NVIDIA's market share in China drops below 60%, with domestic AI chip deliveries reaching 1.65 million units annually, capturing market share.

Author: Deep Tide TechFlow

Deep Tide TechFlow’s Briefing: IDC data shows that in 2025, China’s total shipments of AI accelerator cards will be approximately 4 million units. Domestic vendors will deliver a combined 1.65 million units, accounting for 41%, while Nvidia’s share will drop from about 95% before sanctions to 55%.

Huawei will lead the domestic camp with 812,000 chips. Its newly released Atlas 350 accelerator card claims inference performance that is 2.87 times that of Nvidia’s H20.

Last November, Beijing ordered state-owned capital data centers to fully replace foreign products with domestic ones, accelerating the reshaping of the market landscape.

Three years ago, Nvidia nearly monopolized China’s AI chip market. Today, this landscape has been completely transformed.

According to Reuters, citing IDC data from a market research firm, in 2025 China’s total shipments of AI accelerator cards (dedicated computing chips for AI servers) will be about 4 million units. Nvidia remains the largest single supplier, shipping approximately 2.2 million units for a 55% share. However, this is a significant decline from the roughly 95% market share before the sanctions, shrinking nearly 40 percentage points. Meanwhile, Chinese domestic vendors will ship about 1.65 million units, capturing 41% of the market. AMD will ship around 160,000 units, ranking third with 4%.

The rise of domestic vendors is both a passive result of U.S. export controls and an active outcome of the “domestic replacement” policy.

Huawei Leads the Domestic Camp: Atlas 350 Targets Nvidia H20

Among the Chinese AI chip manufacturers, Huawei is the biggest winner.

IDC data shows that Huawei will ship about 812,000 AI chips in 2025, accounting for roughly 20% of the entire market and nearly half of domestic vendors’ shipments. Alibaba’s chip design division Pingtouge (T-Head) will rank second with about 265,000 units. Baidu’s Kunlun chips and Cambricon will each ship about 116,000 units, tying for third place. Additionally, Hygon, MetaX, and Iluvatar CoreX will account for 5%, 4%, and 3% of domestic vendors’ shipments, respectively.

Last month, at Huawei’s China Partner Conference 2026 in Shenzhen, Huawei announced its next-generation AI accelerator card, Atlas 350, equipped with its self-developed Ascend 950PR chip. Zhang Dixun, head of Huawei’s Ascend computing business, stated at the launch that Atlas 350 achieves 1.56 PFLOPS (quadrillions of operations per second) in FP4 low-precision computing, with performance 2.87 times that of Nvidia’s China-specific H20. The card is equipped with 112GB of self-developed high-bandwidth memory HiBL 1.0, with a memory bandwidth of 1.4TB/s and a power consumption of 600W.

However, this performance comparison has some definitional issues. Nvidia’s Hopper architecture GPUs do not natively support FP4 precision. Atlas 350 is the first domestically optimized accelerator card for FP4, so the two cannot be directly compared at the same precision level. Huawei’s true competitive advantage lies in inference: Atlas 350 is positioned for inference workloads during AI model deployment, not for large-model training.

Seven Huawei partners have already launched complete server products based on Atlas 350, and iFlytek has announced that its next-generation Spark large model will be adapted to the Ascend 910/950 computing platform.

Export Controls and Domestic Replacement Drive Dual Momentum

Nvidia’s shrinking share in China results from both the escalating U.S. export restrictions and Beijing’s proactive domestic replacement policies squeezing from both sides.

The timeline is roughly as follows: Starting October 2022, the U.S. restricted exports of AI chips to China. Nvidia responded by launching compliant, downgraded versions of H20 and products like A800/H800. In April 2025, the Trump administration fully banned all AI GPU exports to China; in July of the same year, export licenses for H20 and AMD MI308 were restored. In October, Nvidia CEO Jensen Huang publicly stated that Nvidia’s share in China’s advanced AI accelerator card market “fell from 95% to zero.” In December, Trump permitted Nvidia to export H200 to China, but Chinese companies were instructed to suspend orders for Nvidia chips.

On the other side, policy momentum is equally strong. According to a Reuters report in November 2025, Beijing issued guidance to newly built data centers using state-owned assets, requiring all to adopt domestic AI chips. Projects less than 30% complete were ordered to remove foreign chips or cancel procurement plans.

Reuters’ data shows that since 2021, China’s AI data center projects have received over $100 billion in state capital investment. Most Chinese data centers are in the construction phase and have received some form of state support, indicating broad policy coverage.

A flagship example is the large data center built by China Unicom in Qinghai, valued at $390 million, powered entirely by domestic AI chips such as those from Pingtouge.

The technology gap is real, but the inference side has already reached a “good enough” threshold

The increasing market share of domestic chips does not mean the technology gap has disappeared.

Most industry analysts estimate that China’s domestic AI chips still lag Nvidia by 5 to 10 years on the data-center training side. When training trillion-parameter-level large language models (LLMs), Nvidia’s high-end GPUs remain the first choice. A concrete example is the 50,000 Hopper-series GPU cluster used to train DeepSeek’s R1 model.

However, on the inference side, the situation is different. Industry observers believe that for 90% of commercial applications—including image recognition, chatbots, and autonomous driving—domestic chips have already reached the “good enough” threshold, making switching from Nvidia to domestic solutions a feasible business decision. Expectations of further sanctions have accelerated this shift.

The real bottleneck lies in the software ecosystem. Nvidia’s CUDA platform, developed over more than a decade, has become the de facto standard for AI development. Domestic chip companies are investing heavily in compatibility: MetaX announced that its C500 series will support CUDA compatibility; Huawei plans to fully open-source the CANN platform in 2025 to expand its developer ecosystem; Cambricon and Moore Threads have each developed translation tools from CUDA to their own programming languages. The pace of ecosystem development will determine the ceiling for domestic chip market share.

Domestic AI chip companies are rapidly rushing into capital markets

The shift in market share is also being reflected in capital markets.

Since early 2026, a wave of IPOs has swept China’s GPU sector. Biren Technology and MetaX have listed on the STAR Market; TianShu Zhixin has gone public on the Hong Kong Stock Exchange’s Main Board; Iluvatar’s application for listing on the STAR Market has been accepted. Baidu announced plans to spin off Kunlun for independent listing, and insiders say Alibaba is considering a similar split for Pingtouge (T-Head).

In 2025, Huawei’s R&D investment reached RMB 192.3 billion, accounting for 22% of revenue, focusing on chips, software, and manufacturing tools to further reduce dependence on U.S. technology. Huawei’s rotating chairman Xu Zhijun stated at MWC 2026 that Huawei will become an “alternative choice to ensure uninterrupted global AI computing capacity supply.” According to Reuters, Huawei’s next-generation Ascend 950PR chip has already attracted orders from giants like ByteDance and Alibaba. The shipment target for 2026 is about 750,000 units, with large-scale mass production starting in the second half of the year.

For Nvidia, even if the H200 has been approved for export to China, the trust foundation has been shaken. Beijing’s pursuit of independent controllability is no longer just a vision—it is a fact established by every domestic chip operating in data centers. When the 2026 market share data is released, whether the 55% figure will rebound or continue to decline will depend on whether Washington’s export policies shift again and how quickly domestic chips catch up on the training side.