DeepSeek plans to deploy over 160,000 Huawei Ascend-950DT chips at a new Inner Mongolia facility dedicated exclusively to inference workloads.

Chinese AI company DeepSeek has announced plans to deploy at least 160,000 of Huawei Technologies’ next-generation AI semiconductors, the Ascend 950DT, at a large-scale data center under construction in the Inner Mongolia Autonomous Region. If realized, the project would rank among the world’s largest single clusters of Huawei-made AI chips, symbolizing China’s broader push toward advanced semiconductor self-sufficiency.

According to multiple sources familiar with the matter, DeepSeek intends to use the 950DT processors specifically for model inference workloads at the planned facility in Ulaan Chab. Although Huawei designs and markets the chip to handle the high-compute demands of training workloads, DeepSeek reportedly has no immediate plans to utilize it for training. The company has previously experimented with Huawei-made components for training but continues to rely on accelerators from U.S. semiconductor giant Nvidia (NASDAQ: NVDA) for its primary training pipeline. This indicates that DeepSeek’s dependence on American technology for its most critical computational processes will persist in the foreseeable future.

The deployment timeline remains contingent on Huawei’s manufacturing capacity. Due to supply constraints on advanced components such as high-bandwidth memory, 950DT production this year is projected to remain limited to a few hundred thousand units. Huawei is simultaneously fulfilling orders for other clients and conducting small-volume exports overseas. Consequently, sources indicate that fulfilling DeepSeek’s full order could take more than a year. While DeepSeek aims to procure additional 950DT units, Huawei’s current supply capacity cannot keep pace with demand.

To fund its infrastructure expansion, DeepSeek raised approximately CNY 50 billion (approximately $7.5 billion) in June and is reportedly in discussions to secure several billion dollars more. By the time the Inner Mongolia data center becomes operational, multiple similarly scaled clusters are expected to already exist across China, highlighting the rapidly evolving competitive landscape for domestically produced AI semiconductors.

This development accelerates a broader industry shift as Chinese firms increasingly adopt domestic alternatives amid tightening U.S. export controls on advanced semiconductors. Beyond its smartphone operations, Huawei has been aggressively advancing its in-house AI semiconductor portfolio, with the 950DT representing its latest generation. However, DeepSeek’s continued reliance on Nvidia products for training underscores that China’s AI sector still faces significant hurdles in developing cutting-edge models without U.S.-made hardware. The company’s hybrid approach—deploying different chips for inference versus training—reflects a pragmatic strategy designed to balance peak performance against persistent supply constraints.

For investors, attention is now focused on Chinese domestic firms participating in Huawei’s AI semiconductor supply chain, alongside ongoing scrutiny of how much of Nvidia’s China revenue can be sustained moving forward. DeepSeek’s substantial order signals that China’s domestic AI infrastructure has officially entered a mass-production phase, even as Huawei’s near-term production bottlenecks continue to constrain immediate scaling.