Big setback for NVIDIA, to take a hit on 25 per cent of revenue as US tightens rules on AI chip sales to China

In a significant development with far-reaching consequences for the global technology landscape, the Biden administration is imposing stricter export controls on cutting-edge artificial intelligence chips.

This move, announced by the US Commerce Department, represents an update to existing regulations that will significantly curtail the ability of companies like NVIDIA and others to sell high-performance semiconductors to China.

NVIDIA in particular is likely to face some turbulent waters. NVIDIA has had a dream run for some time to the point where it became the most valued chip maker. However, during this time, NVIDIA made over 25 per cent of its revenue from the sale of AI chips to data centres in China and other Chinese tech companies.

Related Articles

AMD set to acquire a major AI software company to challenge NVIDIA’s dominance in GPUs

US to tighten the screws on China, close loophole that allowed Chinese tech cos to import sanctioned tech

These export controls were initially introduced in October 2022 but have now been extended to address technological advancements and, more crucially, to close potential loopholes that could be exploited by companies looking to bypass these restrictions.

Commerce Secretary Gina Raimondo explained that the primary aim of this update is to restrict China’s access to advanced chips that have the potential to fuel breakthroughs in artificial intelligence and advanced computing. The concern, she noted, is that such technologies are vital for the Chinese military.

This development comes on the heels of potential talks between US President Joe Biden and Chinese President Xi Jinping, pending Xi’s attendance at the Asia-Pacific Economic Cooperation forum in San Francisco.

Unsurprisingly, China has criticised the US over these export controls. However, US officials have emphasised, during recent visits to Beijing, that while the Biden administration is seeking increased engagement with China, it will not shy away from taking necessary national security actions.

Graphics processing units (GPUs) produced by companies like NVIDIA, AMD, and Intel have become indispensable for training large AI models across various sectors. As a result, the rush to acquire the latest chips has been a key trend in the technology industry.

The situation has prompted NVIDIA to design new versions of its top-tier H100 and A100 GPUs tailored for Chinese customers, with reduced performance capabilities to comply with US regulations. Chinese tech companies have been quick to purchase these modified GPUs, especially those crucial for generative AI, out of concern that the US would tighten restrictions.

One US official disclosed that the updated rule would prohibit NVIDIA from selling A800 and H800 GPUs in China. Additionally, the rules will also impact Intel’s Gaudi2 AI chip. According to another official, these changes consider how entities may attempt to circumvent the established parameters.

NVIDIA’s CEO, Jensen Huang, had previously lamented that the 2022 controls had hamstrung his company by preventing sales of its most advanced chips to China. He also expressed concern that further restrictions could negatively impact US chipmakers’ ability to invest in innovation.

In response, NVIDIA stated that they adhere to all applicable regulations and do not expect an immediate significant financial impact. Intel is reviewing the rules and assessing potential consequences, while AMD has not yet commented on the matter.

The revised export controls are poised to cut off China from advanced AI chips, including the modified versions tailored to last year’s regulations. This leaves Chinese companies facing a situation where the AI chips accessible to them will be significantly inferior to what is available in the West.

Under the 2022 rule, the US prohibited exports of chips exceeding two thresholds related to power and chip-to-chip communication speed. The updated regulations replace the latter with a “performance density” measure explicitly designed to thwart any workaround attempts.

These revamped controls will disallow the sale of data centre chips to Chinese entities capable of operating at speeds exceeding 300 teraflops, equating to a capacity of 300 trillion operations per second and above. For chips operating at speeds between 150 and 300 teraflops, sales will be prohibited if their performance density exceeds 370 gigaflops (billion calculations) per square millimetre. Chips falling within this speed range but with lower performance density will exist in a “grey zone,” requiring companies to notify the government about sales to China.

Secretary Raimondo clarified that the rules will exempt chips intended for consumer products such as smartphones and gaming, but exporters will still need to notify the government when exporting chips with speeds exceeding 300 teraflops.

Furthermore, the updated rules expand the list of chipmaking tools that cannot be sold to China. Additionally, two Chinese groups are set to be added to the “entity list,” making it exceedingly difficult for them to access US technology. These groups have been involved in the design of chips deemed to undermine US national security.

ASML, a top supplier of cutting-edge chip manufacturing equipment, expects the new regulations to affect only a limited number of fabs in China that it supplies. While ASML’s shares saw a decline of approximately 2 per cent after the announcement of the new controls, the company believes these measures will have no material impact on its financial outlook for 2023 and its long-term scenarios for 2025 and 2030.