Six Chip Stocks Shed 360 Billion Dollars Before the Open as Europe’s AI Selloff Deepens

By The Edge Research Team

Six US listed suppliers of artificial intelligence hardware shed a combined 360.5 billion dollars of market value at their Wall Street premarket prices on Monday, measured against their Friday closing capitalisations, on our calculation, after the leaders of three rival AI laboratories agreed in public that the industry should slow the rate at which it advances model capabilities. Arm fell 7.17 percent before the open, Intel 6.57 percent and Micron 5.75 percent, while Nvidia, the largest company in the group at 5.26 trillion dollars on Friday’s close, fell 2.41 percent, which alone is 126.8 billion dollars of value, on our calculation.

The selling began in Asia, where the Kospi closed 3.26 percent lower and SoftBank fell 10.72 percent, and the Stoxx Europe 600 technology index was down 2.30 percent by 12:02 GMT, in the European afternoon, with ASML lower by 6.11 percent, Infineon by 8.28 percent and Nokia by 8.47 percent, all per CNBC quote data. Those three alone lost about 46 billion euros of market value on the day, on our calculation. The move deepened rather than faded as Europe took over from Asia, on our reading.

Europe is carrying the deeper losses

Intraday quotes pulled in a single call at 12:17 GMT on 14 September 2026, against Friday’s closes, on CNBC quote data. The Helsinki line last traded at 12:17 GMT, the Amsterdam and Frankfurt lines at 12:02 GMT.

All three fell further than the index that contains them. ASML sells the lithography machines that make advanced logic, Infineon makes the power semiconductors used around the data centre, and Nokia sells networking equipment. TSMC, which fabricates the chips, closed 1.24 percent lower in Taipei on the same quote data, and its US listing was 3.50 percent lower in premarket trade. The 360.5 billion dollar total covers the six chip names in the premarket table below whose primary listing is in New York. TSMC is left out of it because its primary listing is in Taipei.

The split is between those who sell the chips and those who buy them

US premarket prices captured at 12:17 GMT, which is 08:17 New York time, against the closes of Friday 11 September, on the same quote data. Premarket volumes are a fraction of the regular session and these are not opening prices. TSMC is excluded from the 360.5 billion dollar total because its primary listing is in Taipei, which had already completed Monday’s trading.

Every chip name in that table fell. The two buyers of their output did not move with them: Meta Platforms was 2.08 percent higher before the open, adding 34.3 billion dollars against Friday’s close, on our calculation, and Amazon was only 0.58 percent lower, against falls of 2.41 to 7.17 percent across the chip names. A slower frontier would mean a smaller order book for the vendors and a smaller capital expenditure bill for the buyers, on our reading. The divergence is consistent with investors placing more of the potential slowdown risk on the infrastructure suppliers than on the hyperscale buyers that carry much of the spending, on our reading.

The scale of what is being repriced is set by the spending plans rather than by the models. Morgan Stanley’s Brian Nowak has forecast that AI spending will pass 1.2 trillion dollars by 2027, per Reuters, whose report of the selloff also quoted Swissquote’s Ipek Ozkardeskaya on where a slowdown would land first: the leases, the debt and the power commitments remain even if compute demand and revenue growth slow.

What the statements actually say

Amodei, who is chief executive of Anthropic, wrote on Saturday that “We must slow the pace at which we improve the capabilities of AI models”, and proposed three steps: embedded third party evaluators with access comparable to that of employees, which his company is committing to on its own; coordination on safety standards among frontier developers in democratic countries; and, hardest of all, coordination with China. He was explicit that “pacing does not mean halting model training or technical progress”, and he put a number on the risk he is pacing against, warning that within six to twelve months a misaligned swarm of agents could be capable of taking over the entire internet and causing hundreds of billions of dollars in damage.

Altman, chief executive of OpenAI, endorsed that the same day at 16:30 GMT, posting that he agreed the frontier needed pacing, that it had been a primary topic of discussion inside OpenAI in recent weeks, and that committing to independent evaluators with employee level access was a good idea his company would match. He returned to it at 04:05 GMT on Monday, posting that “When we talk about ‘pacing’, we do not mean ‘stopping’. Progress has been rapid and will continue to be. But it should be slower than it otherwise could be; interventions like safety cases and monitoring have significant costs.” He added that the company now writes explicit safety cases before frontier reinforcement learning runs it expects to raise capability significantly, and that it welcomes a federal framework setting consistent safety requirements.

Per Reuters, Elon Musk, who runs xAI, also said he agrees with Amodei, and Altman said the company will not proceed with a public offering this year on safety grounds. The wire reported separately on 11 September that Amodei’s company is in talks to bring Nvidia in as an anchor investor in a listing expected to complete before November, seeking as much as 100 billion dollars at a valuation of about 2 trillion dollars, citing two sources.

Why it matters: the commitments described above are about process rather than production targets, and none of them names a rate, a capability ceiling or a date. What the market repriced on Monday was therefore not a measured slowdown but the possibility of one, and it repriced it in the part of the chain that gets paid first, the companies that sell chips, tools and networks into data centres already under construction, on our reading. The premarket split is the evidence closest to hand, with six vendors down 360.5 billion dollars, on our calculation, and the two buyers in the same table roughly flat or higher. Vendor valuations rest on both strong demand and a relentless pace of progress, and the 1.2 trillion dollar spending forecast that underwrites them is a plan rather than a contract, on our reading. Nothing in those commitments obliges anyone to spend less; the risk being priced is that safety cases and embedded evaluators lengthen the interval between one generation of models and the next, and with it the interval between one wave of orders and the next.

Outlook: the regular session opens at 13:30 GMT and will settle whether the premarket move survives contact with full volume, and the three names that fell hardest before the open, Arm, Intel and Micron, are where a reversal would show first. The Anthropic listing, expected before November, will be an early market test of whether investors will pay for a developer that has made pacing its public position, on our reading. Neither developer has published a dated implementation timetable for the evaluator arrangements, each pointing only to the near future, so the next hard datapoints are corporate: capital expenditure guidance from the hyperscalers, and order commentary from the equipment makers whose shares carried Monday’s losses.

Sources: Anthropic, OpenAI, Reuters, CNBC, The Edge.

Get Complimentary Insights

Follow us