Chinese government tenders show important institutions, such as Tsinghua University, rely on AI chips from Nvidia and AMD, which are now under US export limits
High-profile universities and state-run research institutes in China have been relying on a U.S. computing chip to power …
Context & Ripple Effects
The 2022 tender records were the first public paper trail showing that elite Chinese institutions — Tsinghua among them — built their AI research capacity on Nvidia and AMD silicon just as Washington's export limits came down. What looked then like an accounting quirk became the template for the next four years of coverage: [[a:848306|dozens of military bodies, institutes, and universities kept buying small batches of Nvidia chips after the ban]], and ten entities obtained advanced chips embedded in Supermicro, Dell, and Gigabyte servers rather than as bare GPUs.
The arc has since inverted. Beijing moved from quietly sourcing restricted chips to actively rationing them — requiring state-funded data center projects to use only domestic AI chips — while still carving out exceptions, including approving H200 purchases only under special circumstances such as university research. The tenders matter because they show the demand side was never in doubt; the fight has been over who controls the supply channel.
First-order effects
- Tsinghua and the state-run research institutes named in the tenders lose a compliant procurement path for the chips their existing AI infrastructure depends on, forcing each new tender toward either licensed variants, gray-market servers, or domestic alternatives.
Second-order effects
- The gap between demand and legal supply spawns the workarounds documented in later reporting — small-batch purchases and advanced chips hidden inside Supermicro, Dell, and Gigabyte products — while Nvidia's A100/H100 cloud rental inside China stays cheaper than in the US, keeping demand alive despite the controls.
Third-order effects
- Universities are emerging as the sanctioned exception in both directions: Beijing restricts H200 approvals except for university research even as defense-linked universities seek H200s, making academic institutions the contested interface where US controls and Chinese rationing overlap.
The trend: AI compute is becoming a state-allocated strategic resource, with export controls, gray-market channels, and domestic substitution converging to turn chip access into an instrument of policy rather than procurement.