Nvidia’s H200 returns to China, but its market share keeps shrinking

Small batches of Nvidia’s H200 AI accelerators have reportedly started arriving in China following the relaxation of Washington’s licensing rules. This development comes at an uncomfortable time for Nvidia since, although Chinese demand for its products is high, the company’s share in the world’s second-largest chip market is dropping below ten percent.

The situation has effects that extend beyond Nvidia. The company’s GPUs enable much of the training of some of the world’s most sophisticated AI models, which means that decisions regarding who gets access to them are going to shape where the next generation of AI technology is developed.

Case-by-case licenses replaced the outright ban

Earlier, the Bureau of Industry and Security of the Department of Commerce announced that export applications for Nvidia’s H200, AMD’s MI325X, and similar processors would be considered individually rather than being banned altogether. This shift happened after President Trump announced on December 8, 2025, that approved Chinese clients could purchase H200 chips if they abide by certain rules.

Exporters will continue to face stringent regulations. It is essential that they convince the authorities that their sales will not lead to a reduction in supply for American clients. In addition, it is required that the Chinese buyers follow export regulations and conduct customer verification procedures, while the chips undergo independent testing in the US.

“Export controls should evolve with changes in technology, while protecting national security,” Under Secretary of Commerce for Industry and Security Jeffrey Kessler said.

Beijing has set its own limitations. Cryptopolitan reported in July that companies such as Alibaba, ByteDance, and DeepSeek were poised to receive the chips, but authorities would only allow orders of about 200,000 H200 chips to be fulfilled, which is significantly less than requested, and they are also limited to AI training on public data only and not for inference or any use in sensitive workloads.

Nvidia budgeted for zero China revenue

Nvidia’s projections are not predicated on a substantial recovery in China.

On May 20, the company announced its results for the first quarter of FY 2027. The company’s revenue reached an unprecedented $81.6 billion, representing an increase of 85% compared to the previous year, while the Data Center revenue reached $75.2 billion. Nevertheless, the management’s guidance for the following quarter, which is around $91 billion, did not incorporate any Data Center compute sales to China.

CEO Jensen Huang has opened up about the reason.

“Huawei is very, very strong,” he said in May, and he claimed that Nvidia had “largely conceded that market to them” after enduring years of increasing restrictions in the US.

For a company at the hub of global AI infrastructure spending, not including China in their forecast shows how much the competitive landscape has changed.

Domestic chips are taking the budget

In a survey cited by TrendForce, executives said Chinese-made chips are expected to account for 46% of their AI accelerator spending over the next year, up from about 30% today.

Bernstein made an even bolder estimate that Nvidia’s share of the AI chip market in China could drop to about 8% by 2026 from nearly 40% just a year before, while Huawei rises above 50%.

The change is being facilitated by government-funded investment. TrendForce noted that China intends to invest around 2 trillion yuan (about $294 billion) in data centers over the next five years, with at least 80% of the core technology, including chips, expected to come from local manufacturers.

This allows Huawei and other Chinese chip manufacturers to have a larger domestic market to develop their products, while also making big buyers, like Tencent and Alibaba, to build more of their artificial intelligence systems using domestic equipment.

Controls that helped build a rival

A March 2026 CSIS commentary found that US and allied export controls, first imposed in 2022, did slow China’s progress at the technological frontier. But the restrictions also gave Beijing’s semiconductor self-reliance push new urgency and created a captive customer base for domestic chipmakers.

At the same time, the gap in performance that those controls intended to safeguard is much smaller now. The Stanford AI Index 2026 found that the gap between the top AI models from America and China is almost gone; as of March 2026, the American model is ahead only by 2.7%.

That means that the shipments of H200s are much more important for the information that they convey than for their total numbers. While Nvidia might be able to have some sales in China again because of new permits, years of bans had already directed clients to local products. America might manage to open the door a bit, but fixing the ecosystem created during its closure will be much more difficult.