Nvidia’s China AI chip market share has fallen to 34% in the first half of 2026, according to new IDC data reviewed by Reuters on October 13, while Huawei has surged to 28% — a six-point gap in what remains the world’s second-largest AI chip market. For a company that once treated China as an unshakeable stronghold, the numbers mark a turning point: export controls, not competitors, have become the force reshaping the market.
The headline figures are stark. Nvidia shipped 678,000 AI accelerators in China between January and June, down 44% from a year earlier, the IDC figures show. Huawei shipped 572,000, up from more than 300,000 in the same period last year. Total AI-accelerator shipments in China reached 2.008 million units in the half — a growing market that Nvidia is increasingly watching from the sidelines.
Nvidia’s China AI chip market share: how export controls rewrote the map
Washington’s restrictions are doing most of the heavy lifting. The United States has barred China’s access to Nvidia’s most powerful AI chips, limiting the range of products the company can sell in the country. Beijing, meanwhile, has actively promoted domestic alternatives — and now requires Chinese companies to seek approval before importing even the Nvidia chips that remain compliant with the rules.
Nvidia CEO Jensen Huang has repeatedly criticised the curbs, arguing the policy forced an American company to cede China’s $50 billion AI chip market to Huawei. Nvidia declined to comment on the IDC figures, according to Reuters.
One caveat worth keeping in mind: IDC did not specify which Nvidia products made up its 34% share, or whether the tally includes cutting-edge chips such as the B200 and B300, which are barred from export to China but have nonetheless been smuggled into the country, Reuters noted. The direction of travel, though, is unambiguous.
It’s not just Huawei
While Huawei is the story’s headline act, the IDC data shows a domestic ecosystem rising almost across the board. Alibaba’s chip-design unit T-Head held 7% of the market, followed by Cambricon with 5%, Baidu’s Kunlunxin with 4%, and Hygon with 3%. Nine of the 11 Chinese chipmakers tracked by IDC increased their shipments from a year earlier.
The shape of demand is shifting too. Shipments of traditional GPUs — the chips most widely used to train AI models — fell 26% to 991,000 units, as supplies eligible for sale in mainland China remained limited. But shipments of other AI accelerators rose nearly 80% to 1.016 million units, overtaking GPUs by volume. These are chips designed for specific AI tasks, such as running trained models, and the category is where China’s domestic players are strongest: Huawei, T-Head, Kunlunxin and Cambricon together accounted for 95% of the non-GPU segment, which IDC said is being adopted by government agencies, internet companies and telecoms providers.
The contest is moving from chips to systems
Perhaps the most telling line in the IDC report is about where the competition is heading. The firm said the contest is shifting from the performance of a single chip to system-level capabilities — including how well chips work together in a larger computing cluster.
That shift favours the home team. Chinese cloud giants and telecom operators are building vast clusters around domestic silicon, and the software stacks to run them. For Nvidia, the risk was never just losing unit sales; it was losing the ecosystem lock-in that made CUDA the default language of AI development. If Chinese labs train and deploy their flagship models on Huawei hardware running domestic software, that lock-in erodes.
IDC expects total accelerator shipments in China to grow 19% this year, even as GPU shipments fall 9% — a forecast that bakes in the assumption that the domestic replacement cycle is now self-sustaining.
What this means for the AI race
A few takeaways for anyone watching the industry:
For AI buyers in China, domestic silicon is no longer a compromise. The 80% jump in non-GPU accelerator shipments suggests Chinese firms are finding Huawei and Cambricon hardware good enough for a growing share of real workloads — especially inference, where running trained models matters more than raw training throughput.
For Nvidia investors, the $50 billion question is open. Huang’s figure for the China market at stake is not a forecast — it’s a description of what’s being contested right now. The company’s global dominance remains intact, as AMD’s record-setting run toward a trillion-dollar market cap shows just how much value the market still attaches to leading-edge AI silicon. But China is the one geography where Nvidia is structurally losing.
For the broader AI buildout, geopolitics is now a design constraint. The same week brought news of Oracle and Tencent’s $7 billion AI cloud deal and Japan’s $15 billion AI data-center buildout — every region is now planning its AI infrastructure around which chips it can actually get.
The open questions: whether Huawei’s next-generation accelerators can close the performance gap in large-scale training clusters, not just inference; whether Washington tightens the screws further in response to smuggling; and what Nvidia’s compliant-chip strategy for China looks like when even compliant chips need Beijing’s approval. The answers will shape not just two companies’ fortunes, but where the world’s AI models get trained for the rest of the decade.
Source: IDC data via Reuters, October 13, 2026. Reporting by Reuters; figures refer to China’s data-centre AI accelerator market in H1 2026.
