Beijing Eases Import Block on Nvidia H200 AI Chips for Tech Giants

Beijing has begun loosening its import restrictions on Nvidia H200 AI chips, allowing small batches of the high-performance processors to enter mainland China. The move signals a strategic shift as China attempts to maintain its AI capabilities despite ongoing trade tensions with the United States.

ByteDance and Tencent have each reportedly received approximately 10,000 H200 processors in recent weeks. The US Commerce Department has approved roughly 10 Chinese firms—including Alibaba, Tencent, ByteDance, and JD.com—to purchase the hardware. Under these licensing terms, approved customers are permitted to purchase up to 75,000 H200 chips each.

The Hardware Gap

The H200 is a Hopper-generation chip that offers a significant performance leap over the H20, the previous export-compliant version designed specifically for the Chinese market. However, the H200 remains two generations behind Nvidia's current Blackwell architecture, which continues to be banned for export to China.

This access to H200s comes at a critical juncture for Chinese AI labs, including DeepSeek, Alibaba, and Moonshot. These firms are currently racing to close the performance gap with leading US models, such as Anthropic's restricted Mythos 5 and the public Claude Fable 5. Without access to high-end compute, Chinese developers risk falling behind in the training of next-generation large language models.

A Delicate Balancing Act

While Beijing is easing the import block, it is doing so with strict caveats to protect its own semiconductor industry. The Chinese government is encouraging companies to keep the majority of these shipments in Hong Kong rather than deploying them on the mainland. This strategy is intended to force a long-term transition toward domestic hardware by limiting the physical footprint of foreign chips within mainland borders.

This directive creates a logistical challenge for Chinese tech giants. Hong Kong possesses limited power infrastructure and data center capacity compared to the mainland, potentially throttling the actual utility of the newly acquired H200 clusters. Despite these constraints, the immediate need for compute power to train competitive AI models appears to outweigh the infrastructure hurdles.

Market Outlook

Nvidia is reported to have approximately 500,000 H200 chips in stock. While supply is available, the primary bottleneck for these sales has been Beijing's own regulatory restrictions rather than US export limits.

Industry observers will now watch whether other licensed Chinese firms follow ByteDance and Tencent in taking delivery of their quotas. The long-term success of this arrangement depends on whether China can successfully cultivate its own chipmakers while relying on a limited supply of aging Nvidia architecture to stay in the global AI race.