Sources: DeepSeek plans to deploy 160K+ Huawei Ascend 950DT chips in an Inner Mongolia data center, which would create one of the largest Huawei chip clusters
DeepSeek plans to deploy at least 160,000 of Huawei Technologies Co.'s top accelerators at a massive data center it's building …
Context & Ripple Effects
DeepSeek’s reported Inner Mongolia buildout was already framed as a 1 GW project, with some capacity targeted for late 2027 or early 2028. The planned accelerator deployment gives that large data-center proposal a specific compute architecture.
The company had reportedly intended to use Huawei Ascend chips for smaller R2 training runs while retaining Nvidia hardware for its largest models. Separately, bulk orders tied to DeepSeek’s V4 were reported to have pushed Ascend 950PR prices higher, making a much larger proposed deployment a consequential demand signal for Huawei’s supply chain.
First-order effects
- If carried out, the 160,000-plus-chip deployment would make DeepSeek a major customer for Huawei’s Ascend 950DT and concentrate a large share of its planned compute capacity in one Inner Mongolia facility.
- Huawei would gain a high-profile workload for its top accelerator line, while DeepSeek’s reported dependence on a mixed Nvidia-Huawei training setup would shift further toward Huawei at this site.
Second-order effects
- A large DeepSeek order would add pressure to Ascend allocation and pricing, following reported price increases for the 950PR after earlier bulk orders linked to DeepSeek V4.
- The project links accelerator availability to the delivery of power-ready data-center capacity: the proposed 1 GW facility needs both infrastructure completion and a sustained Huawei chip supply.
Third-order effects
- If comparable deployments materialize, Chinese AI developers may treat Huawei clusters as a second-source compute path for frontier-scale training rather than reserving them for smaller workloads.
- The architecture also points to a more heterogeneous AI-compute market, in which developers distribute training and inference across proprietary chips and multiple accelerator vendors rather than standardizing on one supplier.
The trend: Chinese AI infrastructure is moving toward domestically supplied, facility-scale accelerator clusters as developers seek second-source compute for larger model workloads.