Due to a severe shortage of high-bandwidth memory (HBM), leading Chinese producers of artificial intelligence chips are significantly increasing the prices of their processors. Reuters reports that companies such as Cambricon and Huawei Technologies have raised prices for both current and future hardware, as the rising costs of sourcing essential memory components undermine efforts to develop domestic alternatives to Nvidia’s technology.
Huawei is said to have increased the price of its high-end Ascend 950DT accelerator card to more than 250,000 yuan (around €32,000). Depending on the specific contract terms, this represents a price rise of 20% to 50% compared with quotes issued just two months earlier.
Although the 950DT is not expected to reach the market until the end of 2026, other companies are following suit; Cambricon has raised the expected cost of its upcoming 690 chip by 20% to 30%, while smaller players such as Iluvatar CoreX and MetaX are making similar adjustments.
These price hikes are a direct consequence of a global shortage of HBM. This specialised memory, which uses vertically stacked chips to enable rapid access to data, is essential for AI applications, but is largely controlled by Micron Technology and the South Korean giants Samsung Electronics and SK Hynix.
Since the United States imposed stricter export restrictions on advanced HBM in December 2024, Chinese companies have been forced to rely on unofficial grey-market channels. Because these alternative procurement channels are much more expensive than standard market prices, the extra costs are passed directly on into the final product.
This trend also extends to older hardware. The Ascend 950PR has undergone a price increase of around 30%, from 60,000 yuan (€7,680) at the beginning of the year to more than 80,000 yuan (€10,240). The Ascend 910C has also gone up from 90,000 yuan (€11,520) to more than 110,000 yuan (€14,080).
Obstacles to AI independence
This financial pressure is a significant obstacle to China’s ambition to build an independent AI infrastructure, especially as the country is trying to fill a $50 billion market gap created by US restrictions on Nvidia. The supply shortage is also changing distribution patterns.
For example, Iluvatar CoreX has reportedly doubled its GPU deliveries to ByteDance – the maker of TikTok – this year to 100,000 units, even diverting chips that were originally intended for internal use in order to meet the tech giant’s needs.
Although Huawei remains the main domestic supplier for ByteDance, followed by Cambricon and Iluvatar CoreX, rising costs and delivery bottlenecks continue to hamper the expansion of AI computing power in the region.
Business AM
Ltd
