China’s grip on minerals threatens AI boom in US

China’s tightening of its chokehold on the supply of rare earths comes at an inopportune moment for American tech companies, which have been riding high on a wave of optimism about artificial intelligence but are also facing questions about whether the largest players, including Nvidia and OpenAI, are becoming too financially interdependent.

“China is signalling: we’re willing to threaten your primary growth driver,” Chris Miller, the author of Chip War, told news site Axios. “People are looking to see whether the Trump administration builds alternative sources of leverage against China that forces them to back down.” Trump has said the additional tariffs would come into force on 1 November, before a summit with Chinese leader Xi Jinping in South Korea.

The five rare-earth elements for which Beijing has chosen to limit licences, in addition to previous curbs, do not go into chip processors in large amounts but are essential for the machines and networks that make and run AI chips. Foreign companies must now seek the Chinese government’s approval to export them, as well as refining technology.

Dysprosium is an element that prevents data centres from overheating, while terbium and samarium are used in the high temperature magnets that are critical to the chip supply chain. For all three, more than 95% of refining capacity is based in China.

“It is the first time [rare earth] export control policies mention semiconductors,” China Merchants Securities, a state-backed brokerage firm, wrote in a research note on Friday, adding that “a new level of regulatory intensity” had been reached.

Analysts say that ASML, the only manufacturer in the world of machines that make the most advanced semiconductors, is now expecting weeks-long delays. Nvidia, which makes chips for training AI, lost $229bn on Friday, while the sell-off wiped out Amazon’s gains for the year. The S&P 500 plunged by 2.7% in its worst day of trading since April.

Defence contractors, smartphone makers and startups focused on the next wave of “physical AI” and robotics are identified by analysts as also being exposed. But the biggest questions may arise for the megacap tech companies that in recent weeks have made a series of eye-popping bets on the rollout of AI infrastructure – often by investing in each other.

Last week OpenAI inked a deal with chipmaker AMD to build a 10% stake in the company over time as it pledges to provide the ChatGPT-owner with 6GW of computing capacity. That’s after rival Nvidia announced plans to invest $100bn in OpenAI, which in turn agreed to buy $300bn in data centre capacity from Oracle.

This web of agreements has spurred major players, including JPMorgan’s Jamie Dimon, James Anderson and even the Bank of England, to warn AI-eager stock markets are heading for “a correction” – or worse, a replication of the dotcom bubble. “Today’s valuations are heading towards levels we saw during the bullishness about the internet 25 years ago,” said Kristalina Georgieva, IMF managing director, on Wednesday.

Anderson, who helped turn Baillie Gifford into a key tech investor, told the Financial Times some of the AI deals “rhymed” with the dotcom-era practice of telecoms equipment makers taking on large debts to help their customers build out internet fibre.

The correction in stocks seen on Friday was driven by trade war worries, rather than doubts about AI financing. Whether it continues this week will depend to what extent investors expect both to grow.

Photograph by Jie Zhao/Getty