US chip inventory backlog seeks shipments to China, Nvidia's H200 export plan attracts attention.

Nvidia has revealed to Chinese customers that it plans to export its second high-performance AI chip, the H200, to the Chinese market before the Lunar New Year in 2026. This chip's performance ranks just below its top model.
According to reports, the initial shipment scale is expected to be between 5,000 and 10,000 chip modules. Based on the conventional specifications where each module contains multiple chips, this order corresponds to approximately 40,000 to 80,000 H200 chips.
There are four core pieces of information worth noting regarding this chip export plan:
1. The H200 ranks second in performance among NVIDIA's AI chip product line, significantly outperforming the previously launched China-specific version H20 in core workloads such as AI inference and training.
2. This is the first plan to deliver high-performance AI chips of this level to China after the adjustment of the United States' related chip export policy.
3. Although NVIDIA has completed the relevant preparations before shipment, whether it can successfully enter the Chinese market ultimately depends on the approval from the Chinese side. The import of such chips may impact the development pace of the domestic chip industry.
4. The United States has loosened its export restrictions on related chips, primarily driven by the pressure of inventory buildup faced by its chip industry; meanwhile, the U.S. chip industry is promoting transformation and upgrading, attempting to influence the development process of China's chip industry through an export strategy focused on mid-to-high-end chips.
Although the H200 is not NVIDIA's most powerful chip, it is capable of meeting the usage needs for scenarios such as daily training of large models, with a processing capability that has improved sixfold compared to the previous H20. It is currently adopted by multiple high-tech industries in the United States.