DeepSeek to deploy 160,000 Huawei chips for AI inference

answer: DeepSeek plans a 160,000-chip Huawei Ascend cluster for running, not training, models.

takeaways:

What happened: Bloomberg reported on 4 September, citing people familiar with the matter, that DeepSeek plans to deploy at least 160,000 Huawei Ascend 950DT chips at a data centre it is building in Ulanqab, Inner Mongolia. Neither company has commented.

The numbers: The site targets about 1 gigawatt of capacity, split between self-built and leased infrastructure. The 160,000 chips fill only part of that. Each Ascend 950DT carries 144 GB of HBM and 4.0 TB/s of memory bandwidth. At roughly $16,000 a chip, an outside estimate puts the hardware cost near $2.56 billion. That's nearly 20 of Huawei's 8,192-chip Atlas 950 SuperPoDs.

The context: A July report said part of the site would come online by late 2027 or early 2028. This would rank among the largest known clusters of Huawei AI chips anywhere, according to tech-ish.

The details: The chips will handle inference — serving users — not training. DeepSeek still trains its models on Nvidia hardware, though its V4 model was tuned to run on Ascend 950. DeepSeek is reportedly also developing its own inference chip with SMIC, according to XenoSpectrum.

The catch: Filling the order hinges on how fast Huawei can manufacture chips under US export controls. That could take more than a year.

Why it matters: US restrictions keep Nvidia's most advanced chips out of China. One forecast sees Nvidia's share of China's AI chip market falling from 40% to 8% in 2026, pushing domestic players like Huawei further into the gap.

What's next: Watch for how quickly Huawei can supply the chips, and whether DeepSeek's SMIC-built inference chip advances alongside the Ascend deployment.