New US AI chip restrictions on China creates massive issues for Baidu, Alibaba’s AI development

Recent measures taken by the United States to tighten export controls on AI chips are poised to present significant challenges for Chinese tech companies, such as Alibaba and Baidu, in their pursuit of artificial intelligence (AI) advancements. These new export restrictions have implications for the development and manufacturing of their latest AI processors.

Alibaba and Baidu have been at the forefront of China’s efforts to create homegrown alternatives to advanced AI products by US-based manufacturer NVIDIA. Currently, these companies manufacture their chips at facilities operated by Taiwan Semiconductor Manufacturing Company (TSMC) and Samsung.

However, the newly introduced export controls by the United States have affected the processing speeds of their cutting-edge AI chips, causing complications for the tech giants.

These tightened restrictions will also impact Silicon Valley-based NVIDIA, which will be required to halt shipments to China of two processors specially tailored to meet earlier export controls, according to an announcement by NVIDIA.

In light of these collective restrictions, Chinese tech companies may need to resort to AI chipsets similar to NVIDIA V100, which was released in 2017 and has since been discontinued. This implies that they would have to use outdated technology to train and run generative AI models, as modern chips have evolved significantly, enabling innovations like OpenAI’s ChatGPT.

One Beijing-based chip consultant characterized the US measures as posing “an existential challenge” to China’s efforts to catch up with AI development as seen in American companies like OpenAI.

The US controls extend to foundries in Taiwan and South Korea, which are contracted to manufacture chips for Chinese groups. This is made possible due to the extensive presence of American hardware and software in the semiconductor supply chain. In contrast, China’s domestic alternatives, including the partially state-owned SMIC, lag behind in chip manufacturing technology by several generations.

These updated rules arrive amidst deteriorating US-China relations and an expanding effort to obstruct Beijing’s technological progress. Some experts, like Gregory Allen from the CSIS think-tank, suggest that these controls aim to limit China’s access to the future of AI.

The updated controls increase the number of advanced AI chips requiring a license, likely leading to denials and effective bans on their export. They also establish a reporting regime that encompasses a wide range of data centre chips, imposing restrictions on selling to subsidiaries of Chinese companies located outside China.

According to Boris Van, a senior analyst at Bernstein, relying on chips similar to NVIDIA’s V100 could at least double data processing costs for Chinese AI firms once their existing chip supplies are depleted.

In August, it was reported that leading Chinese tech companies, including Alibaba, Baidu, ByteDance, and Tencent, had placed orders worth over $5 billion for NVIDIA chips. However, most of these orders have yet to be fulfilled, resulting in significant supply challenges.

While there may be expectations that some banned chips could still find their way into the country through black market channels, experts do not anticipate that such supply would be sufficient to meet the high demand from tech groups focused on training generative AI models. Consequently, the Chinese AI sector may find itself relying on low-end NVIDIA chips, leading to a growing infrastructure gap between China and the rest of the world.