Chinese AI chipmakers, including Huawei Technologies and Cambricon, have sharply raised prices for current and next-generation processors as soaring costs for high-bandwidth memory squeeze Beijing’s drive to build a domestic alternative to Nvidia’s advanced AI accelerators.
Huawei has raised the indicated price of its Ascend 950DT accelerator card to more than 250,000 yuan ($37,255), two people familiar with the matter told Reuters. Depending on contract terms, the new price is 20% to 50% higher than levels quoted to customers just two months ago.
The increase underpins an emerging problem for China’s semiconductor industry: developing a domestic AI-chip supply chain is becoming more expensive just as Beijing is pushing technology companies to reduce their reliance on U.S. suppliers.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Huawei has publicly said its Ascend 950DT, its most advanced AI chip, will become available in the fourth quarter of 2026.
Beijing-based Cambricon has also raised the indicated price of its next-generation processor, tentatively known as the 690, by 20% to 30% from levels discussed with customers two months ago, according to two sources. Smaller Chinese chipmakers MetaX and Iluvatar CoreX have made similar increases, the sources said.
HBM, a specialized type of memory critical to modern AI accelerators, remains at the center of the price increases. HBM stacks memory chips vertically and allows processors to move very large volumes of data at high speeds, making it essential for training and running complex AI models.
The market for advanced HBM is dominated by South Korea’s SK Hynix and Samsung Electronics and U.S.-based Micron Technology. China’s access to some advanced HBM products has been restricted by U.S. export controls, forcing domestic chipmakers to seek alternative supplies.
Since Washington tightened controls on exports of certain advanced HBM products to China in December 2024, Chinese companies have relied on grey-market channels to obtain supplies, according to the sources.
However, that route comes at a premium. HBM procured through such channels can cost several times more than prices paid by buyers outside China, the sources said. Because memory represents a substantial portion of the manufacturing cost of an AI accelerator, the higher component prices are being passed directly to customers.
The problem reveals the difficulty of creating a fully domestic AI-computing ecosystem. China can design processors to replace Nvidia products, but advanced AI systems require more than the accelerator itself. High-performance memory, packaging, manufacturing capacity and supporting software all form part of the supply chain.
Huawei has said its Ascend 950 series will use two proprietary HBM technologies, HiBL 1.0 for the 950PR and HiZQ 2.0 for the 950DT, although the company has not disclosed where or how the underlying memory is manufactured.
The 950DT is designed primarily for developing AI models and generating responses, while the 950PR is intended to process user requests before they are passed to the model. Prices for Huawei’s existing chips have also increased, suggesting the pressure is not limited to the company’s newest products.
The Ascend 950PR, which sold for roughly 60,000 yuan per card at the beginning of the year, now costs more than 80,000 yuan, representing an increase of about 30%, according to two sources.
The older Ascend 910C board has risen to more than 110,000 yuan from about 90,000 yuan at the beginning of the year, they said.
The increases raise the cost of deploying AI computing capacity at a time when demand from Chinese technology companies is accelerating. They also expose a contradiction at the heart of China’s effort to replace Nvidia.
U.S. restrictions have created a large domestic market for Chinese accelerator manufacturers by limiting Chinese companies’ access to Nvidia’s most advanced processors. But the resulting demand is now colliding with shortages of critical components, potentially limiting how quickly domestic suppliers can scale.
China’s AI-chip market is estimated at about $50 billion, making the opportunity substantial. The restrictions have effectively created space for Huawei, Cambricon, MetaX and other domestic suppliers to expand, but capturing that market requires them to compete not only on chip performance but also on price and availability.
The pressure is already affecting the allocation of computing hardware.
Iluvatar CoreX has doubled its shipments of graphics processing units, or GPUs, to ByteDance, the owner of TikTok, to 100,000 units this year as computing constraints have intensified, according to one source.
The company has also diverted GPUs originally intended for its own internal use to meet ByteDance’s requirements, the source said.
ByteDance is a major buyer of domestic AI computing capacity, and Huawei is its largest domestic supplier, followed by Cambricon and Iluvatar CoreX, according to the three sources.
The competition for supply suggests that Chinese AI companies could face a difficult trade-off between expanding computing capacity and controlling costs. Higher accelerator prices could ultimately raise the cost of training and operating AI models, potentially slowing the pace at which companies can deploy them or increasing the cost of AI services.
For chipmakers, meanwhile, higher prices may help offset expensive memory procurement and constrained component supply. But sustained price increases could undermine one of the principal advantages of domestic alternatives: their ability to offer Chinese customers a viable and potentially more economical substitute for restricted Nvidia products. That makes HBM more than a component-level bottleneck. It has become a strategic constraint on China’s broader AI ambitions.
The situation also shows why U.S. semiconductor restrictions can have effects beyond simply preventing Chinese companies from buying particular chips. Restricting access to advanced processors and memory can force Chinese companies to rebuild entire portions of the AI hardware supply chain, increasing costs and creating bottlenecks in areas where domestic alternatives remain limited.