Nvidia's China Grip Is Slipping: Huawei Now Holds 28% of the AI Chip Market

The most lopsided contest in semiconductors just got interesting. Nvidia, the company that defined the AI chip era, saw its share of China's data-center AI accelerator market slide to 34% in the first half of 2026, according to an IDC report seen by Reuters on October 13. Huawei, the domestic champion under US sanctions since 2019, is now within striking distance at 28%. It is the closest anyone has come to Nvidia in the world's second-largest AI infrastructure market, and it happened almost entirely by design: American policy squeezed the supply, and Chinese policy organized the demand.

The numbers are stark. Nvidia shipped 678,000 AI accelerators in China between January and June, down 44% from a year earlier. Huawei shipped 572,000 units, up from just over 300,000 in the same period last year. Total accelerator shipments in the country reached 2.008 million units, and the market is still growing, just not in Nvidia's favor.

What the IDC data actually says

The headline battle is Nvidia versus Huawei, but the deeper story is about everyone else. Alibaba's chip-design unit T-Head took 7% of the market, Cambricon 5%, Baidu's Kunlunxin 4%, and Hygon 3%. Nine of the 11 Chinese chipmakers tracked by IDC increased shipments year on year. The domestic ecosystem is no longer one company propped up by state subsidies; it is a roster of suppliers scaling across the stack.

The more revealing shift is architectural. Shipments of graphics processing units, the chips most associated with training AI models, fell 26% to 991,000 units as products eligible for sale in mainland China stayed scarce. Meanwhile, shipments of other AI accelerators, chips designed for specific AI tasks such as running trained models, rose nearly 80% to 1.016 million units, overtaking GPUs by volume for the first time. Huawei, T-Head, Kunlunxin, and Cambricon together account for 95% of that non-GPU segment, which IDC says is being bought by government agencies, internet companies, and telecom providers.

IDC's read on what comes next is notable for what it no longer prizes. Competition, the firm said, is shifting from the performance of a single chip to system-level capabilities: how well chips work together in a larger computing cluster. That framing favors vendors who sell racks and pods rather than just cards. IDC expects total accelerator shipments in China to grow 19% this year, even as GPU shipments fall 9%, which suggests the non-GPU buildout is where the growth now lives.

For the first time in the AI era, Nvidia is no longer the obvious answer to the question of who powers Chinese data centers.

How export controls engineered the reversal

None of this was accidental. Washington has spent years tightening the screws: restricting China's access to Nvidia's most powerful AI chips, limiting the products the company can legally sell, and layering on licensing regimes that turn every compliant sale into a negotiation. Beijing, for its part, has promoted domestic alternatives while requiring Chinese companies to seek approval before importing compliant Nvidia chips, a bureaucratic friction that functions as a nudge toward local suppliers.

The financial damage to Nvidia is already on the record. In its fiscal first-quarter commentary last year, the company disclosed that Washington informed it in April that exports of its H20, the China-tailored card it built specifically to comply with earlier rules, would require a license. Nvidia took a $4.5 billion charge tied to excess H20 inventory and purchase obligations, reported $4.6 billion in H20 sales before the new requirements bit, and said another $2.5 billion of H20 product simply could not ship.

Nvidia chief executive Jensen Huang has been the most vocal critic of the policy, arguing repeatedly that Washington's curbs are forcing an American company to surrender China's $50 billion AI chip market to Huawei. From his vantage point, the strategy is self-defeating: the restrictions deny Nvidia revenue without denying China chips. Nvidia did not respond to Reuters' request for comment on the IDC findings.

Huawei's machine is accelerating

Huang's complaint gains weight when you look at what Huawei has built in the meantime. At Huawei Connect in Shanghai on September 17, the company pulled forward its Ascend roadmap: the Ascend 960DT will arrive in the first quarter of 2027, three quarters ahead of its previous schedule, with the 960PR following in the third quarter of 2027. Deputy chairman David Wang said Huawei is evolving the Ascend series on a one-generation-a-year cycle, a cadence that echoes the annual release rhythms of the GPU industry's leaders.

The company is also building at the systems level, which matches IDC's read of where competition is headed. Huawei reported more than 1,000 deployed Atlas 900 A3 SuperPods and large-scale commercial use of its Atlas 950 SuperPod, selling clusters rather than chips. Those deployment figures come from the company itself and should be read accordingly, but the direction is clear: Huawei is competing as an infrastructure vendor, not a component supplier. For context on how rivals are responding to the same instinct, Microsoft's Maia 300 push is another bet that AI accelerators are now a full-stack game.

Caveats, smuggling, and the measurement problem

The IDC figures deserve two grains of salt. The firm did not specify which Nvidia products made up the 34% share, or whether the data includes cutting-edge chips such as the B200 and B300 that are barred from export to China but have reportedly been smuggled into the country. Chip smuggling is not a footnote here: in early October, US and Taiwanese authorities separately indicted 11 people in two cases involving the illegal resale of US-manufactured semiconductor chips to China, including servers said to contain hundreds of millions of dollars in export-controlled Nvidia hardware. If smuggled cards are flowing, Nvidia's true footprint in China could be modestly larger than the legal-share figures suggest.

The other caveat runs in the opposite direction. IDC counts shipments, not installed performance, and a Huawei accelerator does not yet trade one-for-one with Nvidia's best on raw capability. But on the metric that matters for sovereignty, availability, the trend is unambiguous: nine of eleven tracked domestic suppliers grew, non-GPU accelerators now outship GPUs, and the system-level race is being fought in Chinese data centers with mostly Chinese silicon.

What comes next

For Nvidia, the question is whether China becomes a lost market or a recoverable one. The company still sells a third of the accelerators shipped there, and any softening of US policy could reopen compliant sales quickly. But each quarter of restricted supply is a quarter in which Chinese buyers rewrite their software stacks, qualify new hardware, and build clusters around domestic parts. Those switching costs harden with time.

For everyone else, the message is that the two-track semiconductor market is here to stay: a Western supply chain dominated by Nvidia and its would-be challengers, and a Chinese one increasingly led by Huawei with a growing supporting cast. Investors, policymakers, and chipmakers are now pricing in a world where the same AI boom is being built on two different sets of silicon, and the October 13 IDC numbers are the clearest evidence yet of how far the split has gone.