China allows small quantities of Nvidia H200 chips to keep domestic AI companies competitive

China is allowing small quantities of Nvidia H200 chips onto the Chinese mainland. That is according to The Decoder, which cites reports of targeted exemptions for domestic AI companies. The move appears aimed at giving Chinese technology companies access to advanced GPUs, amid ongoing US export restrictions that have complicated the import of high-end Nvidia chips into China for years.

The H200 is Nvidia's most recent flagship configuration for AI training and inference and officially falls under US export controls on advanced semiconductors. The fact that small batches are nonetheless reaching the mainland suggests that Chinese authorities are selectively creating room for domestic AI players who would otherwise fall further behind their American competitors.

Background to the export restrictions

The United States tightened its semiconductor export rules toward China in October 2022 and again in November 2023. The rules specifically target chips that exceed certain thresholds for computing performance, a category that clearly includes Nvidia's H100 and its successor, the H200. Washington's aim is to prevent China from deploying these chips for military applications or the development of advanced AI systems with strategic significance.

Nvidia previously brought adapted variants to market designed specifically for the Chinese market, such as the A800 and H800, but those too were later affected by tightened regulations. The H20, a further stripped-down version, has so far remained outside the heaviest restrictions, but carries significantly lower specifications than the H200 sold in the rest of the world.

As a result, Chinese AI companies face a structural disadvantage in access to the computing power needed to train large language models and other compute-intensive AI applications. Companies such as Baidu, ByteDance and a range of fast-growing AI startups are dependent on alternatives, including domestic chips from manufacturers such as Huawei, which are not yet technically on the same level.

How chips reach the mainland

The exact route by which H200 chips are reaching China is not fully documented in the available reporting. Earlier reports on comparable situations pointed to detours via third countries or traders exploiting loopholes in export regulations. Whether the Chinese government in this case is actively granting permission or simply turning a blind eye cannot be established with certainty from the available source material.

What is clear is that the volumes involved are relatively small. A large-scale import of H200 chips would almost certainly attract the attention of US regulators and provoke a diplomatic response. The limited scale suggests a deliberate strategy to provide selected companies with additional computing capacity without triggering a direct confrontation with Washington.

Strategic importance for the Chinese AI sector

For Chinese AI companies, access to the best available hardware is not a luxury but a prerequisite for remaining competitive. The training costs and timescales for large models are directly tied to the quality of the underlying chips. A generational lag in hardware can translate into substantially longer development cycles or higher costs per trained model.

At the same time, China is investing heavily in strengthening its own chip industry. SMIC, China's largest chipmaker, and Huawei's HiSilicon are working on processors intended to reduce dependence on foreign suppliers. However, analysts estimate that China still needs several more years before domestic chips approach the performance of Nvidia's top products, particularly in the area of AI-optimised architectures and their associated software ecosystems.

The reports of H200 chips reaching the mainland fit a broader pattern in which both the Chinese government and Chinese companies are actively seeking ways to limit the impact of Western export controls without further disrupting trade relations.

Relevance for the European and Dutch AI scene

The development also has an indirect bearing on the position of European AI companies and policymakers. The export controls imposed by the US apply equally to European companies seeking to supply or re-export chips to China. ASML, the Dutch manufacturer of chip-making equipment, has long been operating in this field of tension and must apply for export licences for its most advanced EUV machines destined for China.

For Dutch and European AI startups purchasing computing power from cloud providers or building their own infrastructure, this situation illustrates how deeply the global AI race has become intertwined with geopolitical relations. Access to hardware is no longer purely a market matter. Policymakers in The Hague and Brussels are closely monitoring these developments, not least because the position of companies such as ASML is directly dependent on the agreements reached about what may and may not be exported to China.