In July 2025, a statement by NVIDIA CEO Jensen Huang broke the deadlock in the Sino-U.S. AI chip trade—The U.S. government had approved the resumption of H20 chip sales to China and promised to grant export licenses. This decision came just three months after the sales ban in April 2025, reflecting the complex interplay of business interests and political maneuvering. As a customized AI chip designed specifically for the Chinese market, the fate of H20 highlights the inextricable connections in the global tech supply chain.
The U.S. government’s policy reversal was directly driven by economic pressure. Data shows that the H20 ban caused NVIDIA to lose $4.5 billion in Q1 of FY2025, and it was projected to lose another $8 billion in Q2. These consecutive quarters of massive losses forced NVIDIA to lobby the U.S. government to adjust the policy. Meanwhile, the immense demand from the Chinese market became an undeniable variable: in 2024, ByteDance and Tencent each purchased about 230,000 NVIDIA chips, making up a significant portion of NVIDIA’s business in China. The H20 chip, designed specifically for the Chinese market, generated between $12 billion to $15 billion in revenue for NVIDIA in 2024. Such substantial commercial interests made it impossible for the U.S. government to ignore the cost of the restrictive policy.
The deeper logic behind the reversal lies in the U.S. government’s strategic consideration of “precise control.” The H20 chip’s performance is deliberately limited—its FP16 precision can only achieve 296 TOPS, just 15% of the H100’s performance, and can only support specialized model training rather than large-scale trillion-parameter models. This “allowing the low-end, blocking the high-end” strategy both restricts China’s progress in high-end AI technology and preserves a significant market share for U.S. companies, attempting to strike a delicate balance between technological containment and commercial benefits.
Procurement Frenzy and Immediate Industry Reactions
The unban of H20 chips has given Chinese tech companies, previously anxious about the supply shortage, a chance to breathe. ByteDance and Tencent were the first to submit procurement applications and enter the “whitelist” approval process set up by NVIDIA. As giants in China’s internet industry, these two companies have huge AI computing needs, which were clearly shown by their procurement volume in 2024. Although the H20’s performance is significantly lower than the H100, it can still effectively support inference tasks for large models, thanks to NVIDIA’s mature CUDA ecosystem, which is critical for companies to maintain their AI operations.
On an industry level, the resumption of H20 supply has temporarily alleviated the pressure caused by the AI chip supply chain disruptions in China. The previous ban had led to a tight demand for AI servers, driving up prices and risking project delays. After the unban, the supply of H20 chips is expected to stabilize in the short term, helping restore industry order. However, it’s important to note that this resumption primarily applies to existing stock, and the long-term supply remains uncertain. Some sources in the supply chain suggest that H20 might be “sold out with no replenishment,” and companies need to submit detailed application materials to qualify for procurement, adding uncertainty to long-term business planning.
Moreover, the establishment of the whitelist mechanism has made the procurement process more complicated. Companies must register their procurement intentions and submit them for U.S. government approval, which not only extends the procurement cycle but also exposes their procurement strategies to regulatory scrutiny. For large companies like ByteDance and Tencent, how to balance business needs with navigating the complex approval process and potential policy risks has become a critical factor in their procurement decisions.
Long-term Trends and Strategic Choices in Tech Rivalry
The unban of H20 chips is not the end of the Sino-U.S. tech rivalry but rather the beginning of a new round of competition. The U.S. government’s policy adjustment is essentially an attempt to maintain the technological gap by “precisely targeting” Chinese companies. This strategy aims to prevent China from easily gaining high-end technology while continuing to benefit economically from low-end products. In the short term, this strategy may be effective, but in the long run, it will only strengthen China’s resolve to develop its own chip industry.
The process of domestic substitution is accelerating. Although H20 has been unbanned, the development momentum of domestic chips in China is impressive. Chips like Huawei’s Ascend 910B are already close in performance to H20, and by 2025, the share of domestic chips in the AI server market is expected to rise to 40%, potentially competing on equal footing with imported chips. This trend not only reduces China’s dependence on foreign chips but also offers opportunities for the rise of China’s chip industry. As domestic chip technologies continue to improve, this could have a significant impact on the future of U.S. technology restrictions on China.
However, challenges remain. U.S. policies are uncertain, and if Chinese companies manage to bypass H20’s performance limitations through applications like clustered computing, the U.S. could tighten export controls again. Moreover, H20 can only support certain AI tasks; for high-end trillion-parameter model training, Chinese companies still rely on high-end U.S. equipment, which means China’s development in advanced AI technologies will still be constrained by the U.S.
In the long run, the global AI computing power landscape is undergoing reshaping. The U.S. is trying to maintain its technological superiority through restrictions, while China is accelerating the push for domestic substitutes. This ongoing rivalry will continue to influence the direction of industry development. Whether it’s the procurement choices of companies like ByteDance and Tencent or China’s efforts in chip innovation, both will play crucial roles in this rivalry, determining the future competitive landscape of the global AI industry.
