The world’s largest chip foundry is tightening restrictions on semiconductor design companies on the Chinese mainland, in a move that could close one of the remaining routes for producing high-performance AI processors there.

Two people familiar with the matter say TSMC is suspending some advanced-process orders that had previously been approved. Affected customers will generally be restricted to older nodes. The move makes the push for domestic replacements more complicated, since those replacements target Nvidia’s GPUs.
The foundry has instructed that any affected order using 16 nanometres or below must stop unless the customer holds a US export licence. The guidance, seen by two sources, says mainland Chinese design houses will generally be limited to 28 and 40 nanometre processes, on condition the chips do not use high-bandwidth memory, or HBM, and do not require advanced packaging.
Most mainstream AI chips use 7 nanometres or smaller nodes for speed and power efficiency. HBM handles the huge data volumes AI workloads require, while advanced packaging integrates chips and memory to raise compute.
The rule behind the restriction
The limits follow a US Commerce Department rule that took effect on Thursday, 19 November. It removes a compliance path that had allowed some mainland Chinese designers to use advanced foundry services without applying for an individual export licence. Previously TSMC could check the compute and other technical features of a Chinese design to judge whether it fell outside US controls. Under the new rule, TSMC can no longer perform that check for affected clients. Their advanced chips are presumed controlled unless an exception applies, which means an export licence is required.
The Financial Times first reported on Thursday that, after the new limits, TSMC and Samsung Electronics had paused some chip shipments to Chinese customers. The measures are likely to hit mainland companies developing alternatives to Nvidia processors, including both independent design houses and technology giants making their own silicon.
A US government list of approved chip designers includes Nvidia, AMD and Apple, as well as MediaTek and Realtek in the Taiwan region, but no company from the Chinese mainland.
Following the route, not just the chip
One source with direct knowledge says that until Thursday some Chinese AI chip designers could still use TSMC’s 7-nanometre process. Reuters could not independently verify the arrangements under which those chips were made. Two sources say TSMC’s guidance also requires stricter scrutiny of a customer’s place of business, parent company, major shareholders, packaging suppliers and shipping routes.
That scrutiny extends to orders placed through overseas subsidiaries or intermediaries, including entities based in Taiwan, the United States, Europe, Japan and Southeast Asia. The effect is to follow the chip through its entire commercial route, not just to inspect the design on paper.
For Beijing, the constraint cuts both ways. It accelerates the case for domestic tools and domestic capacity, but it also slows the near-term road to competitive AI silicon. For the foundries, it is a commercial cost imposed by geography, as export control moves from the product itself to the network that delivers it.
Editor’s note: This is an adapted translation of the original Sohu Tech report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://www.sohu.com/a/1103966657_121948416.