Sam Altman, Dario Amodei, Demis Hassabis, and Elon Musk agreed over the weekend to slow down AI development. Jensen Huang called the president to say it’s not happening.

This isn’t a minor disagreement. It’s a fundamental split over the future of the industry, and the battle lines tell you everything about who’s winning and who’s worried.

Anthropic CEO Dario Amodei kicked it off Saturday with an essay called “We Must Pace the Frontier.” His proposal: embed third-party auditors in AI labs, regulate domestic development, and work toward a global slowdown agreement. OpenAI’s Altman, Google DeepMind’s Hassabis, and Musk all signed on, at least in principle.

On Monday, Nvidia CEO Jensen Huang took a call from President Trump onstage at the All-In Summit. He put it on speakerphone. “We’re not going to let that happen,” Huang told Trump, referring to the slowdown proposal.

The Cartel Question

The critics called it immediately. If you’re OpenAI or Anthropic, sitting on the current generation of frontier models, a slowdown locks in your advantage. You get time to monetize what you’ve built while competitors can’t catch up. Call it safety if you want. It works like a cartel.

The timing matters. OpenAI just spent $300 million to acquire Glass Imaging, a smartphone camera company founded by former Apple engineers who built Portrait Mode. That’s not the move of a company planning to slow down. That’s the move of a company building toward hardware, toward consumer products, toward locking users into an ecosystem.

Altman and Amodei aren’t idiots. They know how this looks. But they’re also not wrong that AI safety is a real concern. The question is whether you can separate the legitimate safety argument from the convenient competitive one.

The Nvidia Position

Huang’s opposition makes perfect sense. Nvidia sells the shovels in this gold rush. A slowdown means fewer chips sold, lower revenue, worse margins. The company’s entire valuation depends on AI labs buying more GPUs, not fewer.

But there’s a deeper issue. Nvidia doesn’t build the models. It doesn’t face the regulatory scrutiny that OpenAI and Anthropic do. It doesn’t have to answer questions about what happens when AI gets more capable. It just has to make faster chips.

That’s a comfortable position. It’s also why Huang can afford to oppose the slowdown while Altman can’t.

What Happens Next

The White House hasn’t shown much interest in slowing anything down. Trump’s administration has been clear: America competes, China competes, and regulation is something you do after you win.

That puts the pressure back on the companies. If Amodei really believes in pacing the frontier, Anthropic can do it unilaterally. Slow down releases. Add more testing. Build in more safeguards. But that only works if OpenAI does the same thing. And if Google does. And if the dozens of other labs do.

Which brings us back to the cartel question. The only way a slowdown works is if everyone agrees. And the only way everyone agrees is if it’s enforced. And the only way it’s enforced is through regulation.

Amodei’s proposal includes third-party auditors embedded in labs, which is a nonstarter for most companies. It includes domestic regulation, which Congress hasn’t shown any ability to pass. And it includes a global agreement, which requires China to cooperate.

None of that is happening soon.

The Real Story

This isn’t really about safety. It’s not really about competition either. It’s about who controls the pace of development, and whether that control comes from inside the industry or outside it.

Right now, the labs control it. They decide when to release new models. They decide what safeguards to build in. They decide how fast to push.

Amodei’s proposal shifts that control to regulators and third-party auditors. That’s a huge change. It’s also probably necessary if you believe AI poses existential risks.

But it requires the people making money from AI to agree to let someone else set the speed limit. And that’s a harder sell than any safety argument.

Huang’s call with Trump wasn’t about technical capabilities or market dynamics. It was about making sure the speed limit doesn’t happen. At least not yet. At least not while there’s still money to be made.

The split in Big Tech isn’t about whether AI is dangerous. Everyone agrees it could be. The split is about whether you slow down now or wait until you’ve already won.