DeepSeek reportedly plans to deploy at least 160,000 Huawei Ascend 950DT chips at a new data center in Inner Mongolia, according to Bloomberg.
If completed, the installation could become one of the largest known clusters of Huawei AI accelerators and another major step in China’s push to reduce its dependence on Nvidia hardware. For infrastructure partners, it also points to growing demand for suppliers capable of deploying and supporting alternative AI computing platforms.
Bloomberg, citing people familiar with the matter, reported that DeepSeek intends to use the processors for AI inference rather than model training at the facility.
DeepSeek has previously tried using Huawei hardware for training but has continued relying on Nvidia accelerators for that demanding workload. Neither DeepSeek nor Huawei confirmed the reported plan. Bloomberg also reported that Huawei does not currently have enough production capacity to fulfill the order quickly.
Huawei faces a supply bottleneck
Huawei may need more than a year to supply the full reported allocation. Bloomberg’s sources said shortages of advanced memory components could limit total 950DT production this year to the low hundreds of thousands of chips, which would be divided among domestic customers and some overseas buyers.
Bloomberg also reported that DeepSeek has asked Chinese officials to encourage Huawei to accelerate delivery.
Huawei unveiled the 950DT as a more capable successor to its earlier Ascend processors. However, its availability and performance have not yet been demonstrated at the scale DeepSeek reportedly wants.
The backdrop is a Chinese AI accelerator market increasingly shifting toward domestic suppliers. Huawei is projected to control roughly half of that market by year’s end, according to Bloomberg, while US export restrictions and Chinese procurement policies have weakened Nvidia’s position.
Jensen Huang, Nvidia CEO, told CNBC in May: “We have largely conceded the China market to Huawei,” calling China a “$50 billion opportunity this year alone.”
Why the number matters
DeepSeek’s planned Inner Mongolia infrastructure is reportedly designed at a gigawatt scale. The 160,000 Huawei chips would represent only part of that capacity, leaving an important question over what hardware will power the rest.
Still, the proposed cluster would be dramatically larger than China’s previously disclosed Huawei deployments. Shenzhen recently reported the launch of a 10,000-chip Ascend computing cluster, while Meituan said it had trained an AI model using 50,000 domestic chips, Bloomberg reported.
For channel partners, the reported deployment points to a more fragmented AI infrastructure market in China, where domestic accelerators could increasingly handle inference while Nvidia hardware remains important for training. The next question is whether Huawei can supply enough chips — and the supporting systems and software — to make that split sustainable at enterprise scale.
Read more: As China expands its domestic AI hardware ecosystem, Nvidia’s $500 billion infrastructure financing plan shows how Western providers are helping customers fund increasingly expensive AI deployments.