Nvidia falls to 34% of China's AI chip market as Huawei takes 28%
An IDC shipment ledger seen by Reuters puts Nvidia at 34% of China's data-center AI chips in the first half of 2026, on shipments 44% below a year earlier. Non-GPU accelerators outsold GPUs for the first time, and Huawei took 28% of a market that kept growing.
Richard Tang · 3 min read
The shipment ledger IDC, the market-research firm, compiled for China's data-center AI chips in the first half of 2026, seen by Reuters and published on October 13, settles the argument over America's export controls 1. Nvidia's−0.25% — Nvidia, down 0.25 percent today share fell to 34%, on 678,000 accelerators shipped, 44% fewer than a year earlier 12. Jensen Huang, Nvidia's chief executive, has argued repeatedly that Washington's rules force his company to cede China's $50 billion AI chip market to Huawei 2. The ledger shows the ceding is well advanced.
The growth moved onto silicon Nvidia cannot ship
Total accelerator shipments in China reached 2.008 million units in the first half, so the market itself kept growing 1. GPUs, the chips most widely used to train AI models and the category Nvidia dominates, fell 26% to 991,000 units, squeezed by export rules and by Beijing's requirement that companies obtain approval to import even compliant Nvidia chips 12.
Every other kind of accelerator, silicon often tuned to running already-trained models, rose almost 80% to 1.016 million units and passed GPUs by volume for the first time in IDC's reported period 127. Huawei, the Chinese equipment maker, shipped 572,000 units against just over 300,000 a year earlier, taking 28% of the whole market 1. Huawei, Alibaba's+0.40% — Alibaba, up 0.40 percent today chip-design unit T-Head, Baidu's−0.29% — Baidu, down 0.29 percent today Kunlunxin unit and Cambricon, a Chinese AI-chip designer, together hold 95% of that non-GPU segment, and nine of the eleven Chinese chipmakers IDC tracks grew shipments 12.
Nvidia's China lead narrows to 34%, with Huawei at 28%
Data
| Value | |
|---|---|
| Nvidia | 34% |
| Huawei | 28% |
| T-Head | 7% |
| Cambricon | 5% |
| Kunlunxin | 4% |
| Hygon | 3% |
Huawei's answer is to wire weaker chips into one machine
IDC says competition in China is shifting from the performance of a single chip to how well chips work inside large computing clusters 1. That is Huawei's play exactly. Its Peerium architecture uses UnifiedBus links to make individually weaker domestic chips behave as one machine, and its Atlas 950 SuperCluster is built to connect up to 256,000 accelerator cards 4. The next Ascend chip, the 960DT, has been pulled forward to the first quarter of 2027 from a planned third-quarter debut 45.
The binding constraint on China's build has therefore moved from American supply to Huawei itself. TechRepublic, citing Reuters, reports the company cannot produce enough AI computing equipment to meet Chinese demand 5. Its Ascend software ecosystem counts about 5,270 monthly active developers 6, against the years of tooling Nvidia's CUDA platform has accumulated 5.
Nvidia's China demand has learned to buy elsewhere
Nvidia has already booked this door closing once, with a $4.5 billion charge on China-bound H20 inventory, the chip it designed for China, after Washington imposed license requirements in April 2025 3. The customers it lost in the first half bought domestic silicon instead, under a Beijing approval regime that still gates Nvidia imports 1. Winning them back would take more than a policy reversal.
Two numbers frame what to watch next. IDC expects China's total accelerator shipments to grow 19% this year even as GPU shipments fall 9% 17. And the Ascend 960DT's arrival in the first quarter of 2027, pulled forward from the third, is the next data point on whether Huawei can supply the growth it is being handed 45.
Deepdive
AI-generated from this story and its cited sources. Not investment advice.
