Login to Continue Learning
China faces significant challenges in scaling up AI chip production not just due to capacity constraints from firms like SMIC, but also because of a lack of domestic High Bandwidth Memory (HBM) technologies. Beijing is rapidly enabling domestic firms to switch towards in-house tech stacks such as those from Huawei and Cambricon. However, the nation lacks sufficient resources to cater to the enormous demand for AI compute capabilities.
According to a report by SemiAnalysis, China’s reliance on foreign HBM has played a major role in its AI development over the past few years. The report claims that China is facing an ‘HBM bottleneck,’ with the country heavily dependent on the stockpile of HBM accumulated before US export controls were imposed. Samsung was one of the significant beneficiaries, shipping out 11.4 million stacks of HBM to China, making up a large portion of its inventory.
While demand for foreign HBM can be catered through ‘grey channels,’ the flow from other nations to China has reduced dramatically due to US export restrictions. Huawei, which can produce 805,000 units of Ascend 910C chips based on TSMC and SMIC capacity, still lacks enough HBM to meet production requests. This indicates that memory shortages have limited Beijing’s efforts to expand its influence in the AI compute market.
China is making domestic efforts to scale up HBM production, but firms like CXMT face constraints regarding equipment required to convert standard DRAM into HBM technologies. With investments being poured into the domestic HBM sector and ‘unspecific’ sanctions related to this segment, China could scale up to HBM3E by 2026, as predicted by SemiAnalysis.
It will be interesting to observe how China’s AI chip industry evolves under US export controls, considering that domestic firms are ramping up efforts to reduce their reliance on Western alternatives. However, they have yet to establish a suitable supply chain to meet the demand of the Chinese AI industry.
