News by Nicholas Cabel

AI desk

Nvidia's Huang rejected extinction warnings, not an AI bubble

AIAI summaryNicholas Cabel

Jensen Huang told CBS News there is no chance AI ends the world by 2030, answering Anthropic's Dario Amodei, OpenAI's Sam Altman and a former Anthropic researcher - not the Bank of England, which has warned that AI valuations look stretched.

A coiled microphone and a folded card on the lip of an empty, dimly lit stage
AI-generated illustration

Jensen Huang told CBS News there is no chance AI ends the world by 2030, answering Anthropic's Dario Amodei, OpenAI's Sam Altman and a former Anthropic researcher - not the Bank of England, which has warned that AI valuations look stretched.

Key points

  • Nvidia's chief executive sat down with CBS News last Friday and dismissed the idea that AI could wipe out humanity by the end of the decade. "There is 0% chance that's going to be the end of the world," he said, according to the CBS interview by Jo Ling Kent, Emily Pandise and Aimee Picchi, published Sept. 20. He called such warnings ungrounded in science and said frightening people is both unnecessary and irresponsible. CBS puts Nvidia's market value at $5.3 trillion, the largest of any company.
  • The argument he is answering is about safety, not share prices. CBS traces the exchange to social media posts by Jacob Coxon, a former Anthropic researcher, who argued that the people building the technology genuinely think it could kill everyone before the decade is out. CBS also names Anthropic chief executive Dario Amodei and OpenAI chief executive Sam Altman among the industry figures who have called for slowing development.
  • Amodei's case is set out in an essay published in September on his personal site, not Anthropic's. It asks the industry to slow the rate at which model capabilities improve so that alignment and interpretability work can catch up, and proposes three steps: outside evaluators embedded with ongoing, employee-level access, which Anthropic says it is committing to unilaterally; common safety standards agreed among frontier labs in democratic countries, helped along by government mediation or an antitrust waiver; and, beyond that, democratic governments attempting to coordinate with authoritarian ones as far as that proves possible, which the essay itself treats as the step hardest to verify. The risks it names are loss of control, AI-enabled cyberattacks and bioterrorism, economic disruption, and swarms of misaligned AI agents acting without authorisation.
  • That essay says nothing about an AI bubble, valuations, stock prices or investment levels. This is the detail that gets lost in aggregation: the fight Huang picked is over whether the technology is dangerous, not over whether the money comes back.
  • He had made the same argument three days earlier. At Salesforce's Dreamforce conference, which opened in San Francisco on Sept. 15 with about 43,000 people expected, Huang said the industry needs no new laws or regulations and that market forces are sufficient, framing safety as an engineering problem rather than a legal one, TechCrunch reported from the event. Amodei appeared at the same conference, on a panel, restating his case for caution.
  • The bubble warnings come from somewhere else entirely. The Bank of England's Financial Stability Report of July 7 says equity gains have been driven partly by a narrow set of AI-related companies, concentrating global indices, and that valuations look more stretched on some measures. It adds that a reassessment could be amplified by correlated momentum positions and by a significant rise in hedge fund leverage in equity markets. Huang did not address valuations, index concentration or leverage in either the CBS interview or the Dreamforce session.
  • Nvidia's own numbers are the part of this anyone can check. Its results for the quarter ended July 26, released Aug. 26, show $96.2 billion of revenue, up 106% from a year earlier, with $89.0 billion of that from data centres, up 117%. Gross margin was 75.0% on both a GAAP and non-GAAP basis, and the company guided to $108.0 billion for the current quarter while assuming no data centre compute revenue from China at all.
  • The commitments behind those sales are in the filing, not the press release. Nvidia's Form 10-Q reports supply and capacity commitments of $279 billion as of July 26, up from $119 billion three months earlier. It also discloses guarantees entered in August, capped at $105 billion, backing leases for roughly 4.25 gigawatts of IT load at SB Energy's PORTS Technology Campus in Pike County, Ohio, with a further $3.5 billion of maximum exposure from guarantees for AI cloud partners.
  • On the Aug. 26 earnings call the word bubble is never said once across sixteen pages. The closest anyone comes is Nvidia raising the criticism itself. Chief financial officer Colette Kress, describing selective credit support for nearly 2 gigawatts of compute for an unnamed frontier AI lab, said the company knows some will call that circular financing, and argued its own risk is limited because the compute is fungible and can be redeployed to other customers. On the same call Kress guided to roughly 70% revenue growth in fiscal 2028 and called that a supply-constrained outlook, saying customer forecasts point to the company's growth doubling next year while supply allows about 70%. Asked by Morgan Stanley's Joe Moore what gave him the confidence to guide a full year out, Huang said Nvidia had never forecast or guided a year in advance before.

Why it matters

If you read the aggregator headline and filed this as a market story, you read the wrong story. Two separate arguments are being run together, and they have different participants. One is about whether frontier models are getting dangerous faster than anyone can check them, and it is being had between Huang and people like Amodei, Altman and Coxon. The other is about whether the buildout's revenue arrives, and it is being had by institutions like the Bank of England. Huang answered the first at length, twice in a week, and did not take up the second on either occasion. The interview settles neither. The filing is the document with numbers attached: the jump from $119 billion to $279 billion in supply commitments and the $105 billion guarantee cap are what Nvidia has actually signed, in a 10-Q it can be held to.

This is an AI-written summary of the sources credited above and linked in the text, read and edited by Nicholas before publishing. The facts and the quote belong to those sources; the wording is ours. Read the original.