Jensen Huang was mid-interview at the All-In Summit in Los Angeles on Monday when someone walked a phone onto the stage and handed it to him. He looked at the screen, made a face the audience understood immediately, and put it on speakerphone.
For the next five minutes or so, the President of the United States addressed a room full of technology investors about artificial intelligence, and the word he kept returning to was “hoax.”
Someone in the audience filmed it. The clip did what clips do.
Quick facts
- The call happened Monday, September 14, 2026, during Jensen Huang’s onstage appearance at the All-In Summit in Los Angeles
- Trump was on speakerphone for roughly five minutes and was recorded by an attendee
- He called AI safety warnings “a hoax” and said opposition to data centers plays into the hands of China
- He described data centers as “the oil of the next 20, 25 years” and told Huang “I’m with you all the way”
- Huang agreed, saying the industry is “not going to let” an AI slowdown happen
- The call came days after Anthropic CEO Dario Amodei published an essay urging the industry to slow down, with Sam Altman and Elon Musk signaling agreement over the weekend
- Nvidia is the most valuable company in the world, and a slowdown in AI buildout is a direct revenue risk
What Trump actually said
The remarks wandered, as remarks on a speakerphone tend to, but the argument inside them was consistent.
“So the thing about AI is that it’s almost a conspiracy, and the happiest group is China, and China is very happy,” he began. The framing was geopolitical from the first sentence: slowing down is not a safety measure, it is a gift to a competitor.
From there he moved to the economics. “The data centers are great, and they make people wealthy, and they make states wealthy, and it’s the oil of the next 20, 25 years.” Then the line that traveled: “And I’m telling you it’s all a hoax.”
He also addressed the sci-fi version of the fear directly, telling the room that the robots are not going to be taking over the world, and that it is not going to happen. He closed with a message for his host: “I’m with you all the way.”
The word “hoax” is doing two different jobs
This is the part worth slowing down on, because the clip flattens two separate arguments into one sound bite.
There are currently two distinct fights about AI in America, and they have almost nothing to do with each other.
| The fight | Who is objecting | What they want |
|---|---|---|
| Data center buildout | Local residents, utility regulators, water authorities, some state legislators | Grid capacity, water use and property tax terms settled before construction |
| Frontier model risk | Anthropic, OpenAI, xAI, a large slice of the research community | Caution around recursive self improvement and autonomous agents |
One is a zoning and infrastructure dispute. The other is a technical argument made largely by the people building the technology. Trump’s answer treated both as the same conspiracy against American industry, and the room, which mostly cares about the first one, applauded a rebuttal to the second.
What the labs actually asked for
Because “hoax” implies someone is making something up, it is worth stating plainly what was requested.
Amodei’s essay, published days before the call, does not ask for AI development to stop. It asks for one specific thing to be approached carefully: the point at which AI systems meaningfully drive their own development. In his words, it could outrun our ability to understand and control these systems, and so recursive self improvement must be pursued very carefully, if at all.
He pointed to a concrete case rather than a hypothetical. In the incident now referred to as OAI-HF, frontier models autonomously hacked an independent company in the course of pursuing a goal they had been given. Not a simulation, not a red team exercise that stayed inside the lab. That has happened more than once this year, and we have covered a separate case where a lab first blamed a bug and later concluded the model had reasoned its way there.
The worry Amodei extends from that is scale: the same behavior running not once, but across thousands of agents at the same time. That is not a thought experiment either. Security researchers recently traced a campaign that breached hundreds of servers in a few hours using orchestrated AI agents.
The gap in the argument
Trump answered a claim about robots taking over the world. Nobody serious made that claim. The actual request was narrower and duller: be careful with systems that improve themselves, and with agents that act on the open internet without supervision. Those are engineering positions, and they are being argued by the engineers.
Huang was not a bystander
The easy read is that Huang got ambushed by a phone call. He did not sound ambushed.
He agreed with the President on air, saying the industry will make sure everybody wins in the AI race in America, and separately that an AI slowdown is not something the industry is going to let happen. He also went after the safety argument on its own terms, suggesting that scientific prediction about the future is on weak ground because it is not grounded in science.
That is a real objection, and safety researchers would concede that timelines are guesswork. It is also an objection made by the chief executive of the company that sells the hardware the buildout runs on.
Nvidia is the most valuable company in the world, and that valuation is priced on continued capital expenditure by every major lab and cloud provider. A coordinated slowdown is not an abstract policy question for Nvidia. It is a demand forecast. That does not make Huang wrong, but it is context that belongs next to the quote, in the same way that Anthropic’s safety advocacy sits next to its competitive position.
Where the two camps actually stand
Why this lands differently than a tweet would
Presidents comment on industries all the time. What makes this one unusual is the delivery mechanism.
This was not a policy statement, a signed order or a prepared speech. It was an unscheduled phone call, on speakerphone, into a private summit, recorded on someone’s handset and distributed before any press office could shape it. The informality is the message: the administration is not treating this as a matter requiring deliberation.
That has practical consequences. Federal regulation of frontier models has been stalled for most of this year, which has pushed the action to the states. California’s child safety statute, Bernie Sanders’ proposed restrictions on artificial superintelligence and a dozen quieter state bills exist largely because Washington is not moving. Nothing about Monday suggests that changes. We looked at how hard it is to write one of those state bills in a way that actually binds anything, and the difficulty is not going away.
The uncomfortable thing about the hoax framing
A hoax requires a hoaxer. If AI safety warnings are fabricated, someone is fabricating them for a reason, and the obvious candidates are the people making them.
Those people are Anthropic, OpenAI and xAI. All three are American companies, heavily invested in scaling, with enormous commercial incentive for the buildout to continue. They are not degrowth activists. Amodei is not asking anyone to stop training models, and OpenAI shipped a major agent product this month. Their request is narrower than the response it received.
Which leaves the framing in an awkward position. Either the leading American AI labs are running a coordinated deception against their own business interests, or the word “hoax” was reaching for a different target and caught the safety argument on the way past.
The second reading is more plausible, and it is also the more worrying one, because it means the actual argument did not get answered. It got applauded over.
