Small batches of Nvidia’s H200 AI accelerators, a Hopper-generation chip that trails the company’s newer Blackwell systems, have reportedly begun arriving in China after Washington relaxed its licensing rules. The move comes at an awkward moment for Nvidia: Chinese demand for its products remains strong, but its share of the world’s second-largest chip market is falling below 10%.

The implications reach beyond Nvidia. The company’s GPUs power much of the training behind some of the world’s most advanced AI models, meaning decisions about access to those chips will help shape where the next generation of artificial intelligence is built.

Case-by-case licenses replaced the outright ban

The Bureau of Industry and Security at the U.S. Department of Commerce earlier said export applications for Nvidia’s H200, AMD’s MI325X, and similar processors would be reviewed individually rather than barred outright. The shift followed President Trump’s December 8, 2025 announcement that approved Chinese customers could buy H200 chips if they met certain conditions. It is the latest turn in a rolling set of rules: earlier restrictions had already pushed Nvidia to sell China reduced-specification chips such as the H20, which Washington in turn subjected to licensing requirements in 2025.

Exporters still face strict requirements. They must show regulators that sales will not reduce supply for American customers. Chinese buyers are also required to comply with export rules and customer verification procedures, while the chips must undergo independent testing in the United States.

“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 imposed its own constraints. Cryptopolitan reported in July that companies such as Alibaba, ByteDance, and DeepSeek were positioned to receive the chips, but Chinese authorities would allow only about 200,000 H200 chips to be fulfilled, far fewer than requested. The chips are also restricted to AI training on public data, not inference or any sensitive workloads.

Nvidia budgeted for zero China revenue

Nvidia is not counting on a major rebound in China. On May 20, the company reported first-quarter FY 2027 results, with revenue reaching a record $81.6 billion, up 85% from a year earlier. Data Center revenue totaled $75.2 billion.

Even so, management’s guidance for the next quarter, about $91 billion, did not include any Data Center compute sales to China.

CEO Jensen Huang has explained why. In May, he said, “Huawei is very, very strong,” and said Nvidia had “largely conceded that market to them” after years of tightening U.S. restrictions.

For a company at the center of global AI infrastructure spending, excluding China from its forecast underscores how much the competitive landscape has changed.

Domestic chips are taking the budget

A TrendForce-cited survey found that executives expect Chinese-made chips to account for 46% of AI accelerator spending over the next year, up from about 30% today.

Bernstein offered an even sharper estimate, saying Nvidia’s share of the AI chip market in China could fall to about 8% by 2026 from nearly 40% just a year earlier, while Huawei, whose Ascend 910-series accelerators have become the main domestic alternative, rises above 50%.

Government-funded investment is helping drive the shift. TrendForce said China plans to invest about 2 trillion yuan, or roughly $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.

That gives Huawei and other Chinese chipmakers a larger domestic market in which to develop their products, while also pushing major buyers such as Tencent and Alibaba to build more of their AI systems with domestic equipment.

Controls that helped build a rival

A March 2026 CSIS commentary found that U.S. and allied export controls, first imposed in October 2022 to block the A100 and H100 and later tightened to close off downgraded China-specific variants such as the H800, did slow China’s progress at the technological frontier. But the same restrictions also intensified Beijing’s semiconductor self-reliance drive and created a captive customer base for domestic chipmakers.

At the same time, the performance gap those controls were meant to preserve has narrowed significantly. The Stanford AI Index 2026 found that the gap between the top AI models in the United States and China has nearly disappeared; as of March 2026, the American model led by just 2.7%.

That makes H200 shipments more important for the signal they send than for their volume. Nvidia may be able to resume some sales in China under new permits, but years of restrictions have already pushed customers toward local alternatives. The U.S. may be reopening the door, but rebuilding the ecosystem that formed while it was closed will be much harder.