AI‑chip controls fraying
U.S. export controls on advanced AI chips are both slowing American approvals and leaking abroad: approvals for Nvidia and AMD exports to China have stalled after the Bureau of Industry and Security lost nearly 20% of its licensing staff. At the same time, a U.S. probe found roughly $92 million worth of restricted AI chips reached a Chinese firm, and reporting says Chinese developers are working around chokepoints by using foreign data centres and obscuring ownership. (tomshardware.com) (arbiterz.com) (nytimes.com)
The United States is taking longer to approve some artificial-intelligence chip exports even as restricted chips keep showing up inside China. (bloomberg.com 1) (bloomberg.com 2) At the Commerce Department’s Bureau of Industry and Security, staffing fell by 101 employees since 2024, a 19% drop, according to a Bloomberg analysis of federal personnel records, LinkedIn updates and agency records. Bloomberg reported that turnover among rulemaking and licensing staff was close to 20%, while processed licenses fell about 25% and some reviews stretched for months. (finance.yahoo.com) (ttnews.com) The backlog has hit Nvidia and Advanced Micro Devices applications to ship artificial-intelligence accelerators to China, according to Bloomberg’s reporting on April 10 and April 13. The same reporting said billions of dollars in export deals, including shipments to some United States allies, are now stuck in the queue. (bloomberg.com) (finance.yahoo.com) An artificial-intelligence chip is the processor that trains and runs large models, the way an engine powers a truck. Washington has spent the past four years trying to keep China from buying the fastest versions, starting with broad semiconductor controls announced in October 2022. (congress.gov) (csis.org) Those controls are supposed to work through licenses, product thresholds and end-user checks. But Bloomberg reported on April 10 that a Shenzhen company, Sharetronic Data Technology, disclosed sales of 276 Super Micro server systems worth 632 million yuan, about $92 million, to a local subsidiary, and the systems were configured for Nvidia H100 and H200 chips that require Washington’s permission. (bloomberg.com) (theedgesingapore.com) Bloomberg said the disclosure surfaced hours after the United States charged a Super Micro Computer co-founder with illegally smuggling billions of dollars’ worth of Nvidia chips to China. Sharetronic’s Shenzhen-listed shares fell by the daily 20% limit after the report, according to Bloomberg. (finance.yahoo.com) (bloomberg.com) Nvidia’s China business was already under pressure before this month’s staffing story. In April 2025, Nvidia said the United States would require a license for H20 exports to China for the indefinite future, and the company warned it would take a $5.5 billion charge tied to inventory and commitments for that chip. (bloomberg.com) (cnbc.com) Chinese buyers have also been shifting where they get computing power. The New York Times reported on April 13 that some Chinese developers are renting capacity in overseas data centers and masking who ultimately controls the accounts, a workaround that uses remote access instead of importing the chips themselves. (nytimes.com) Beijing has been moving the other way at home. Reuters reported in November 2025 that Chinese authorities told new state-funded data-center projects to use only domestically made artificial-intelligence chips, and projects less than 30% complete could be ordered to remove foreign processors. (reuters.com) (newsbreak.com) The result is a system under pressure at both ends: licenses are taking longer in Washington, while enforcement gaps and offshore workarounds keep weakening the wall those licenses are meant to build. (bloomberg.com) (nytimes.com)
