Highlights

• China has instructed customs to block Nvidia H200 AI chips despite US export approval
  
• Chinese firms face a choice between costly black-market GPUs and weaker domestic chips
  
• The move signals Beijing’s push for long-term chip self-reliance over short-term performance
  

H200 Caught in the Middle of Politics

China has quietly blocked Nvidia’s H200 AI chips at customs, days after the United States approved their export under strict conditions. According to multiple sources, Chinese customs officials were told not to allow H200 shipments into the country. No formal ban has been announced, but the instructions are being treated as a de facto restriction.

Pressure on Chinese Tech Firms

Chinese technology companies have also been told not to buy H200 chips unless absolutely necessary, with narrow exceptions for university or joint research projects. This has left firms with urgent AI needs considering black-market purchases, where an eight-chip H200 server reportedly costs about 2.3 million yuan, around 50% above official prices. Even this supply is becoming harder to secure due to tighter enforcement.

Why the H200 Matters

The H200 is Nvidia’s second-most powerful AI chip and is critical for training large AI models. It is far more capable than most Chinese alternatives, especially for training tasks, though domestic chips are improving in AI inference workloads.

Analysts say Beijing is willing to accept short-term disruption to reduce reliance on US-controlled technology. The H200 episode shows how advanced chips are no longer just products, but tools of geopolitics, with market logic increasingly shaped by national strategy.

Read More:

• China’s SuperTac Sensor Enables Human-Like Touch in Humanoid Robots
• Lei Jun Guarantees Redmi Turbo 5 Max Battery for Five Years With Free Replacement
• OpenAI Wants a Cut of Your Profits: Inside Its New Royalty-Based Plan and Other Business Models

(via)

The post Tech War: Why China Has Blocked Nvidia H200 AI Chip Imports Despite US Approval ? appeared first on Gizmochina.