A wave of safety warnings from leading artificial-intelligence labs, together with OpenAI’s decision to shelve its IPO this year, pushed markets to price in a simple equation: slower frontier-model iteration means weaker chip demand. On Monday the selling swept Asia, Europe and the Americas, and early tallies put the market value wiped from the global semiconductor and AI-compute hardware chain at more than US$500 billion.
In Asia, SoftBank, a major OpenAI backer, closed down 11 per cent. Flash-memory maker Kioxia fell 6.4 per cent, and South Korean giants SK Hynix and Samsung dropped 6.4 per cent and 4.1 per cent, dragging the Kospi down 3.3 per cent. Europe was no calmer: Dutch equipment maker ASML lost 6 per cent, while ASMI and BE Semiconductor fell 10 per cent and 8.2 per cent. Infineon and STMicroelectronics each dropped more than 6 per cent. Siemens Energy, which builds turbines for data centres, slid over 8 per cent and led the European selloff.
The shock reached American shores with the heaviest damage in the very sector that led the AI rally. The Philadelphia Semiconductor Index fell nearly 6 per cent: Nvidia dropped 3.3 per cent, Broadcom, AMD and SanDisk more than 4 per cent, and Micron 5.3 per cent. “If this ultimately triggers a slowdown in AI capital spending and a rethink of the sector, it will have a deep impact on the economy and on a critical slice of the equity market,” said Steve Sosnick, chief market strategist at Interactive Brokers. “The hot run we have seen was, fundamentally, driven by AI capital spending.”
The panic was lit by a blog post from Anthropic chief executive Dario Amodei, who argued the industry should avoid a reckless race and take the time to build frontier models responsibly, even bringing in third-party assessors. Sam Altman backed his rival’s view and said OpenAI would not pursue an IPO in 2026 on safety grounds. Microsoft published an interim code of conduct for its AI models, barring them from weapons, hazardous-material advice or policy-violating content.
Then President Trump weighed in. He said he had rarely seen an industry leader call for regulation that, if enforced strictly, would silence or bankrupt the firms themselves, and dismissed talk of AI destroying humanity as a hoax. Artificial intelligence and data centres, he said, will be the most powerful economic engine in history, eclipsing oil, gold, diamonds and even the internet, and will not be stopped on his watch. Several US senators are already debating legislation that would require AI firms to prove they have reasonable safeguards in place.
Not everyone bought the fear. Michael Burry, the investor who famously shorted the US housing market in 2008, called the warnings a tool for big labs to suppress smaller rivals. Morgan Stanley still expects global AI spending to top US$1.3 trillion by 2027, and Deutsche Bank notes competition between firms and nations remains too fierce for anyone to step back while rivals press ahead.
The build-out continues regardless. Samsung and SK Hynix plan a chip cluster in Korea with total investment above US$500 billion, while Kioxia and SanDisk are spending more than US$31 billion to expand flash output. “The next phase of AI demand may depend less on how often new models are trained and more on how densely they are used,” said James Voi of Tiger Securities. Inference and agentic workloads will keep hardware and infrastructure suppliers busy.
Editor’s note: This is an adapted translation of the original Shenzhen News report. It has been trimmed and restructured for readability for an international business audience.
Translated and adapted from Shenzhen News (https://www.sznews.com/news/content/2026-09/15/content_32172022.htm).