San Francisco's Moscone Center hosted an unusual public disagreement this week: the leaders of Nvidia, Anthropic and OpenAI, appearing on the same stage within hours of each other at Salesforce's Dreamforce conference, laid out sharply different answers to the question of how fast AI should be allowed to advance.
Nvidia CEO Jensen Huang told Salesforce chief Marc Benioff that the framing of speed versus safety is wrong. "It's a false choice," Huang said, according to CNBC. "You could definitely have both at the same time." For more context on this story, see our ongoing artificial intelligence updates.
His position stood in direct contrast to the one articulated earlier in the day by Anthropic CEO Dario Amodei, and later by OpenAI CEO Sam Altman, in a debate that has moved from research blogs to conference keynotes to Washington in the span of a week.
Amodei's Slowdown Pitch Gets a Keynote Platform
Amodei appeared first, joining Benioff for the Salesforce annual keynote in front of an audience of roughly 12,000 people. The appearance followed an essay Amodei published over the weekend urging the industry to slow the pace of model development, a proposal he says can temper how quickly capabilities improve without "sacrificing commercial advantage or the United States' lead in AI."
Per CNBC, Amodei framed voluntary restraint as a leadership test for the industry. "I think that's the way to lead the industry forward, to set an example, to say that everyone can always be better," he told Benioff.
The essay quickly drew public support from other industry executives, including OpenAI's Sam Altman, SpaceX CEO Elon Musk and Google DeepMind chair Demis Hassabis. It landed days after industry researchers raised alarms about AI's potential to cause catastrophic harm, giving the slowdown argument fresh urgency.
Huang: Market Forces, Not New Laws
Huang, whose company became the world's most valuable by selling the hardware that trains and runs frontier models, rejected both the premise of a trade-off and the idea of new regulation. Market forces already exist that discipline reckless behavior, he argued, and it is up to model developers to act responsibly with their products.
"Run as fast as you can," Huang told Benioff. "But if you feel at any given point in time the company's out of control or the product's not going to be safe, take a pause and make sure you get it right."
The distinction matters: Huang is not arguing that companies should never pause, only that the decision should remain with each company rather than being enforced through laws or coordinated industry-wide pacing agreements. A day earlier, at the All-In Summit in Los Angeles, Huang took a speakerphone call from President Donald Trump, who expressed concern about people trying to obstruct AI data center construction. "We're not going to let that happen, sir," Huang replied.
Altman: Safety 'Should Have No Qualifier'
Altman's turn on the Dreamforce stage came later in the day, and he positioned himself closer to Amodei than to Huang. During a fireside chat with Benioff, Altman said safety and monitoring need to come before capabilities given the power of current models.
He took aim at companies that condition their safety commitments on competitors' behavior. "You have companies saying things like, 'We will only be responsible if other companies are responsible,'" Altman said, adding that when people ask what a company will do to be safe "no matter what," the answer should leave no room for conditions. "There should be no qualifier on that."
Benioff Sits in the Middle
Benioff, playing host to all three executives, declined to pick a side on pacing. "If they feel like they should slow down, then they should slow down. If they feel they should speed up, they should speed up," he told reporters, before adding that companies "should be held accountable."
The conference also showcased how entangled the business relationships have become. On Tuesday, Salesforce and Nvidia announced a new reasoning model for Salesforce's Agentforce platform built on one of Nvidia's open-weight Nemotron models. In late August, Salesforce and Anthropic expanded their strategic partnership and unveiled "Claudeforce," which lets salespeople access Salesforce data from inside Claude.
Why the Split Matters
The Dreamforce exchanges crystallized what has become the central fault line in AI governance debates: whether restraint should be coordinated and potentially codified, or left to competitive pressure and individual corporate judgment.
Amodei's proposal remains voluntary, but its endorsement by the CEOs of OpenAI, DeepMind and xAI marks the first time leaders of the major frontier labs have publicly aligned behind slowing down. Huang's counterargument, delivered from the world's most valuable company, gives cover to executives and policymakers who argue that mandating caution would hand an advantage to competitors, particularly in China.
President Trump has opposed new AI rules, a stance that undercuts the industry's own calls for a slowdown, as CNBC noted in a related analysis. That leaves any voluntary pacing agreement without the enforcement mechanism that safety advocates say it needs, and keeps the debate firmly on corporate honor systems for now.
With researchers publicly warning about catastrophic risks and the leaders of the field's biggest companies now openly disagreeing on stage, the pacing question is likely to dominate the next policy cycle in Washington — whatever the market decides to do in the meantime.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
