NVIDIA has secured approval to resume selling its powerful H200 accelerators to China, marking a significant shift after years of export restrictions that nearly eliminated the company's presence in the world's second-largest AI market. However, the comeback comes with unusual conditions: a 25% levy on all China revenue, case-by-case government approval for each sale, and a competitive landscape transformed by Huawei and other domestic rivals who gained ground during the freeze. Whether NVIDIA can reclaim the billions it once earned in China remains uncertain.

What Happened to NVIDIA's China Business?

China was once a goldmine for NVIDIA. The company's data center revenue from China ran at roughly $12 billion annually before restrictions tightened, and CEO Jensen Huang estimated the total addressable opportunity in the country at around $50 billion. But starting in 2022, the US government imposed escalating export controls that cut off China's access to NVIDIA's most powerful chips, including the A100 and H100.

NVIDIA designed lower-performance alternatives specifically for China, most notably the H20. But in April 2025, the US required a license for H20 exports too, effectively halting sales overnight. The impact was devastating: NVIDIA took a $4.5 billion charge on excess H20 inventory and reported a 45% year-over-year drop in China revenue in the second half of 2025. For several quarters, the company reported essentially zero data center compute shipments to China.

Why Did Chinese Competitors Gain So Much Ground During the Freeze?

The export freeze created an opening that domestic rivals seized aggressively. Huawei's Ascend 910C, fabricated domestically, became the default training accelerator for many of China's largest AI labs. Other domestic players advanced rapidly too, with chip designer Cambricon reporting a 300% revenue surge in 2025, though from a small base. By early 2026, domestic champions like Huawei and Baidu controlled a large majority of China's home-grown AI cloud market.

Beijing's push for semiconductor self-sufficiency reinforced this shift, encouraging Chinese firms to buy domestic. In effect, the export controls intended to slow China's AI progress inadvertently accelerated the growth of the very competitors NVIDIA now faces. The restrictions handed Chinese firms a captive customer base and years of guaranteed demand, allowing a domestic industry to mature that might otherwise have struggled to compete with NVIDIA on the open market.

How Does the H200 Compare to Huawei's Chips?

On raw performance, NVIDIA's H200 retains a commanding lead. Its advanced memory offers roughly double the bandwidth of Huawei's Ascend 910C, and its FP8 compute performance is more than twice as fast, keeping it well ahead on specifications. Some analysts assess the H200 as more capable than anything Huawei plans to produce for at least two years, an eternity in fast-moving AI. China's domestic manufacturing capacity also remains a fraction of what NVIDIA can supply.

NVIDIA's other advantage is software. Its CUDA ecosystem is deeply entrenched in Chinese development teams, and maintaining access to NVIDIA hardware helps preserve that software dominance. For many Chinese developers, the cost of retraining teams and rebuilding software around a domestic chip is itself a reason to prefer NVIDIA hardware whenever policy allows them to buy it.

What Are the Conditions of the H200 Approval?

The reopening of China's market comes with strings attached. Each H200 sale requires individual government approval, and a reported 25% of the revenue from these China sales goes to the US government as a levy. This arrangement reduces NVIDIA's margins on those chips compared to sales elsewhere in the world. Despite the reduced profitability, NVIDIA judged the arrangement worthwhile, calculating that selling capable chips at a reduced margin beats ceding the market entirely to domestic rivals.

The approval framework represents a middle ground between complete market closure and unrestricted access. It allows NVIDIA to maintain a foothold in China while giving the US government oversight and financial benefit from the arrangement. However, the levy and approval requirements make this a more complicated and less profitable channel than NVIDIA's business in other regions.

How to Assess NVIDIA's China Market Recovery Prospects

• Demand Uncertainty: Even with the H200 approved, it is not certain Chinese customers will buy in large volumes, since Beijing has discouraged reliance on US chips and Chinese authorities had not immediately cleared the shipments at the time of approval.
• Political Headwinds: Some buyers may prefer to support domestic suppliers for strategic reasons, while others will want the H200's superior performance for demanding workloads. The outcome will depend on a mix of politics, price, and how much the performance gap matters to each customer.
• Market Share Recovery Challenge: The market share lost to Huawei during the freeze does not automatically come back simply because licenses are granted. NVIDIA must effectively win back customers it lost, competing not only on chips but against a political current that favors home-grown technology.

For NVIDIA, a full return of Chinese demand would add meaningful revenue, but it is far from guaranteed. The scale of any recovery is genuinely hard to predict given the geopolitical complexity and the entrenched position of domestic competitors.

What Does This Mean for the Broader AI Chip Market?

The China situation illustrates a broader tension in the AI hardware industry: the collision between commercial demand, geopolitical strategy, and the rise of alternative chip makers. NVIDIA's dominance in AI accelerators remains unmatched globally, but the company's inability to freely access China, the world's second-largest AI market, has created space for competitors to build scale and improve their designs. The H200 approval represents a partial thaw, but the 25% levy and case-by-case approval process signal that access to China will remain conditional and constrained for the foreseeable future.

The stakes extend beyond NVIDIA's quarterly earnings. A strong presence in China shapes global standards and keeps NVIDIA's software ecosystem embedded in the world's largest pool of AI developers. Losing that influence entirely would reshape the competitive landscape for AI chips and software for years to come. The reopened market offers NVIDIA a chance to preserve its position, but success is far from assured.