DeepSeek Plans Massive 160,000 Huawei Ascend 950DT Chip Deployment in Inner Mongolia

DeepSeek plans to deploy at least 160,000 Huawei Ascend 950DT chips at a new gigawatt-scale data center in Ulanqab, Inner Mongolia, creating the largest known cluster of Huawei AI chips for inference workloads while continuing to use Nvidia hardware for training. For broader context, explore our AI News. For broader context, explore our Top 100 AI Tools.

Strategic Investment in Domestic AI Infrastructure

The proposed data center in Ulanqab is designed for substantial capacity, estimated at approximately one gigawatt. This scale underscores the growing demand for robust infrastructure to support advanced AI applications. While DeepSeek intends to use Huawei's Ascend 950DT chips specifically for inference tasks, the company is expected to continue utilizing Nvidia hardware for its more demanding AI model training workloads. This dual-vendor strategy highlights the current landscape where specialized hardware excels in different aspects of the AI development pipeline.

The initial phase of the facility is projected to commence operations by late 2027 or early 2028. This timeline suggests a long-term strategic commitment to building out domestic AI capabilities, aligning with broader governmental initiatives in China to foster self-sufficiency in critical technology sectors.

Challenges in Chip Supply and HBM Production

Despite the scale of DeepSeek's plans, Huawei reportedly faces challenges in fulfilling such a large order promptly. Production limitations and ongoing shortages of high-bandwidth memory (HBM) chips are cited as potential hurdles, possibly delaying full delivery for over a year. This situation reflects broader supply chain pressures impacting the global AI hardware market.

The availability of advanced memory components, particularly HBM, is crucial for high-performance AI accelerators. While domestic HBM maker CXMT is currently producing small batches of HBM3E, it lags behind international leaders like Samsung, SK Hynex, and Micron, which are already engaged in mass production of the more advanced HBM4. This disparity in HBM technology underscores a key area where China's domestic chip ecosystem is still developing to meet the demands of cutting-edge AI hardware.

Why This Matters Now

This development is significant for several reasons. Firstly, it represents a substantial commitment to Huawei's Ascend series, potentially solidifying its position as a viable alternative to Western hardware for specific AI applications within China. Secondly, the sheer scale of the planned deployment could accelerate the optimization and adoption of Huawei's AI ecosystem. Lastly, it highlights the ongoing geopolitical dynamics influencing technology supply chains, pushing countries to invest heavily in domestic capabilities.

For developers and enterprises, understanding the capabilities and limitations of different AI hardware platforms, including those from Huawei, becomes increasingly important. As the global AI landscape evolves, the strategic choices made by major players like DeepSeek will shape future infrastructure trends.

Key Takeaways

• DeepSeek plans to deploy 160,000 Huawei Ascend 950DT chips for AI inference.
• The project involves a gigawatt-scale data center in Inner Mongolia.
• Huawei faces production and HBM shortages, potentially delaying delivery.
• DeepSeek will continue using Nvidia for AI model training workloads.
• This represents the largest known cluster of Huawei AI chips globally.

Sources

• Deepseek plans the largest known Huawei chip cluster with 160,000 processors in Inner Mongolia
• DeepSeek turns to Huawei for 160,000 AI chips as Nvidia stays locked out of China - tech-ish