China wants its tech giants to run on homegrown silicon. Its top AI developers keep reaching for Nvidia anyway.
That tension sits at the center of Beijing’s push for semiconductor self-sufficiency. US export controls have made the effort urgent, and the gap between selling chips and changing habits is proving wider than policymakers might like.
Huawei is winning on volume, not on habits
Huawei, China’s leading domestic chip player, claims its Ascend AI chips have outsold Nvidia within China as of October 2026. For a company operating under heavy US restrictions, that is a notable milestone.
Eric Xu, Huawei’s rotating chairman, has framed the company’s pitch around dependable domestic supply rather than raw peak performance. The argument is simple: a chip you can actually buy beats a faster one you might not be allowed to get.
Many of China’s leading AI developers still rely on Nvidia hardware to train advanced large language models, the expensive, compute-hungry process of building an AI system from scratch.
DeepSeek offers the clearest example. The company plans to deploy at least 160,000 Huawei Ascend 950DT chips for inference, while keeping Nvidia hardware for its training workloads.
The software moat nobody can chip away at quickly
The stickiness comes down to software. Nvidia’s CUDA platform is the toolkit that developers use to write code that runs efficiently on Nvidia chips, and it has become deeply entrenched across the AI industry.
Huawei offers its own alternative, called CANN. A transition for training workloads is predicted to carry roughly 50% higher costs and time.
Inference deployments appear to be rolling out with more agility, which helps explain why Huawei’s sales numbers can climb even as Nvidia keeps its foothold at the top end.
Washington’s controls are shaping the board
All of this plays out against tight US restrictions on advanced chip exports to China. Enforcement has teeth. US authorities have arrested individuals involved in smuggling Nvidia chips into China, with cases still ongoing as of October 2026.
At the same time, Washington has granted limited approvals for certain Nvidia models despite the broader restrictions. The result is a patchwork: some Nvidia hardware moves legally, some moves illegally, and Chinese firms plan around the uncertainty of both.
What the market share math could look like
The projections are striking. Domestic AI chipmakers could capture approximately 50% of China’s AI accelerator market by 2026.
Nvidia’s share, meanwhile, may fall to around 8% in 2026, down from about 40% in 2025. Both things can be true at once: Nvidia could lose most of the unit volume in China while still powering a disproportionate share of the country’s frontier model training.
What this means
For Nvidia, the risk is less about one quarter’s revenue and more about the long game. China has been a lucrative AI market, and losing its dominant position there could ripple through the company’s global strategy and profitability.
For Huawei, winning inference is a meaningful foothold. Winning training, where switching costs are highest and performance matters most, is the harder prize, and the source of Nvidia’s remaining leverage.
Investors watching the sector should track a few signals. First, whether more top Chinese labs announce training runs on domestic hardware, which would indicate the CUDA moat is narrowing where it counts. Second, how Washington handles its limited Nvidia approvals going forward. Tighter rules would accelerate the shift to Huawei, while looser ones could give Chinese labs a reason to stay put. Third, whether the projected collapse in Nvidia’s China share actually materializes in 2026. A drop from about 40% to around 8% would reshape not just one company’s China business but the broader map of global semiconductor supply chains.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.