Global stock rout deepens as AI data centre fears fuel worst US chip sell-off since 2020

US technology and chip stocks suffered their sharpest broad sell-off in years. The Philadelphia Semiconductor Index fell 12.2% on Thursday, its biggest one-day drop since 2020. The Nasdaq Composite lost 5.5%, while the S&P 500 and Dow fell 4.4% and 3.4%. The Nasdaq ended 11.9% below its record high.

The losses spread quickly through Asian markets. Japan’s Nikkei 225 fell about 7% at one point and was down 5.7% by midday. South Korea’s Kospi dropped 8.8%, triggering a 20-minute trading halt. Taiwan’s Taiex closed 4.5% lower after an early fall of more than 8%.

Individual chip and technology companies also suffered. Kioxia and SoftBank fell more than 10% in Tokyo, while SK Hynix and Samsung Electronics dropped about 10% and 8%. Investors were reassessing whether AI expansion can deliver profits quickly, especially if data-centre construction faces political and financial limits.

An AI data centre is a large facility containing servers, networking equipment, storage, cooling systems and power infrastructure. Its computers train AI models and answer requests from users. Unlike ordinary office servers, AI systems often perform enormous numbers of calculations on text, images, video or other data.

Specialised chips carry out these calculations in parallel, so a useful AI service may need thousands of them working together. The servers also generate substantial heat. Cooling equipment must run continuously, while the chips, networking hardware and storage systems draw electricity around the clock. That makes power availability a central part of building an AI facility.

The source links this infrastructure to hundreds of billions of dollars in company investment. It also warns that slower construction could reduce demand for chips and electricity. Higher borrowing costs make the projects harder to finance, while proposed pauses or restrictions could delay the facilities needed for AI expansion.

The sell-off erased a record amount of value from major US companies. S&P 500 companies lost $2.9 trillion (€2.6 trillion) in market value on Thursday. The scale shows how quickly changing expectations about AI investment can affect the wider stock market, not only companies that directly build or sell AI systems.

The political risk is also substantial. Cleanview estimates that about 40% of planned US data-centre capacity is in states where governors or governors-elect have pledged, or already imposed, a pause on new projects. That does not mean all 40% will be cancelled. It shows that a large share of future capacity could face delays, restrictions or additional uncertainty.

The figures help explain investor concern. Companies have committed hundreds of billions to AI, but projects may take longer to build or earn returns. If construction slows, expected chip sales, electricity demand and future technology profits may all be revised lower.

Investors are trying to connect the US midterm election results with the future of AI infrastructure. Democrats won sweeping victories, and many successful candidates had pledged to slow or halt new data-centre construction. New rules could make projects harder to approve, increasing delays, costs and uncertainty for companies planning large facilities.

The concern is especially important because data centres require enormous upfront spending. Companies have already poured hundreds of billions into AI, while US Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez proposed a federal moratorium earlier this year. About 40% of planned US capacity is in states where governors or governors-elect support or have imposed pauses.

Investors therefore questioned when companies will recover their spending. Higher borrowing costs add another obstacle by making construction more expensive. If projects cannot expand as quickly as expected, forecasts for chip sales, electricity demand and AI-related profits may fall, putting pressure on technology share prices.

Data centres connect several parts of the economy. Building them creates demand for advanced chips, servers, power equipment and electricity. Once operating, they continue using electricity for computing and cooling. If companies build fewer facilities, those purchases and ongoing energy needs could be smaller than investors expected.

The article gives a clear chain of risk. Political pauses or stricter rules could delay new projects, while higher borrowing costs could make construction less affordable. A slower buildout would reduce orders for chipmakers and weaken demand from electricity providers. Companies that expected rapid AI expansion might then earn revenue later or invest less.

The wider effect could reach the US economy. The article warns that a slowdown could weigh on US growth and dent demand for chips and electricity. It also raises questions about when firms will recoup their hundreds of billions in AI spending, potentially lowering confidence in technology profits and valuations.

The US data-centre question matters globally because chipmakers in Asia supply the technology needed for AI infrastructure. Their share prices reflect expected future sales, not only current orders. If US companies build fewer facilities, investors may expect weaker demand for chips made in Japan, South Korea and Taiwan.

That mechanism appeared during the sell-off. In Tokyo, Kioxia and SoftBank fell more than 10%, while Advantest dropped more than 8%. South Korea’s SK Hynix fell about 10% and Samsung Electronics about 8%. Taiwan’s TSMC fell as much as 8.5%, and the Taiex closed 4.5% lower.

The market reaction shows how connected the AI supply chain has become. Investors were not responding only to local political news. They were revising expectations for worldwide AI investment, chip orders and profits. Neil Wilson of Saxo Markets called the sell-off overdone, but said investors would want markets to stabilise first.

A stock price is largely a bet on a company’s future profits. Investors estimate how much money a business may earn and translate those future amounts into today’s value. Stronger expected growth can support a higher share price, while doubts about sales, costs or regulation can lower it before profits actually change.

Higher borrowing costs affect both sides of that calculation. They make it more expensive for companies to finance data centres, chips, power systems and other expansion. They also make future profits less valuable when investors compare them with safer returns available elsewhere. Fast-growing technology companies are especially sensitive because much of their expected value may depend on profits far in the future.

The article connects these pressures directly to AI. Companies have invested hundreds of billions, but investors questioned when returns will arrive. Possible construction pauses and higher financing costs could delay growth, reduce expected demand and make technology shares look less valuable today.