(Bloomberg) — Huawei Technologies Co. has notified some clients that it’s significantly increasing the prices of its most advanced AI accelerators, reflecting how demand is rapidly outstripping supply during the data center boom.
Most Read from Bloomberg
The suggested price for Huawei’s Ascend 950DT jumped roughly 60% to 250,000 yuan ($37,300) apiece over the past three months, according to people familiar with the matter, who asked not to be named as discussions aren’t public. That price was set in line with Nvidia Corp.’s B200, according to one of the people. Both Nvidia and Huawei sell their artificial intelligence processors in batches of eight or more and their pricing varies with the size of the order and how much each customer buys of auxiliary equipment like networking.
Huawei’s decision underscores growing demand for a Chinese alternative to Nvidia. DeepSeek plans to deploy at least 160,000 of the top-end accelerators in a large data center project in Inner Mongolia, potentially creating one of the largest known clusters of Huawei AI chips.
But the company is grappling with production hiccups. Huawei’s sales representatives have told customers that the supply of high-bandwidth memory is particularly limited and pushing up costs, according to one of the people. The use of HBM in AI systems is at the root of the global memory shortage of the past year, and Huawei earlier this week also launched a new flagship smartphone with otherwise premium specs but limited memory.
Reuters earlier reported the Ascend 950DT price jump and said Beijing-based Cambricon Technologies Corp. is also increasing prices by as much as 30%, citing unidentified people familiar with the pricing discussions.
Huawei did not immediately respond to an emailed request for comment.
The Shenzhen-based telecom and electronics giant is prioritizing the biggest Chinese tech companies and AI infrastructure builders for deliveries of its Ascend 950DT. The company was able to manufacture somewhere in the low hundreds-of-thousands of the 950DT this year, Bloomberg News reported last week.
Most Read from Bloomberg Businessweek
©2026 Bloomberg L.P.