Forced by US chip ban! FT reveals Chinese company went shopping for millions of Huida eSports cards to modify AI

Nvidia CEO Jen-Hsun Huang held up a display of the GeForce RTX series of high-end gaming graphics cards at the press conference. (Reuters photo)

[Compiled by Chen Chengliang/Comprehensive Report] The latest investigation by the British "Financial Times" (FT) revealed that in the face of the increasingly stringent export controls of high-end artificial intelligence (AI) chips in the United States and the dual dilemma of domestic substitute chips that cannot meet the performance standards, Chinese technology giants have fallen into an extremely severe "hungry for computing power." In order to maintain the operation of AI models, Chinese technology companies have purchased more than 1 million Nvidia gaming and workstation-level processors on the market this year, and then dismantled and modified them to form AI computing clusters.

The report pointed out that these large quantities of chips flowing into China are mainly consumer-grade e-sports graphics cards such as Huida's GeForce RTX series. They were originally designed to drive high-end computer games and 3D graphics, and were not used to build complex large-scale language models for data centers. However, because the U.S. Department of Commerce has completely cut off the channels for top data center GPUs such as Huida H100 and Blackwell to be directly sold to China, and because Chinese local chips such as Huawei are still unable to replace Huida's CUDA architecture in terms of production yield and software ecosystem, Chinese companies have been forced to take the wrong path and turn to the consumer-grade hardware market that has not been completely sealed.

Please read on...

Disassembly and assembly of local steelmaking methods are difficult to overcome the computing power gap

Industry engineers and supply chain insiders revealed that factories across China are using "homemade steelmaking" modified production lines to remove the core computing components and memory particles of these e-sports graphics cards from their original circuit boards, and then re-solder and install them onto customized server substrates.

However, semiconductor experts bluntly stated that consumer-grade game chips lack high-bandwidth interconnect technology dedicated to data centers, and the speed of data transmission between chips is subject to physical limitations. Assembling a large number of game cards into clusters is not only difficult to dissipate heat, but also consumes extremely high energy. When conducting large-scale model training, the actual computing efficiency lags far behind that of regular AI servers.

This wave of acquisition frenzy triggered by large Chinese technology companies and start-up teams has significantly increased the quotations and smuggling risks of related high-end gaming graphics cards in the Asian market. The Financial Times analyzed that this investigation directly debunked Beijing’s official propaganda myth that “domestic chips are sufficient to support the development of AI” and highlighted that China is still unable to get rid of its high dependence on American advanced technology for its computing power. Chinese companies’ use of consumer-grade product modifications to evade controls is also triggering heightened vigilance from Washington’s national security and trade regulatory authorities.

Grasp the economic pulse with one hand. Click here to subscribe to Free Finance Youtube Channel