🎙 Listen to a summary of this story
Nvidia’s long-standing dominance of China’s artificial-intelligence chip market is weakening as Huawei and a widening field of domestic suppliers gain ground, offering some of the clearest evidence yet that US export controls are accelerating the creation of a separate Chinese computing ecosystem.
Nvidia held 34% of China’s data-centre AI-accelerator market in the first half of 2026, while Huawei reached 28%, according to IDC data reviewed by Reuters.
Nvidia shipped 678,000 accelerators in China between January and June, a decline of 44% from a year earlier. Huawei shipped 572,000, rising from slightly more than 300,000 during the same period in 2025.
The figures do not mean Huawei’s chips have matched Nvidia’s most advanced processors. But they show how restrictions intended to constrain China’s access to cutting-edge computing are also creating protected demand for domestic alternatives.
A market fragments beyond Nvidia
China received 2.008 million AI accelerators during the first half, according to IDC. Nvidia remained the largest individual supplier, but it now faces competitors extending beyond Huawei.
Alibaba’s T-Head chip division captured 7% of the market, followed by Cambricon with 5%, Baidu’s Kunlunxin with 4% and Hygon with 3%. Nine of the 11 Chinese manufacturers tracked by IDC increased shipments from the previous year.
That emerging field challenges the idea that China will replace Nvidia with a single national champion. Instead, different suppliers are building processors for cloud computing, telecommunications, government systems and specialised AI services.
The change follows years of US restrictions designed to prevent China from acquiring advanced chips that Washington says could support military and surveillance capabilities. Those controls have limited which Nvidia products may be sold legally in mainland China.
Nvidia CEO Jensen Huang has criticised the policy, arguing that it is pushing Chinese customers toward domestic technology while denying the American company access to a market he values at $50 billion.
Specialised chips overtake GPUs
The most consequential finding may concern the type of hardware China is buying. Shipments of graphics processing units, the general-purpose accelerators widely used to train AI models, fell 26% to 991,000 units.
Shipments of other accelerators rose almost 80% to 1.016 million, overtaking GPUs by volume. Huawei, T-Head, Kunlunxin and Cambricon supplied 95% of that non-GPU segment.
These processors are frequently designed for narrower tasks, including operating — or performing inference with — models that have already been trained. Their growth may indicate that China’s AI industry is moving beyond laboratory development toward deployment across commercial and public-sector systems.
Shipment counts alone cannot confirm that shift, however. One high-end training processor may provide more computing power than several specialised chips, making simple unit comparisons potentially misleading.
IDC said competition is increasingly moving from individual processor performance to system-level capability: how effectively chips, networking equipment and software work together inside large computing clusters.
Self-sufficiency has limits
Huawei has pursued that systems approach by connecting large numbers of Ascend processors to compensate for limitations in individual chips. Alibaba, Baidu and other companies are similarly combining proprietary hardware with their cloud platforms and AI software.
Yet China remains dependent on parts of the international semiconductor supply chain. Domestic companies face constraints in advanced manufacturing equipment and high-bandwidth memory, while Nvidia’s CUDA software remains deeply established among developers.
A recent Financial Times report said Chinese technology companies have continued buying Nvidia gaming and workstation processors, adapting products not originally designed for large AI systems. That persistent demand suggests domestic alternatives have not fully replaced Nvidia’s performance or software advantages.
The underlying IDC report is proprietary and its full methodology has not been made public. IDC did not identify which Nvidia products were included or whether its estimates captured restricted processors that entered China through unofficial channels.
Market share by shipments also does not measure revenue, usable computing power or model-training performance. Huawei’s growing volume therefore demonstrates commercial adoption rather than technological parity with Nvidia’s leading systems.
Even with those qualifications, the trend is difficult to dismiss. US controls have reduced Nvidia’s available product range, Chinese procurement policies favour local suppliers and domestic manufacturers are producing chips at greater scale.
The result is not an isolated substitute for Nvidia but an expanding network of Chinese processors, software and computing clusters. Whether that network can support frontier AI development remains uncertain. Its emergence as a durable commercial ecosystem is becoming much harder to dispute.
