AI‑chip controls fraying

U.S. export controls aimed at stopping advanced AI chips from reaching China are showing enforcement gaps and creating bureaucratic headaches. Reporting suggests Chinese developers are finding workarounds despite the rules, while investigators say about $92m in banned AI chips moved from Super Micro to a little‑known Chinese tech firm and former executives face probes tied to export violations (nytimes.com) (zerohedge.com) (simplywall.st). At the same time, unrelated strains on U.S. policy—like delays tied to the Iran war—are stretching approval times for licenses and adding to the choke‑point effect the rules were supposed to enforce (moneycontrol.com) (timesofindia.indiatimes.com).

The United States built a licensing wall around top artificial intelligence chips for China, but prosecutors and company records show the hardware is still getting through. (justice.gov) (bloomberg.com) Federal prosecutors said on March 19, 2026 that Yih-Shyan “Wally” Liaw, Ruei-Tsang “Steven” Chang, and Ting-Wei “Willy” Sun conspired to divert United States-assembled servers with restricted Nvidia chips to China in violation of export-control law. The Justice Department said the shipments generated billions of dollars in sales and used false records and transshipment through a Southeast Asian company. (justice.gov 1) (justice.gov 2) Bloomberg reported on April 10 that Shenzhen-listed Sharetronic disclosed about $92 million of Super Micro systems containing Nvidia H100 or H200 processors, both chips covered by United States restrictions for China without a license. Sharetronic said it complies with regulations and has “no business cooperation or relationship” with Super Micro, while invoices reviewed by Bloomberg tied the systems to Super Micro hardware. (bloomberg.com) These controls are not a general technology ban. The Commerce Department’s Bureau of Industry and Security first tightened rules on advanced computing chips for China on October 7, 2022, then expanded them again on October 17, 2023 to cover more performance thresholds, more destinations, and more ways to route products through third countries. (bis.gov 1) (bis.gov 2) (bis.gov 3) Chipmakers responded by designing lower-spec products for China, then Washington tightened again. Nvidia told investors on April 9, 2025 that the government now required a license for exports of its H20 chip to China, and said on April 14, 2025 that the requirement would remain in effect indefinitely. (sec.gov 1) (sec.gov 2) Nvidia later said the H20 restriction produced a $4.5 billion inventory and purchase-obligation charge, after $4.6 billion in first-quarter fiscal 2026 H20 sales before the new license rule took hold. Those numbers showed how much business had built up around a chip designed mainly for the China market. (sec.gov) (sec.gov) The enforcement side is now colliding with the paperwork side. The Times of India reported on April 13, 2026 that license approvals for chipmakers and other companies were stretching out as the small office handling export reviews was pulled into wider national-security demands tied to the Iran war. (timesofindia.indiatimes.com) Super Micro has tried to separate the company from the criminal case. CNBC reported that Liaw resigned from the board after the indictment, that Super Micro placed Liaw and Chang on administrative leave, and that the company said it stopped working with Sun; Bloomberg later reported that independent directors hired a law firm to investigate the server sales. (cnbc.com) (bloomberg.com) The result is a system trying to block a class of hardware that is small, high-value, and easy to reroute through intermediaries. Washington can keep tightening the thresholds, but the March indictments and the April disclosures show that writing the rule and stopping the shipment are two different jobs. (justice.gov) (bloomberg.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.