US Tightens Chip Export Controls: AI Sector Faces a Winter

The U.S. Department of Commerce has officially announced new chip export control regulations, further tightening the export of high-performance chips to China. This move has raised concerns across the entire artificial intelligence industry.

The new regulations use computing power as the main evaluation criterion, covering almost all current high-performance chips. Even NVIDIA's consumer-grade gaming graphics card RTX 4090 is not exempt, as its performance density falls within the scope of the control list. Even chips with slightly lower performance than the standard need to notify the government before export.

This means that all high-performance chips, including Nvidia's special A800, H800, L40S, AMD's MI250, and Intel's Gaudi2, are subject to restrictions. The severity of the new regulations has exceeded market expectations, with reports suggesting that Nvidia has even urged customers to place orders for the H800 chip before the new regulations take effect.

The U.S. Department of Commerce's recent regulatory measures can be described as comprehensive, not only restricting chip exports but also imposing more barriers on the export of semiconductor manufacturing equipment. At the same time, some Chinese companies have been placed on the so-called "entity list." The Department of Commerce also plans to restrict China's access to cloud computing power, which will affect the world's largest cloud service providers.

This policy has sparked differing reactions within the industry. The Semiconductor Industry Association (SIA), which represents most chip companies in the United States, believes that overly broad unilateral controls would harm the U.S. semiconductor ecosystem. However, some lawmakers argue that stricter measures are still needed.

For chip companies, their lobbying during this policy-making process has clearly failed. Intel originally planned to promote the Gaudi2 chip in China, but may now need to reconsider. Nvidia is also facing uncertainty regarding its revenue from the Chinese market, which has consistently accounted for about a quarter of its total revenue.

This situation reflects a paradox in the globalized economy. The semiconductor industry's division of labor, once seen as a model of globalization, has now become a tool for certain countries to implement "weaponized dependence." The United States is leveraging its dominant position at critical junctures to exert influence in the name of national security.

In the face of this situation, Chinese artificial intelligence and large model companies have taken some countermeasures, including ordering a large number of chips in advance and using domestic alternatives. Huawei's Ascend ecosystem has shown some competitiveness in this regard. However, the current capabilities of domestic chips still struggle to fully meet the huge market demand for computing power.

Despite facing challenges, China's strong capabilities in the field of large models indicate that chip restrictions may slow down the pace of development, but will not completely hinder the advancement of artificial intelligence technology. This trial may become a driving force for accelerating the development of China's chip industry.