Semiconductor geopolitics heats up
- Export controls, smuggling probes, and tighter audits are reshaping global AI-chip supply chains and sales channels. - US officials say Nvidia’s H200 chips haven't been sold to China, and Nvidia stepped up supply-chain audits after a smuggling case. - At the same time Google unveiled two new AI chips, underscoring rising competition amid persistent geopolitical constraints. (en.bloomingbit.io) (digitimes.com) (finance.yahoo.com)
Nvidia’s China business is being squeezed from both sides: Washington says no H200 chips have reached Chinese buyers, while smuggling probes are forcing tighter checks across the supply chain. (usnews.com) (digitimes.com) U.S. Commerce Secretary Howard Lutnick said on April 22 that Nvidia’s H200 artificial intelligence chips “have not yet been sold” to Chinese companies. Reuters reported the Trump administration had cleared some China-bound H200 sales in January, but shipments stalled over licensing conditions and approval problems on both the U.S. and Chinese sides. (usnews.com) At the same time, Nvidia has intensified supply-chain audits after a March smuggling case tied to Super Micro Computer, according to DigiTimes. The publication said Nvidia tightened scrutiny of shipment routes and customer checks after the case exposed gaps in how advanced graphics processors move through server makers and distributors. (digitimes.com) The Justice Department said on March 19 that Yih-Shyan “Wally” Liaw, Ruei-Tsang “Steven” Chang, and Ting-Wei “Willy” Sun were charged with conspiring to divert U.S.-assembled high-performance servers with American artificial intelligence technology to China in violation of export-control laws. Supermicro said on April 7 that it had opened an independent investigation and that the company itself was not named as a defendant. (justice.gov) (ir.supermicro.com) The dispute sits inside a larger fight over AI chips, the processors that train models and answer prompts by handling huge volumes of math in parallel. Since 2022, U.S. export controls have tried to keep the fastest accelerators and the servers built around them out of China without a license, pushing sales into narrower, more heavily monitored channels. (justice.gov) (usnews.com) That pressure is colliding with competition. Google said on April 22 that it is introducing two new Tensor Processing Unit chips, TPU 8i for fast-response agent workloads and TPU 8t for training and serving large models, at Cloud Next ’26. (blog.google) Google framed the new chips as infrastructure for “autonomous AI agents” and other demanding workloads, extending its push to offer customers alternatives to Nvidia-based systems in the cloud. Sundar Pichai said Cloud Next ’26 also featured Google’s eighth-generation Tensor Processing Units as part of a broader artificial intelligence infrastructure buildout. (blog.google 1) (blog.google 2) For Nvidia, the immediate picture is not just demand in China but control over where chips go after they leave the factory. The opening move was a blocked sale; the next one is a tighter audit trail. (usnews.com) (digitimes.com)
