DeepSeek Bets On 160,000 Huawei Chips To Power AI Push

Chinese artificial intelligence (AI) startup DeepSeek plans to deploy at least 160,000 advanced chips from Huawei Technologies at a large data centre it is building in Inner Mongolia, according to a Bloomberg report. If completed as planned, the facility could become one of the largest known clusters of Huawei AI chips.

DeepSeek intends to use Huawei’s next-generation Ascend 950DT processors to run its AI models, although the timeline will depend on Huawei’s ability to manufacture the chips, the news agency reported. The startup does not currently plan to use the 950DT chips for training its models, even though Huawei designed them for more demanding workloads.

Huawei Faces Production Constraints

DeepSeek is seeking more Huawei chips, but production capacity remains a major hurdle. The news agency reported that shortages of components such as advanced memory could limit Huawei’s 950DT output to the low hundreds of thousands this year.

The company is also balancing demand from other customers and seeking to supply small quantities to overseas markets. As a result, fulfilling DeepSeek’s order could take more than a year, according to the report.

The proposed Inner Mongolia facility is part of DeepSeek’s broader plans to expand its AI infrastructure. Its computing infrastructure plans in the region are expected to reach a gigawatt scale, with the 160,000 Huawei chips forming only part of the planned capacity.

China Pushes Domestic AI Chips

The planned cluster comes as Beijing encourages Chinese AI companies to rely more heavily on domestic processors. Huawei has emerged as a key supplier in that effort as China seeks to reduce its dependence on Nvidia chips amid US export restrictions.

Nvidia’s AI accelerators remain widely used for training and running advanced models, but sales of its most powerful chips to China are restricted by Washington. DeepSeek has previously relied on Nvidia hardware for training, while also working to optimise its models for domestic chips.

Other Chinese companies are also expanding their use of local processors. Meituan said in June that it had trained an AI model using 50,000 domestic chips, while Z.AI has completed a data centre designed to run on Chinese-made processors.

The planned DeepSeek facility therefore represents more than a single company’s hardware decision. It could also signal the growing scale of China’s effort to build AI infrastructure around domestic chips as technological competition with the US intensifies.