US halts Nvidia chip shipments

- The U.S. Commerce Department moved on May 31 to block advanced Nvidia AI chip shipments to Chinese companies’ overseas subsidiaries, closing a loophole. - Reuters reported hundreds of thousands of chips may have reached Chinese subsidiaries abroad, while Axios called electricity the “gold rush” beneath AI. - The Commerce guidance leaves existing data center use and servicing untouched, while regulators and cloud operators absorb the new restrictions.

The U.S. Department of Commerce moved on May 31 to stop exports of the most advanced AI chips to subsidiaries of Chinese companies located outside China, according to Reuters. The guidance targets a gap that had allowed top processors from Nvidia and others to reach Chinese-linked entities in places such as Malaysia despite broader U.S. restrictions on sales into China. Reuters reported the move does not require data centers already using the affected chips to stop operating or servicing them. Axios reported the same weekend that the AI buildout is also colliding with a second constraint: electricity. Data-center demand is pushing technology companies, utilities and investors deeper into power procurement, storage and grid planning, Axios said, describing the scramble for supply as the “gold rush” beneath the AI boom. A separate Reuters report on June 1 said the European Commission has held “productive meetings” with Anthropic on possible future access for EU bodies to its Mythos AI product, another sign governments are treating advanced AI systems as strategic tools. ### Which shipments did Washington move to stop? Reuters said the Commerce Department’s new guidance is aimed at the world’s most advanced AI chips, including Nvidia’s highest-end Blackwell processors. The step closes what Reuters described as a potential loophole created about a year earlier, under which those chips could still be exported to Chinese companies’ subsidiaries outside mainland China. (money.usnews.com) CNBC, citing the Reuters report, said the guidance suggested the chips may have been reaching those subsidiaries for almost a year. Reuters also said an industry source estimated that hundreds of thousands of chips may already have gone to Chinese subsidiaries abroad. ### Why does a subsidiary in Malaysia matter here? Malaysia was one of the locations Reuters cited as an example of where Chinese AI firms’ overseas subsidiaries were based. (money.usnews.com) The issue for U.S. officials is not a shipment into China itself, but whether a Chinese company can obtain restricted computing power through an affiliate incorporated elsewhere. (cnbc.com) The Commerce action, as described by Reuters, tries to align the rule with the ownership and control of the buyer rather than only the geography of the shipment. That is a narrower but more direct way to police access to high-end chips. ### Does this mean data centers have to shut machines off? Reuters said no. The guidance does not require data centers that already have the affected chips to stop using them, and it does not bar servicing of those systems. (money.usnews.com) That distinction matters for cloud operators and facility owners that may already be running installed Nvidia hardware for customers linked to Chinese firms. The immediate effect is on future shipments, not forced removal of chips already deployed. Reuters framed the step as a tightening of export controls rather than a retroactive seizure or operating ban. ### Why is power now part of the same story as chips? Axios reported on May 31 that AI’s expansion is turning energy into one of the hottest businesses in the United States because data centers need huge and reliable electricity supplies. (tbsnews.net) The outlet said companies from large technology groups to automakers are moving into energy deals, storage and infrastructure as they race to secure enough power. (money.usnews.com) The International Energy Agency made a similar point in its analysis of AI and energy, saying there is no AI without electricity for data centers. That does not change the export-control story, but it helps explain why access to compute now depends on two bottlenecks at once: chips and power. ### Where do governments fit beyond export controls? The European Commission said on June 1 it had several productive meetings with Anthropic about possible future access for EU bodies to Mythos, according to Reuters. (axios.com) Politico, citing Commission spokesperson Thomas Regnier, reported the same talks and said the EU cybersecurity agency ENISA is expected to be the body that could receive access if an agreement is reached. (iea.org) That EU discussion is separate from the U.S. export move, but both developments show public authorities trying to shape who can use frontier AI capabilities and for what purpose. The next concrete step in the U.S. case is implementation of the Commerce guidance by exporters, distributors and data-center operators handling future chip orders. (money.usnews.com) (msn.com)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.