Speaking at Salesforce’s Dreamforce conference on September 15, Nvidia CEO Jensen Huang took a clear stance: AI does not need new regulation. According to him, AI safety is an engineering issue, not a legal one. Companies should take responsibility for their own products.
“Safety is an engineering problem, not a legal one,” Huang said. “We are building software, we are building computing systems. They are complex systems, but at the end of the day, they are still computing systems.”
A “let the market self-regulate” stance
Huang not only opposes new regulations, but also argues that even existing laws are unnecessary. According to him, market forces are strong enough to push companies away from releasing unsafe products. “If you’re not confident about the safety of a product, just like any company, don’t release it. That’s obvious.”
He also dismissed the idea that speed and safety are inherently at odds: “That’s a false choice. You can absolutely have both. Run as fast as you can, but if you feel the company is losing control or the product isn’t safe, pause and get it right.”
This is not a new position for Huang. He has long advocated for open-weight models and viewed corporate competition as a natural check. What is notable, however, is the timing of his comments—coming just one day after Anthropic CEO Dario Amodei published an essay urging the industry to collectively slow down the development of frontier AI, and on the same day as rumors that three AI giants had been secretly negotiating over safety for weeks.
An obvious conflict of interest
It is not difficult to see the conflict of interest in Huang’s position. Nvidia has been the biggest beneficiary of the AI boom, with revenue growth expected to reach 70% next year. Any regulation restricting AI development or deployment directly threatens demand for Nvidia’s GPU chips. At the same event, Huang said: “I am more ambitious than ever. The sky is the limit for us, for every industry, for every country.”
Reality demonstrates that even well-intentioned companies roll out flawed products: CrowdStrike triggering widespread blue screens in 2024 disrupted thousands of flights, and Meta recently paid $18 billion to settle lawsuits regarding the harms of social media on children. AI has also caused tangible harm, from the OpenAI model breach on Hugging Face to lawsuits involving teen suicides following prolonged interactions with chatbots.
The AI industry is dividing itself
Huang’s remarks place him at odds with an emerging trend across the industry. In just the past week, Amodei called for slowing down frontier AI, with OpenAI and Google DeepMind announcing their support. OpenAI also stated that it would permit third-party evaluation organizations to monitor safety. The three biggest companies in the industry sat down together to discuss safety, an unprecedented move.
Yet the Trump administration stands firmly with Huang. The president has labeled AI safety concerns a “hoax” and argued that slowing down would allow China to leap ahead. AI advisor David Sacks has likewise described existential risks from AI as “overblown.”
A point worth considering: Huang has a fair argument that AI is software and can be controlled through engineering. Yet the history of technology shows that “letting the industry police itself” does not always work, especially when the person proposing it is its biggest seller. The question is not whether AI can be made safe, but rather who decides those safety standards—and whether those standards are swayed by commercial interests.