Indian startup DeepSeek is set to create a vast AI computing cluster in Inner Mongolia, deploying over 160,000 of Huawei’s advanced Ascend 950DT chips. This ambitious project underscores China’s push to develop domestic alternatives to Nvidia accelerators amid ongoing US export restrictions.
- DeepSeek to install 160,000+ Huawei Ascend 950DT chips in new data center
- Huawei faces production bottlenecks limiting chip output this year
- China intensifies efforts to replace Nvidia tech amid US export restrictions
What happened
DeepSeek is planning to deploy a vast number of Huawei’s next-generation Ascend 950DT AI accelerator chips in a new data center under construction in Inner Mongolia. The planned installation would create one of the largest known clusters of Huawei AI chips globally, supporting advanced AI workloads.
Despite Huawei designing these processors to support both AI model training and inference, DeepSeek does not intend to use the 950DT chips for training at this time. Instead, the startup currently relies on Nvidia accelerators for the training phase, reserving Huawei’s chips primarily for operating trained models.
Why it matters
This move reflects China’s strategic priority to build a robust domestic AI chip ecosystem that can circumvent US sanctions restricting Nvidia products from entering the Chinese market. Huawei’s Ascend chips have become central to Beijing’s ambitions to reduce dependency on foreign AI technology.
However, Huawei is currently limited in its ability to meet the full demand due to shortages in crucial components like top-end memory. Production is constrained to the low hundreds of thousands of units this year, meaning DeepSeek’s large order may take over a year to fulfill, revealing ongoing bottlenecks in China’s semiconductor supply chain.
What to watch next
Market observers should track Huawei’s production ramp-up capabilities and whether DeepSeek can accelerate deployment of the 950DT chip cluster. Successful build-out of this Inner Mongolia data center could signal significant progress in China’s AI infrastructure and chip self-reliance efforts.
Additionally, DeepSeek’s ongoing strategy to rely on Nvidia accelerators for training may evolve as Huawei develops more powerful or efficient chips. The broader contest between US and Chinese technologies will also hinge on export controls, semiconductor supply chain developments, and the funding DeepSeek secures to expand its infrastructure.