On December 23, Reuters, citing multiple sources familiar with the matter, reported that NVIDIA aims to export its H200 chips to China before the Year of the Horse Lunar New Year—February 17, 2025—with an initial shipment volume estimated between 40,000 and 80,000 units, primarily drawn from existing inventory.
Additionally, NVIDIA has already planned to ramp up H200 production to meet Chinese market demand in Q2 2025.
As of publication, NVIDIA has not officially responded to these reports.
However, significant uncertainty remains over whether the H200 will be successfully exported to China. One key factor is that there has been no indication so far that Chinese authorities have approved any procurement plans related to the H200.
Since the Trump administration imposed chip export restrictions on China in April this year, NVIDIA products have gradually withdrawn from the Chinese market. Subsequently, controversy over alleged "security backdoors" further clouded the company’s business prospects in China. CEO Jensen Huang stated that NVIDIA’s AI chip market share in China had “plummeted from 95% to zero.”
A major breakthrough finally emerged on December 8, when Trump announced via social media that, under the premise of ensuring U.S. national security, NVIDIA would be allowed to deliver H200 chips to China—but required to remit 25% of sales proceeds to the U.S. government.
Although subject to a hefty tax, the approved product for export has upgraded from the less competitive China-specific H20 to the H200—a significant signal that NVIDIA may be returning to the Chinese market.
At the time, NVIDIA publicly responded, calling the move to supply H200 to commercial customers “a positive step.”
On the policy front, Reuters reported last week that the Trump administration has formally initiated a review of H200 exports to China, with the U.S. Department of Commerce submitting the export license application to the State Department, Department of Energy, and Department of Defense.
While the review process is not public, it is expected to take no more than 30 days. Ultimately, Trump—who has shown a relatively favorable stance toward NVIDIA—holds the final decision-making authority.
Several key points remain worth watching regarding NVIDIA’s future in the Chinese market.
Can They Actually Sell?
Reports indicate that U.S. reviews have accelerated, but opposition persists within Congress and government agencies.
When the news broke, Democratic Senator Elizabeth Warren and Representative Gregory Meeks publicly urged the U.S. Department of Commerce to disclose more details about the license to assess “whether the approved NVIDIA chips could be used for military purposes” and to monitor reactions from U.S. allies and partners.
Earlier reports even suggested U.S. officials demanded NVIDIA install trackers on exported chips—an arrangement that could reignite “backdoor” suspicions.
On July 31, China’s Cyberspace Administration summoned NVIDIA over security risks related to alleged backdoors in the H20 chip. NVIDIA emphasized that it “does not and will never include backdoors or surveillance software.” However, no further official updates on the matter have been released.
Against the backdrop of rising domestic chip capabilities and strong policy support for localization, it remains uncertain whether NVIDIA can swiftly regain a foothold in China.
Nonetheless, multiple signals suggest that Chinese authorities maintain an open attitude toward importing advanced AI chips.
When Trump announced the potential approval of H200 exports to China this month, Foreign Ministry spokesperson Guo Jiakun stated, “We have noted relevant reports. China consistently advocates mutually beneficial cooperation between China and the U.S.”
Moreover, AMD Chair and CEO Lisa Su recently led a senior executive team on a visit to China, meeting with Chinese clients and officials—apparently laying groundwork for AMD’s own return to the Chinese market.
Who Will Be the First Buyers?
It is almost certain that the first H200 customers will be Chinese internet giants such as Alibaba, ByteDance, and Tencent.
Government-led projects like China’s intelligent computing centers and IT innovation (Xinchuang) sectors have rapidly completed domestic substitution, especially in inference chips, where local alternatives have already demonstrated validated performance. With strong policy backing, cost barriers for domestic chips are expected to ease in the short term.
In fact, internet tech giants fiercely competing in large model training and AI infrastructure remain highly dependent on NVIDIA’s offerings. Following Trump’s announcement permitting H200 exports, numerous reports indicated that major Chinese firms have already engaged NVIDIA on potential purchases.
At the time H200 clearance was announced, Alibaba, ByteDance, and Tencent all expressed procurement interest and submitted detailed explanations to relevant regulatory bodies.
Furthermore, reports suggest that some Chinese companies, constrained by Trump-era bans, have resorted to procuring NVIDIA chips through Japan, Singapore, and other channels—or directly deploying them overseas for model training—highlighting the intense market demand for NVIDIA’s chips.
How Much Revenue Can Be Generated?
By 2025, NVIDIA’s mainstream U.S. shipments will have transitioned to the Blackwell series, with H200 production scaled back and pricing volatile. The final selling price for the Chinese market remains unclear.
Based on rough market estimates of 20,000–40,000 per card—and accounting for the mandatory 25% remittance to the U.S. government—the initial shipment could generate between 1billionand4 billion in revenue.
NVIDIA’s Q3 earnings report previously disclosed that Hopper-series chip sales in China totaled only $50 million that quarter, with major orders unrealized.
NVIDIA’s current Q4 FY2025 (November 2025–January 2026) guidance excludes any China data center revenue. In FY2024, NVIDIA’s China revenue reached $17.1 billion. In 2025, amid NVIDIA’s withdrawal, multiple institutions project China’s AI chip market to grow by approximately 30% year-over-year.
Does H200 Still Hold an Edge?
In terms of performance and specifications—especially for large model training—the H200 still holds a clear advantage over domestic alternatives, particularly in memory capacity and bandwidth, reflecting a generational gap.
More importantly, most Chinese large models were developed on NVIDIA’s GPU architecture and deeply integrated into the CUDA ecosystem. Using H200 directly would significantly reduce migration costs and compatibility risks.
However, domestic chip performance continues to improve. On December 20, Moore Threads—dubbed “China’s first GPU stock”—unveiled its new “Huagang” GPU architecture, boasting substantially enhanced compute density and energy efficiency.
Moore Threads claims its new AI training-and-inference chip “Huashan,” based on this architecture, surpasses NVIDIA’s Hopper series across key metrics including floating-point performance, memory bandwidth, memory capacity, and high-speed interconnect bandwidth.
Zhang Jianzhong, founder, chairman, and CEO of Moore Threads—and former NVIDIA global vice president and Greater China general manager—stated confidently: “If you trained your models on Hopper before, switching to our new product today will yield better, not worse, results.”
Moore Threads is aggressively building its MUSA ecosystem to challenge—and potentially replace—NVIDIA’s CUDA.
Indeed, Huang has long recognized that exporting downgraded AI chips to China is unwise and now admits he is “unsure whether China will even accept the H200.”
Is AMD Gaining the Upper Hand?
While all eyes are on NVIDIA’s H200, AMD and Intel are also targeting the Chinese market. Both companies fall under the same export rules recently announced by Trump.
As the world’s second-largest GPU vendor, AMD is equally eager to re-enter China. In 2024, China accounted for 24% of AMD’s total revenue; however, its China business also froze starting Q2 2025.
In November, Lisa Su revealed that AMD’s MI308 AI chip had already secured a U.S. export license and that the company was prepared to pay the required tax for Chinese shipments.
Following Trump’s policy announcement, AMD clearly accelerated its efforts:
On December 11, Su met with Chinese Ambassador to the U.S. Xie Feng to discuss AMD’s operations in China and China-U.S. economic cooperation.
On December 16, Su led an AMD executive delegation to Beijing, visiting partner Lenovo’s headquarters.
On December 17, according to an official statement from China’s Ministry of Industry and Information Technology (MIIT), Minister Li Lecheng met with Su to discuss deepening cooperation in digital economy and AI. Li affirmed China’s commitment to offering more opportunities to foreign firms like AMD and encouraged continued investment in the Chinese market.
On December 18, China’s Ministry of Commerce announced that Minister Wang Wentao met with Su to discuss AMD’s business development and enhanced cooperation in China.
Based on current public information, at least at the policy and regulatory levels, Lisa Su appears to be moving faster, making more progress, and receiving more positive feedback than Jensen Huang. AMD may well seize the initiative in the race to re-enter the Chinese market.
|
3
1