Why DeepSeek is building a huge Huawei chip cluster in China

DeepSeek, the Chinese AI company, is building one of the biggest known clusters of Huawei's AI chips, according to Bloomberg. The move shows how Chinese firms are adapting to being cut off from Nvidia's most powerful hardware.

Bloomberg reported on 4 September 2026, 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, in China's Inner Mongolia region. Neither DeepSeek nor Huawei commented, according to tech-ish.

The site is designed to hold around 1 gigawatt of computing capacity, built partly by DeepSeek itself and partly leased. The 160,000 chips would only fill part of that space. A July report said part of the facility could come online by late 2027 or early 2028.

Crucially, these chips are for inference — the stage where a trained AI model actually answers user questions — rather than for training new models from scratch. DeepSeek still trains its models on Nvidia hardware, though its V4 model was tuned to run well on Ascend chips. According to XenoSpectrum, DeepSeek is also reportedly developing its own inference chip with China's SMIC.

Each Ascend 950DT chip has 144 GB of high-bandwidth memory and a data transfer speed of 4.0 TB per second. At an estimated $16,000 per chip, the full order would cost around $2.56 billion. That's nearly 20 times the size of Huawei's standard 8,192-chip SuperPoD cluster unit.

Getting hold of all 160,000 chips won't be quick: supply depends on how fast Huawei can manufacture them under US export controls, and filling the order could take more than a year.

This story is mainly about infrastructure, not a product you'll use directly. But it reflects a bigger shift: because US export controls keep Nvidia's best chips out of China, Chinese AI companies are increasingly running their services on homegrown Huawei hardware instead. One forecast cited in the reporting suggests Nvidia's share of China's AI chip market could fall from 40% to just 8% in 2026.

If you use DeepSeek's models, this build-out is ultimately about the company having enough computing power to serve more users more reliably, even if Nvidia chips remain part of how its models are trained.

The Inner Mongolia data centre is still under construction, with parts expected online by late 2027 or 2028 at the earliest. Whether DeepSeek receives the full 160,000 chips on schedule will depend on Huawei's production capacity under continuing US restrictions.