Nvidia retained a commanding 55% share of China’s artificial intelligence chip market in 2025, shipping 2.2 million AI accelerators to Chinese customers despite the progressive tightening of US export controls, according to IDC data reviewed by Reuters. The figures present a more nuanced picture of Nvidia’s position in China than the company’s own public statements have suggested: Nvidia CEO Jensen Huang has stated that the company’s China market share effectively dropped to zero following the implementation of US restrictions, a characterization that the IDC data directly contradicts.
The 2.2 million units shipped by Nvidia in 2025 represent a significant commercial presence, even if the products involved are tailored versions of the company’s hardware designed to comply with US export regulations. The H20 chip, developed specifically for the Chinese market to meet the performance thresholds permitted under US export rules, has been the primary product driving these shipments. Despite its reduced specifications relative to Nvidia’s flagship H100 and H200 accelerators, the H20 has found substantial demand among Chinese cloud providers and AI companies who value the chip’s compatibility with the CUDA software ecosystem that underpins the vast majority of global AI development.
(Related: China’s Domestic Chipmakers Seize 41% of the Local AI Market)
Huawei Leads Domestic Challengers
Among Chinese chip suppliers, Huawei Technologies emerged as the dominant domestic alternative, shipping 812,000 AI accelerator units in 2025 — approximately 20% of the total market. Alibaba Group Holding ranked second among domestic providers, shipping approximately 265,000 units through its T-Head semiconductor division. Together, Chinese domestic suppliers accounted for 1.65 million units, representing the remaining 45% of the market not captured by Nvidia.
The distribution of domestic market share is significant. Huawei’s 812,000 units represent a substantial ramp from earlier periods and reflect the growing adoption of Ascend series chips by Chinese cloud providers and AI developers who are either mandated to use domestic alternatives under Xinchuang procurement rules or who are proactively diversifying their supply chains in anticipation of further US restrictions. Alibaba’s 265,000 units from its T-Head division reflect the company’s strategy of developing proprietary AI chips for internal use and selective external deployment.
(Related: Huawei Ascend 910C Demand Surges 20% as DeepSeek V4 Pushes to Break CUDA Dependence)
The CUDA Moat and Software Ecosystem Lock-In
Nvidia’s ability to maintain a 55% market share despite export restrictions reflects the depth of its software ecosystem advantage. The CUDA programming framework, developed by Nvidia over more than 15 years, is the foundation on which virtually all commercial AI software is built. Migrating large-scale AI training and inference workloads from CUDA to alternative software stacks, such as Huawei’s CANN framework or the open-source ROCm ecosystem, requires significant engineering effort and carries performance risks. Many Chinese AI companies have concluded that the cost of migration outweighs the benefits, particularly when Nvidia’s export-compliant H20 chip remains available.
This dynamic is changing, however. The National Development and Reform Commission’s recent directive guiding domestic large language models to better adapt to homegrown chips represents a policy-level push to accelerate migration away from CUDA. Simultaneously, Huawei has been investing heavily in improving the developer experience for its CANN framework, and several major Chinese AI labs, including DeepSeek, have publicly committed to optimizing their models for Huawei Ascend hardware.
Long-Term Trajectory and the Post-Nvidia Scenario
The IDC data captures a market in transition. Nvidia’s 55% share in 2025 almost certainly represents a peak: the combination of policy pressure, improving domestic alternatives, and the possibility of further US export restrictions makes it difficult to sustain that level of market penetration over the medium term. JPMorgan’s April 2026 projection of 40% compound annual growth in China’s indigenous AI chip demand through 2030 implies a substantial shift in market share toward domestic suppliers over the next four years.
Huawei’s announcement of its Tau Scaling Law and LogicFolding architecture, covered separately this week, signals the company’s ambition to close the performance gap with Nvidia’s most advanced products by 2031. If Huawei succeeds in delivering chips that match the performance of 1.4-nanometer technology without relying on Western manufacturing tools, the rationale for Chinese companies to maintain CUDA-dependent workflows would diminish significantly. The 2025 market share data thus represents a baseline against which the effectiveness of China’s semiconductor independence strategy will be measured in the years ahead.
It is also worth noting that the IDC figures cover units shipped, not revenue. Nvidia’s H20 chip, while export-compliant, commands a significant price premium over comparable domestic alternatives, which means Nvidia’s revenue share in China’s AI chip market is almost certainly higher than its 55% unit share suggests. For Nvidia’s investors, the persistence of meaningful China revenue despite export restrictions is a more positive signal than the company’s own public statements have implied. For US policymakers, it raises questions about the effectiveness of the current export control regime in limiting China’s access to advanced AI computing capabilities.