Huawei has raised the indicated price of its Ascend 950DT accelerator card to more than 250,000 yuan, roughly 20% to 50% above quotes given about two months earlier depending on contract terms, Reuters reported citing two people2. Bloomberg separately reported that suggested pricing jumped about 60% over three months to around the same 250,000-yuan level, with one person saying the quote was set in line with Nvidia's B200 for comparable multi-card deals3.
The price increases extend beyond Huawei's next-generation part. The Ascend 950PR has moved above 80,000 yuan from about 60,000 yuan at the start of the year, and the Ascend 910C has moved above 110,000 yuan from about 90,000 yuan over the same period. Both outlets note that list economics still vary with order size and how much customers buy in boards, networking, and other auxiliaries.
HBM as the cost driver
High-bandwidth memory costs are the common factor. HBM stacks now dominate finished-card bills of materials, according to people familiar with customer quotes. Huawei has described proprietary memory stacks for the Ascend 950 series: HiBL 1.0 on the 950PR and HiZQ 2.0 on the 950DT. The 950DT, Huawei's most advanced Ascend product for model development and decode-heavy inference, remains scheduled for the fourth quarter of 2026.
Rivals follow suit
Cambricon has repriced its next-generation part, tentatively called the 690, about 20% to 30% above levels indicated two months ago. Smaller rivals MetaX and Iluvatar CoreX have made similar increases1. Reuters reported that the four chipmakers — Huawei, Cambricon, MetaX, and Iluvatar CoreX — have raised prices for current and next-generation chips by 20% to 50% due to rising HBM costs.
On the demand side, Iluvatar has roughly doubled GPU shipments to ByteDance to about 100,000 units this year. Huawei remains ByteDance's largest domestic supplier, followed by Cambricon and Iluvatar.
ANALYSIS The simultaneous repricing across four vendors points to a supply-side constraint rather than individual product-positioning decisions: if HBM costs are lifting bills of materials industry-wide, Chinese AI infrastructure buyers face higher capital expenditure regardless of which domestic accelerator they choose. The Bloomberg-sourced detail that Huawei's 950DT quote was set in line with Nvidia's B200 for comparable deals suggests Huawei is using the memory squeeze to close the price gap with its U.S. competitor from below rather than competing on discount.