Nvidia's China AI chip share falls to 34% as Huawei closes gap to six points

Nvidia's lead narrows to six points

Nvidia's share of China's data-center AI chip market fell to 34% in the first half of 2026, according to an IDC report seen by Reuters and covered by dev.to. The company shipped about 678,000 AI accelerators in mainland China between January and June, a 44% decline compared with the same period a year earlier.

Huawei shipped 572,000 units over the same window, close to double the just-over-300,000 units it shipped a year before, and now holds 28% of the market. That leaves a six-percentage-point gap between the two companies in a market Nvidia once counted among its biggest growth opportunities.

Export controls meet a state-backed push

The dev.to write-up attributes the shift to two overlapping forces. Washington's export control regime has progressively blocked Nvidia's most advanced data-center GPUs from China, while Beijing's well-funded drive for semiconductor self-sufficiency has turned domestic silicon into a commercially credible option.

Huawei has moved fastest to fill the space. In September it unveiled more than ten new chipsets spanning AI accelerators, general processors and connectivity components, and it is working with partners to build software alternatives to Nvidia's CUDA stack, the ecosystem that has long been the company's strongest lock-in.

The change is broad rather than a single-vendor story. Of the eleven Chinese semiconductor suppliers IDC tracks, nine increased shipments year over year. Alibaba's T-Head unit held 7% of the market, followed by Cambricon at 5%, Baidu's Kunlunxin at 4% and Hygon at 3%.

Non-GPU accelerators outsold GPUs

Total AI accelerator shipments in China reached 2.008 million units in the first half of 2026, and IDC projects 19% growth for the full year, so demand is not the constraint. What changed is the mix: GPU shipments fell 26% to 991,000 units, while non-GPU accelerators built for inference and specific model workloads jumped nearly 80% to 1.016 million. For the first time, non-GPU parts outsold GPUs by total volume in China, which suggests buyers are designing infrastructure around purpose-built domestic silicon rather than hunting for substitutes for restricted Nvidia hardware.

There are caveats. IDC did not specify which Nvidia products make up its 34% share, or whether the figure includes restricted chips such as the B200 and B300 that have reportedly been smuggled into the country. Nvidia did not respond to a request for comment, according to the dev.to report.

Capex keeps climbing

The build-out behind these numbers is enormous. According to figures cited by dev.to, combined capital spending by Alibaba, ByteDance and Tencent has more than doubled to roughly RMB 700 billion, about $104 billion, and reports indicate Chinese firms have bought more than a million Nvidia gaming and workstation chips this year to plug compute shortfalls. Demand remains insatiable; the supply is being redirected.

Why it matters

The data suggests export controls are accelerating the outcome they were partly meant to prevent: a domestic Chinese accelerator industry with real shipment volume and a widening product line. For Nvidia, losing ground in what was expected to be one of its largest growth markets adds pressure to its data-center revenue story. For the wider industry, the crossover point where inference-oriented non-GPU chips outsell GPUs signals that AI compute is fragmenting into regional ecosystems, each with its own hardware and software stacks. If Huawei closes the remaining six points, or takes the lead outright, it will confirm that the CUDA moat is contestable at scale.

What to watch

The second half of 2026 should answer three questions: whether Nvidia's share stabilizes as it ships whatever compliant products it can, whether Huawei overtakes it in its home market, and whether non-GPU accelerators extend their volume lead over GPUs. Any of those outcomes would mark a structural change in how AI compute is built and bought, not just a bad quarter for one vendor.