China's racing to build an AI ecosystem as U.S. tech curbs bite. Here's how its supply chain stacks up

In light of U.S. restrictions on advanced semiconductor purchases, China is increasingly turning to domestic alternatives, particularly Huawei, to bolster its artificial intelligence capabilities. The challenge has intensified as these export controls not only limit access to top-tier chips but also hinder the technology necessary for establishing a robust AI chip ecosystem. These constraints affect the entire semiconductor supply chain, from design and manufacturing equipment to essential components like memory chips. Beijing's response has been to invest tens of billions of dollars in efforts to bridge these gaps, but experts suggest that while there have been some noteworthy advancements, significant hurdles remain. Paul Triolo, a partner and senior vice president at DGA-Albright Stonebridge Group, points out that the U.S. export limitations on leading Nvidia AI chips have pushed China's industry to seek alternatives, albeit with added difficulties. Nvidia stands as the dominant player in the AI chip landscape; however, it primarily designs the chips rather than producing them. The company’s graphics processing units (GPUs) are the industry benchmark, with demand so high that Chinese firms continue to acquire any available Nvidia chips despite ongoing restrictions. In April, Nvidia disclosed that new regulations prevented it from selling its H20 processor to Chinese customers, highlighting the ongoing friction between U.S. policies and China's tech ambitions. In the wake of these restrictions, several Chinese semiconductor companies, including Enflame Technology and Biren Technology, have emerged, aiming to capture the GPU market left by Nvidia. Notably, Huawei's design arm, HiSilicon, is making strides with its Ascend chip series. The Ascend 910B is currently in mass production, and while the anticipated Ascend 910C shows promise, experts believe it still lags behind Nvidia's offerings. Dylan Patel from SemiAnalysis indicated that the performance gap is narrowing, with Huawei's progress evident over the past year. Nevertheless, developing competitive AI chips is not solely about design; manufacturing poses significant challenges as well. Nvidia relies on TSMC, the largest global chip foundry, which adheres to U.S. regulations and is prohibited from accepting orders from firms on the U.S. trade blacklist, including Huawei. Consequently, Huawei and other Chinese chip designers are turning to local foundries like SMIC, which currently lacks the advanced capabilities of TSMC. Despite some advancements, such as SMIC's rumored production of a 5-nanometer 5G chip, the foundry's overall capacity and technology are still limited compared to TSMC. Analysts suggest that while Huawei excels in chip design, the lack of capable domestic manufacturers hampers progress. The situation is further complicated by export controls on essential manufacturing equipment, particularly lithography machines crucial for producing advanced semiconductors. The Netherlands, home to ASML—key suppliers of advanced lithography technology—has joined in U.S.-led export restrictions, limiting access to critical tools needed for chip production. In addition to GPUs, memory chips play a vital role in AI operations. High Bandwidth Memory (HBM) has become the standard for AI applications, with South Korea's SK Hynix leading the market. U.S. restrictions have affected China's ability to procure HBM, prompting companies like ChangXin Memory Technologies to initiate production, although they face significant delays and challenges. Despite these obstacles, Chinese firms continue to explore partnerships and alternative solutions, aiming for greater self-sufficiency in the semiconductor sector. However, as the reliance on foreign technology remains pronounced, the journey toward an independent AI chip ecosystem is fraught with complexity and uncertainty.