How warnings over AI became a market-moving battle between tech bosses, investors and politicians
By
On the afternoon of Monday 14 September, at a technology conference called the All-In Summit, Nvidia’s chief executive Jensen Huang took a phone call and put it on speaker for the room. It was Donald Trump. For five minutes the president told an audience of investors and engineers that fears about artificial intelligence were a ‘sick conspiracy’, that data centres were ‘the oil of the next twenty, twenty-five years’, and that nobody was going to be allowed to slow them down. Huang nodded along. It was, by any measure, an unusual use of presidential time.
Three days earlier, the men Trump was implicitly rebuking had said something rather different. Dario Amodei, chief executive of Anthropic, published a long essay warning that AI companies had lost their grip on the pace of their own progress, and that within six to twelve months, autonomous agents could be capable of seizing control of large parts of the internet, at a cost running into hundreds of billions of dollars. Sam Altman of OpenAI agreed within hours. So, more surprisingly, did Elon Musk, who days before had dismissed similar warnings from a departing Anthropic researcher as a ‘psy op’.
Markets did what markets do when three of the most powerful men in technology say the technology might be dangerous: they sold. The Philadelphia Semiconductor Index fell more than five per cent in a single session. Nvidia lost three per cent, Micron over five, SoftBank more than ten. European chip stocks were dragged down with them. By the time Trump reached for the phone, the question about the safety of frontier AI had already become a live pricing event for the most concentrated bet in global markets.
Whose alarm is it?
It is worth asking who benefits from the world believing, even for a week, that artificial intelligence is close to running out of control. Amodei’s essay landed days before Anthropic’s expected public listing, at a moment when the company was reportedly courting Nvidia as an anchor investor. A well-timed warning about moving too fast turns caution into a credential rather than a hedge against a market that many analysts already suspect is overheated.
Altman’s decision to shelve OpenAI’s own listing for 2026, citing safety concerns, came in the same week that markets were newly nervous about AI spending built on debt and on circular deals between the same small circle of companies. An IPO postponed for moral reasons reads very differently to investors than one postponed because the timing has simply turned bad.
Enjoying this article? Check out our related reads…
None of this requires bad faith on Amodei’s or Altman’s part. It is, however, worth noting that their sincerity and commercial interest happened to point in exactly the same direction, and that markets have no reliable way of telling the two apart. A warning that would once have circulated as an internal safety memo is now published on social media and treated as market information.
A president as market maker
Trump’s intervention deserves closer attention than it received. Having called AI safety concerns a ‘hoax’ on social media that morning, he then spent five minutes on a competitor’s stage promising that nothing would be allowed to interrupt the sector’s growth, with the head of the world’s most valuable chipmaker standing beside him as proof. This was not regulatory restraint, nor even really an argument; it was a head of state using the authority of his office to defend a sector’s valuation in real time, nine days before he is due to host Xi Jinping in Washington with AI on the formal agenda. Whatever the merits of the underlying safety case, treating a market wobble as a national humiliation to be corrected by a presidential phone call is an intervention of its own.
Legislating around the edges
The legislative response on both sides of the Atlantic mostly reveals how little machinery exists to act on any of this, whichever way the evidence points. Senator John Kennedy’s proposed ‘kill switch’ bill explicitly bars the federal government from holding any such switch, has no cosponsors, and is being introduced by unanimous consent.
Conversely, in London, a Liberal Democrat amendment to the Cyber Security and Resilience Bill and a separate AI Security Bill from the Labour backbenches would, if passed, make Britain the first G7 country to legislate directly against the development of superintelligent AI, but both still need government backing that has not been offered. China’s state press dismissed Amodei’s essay as a Cold War tactic dressed up as ethics, aimed at slowing a rival’s progress rather than protecting anyone. Even the vocabulary of existential risk, it turns out, is already fluent in the language of great-power competition.
What connects the essay, the selloff, and the phone call from Los Angeles is not a hidden conspiracy. It is something more ordinary, and in its way more unsettling: a market so dependent on the fortunes of a handful of companies that it will move sharply on any statement from anyone with a plausible claim to inside knowledge, regardless of what is actually motivating them to speak. Nobody needed to manipulate a market this credulous. It had already arranged itself to be moved.
