Nvidia's share of China's AI chip market has fallen to 34%, according to new figures from IDC, with Huawei closing the gap as Washington's export controls continue to reshape who sells accelerators into the world's largest AI market.

The number marks a sharp comedown for a company that, until a few years ago, dominated the Chinese market for AI training and inference hardware. Huawei, barred from leading-edge fabs by US sanctions of its own, is the principal beneficiary, IDC's data shows.

By Rebecca Stone

What does the 34% figure actually tell us?

It tells us the export-control era has produced a real substitution effect, not just headlines. When the US first restricted Nvidia's top data-center GPUs for Chinese customers, the expectation inside the industry was that modified, compliance-tuned parts would hold most of the ground. That held for a while. It isn't holding now.

Huawei's Ascend accelerator family has moved from a patriotic purchase to a practical one. Chinese cloud and AI labs that once treated domestic silicon as a fallback now qualify it as a primary option, and every percentage point IDC records shifting from Nvidia to Huawei reflects a design win that will be hard to reverse. Procurement cycles in AI infrastructure run for years.

The share loss also compounds. A customer that ports its software stack to a domestic platform accrues tuning, tooling and operator expertise on that platform. Switching back later carries real engineering cost, whatever Nvidia's performance advantage on paper.

Why is Huawei gaining now?

Timing matters. Huawei spent years building out its compute software ecosystem while its access to advanced manufacturing was constrained. Meanwhile, successive rounds of US rules pushed Nvidia's China-specific parts further from its flagship products in capability. The gap between "what Washington allows" and "what Huawei can ship" narrowed from both directions at once.

IDC's findings land amid continued friction over the terms of Nvidia's China business. Whatever the diplomatic weather, Chinese buyers have a standing instruction of sorts: assume the next restriction is coming, and hedge accordingly. That hedging shows up directly in market-share data like this.

Geopolitics changes the commercial picture here more than any spec sheet does. A 34% share for Nvidia is not primarily a verdict on silicon. It is a verdict on export policy, and on how quickly a determined domestic challenger can industrialize an alternative.

What does this mean for the AI supply chain?

For Nvidia, China was once a multi-billion-dollar revenue pillar. Each share-point slide translates into revenue that either goes unbilled or flows to Huawei — and Huawei's gain strengthens the very competitor US policy was designed to prevent from scaling.

For Chinese AI developers, the shift is less painful than outsiders often assume. Inferencing workloads, which make up a growing share of deployed AI compute, are more tolerant of hardware diversity than frontier training runs. That tolerance gives domestic parts a wider landing zone than raw performance comparisons suggest.

For the broader supply chain, the lesson is that export controls act as industrial policy for the target country. Capacity, tooling and ecosystem investment flow toward whoever can legally serve the demand. In China's AI chip market, that is increasingly Huawei.

What comes next?

IDC's numbers describe momentum, and momentum in this market has been one-directional for several reporting periods. Nvidia will keep pushing whatever compliance-grade products it can ship into China; Huawei will keep converting each opening into installed base. Unless policy in Washington changes materially, the competitive question in China is no longer whether Huawei takes share — it is how fast, and whether anything stalls it.