China’s AI chipmakers raise prices as HBM shortage deepens

Rising costs and constrained access to high-bandwidth memory are pushing Chinese AI chip suppliers to raise prices, adding pressure to Beijing’s efforts to develop domestic alternatives to Nvidia’s advanced processors.

Chinese AI chip manufacturers are reportedly raising prices for both existing and upcoming processors as the cost of high-bandwidth memory (HBM) increases, creating another challenge for the country’s push to build a domestic AI computing ecosystem.

Huawei Technologies has reportedly raised the indicated price of its upcoming Ascend 950DT accelerator card to more than 250,000 yuan, or about $37,255. The revised price is estimated to be 20% to 50% higher than quotations given to customers around two months earlier, depending on contractual terms.

Huawei has said that the 950DT, positioned as its most advanced AI processor, is expected to become available in the fourth quarter of 2026.

Cambricon has also increased the indicative pricing of its next-generation processor, tentatively known as the 690, by around 20% to 30% compared with levels indicated two months ago. The company has not yet formally launched the chip. Smaller Chinese chipmakers MetaX and Iluvatar CoreX have reportedly made similar pricing adjustments.

The information reportedly comes from people familiar with private pricing discussions, who requested anonymity because the negotiations are not public. Huawei, Cambricon, MetaX and Iluvatar CoreX did not respond to requests for comment.

HBM shortage raises pressure on domestic AI chips

The pricing changes underline the growing importance of HBM in China's AI hardware supply chain. HBM uses vertically stacked memory components to provide AI processors with rapid access to large volumes of data, making it a critical part of high-performance accelerators.

The advanced HBM market is dominated by SK Hynix, Samsung Electronics and Micron Technology. Following tighter US restrictions on exports of certain advanced HBM products to China in December 2024, Chinese chipmakers have increasingly turned to alternative and grey-market supply channels, according to the sources.

Memory procured through such channels can cost several times more than comparable supplies available outside China. Since HBM represents a significant portion of the manufacturing cost of an AI accelerator, higher memory prices are being reflected in the final price of accelerator cards.

Huawei has said its Ascend 950 family will use proprietary memory technologies, including HiBL 1.0 for the 950PR and HiZQ 2.0 for the 950DT. The company has not disclosed where the memory is produced or how it is sourced.

The 950DT is designed primarily for AI model development and response generation, while the 950PR is intended to handle incoming user requests before they reach the underlying AI model.

Older Huawei accelerators also see price increases

Pricing pressure has also extended to Huawei's earlier-generation products. The Ascend 950PR, which was reportedly priced at around 60,000 yuan per card at the beginning of 2026, is now selling for more than 80,000 yuan, representing an increase of roughly 30%.

The older Ascend 910C has also become more expensive, with its price rising above 110,000 yuan from approximately 90,000 yuan earlier this year.

The increases could raise the cost of expanding AI computing capacity in China at a time when domestic chipmakers are being asked to fill the gap created by restrictions on access to Nvidia's most advanced processors.

The supply squeeze is also affecting allocation decisions. Iluvatar CoreX has reportedly doubled its planned GPU shipments to ByteDance, the company behind TikTok, to around 100,000 units this year. It has also redirected GPUs originally intended for its own requirements to meet the customer's demand.

Huawei remains ByteDance's largest domestic AI chip supplier, followed by Cambricon and Iluvatar CoreX, according to people familiar with the supply arrangements.

The developments highlight how access to advanced memory has emerged as a significant constraint for China's AI chip industry, potentially increasing costs even as domestic manufacturers expand their role in the country's AI computing market.