Skip to content
Markets & Trends

Three Rivals Asked AI To Slow Down. Chips Fell 5.8%. A Week Later The Nasdaq Closed At A Record. What Just Happened?

On Saturday 12 September, Anthropic's Dario Amodei published an essay arguing AI companies must slow the pace at which they improve their most capable models. Within hours Sam Altman replied 'I agree with Dario' and ruled out an OpenAI IPO this year as 'ill-advised'; Elon Musk posted 'Dario is right'. On Monday the Philadelphia Semiconductor Index dropped 5.8% - its worst day since July - with Arm down 9.7%, ASML 7.2%, Nvidia 3.4%, SK Hynix over 7% and SoftBank 10.7%. Then, on 21 September, the Nasdaq closed at a record 27,122.09. For anyone running money, building trading systems or deploying AI inside a financial firm, the whiplash is the story - and it says something specific about how markets are now pricing AI risk.

AlchmAI Editorial13 min read

-5.8%

Philadelphia Semiconductor Index on Monday 14 September - its worst day since July - after the three CEOs' weekend statements

27,122

Nasdaq Composite record close on 21 September, up 2.26% on the day, with the S&P 500 at 7,764.70

~1/3

Share of 2026 US GDP growth attributable to AI investment, per ING Markets research cited on 18 September

95%

Of businesses that invested in AI have failed to make money from it, per the MIT study - on roughly $40bn of combined spend

The most unusual thing about the second week of September was not that a technology executive warned about AI risk. It was who agreed with him. On Saturday 12 September, Anthropic chief executive Dario Amodei published an essay arguing that AI companies must slow the pace at which they improve the capabilities of their most advanced models, citing the risk of losing control of AI systems and the misuse of AI for cyberattacks and bioterrorism. His proposal was framed as a three-step plan to temper the pace of development without sacrificing commercial advantage or the United States' lead. Within hours, OpenAI's Sam Altman - a bitter commercial rival - posted 'I agree with Dario', adding that pacing did not mean stopping. Elon Musk, who has sued both, posted 'Dario is right'.

Altman went further than agreement. He said that going public this year would be an ill-advised moment for OpenAI, given the safety concerns - a remark that landed on a market that had spent months speculating about the size and timing of an OpenAI listing. That single sentence did as much damage to the AI IPO pipeline as the essay did to the chip complex, and the two together produced the ugliest AI trading session since July.

Why A Safety Essay Moved The Chip Index

Read as a piece of policy the essay is about existential risk, but the market read it as a statement about capital expenditure. The AI trade of the past two years has one load-bearing assumption: that frontier labs will keep buying compute at an accelerating rate because the race to more capable models cannot be paused. Every hyperscaler's capex guidance, every data-centre lease, every memory-chip order book rests on it. If the three most influential figures in the industry are prepared to say - in public, on the same day - that the pace of capability improvement should slow, then the assumption is no longer unquestioned, and the first thing to reprice is the equipment that pace would have required.

That is why the damage was so lopsided. Nvidia, the direct beneficiary of the capex race, fell 3.4%. Arm and ASML, further up the supply chain and more sensitive to any change in the slope of demand, fell two and three times as much. SoftBank, which owns Arm and has staked its balance sheet on OpenAI, fell more than any of them. The market did not panic about AI. It repriced the second derivative of AI spending, which is a far more specific and far more rational reaction.

“The selloff was not a verdict on whether AI works. It was a verdict on whether the spending curve is a straight line. For a week, the market decided it might not be.”

Then The Record

A week is a long time in a market this concentrated. By 18 September, ABC News was reporting that AI investment accounted for roughly a third of US GDP growth in 2026 according to ING Markets, alongside the sobering MIT finding that roughly 95% of businesses that invested in AI had failed to make money from it, on combined spending of around $40bn. Both facts are true at once, and they frame the entire debate: AI is now macroeconomically load-bearing, and most of the money spent on it by ordinary companies has not yet produced a return.

On Monday 21 September the Nasdaq Composite jumped 2.26% to a record close of 27,122.09 and the S&P 500 gained 1.49% to 7,764.70, as oil and Treasury yields fell and chipmakers regained ground. Year to date, the Dow is up around 8%, the S&P 500 around 11% and the Nasdaq around 13%; Nvidia is up about 18% and AMD roughly 155%. The AI trade did not merely recover from the slowdown warning. It went on to make a new high inside seven trading days.

What This Means If You Run Money

  • Your AI exposure is larger than your AI allocation. A UK pension fund with a passive global equity sleeve is a chip-cycle investor whether it intended to be or not. The 14 September session is the cheapest possible demonstration of what a genuine repricing would do to that sleeve.
  • The second derivative is now a tradeable variable. For two years the question was 'is AI real?' The question this month was 'is the slope of AI spending fixed?' That is a more answerable question, and it will be answered by hyperscaler guidance rather than by essays - which makes the next round of capex commentary the most important macro release of the quarter.
  • The IPO pipeline is a sentiment gauge. Altman's 'ill-advised' comment did not just delay one listing. It signalled that the most valuable private company in the sector believes public-market scrutiny would currently be unhelpful, and every other AI IPO candidate's bankers heard it.
  • Energy and rates still matter more on a given day. The record close on 21 September was driven as much by falling oil and yields as by AI enthusiasm. In a market where AI is a third of growth, the AI trade and the macro trade have become the same trade.

What This Means If You Build Financial Systems

We build AI, trading and automation systems for financial firms, and the week produced three practical lessons that have nothing to do with portfolio positioning.

  1. 01Volatility clustering around AI news is now a market-structure feature. Trading platforms and charting systems need to be tested against sessions where a handful of names move 7-10% in the first hour on a headline, because those sessions are arriving with increasing frequency. The 14 September tape is a good replay day for exactly that.
  2. 02Model-provider concentration is a business risk, and the market just said so. If the CEOs of the frontier labs are openly discussing pacing, the assumption that every model generation arrives on schedule and at falling prices is weaker than a year ago. Any firm whose automation depends on a single provider's roadmap should have a tested fallback - not because the roadmap will fail, but because the market has now put a price on the possibility.
  3. 03The MIT 95% figure is the number to build against. If most firms that spent on AI have not made money from it, the differentiator is not access to models - everyone has that - but the integration and process work that turns a capability into a measurable return. The firms in the 5% did the unglamorous part. That is the part worth doing regardless of what the chip index does next.

The Bottom Line

A safety essay from one CEO, endorsed by two rivals, produced the worst semiconductor session since July - Arm down 9.7%, ASML 7.2%, SK Hynix over 7%, SoftBank 10.7% - and an OpenAI IPO pushed out of 2026 as 'ill-advised'. Seven trading days later the Nasdaq closed at a record 27,122.09. The whiplash is not noise; it is the clearest picture yet of how the market is treating AI in 2026. AI investment is now roughly a third of US growth, exposure to it runs through stocks, pension funds, bonds and private markets alike, and the only variable anyone can genuinely trade is the slope of the spending curve - which three people briefly put in doubt and the market then decided to ignore. For investors that means the AI trade and the macro trade have merged. For firms building financial systems it means volatility around AI headlines is a permanent feature to engineer for, provider concentration is a risk the market has now priced, and the MIT finding that 95% of AI spend has not paid off is the real benchmark to beat. The record high says the market is not worried. The 14 September tape says it should occasionally check.

References & Further Reading

AI stockssemiconductor selloffNasdaq recordAI safetyFintech AI Agency LondonAI Agency UKmarkets 2026
Share Email
AI

AlchmAI Editorial

Research and analysis, London

The AlchmAI team writes about the markets, technology and regulation we work with every day. We build trading platforms, real-time charts and AI analysis tools for brokers, prop firms and fintech teams from our office in Mayfair, London. Every article lists its sources. Nothing we publish is investment advice.

This article is general information and commentary. It is not investment advice or a recommendation to buy or sell any investment. Important information

Keep reading

All articles