Jensen Huang was onstage at the All-In Summit in Los Angeles on Monday morning when Donald Trump called his cell phone. The Nvidia boss put the call on speaker for the room, listened to the president dismiss the case for slowing AI down as a hoax, and then told him nothing of the sort would be allowed to happen.
Key Takeaways
- Donald Trump phoned Jensen Huang mid-panel, in front of a room packed with investors in Los Angeles.
- The president called the AI alarm a hoax, and Jensen Huang picked up his exact wording to back him.
- The scene lands two days after Dario Amodei’s essay, which Elon Musk and Sam Altman both endorsed.
Have an AI Sum Up This Article
ChatGPTThe Trump Call Jensen Huang Put on Speaker
Monday morning in Los Angeles, Jensen Huang held the stage at the All-In Summit next to Chamath Palihapitiya, Jason Calacanis, David Friedberg and David Sacks. The phone rang mid-conversation, and Donald Trump was on the other end. The Nvidia chief, who argued back in May at the Milken Institute that AI is creating an enormous number of jobs rather than wiping them out, put the call on speaker for the whole room to hear.
Trump opened with a jab at the phone itself. The man who can design the most complex computer chip on earth, he said, cannot work out how to hit the speaker button. The room laughed. The tone dropped straight after, because the president had not called to talk about hardware.
He moved on to the people asking for a brake on AI development. They’re playing right into the hands of a lot of people who don’t want to see it happen, he told the room, before closing it out in two beats. Nobody is going to let that happen. The whole thing, in his words, is a hoax.
Jensen Huang didn’t reach for a softer line. He took the president’s phrasing word for word, opened with a flat “you’re right,” and confirmed that no slowdown would be allowed through. The audience applauded. There was no hedge about safety work happening in parallel, no nod to the argument on the other side.
What makes the moment unusual is the venue. The All-In stage is a podcast panel with no procedural standing, and the exchange ran live in front of the venture crowd that funds most of the companies now arguing about pacing. A conference call between a sitting president and a chip CEO would normally happen off the record, with a readout afterwards if anyone bothered.
Jensen Huang did not have to answer at all. Putting a caller on speaker is a choice, and answering a political question in front of that particular audience is another one. He made both inside a few seconds, with the president still on the line, in a week where every other frontier lab boss had already picked the opposite side. Once spoken into a live microphone, that answer stops being a personal view and becomes the public position of the company supplying the entire industry.
What the Slowdown Camp Is Actually Asking For
The exchange doesn’t land in a vacuum. Two days earlier, Dario Amodei published a 3,800-word piece on his personal site, an essay telling the whole industry to slow the rate at which it makes models more capable. The Anthropic CEO wrote that the time bought has to go into alignment work, and that progress would still feel fast from the outside.
The essay stops well short of a moratorium. Amodei laid out a three-step plan that starts with something Anthropic says it will do on its own, whatever competitors decide, and climbs toward international coordination from there. He framed the whole thing as compatible with keeping both the commercial advantage and the American lead intact, which is the argument aimed squarely at Washington.
The trigger he named is recursive self-improvement, the point at which AI systems start building the next generation of AI systems without much human involvement. Amodei’s position is that an extra year or two before models reach critical capability levels, spent on alignment, would cut the odds of something going badly wrong by a meaningful margin.
Elon Musk and Sam Altman backed the piece publicly within hours, a rare show of agreement between three direct rivals. Amodei had already, back in August, tied the public backlash against AI to a trust problem rather than a communication one, so the pacing essay reads as the operational follow-up to a position he had been building for a month.
Jensen Huang stayed out of that consensus, and Monday’s call is what made the split public. The gap matters because it isn’t really about safety philosophy. It’s about who absorbs the cost of pacing, and the answer to that question is not the same for a model lab and for the company that sells them silicon.
Nvidia sells the chips every one of those three labs trains on. A coordinated slowdown in training cadence hits the supplier’s order book first, long before it shows up in anyone’s model release calendar. That asymmetry also explains why three fierce competitors could afford to agree: a pacing pact they all sign leaves their relative positions untouched, and Jensen Huang is the one participant with no equivalent upside. Slower training runs mean fewer racks ordered, and no amount of alignment progress shows up on that line of the balance sheet.
More articles on Horizon
- Obama Wants Democrats to Get an AI Plan
- Claude’s Founder Amodei Wants to Slow AI Down
- Astra Agent Runs a Shop and Flies a Drone
Data Centers, the Blind Spot in Refusing to Slow Down
Seven in ten Americans say they would oppose a data center being built where they live, according to the Gallup reading on local opposition to AI data center construction. Just under half are strongly opposed, and the objection holds across every major demographic group the pollster broke out, party lines included.
More than half of the people who object point to the environment. Around one in five point to the cost of living, which is the number worth watching, because it moves with electricity bills rather than with opinions about technology. For scale, the same survey found fewer Americans opposed to a nuclear plant going up nearby than to a data center.
Trump still framed the file as a straight competition story, noting that data centers are great and that AI is bigger than the internet. Public opinion does not track that framing on the local question, and the local question is where permits actually get issued. National enthusiasm does not pour a foundation in a county that has already voted the project down.
For teams building on these models, the promise from Jensen Huang reads as a guarantee on cadence. Nothing gets throttled on the supplier side, so release cycles and capacity jumps stay pinned to the current rhythm, the one that forces an integration rewrite every few months. Product teams quietly hoping for a breather now have their answer, and it came from the person selling the compute. Planning a roadmap on the assumption that model capability plateaus for a year is the wrong bet to make this quarter.
On the competitive side, the position taken by Jensen Huang makes Anthropic’s move harder to land. A unilateral slowdown only works if it turns collective, and the primary infrastructure supplier has now said out loud that it won’t take part. Rival labs have something concrete to point at if they decide to keep their own cadence exactly where it is. Watch what Anthropic does with step one over the coming weeks, because a commitment to move first alone gets tested the moment a competitor ships faster.
That leaves the political ground, where the question doesn’t get settled between chief executives. Barack Obama pushed his own side this weekend to come up with a clear plan on the guardrails AI should face. Between a White House calling the alarm a hoax and an opposition still drafting its doctrine, nothing in Washington forces a pace on anybody, and Jensen Huang spent Monday morning saying what he intends to do with that opening.
Follow the story on Horizon.
