In Focus
DeepSeek plans to order at least 160,000 Huawei Ascend 950DT chips.
The AI developer plans to use the chips for AI inference tasks.
DeepSeek will continue using Nvidia chips for AI training tasks
Chinese AI startup DeepSeek plans to deploy 160,000 advanced Huawei AI chips to power its upcoming AI data center in Ulanqab, Inner Mongolia Region. If completed, the deployment would create one of the largest known clusters of Huawei AI chips, which is expected to boost China’s AI push. Reports about the DeepSeek data center have been circulating in recent months. The one gigawatt AI facility is expected to become operational in late 2027 or 2028.
Why is DeepSeek Deploying Huawei Chips?
DeepSeek is shifting to Huawei chips mainly because U.S. restrictions limit its access to advanced AI hardware. Nvidia’s most capable AI accelerators cannot legally be sold in China, making Huawei’s Ascend chips a key alternative for Chinese AI companies.
Using Huawei chips, DeepSeek can expand its AI infrastructure while reducing its reliance on Nvidia processors. DeepSeek had planned to order larger volumes of Huawei’s advanced chips. However, supply constraints have made this difficult.
Sources familiar with the matter said Huawei’s 950DT production is expected to remain low this year due to shortages of critical components, such as high-bandwidth memory. Given Huawei’s other supply commitments, fulfilling DeepSeek’s order could take more than a year.
DeepSeek’s data center in Mongolia is part of China’s wider plan to shift energy-intensive computing infrastructure toward western regions where the cost of electricity is lower. Inner Mongolia has signed deals valued at about $27 billion in AI computing projects as part of the push.
Which Huawei AI Chips Will DeepSeek Deploy?
A Bloomberg report showed that DeepSeek plans to use Huawei’s Ascend 950DT chips for AI model inference workloads. Huawei has positioned the 950DT processor as a high-performance chip capable of supporting AI model training.
But DeepSeek reportedly does not plan to use the processor for AI training. The company has tested Huawei hardware for training before, but it still relies mainly on Nvidia chips. This means DeepSeek’s reliance on Nvidia chips for critical workloads will likely continue in the foreseeable future.
DeepSeek has been increasing its reliance on Huawei’s AI hardware for some time now. The company tuned its V4-Flash AI model, which launched in April, to run on Huawei’s Ascend 950 chips. Chinese technology companies, including Alibaba, ByteDance and Tencent, have since reportedly placed orders for hundreds of thousands of the processors.
The gradual shift reflects Nvidia’s views about export controls. CEO Jensen Huang has termed DeepSeek’s use of Huawei chips a “horrible outcome” for the U.S. Huang previously argued that export controls would push AI developers in China towards Chinese hardware.
.webp&w=3840&q=75)

.webp&w=750&q=75)
.webp&w=750&q=75)
.webp&w=750&q=75)
