Nvidia H200 chips stuck in Hong Kong bottleneck

About 600,000 of Nvidia’s advanced H200 artificial intelligence chips are currently stored in Hong Kong, waiting for shipment to mainland China while import approvals are withheld. Despite strong demand from Chinese technology firms, Beijing has not authorised large-volume imports of the processors into the mainland. The backlog has turned Hong Kong warehouses into a holding point for high-end AI hardware that cannot yet reach its intended buyers.

Chinese authorities have so far declined to clear large-scale customs entry for the H200, leaving the chips in limbo. Regulators including the Ministry of Industry and Information Technology and the National Development and Reform Commission have not commented publicly on the situation, and Nvidia and major Chinese internet platforms have also remained silent. The scale of the stockpile underscores the gap between commercial demand and current regulatory decisions.

While large-scale approvals are on hold, China has allowed some H200 chips to enter the country in recent months. Several major Chinese internet groups, including Alibaba, Tencent (0700.HK), ByteDance and DeepSeek, have each obtained about 10,000 H200 chips over this period. These allocations provide only partial relief compared with the broader demand for hardware to train large-scale AI models.

Beyond formal approvals, small quantities of H200 chips have been shipped into China through intermediaries and indirect routes. People familiar with these flows have not specified the volumes involved, but describe them as limited. This pattern has created a two-track market in which officially constrained supply coexists with grey-channel imports.

The H200 is not Nvidia’s newest AI processor, yet it offers capabilities that Chinese alternatives have not matched for training advanced AI models. Its high memory bandwidth and integration with Nvidia’s CUDA software ecosystem make it a key component for building and training large language models and other compute-intensive systems. These technical advantages help explain why Chinese internet companies are seeking access even as regulatory hurdles increase.

US export controls imposed since 2022 have restricted China’s access to Nvidia’s most advanced chips and the equipment needed to manufacture them. In response to constrained authorised supply, Chinese buyers have turned more actively to intermediaries and indirect import routes. This shift has tightened effective supply in the domestic market and amplified price pressures for high-end AI hardware.

The constrained availability of H200 chips has contributed to steep price increases in China’s unofficial market. By the end of October, an eight-card H200 server in mainland China was offered for more than 6 million yuan, or about $900,000. That level was around three times its price a year earlier, reflecting how regulatory limits and strong demand have reshaped pricing dynamics.

Servers incorporating Nvidia’s newer Blackwell chips, which are subject to US export controls, have become even more expensive. Offers in China have exceeded 20 million yuan per server for these systems. At the same time, Chinese internet companies including ByteDance, Alibaba and Tencent (0700.HK) have ordered more than 3 million Nvidia RTX Pro 5500 graphics cards, a less powerful but export-compliant product, with a combined sticker price above $35 billion, highlighting a shift toward hardware that can be more readily supplied under current rules.

The H200 stockpile in Hong Kong sits at the intersection of US export controls and China’s efforts to limit reliance on foreign technology. Earlier US measures barred exports of the H200 to China, before US authorities later allowed sales to approved Chinese customers under specific conditions. Beijing has since imposed its own constraints on H200 imports, reflecting a cautious approach to deepening dependence on a US supplier.

The Information reported that total approvals could amount to fewer than 200,000 H200 chips and that authorities wanted companies to prioritise Chinese processors for AI inference tasks. Against that backdrop, the roughly 600,000 H200 chips now in Hong Kong illustrate how policy on both sides of the Pacific is shaping the flow, pricing and composition of AI computing resources available to Chinese firms.