China has turned NVIDIA's H200 into inventory!

From January 16 to 17, the Financial Times and Reuters, among other media outlets, reported successively that NVIDIA's core supplier had suddenly announced a halt in the production of the H200 high-performance AI chip. This shutdown was not an accident but an inevitable result of the complex interplay of Sino-US chip rivalry, commercial risks, and strategic considerations. Within 36 hours, the dramatic shift from U.S. policy relaxation to the cancellation of Chinese orders fully exposed the fragility of the global AI chip supply chain.

The core of this reversal is not an "unexpected supply chain issue," but an inevitable outcome of the collision between U.S. policy fluctuations and China's independent rise. By the end of 2025, the U.S. first relaxed restrictions on H200 exports to China, but with harsh conditions: a 25% sales revenue share, export quantities not exceeding 50% of domestic levels, and requiring Chinese customers to prepay the full amount. Such "money-grabbing" terms have already caused widespread dissatisfaction.

More importantly, China is no longer in the same dependent state as before—NVIDIA once dominated 95% of the Chinese market in 2022, but by 2025, its share had dropped to 54%. The proportion of domestic AI chips in AI servers in China rose from 8% in 2023 to 27%, and Huawei's Ascend 910C has performance reaching 60%-80% of NVIDIA's comparable products, with prices half as low, directly meeting corporate demand with significant cost-effectiveness advantages.

NVIDIA's miscalculation further accelerated inventory buildup. Huang Renxun once claimed at the CES exhibition that "Chinese demand was strong," but he did not anticipate that Chinese companies had already shifted toward domestic alternatives. Additionally, the previously disclosed "remote shutdown" security backdoor in H20 had significantly reduced trust in H200.

Even more critical is that NVIDIA has shifted TSMC's capacity toward the next-generation Blackwell architecture chips. With only 700,000 H200 units in inventory, it tried to lock in orders with full prepayment, ultimately facing order "delays."

The U.S. assumed that relaxing restrictions would allow it to control China's computing power, but it failed to realize that China had long seen through the "technological hegemony + profit exploitation" strategy. While moderately purchasing to ease short-term needs, China has accelerated independent R&D. The Ministry of Industry and Information Technology has clearly set a target for domestic AI chip market share to reach 30% by 2026.

In the end, this shutdown is essentially a struggle between technological hegemony and independent innovation. U.S. policy fluctuations have made the global supply chain lose stable expectations, and NVIDIA sacrificed the interests of Chinese customers to please the U.S., ultimately harming itself.

Meanwhile, China has proven through real R&D investment and changes in market share that core technologies cannot be bought; only self-control can secure the initiative.

The transformation of H200 from a "hot commodity" to inventory is just a microcosm of this game. In the future, the global chip landscape will eventually restructure toward multipolarity and autonomy, and the U.S.'s "wanting everything" hegemonic mindset will ultimately be eliminated by the market and technological progress.

Original article: toutiao.com/article/1854546988984384/

Statement: This article represents the views of the author alone.