TechNews reported that Chinese AI chip manufacturers are facing severe cost pressures, with the quoted price for Huawei’s Ascend 950DT accelerator rising by up to 50% in just two months. This surge, pushing prices above 250,000 RMB (approx. US$37,255), is attributed to a global High Bandwidth Memory (HBM) shortage exacerbated by US export controls that began in December 2024. For storage and memory readers, this highlights how HBM scarcity is no longer just a component cost issue but a primary driver of final AI hardware pricing in restricted markets.
Ascend 950PR and 910C price hikes
The price inflation is directly impacting the final product selling price because memory constitutes a high proportion of the production cost for AI accelerator chips. Huawei’s older products have also seen marked increases; the Ascend 950PR rose from approximately 60,000 RMB to over 80,000 RMB, an increase of about 30%, while the Ascend 910C rose from about 90,000 RMB to over 110,000 RMB. This trend is not limited to Huawei, with competitors like Cambricon also raising the quoted price for their next-generation AI chip, temporarily named “690,” by 20% to 30% compared to two months prior.
Gray market HBM sourcing costs
Chinese firms are increasingly relying on the gray market to secure necessary HBM supply, purchasing it at prices several times higher than those paid by buyers outside China. This reliance on illicit or secondary channels indicates that the supply constraints remain a major determinant of pricing power in the short term. While major players like SK hynix are establishing new production hubs, such as the Indiana fab, to secure future supply, the immediate reliance on gray market sourcing suggests that geopolitical controls are creating a bifurcated market with significant price disparities.
HBM as a critical AI bottleneck
The reported price increases for AI accelerators in China highlight a critical market bottleneck: the availability and cost of advanced HBM. Since HBM is a foundational component for AI computing, the inability to source it reliably and affordably due to geopolitical controls and global shortages forces cost inflation upstream, directly impacting the final product pricing. This suggests that the market for AI compute is highly sensitive to memory supply chain stability, with memory vendors holding significant leverage over chipmakers in restricted regions.
SK hynix Indiana HBM fab
SK hynix, a dominant player in the HBM market, recently broke ground on its first AI memory production hub in the United States in West Lafayette, Indiana. The fab, which will house an HBM production line and Advanced Packaging R&D Testbed, is planned to supply next-generation HBM to US customers starting in the second half of 2029. This move underscores the strategic importance of securing local AI memory production bases in the US, further complicating the supply landscape for Chinese manufacturers who are increasingly cut off from direct channels.
SanDisk HBF as a capacity alternative
The broader context of AI memory demand is also driving innovation in alternative storage technologies. SanDisk has been promoting High Bandwidth Flash (HBF), a module that stacks NAND flash memory dies to offer 10 to 100 times the capacity of DRAM-based HBM. While HBF is not a direct replacement for HBM in all AI workloads, it represents a potential pathway for high-capacity, low-cost memory solutions that could alleviate some of the pressure on HBM supply chains, particularly for inference-heavy applications where large memory footprints are required.