerful to control while seeking regulatory or antitrust exemptions, and made clear that what he opposes is "seeking liability exemptions for oneself on the grounds that the technology cannot be controlled," not regulation itself. Huang used the chip industry as an analogy: NVIDIA invests 80% of its R&D resources in verification, while AI labs currently do the opposite, with about 80% going to capability improvement and 20% to safety evaluation, and this ratio needs to be reversed.

On computing power and industry judgments, Huang made several shocking assertions: future compute volume may increase by "a billion times" compared with the past, because hundreds of billions of AI agents will autonomously call large models; NVIDIA's computing power systems are evolving into an "asset class" similar to commercial aircraft, with building a 1-gigawatt AI factory requiring $50 billion, but it can be leased for $40 billion to $50 billion per year.

Huang also disclosed that the open-source versus closed-source landscape has reversed: at the beginning of the year, closed-source models accounted for about 70% of token usage and open-source about 20%, but now it has flipped to about 70% open-source and about 30% closed-source, and he emphasized that open source is the safest cybersecurity strategy. On employment, he used radiologists as an example to refute the job-destruction theory, pointing out that in the past six months, $500 billion in venture capital has flowed into AI-native companies, which will create a large number of new companies. On the energy issue, he also proposed a highly disruptive reverse-thinking idea: AI is precisely the "once-in-a-century catalyst" driving global sustainable energy investment, and in the short term it may be necessary to increase fossil fuel use, but this is like the short-term pain of surgery: "To save you, you must first be made to feel pain."