Nvidia concedes China AI chip market to Huawei

Nvidia has largely conceded China’s advanced artificial-intelligence chip market to Huawei, chief executive Jensen Huang said — the starkest acknowledgment so far that Washington’s export controls are reshaping the global AI contest.

“We’ve really largely conceded that market to them,” Huang told CNBC in an interview published Wednesday. “The demand in China is quite large. Huawei is very, very strong. They had a record year, they’ll likely, very likely, have an extraordinary year coming up, and their local ecosystem of chip companies are doing quite well, because we’ve evacuated that market.”

We’ve really largely conceded that market to them.
— Jensen Huang, Nvidia chief executive

Months of corporate hedging have now hardened into a plain loss. Nvidia’s share of China’s AI accelerator market has collapsed from roughly 95 per cent in 2023 to near zero, according to Bernstein analysis. Over the same period, Huawei’s share has surged from nothing to roughly half the market, and the Chinese company is now forecasting $12bn in AI chip revenue for 2026 — a 60 per cent jump from $7.5bn a year earlier.

The numbers understate what is happening. Analysts at Morgan Stanley project China’s AI chip market will reach $67bn by 2030, with 86 per cent of that captured by domestic suppliers. Export controls designed to contain China’s AI capabilities are instead accelerating a parallel semiconductor supply chain — one built around Huawei’s Ascend processors rather than Nvidia’s graphics processing units.

Inside the supply-chain shift

That shift is already taking shape. DeepSeek, the Chinese AI lab whose models have repeatedly matched Western benchmarks, optimised its latest V4 model for Huawei’s Ascend architecture, not Nvidia’s CUDA platform. ByteDance, the parent of TikTok, is now Huawei’s largest Ascend chip customer. Chinese hyperscalers are building their AI infrastructure on Huawei’s CANN software framework — a direct alternative to the CUDA stack that has been Nvidia’s unbreachable competitive moat for more than a decade.

The software is where the lasting damage sits. Each year that Chinese AI developers build on Ascend and CANN deepens the divergence from CUDA, making it steadily harder for Nvidia to reclaim the market even if export controls were lifted tomorrow. Huang’s remark that “their local ecosystem of chip companies are doing quite well, because we’ve evacuated that market” reads like a warning dressed as an observation.

The door is shut from both sides

Nvidia’s path back to China depends on a policy reversal neither government appears willing to deliver. President Trump’s administration cleared 10 Chinese firms to purchase Nvidia H200 chips, the most advanced AI processor the US is willing to export under current rules. Beijing blocked the imports anyway. Trump said China “chose not to” approve the purchases because it wanted to develop its own chip industry.

Huang has told investors to expect nothing. “I don’t have any expectation, which is the reason why we put all of our guidance, all of our numbers, all the expectations that I’ve set with all of our analysts and investors to invest nothing, to expect nothing,” he said. Nvidia wrote off $4.5bn on export-restricted chips in the first quarter of its 2026 fiscal year.

Huawei sceptics warn against reading the revenue figures as a straightforward market victory. A Council on Foreign Relations analysis argues that much of the $12bn target may reflect government-directed procurement rather than organic commercial demand, and that Huawei’s chip fabricator SMIC remains stuck at a 7-nanometre process node — several generations behind TSMC’s leading edge. Huawei’s own roadmap reportedly shows its next-generation chip being less performant than the current one, a sign that SMIC’s process ceiling is binding.

Nvidia’s answer: a new $200bn market

Nvidia is not waiting. On the same call where Huang conceded the China market, he unveiled the Vera central processing unit — a new chip architecture that he claimed opens a $200bn total addressable market. The company has already booked $20bn in standalone Vera sales this year. Nvidia posted $81.62bn in quarterly revenue, up 85 per cent year on year, even as China evaporated from its sales ledger.

The pivot reframes Nvidia’s growth story around agentic AI infrastructure — data centres that run autonomous reasoning workloads — rather than around the Chinese hyperscalers that once drove its data-centre segment. It is a bet that demand everywhere else will grow faster than China’s loss can subtract.

Two stacks, two futures

For Beijing, the calculus is blunt. Huang had long argued that it was better for Chinese AI to run on an American technology stack than for a total ban to nurture homegrown competitors. The argument has now been tested and, from China’s perspective, vindicated: a ban created the competitor anyway, and the competitor now holds half the domestic market. An analysis by Rest of World noted earlier this month that the version of US-China relations sold to investors is not the one major American technology companies are actually operating with.

Whether the split can be contained is an open question. Morgan Stanley’s forecast of a $67bn Chinese AI chip market dominated by domestic suppliers by 2030 suggests the separation is structural, not cyclical. Beyond that sits a harder question: whether a world divided into two AI chip camps — one on CUDA for the West, the other on CANN for China — ends up producing two different kinds of artificial intelligence, each shaped by the geopolitical assumptions of the bloc that built it.