Taipei restricts China chip sales to appease the US

Rakesh Kumar, a professor at the University of Illinois Urbana-Champaign, has been an outspoken critic of the U.S. export controls on advanced chips and chipmaking equipment targeting China. In October 2022, days before the Biden administration announced the restrictions, Kumar warned that the crackdown would not work, as it was not a smart strategy to "trip your opponent."

However, four years later, American chipmakers like Nvidia are experiencing a decline in market share in China, while Chinese processors are closing the gap with their U.S. counterparts, and Chinese AI models are already matching some of the U.S.'s best models.

Kumar's question is what these export controls were supposed to achieve. He argues that for most AI applications, high-end chips are not necessary, and by cutting off China's access to high-end chips, the U.S. inadvertently gave Beijing the impetus to develop its own alternatives. He further explains that the export controls ended up backfiring as they spurred China to invest heavily in setting up an ecosystem for locally produced AI processors.

Chinese companies like Huawei, formerly on the brink of death under U.S. sanctions, are now releasing premium smartphones and AI processors, shipping more than 100 million devices powered by their homegrown Kirin processors since 2023.

Chinese AI labs, such as DeepSeek and Z.ai, have adapted their flagship models to run on Huawei's Ascend processors, with one Chinese food delivery giant, Meituan, training a 1.6-trillion-parameter AI model entirely on domestically made chips. Despite this, Huawei's share of data center chips in China has fallen to 34%, and Nvidia's shipments of AI accelerators have dropped by 44% in the first half of 2026 compared to the previous year.

Chinese chipmakers claim to have developed a new memory chip architecture that, if proven, may put them ahead of current leaders like Samsung, SK Hynix, and Micron.

U.S. officials are now shifting their focus from preventing Chinese AI development to keeping the U.S. at the forefront, pointing to workarounds such as the use of Nvidia's top-end B200 and B300 processors despite their export bans and accusations of Chinese labs "distilling" American models to close the gap cheaply. Despite these efforts, Kumar remains skeptical, pointing out that the U.S. policy seems to be constantly shifting the goalposts.

He suggests that China can compensate for its lack of advanced chipmaking equipment by using algorithmic techniques, open-sourcing, and advanced packaging to achieve strategic goals, even if it means sacrificing some performance. Huawei is also exploring alternative approaches, such as stacking chips in three dimensions, with its "Tau Scaling Law" targeting 1.4-nanometer-equivalent chip density by 2031.

While China's self-sufficiency in chip production is a strategic success, it may not necessarily translate into commercial competitiveness against Western inference providers, as most models are developed in China but served on American chips.