DeepSeek is going big on Huawei silicon. According to Bloomberg, the Chinese AI lab plans to deploy at least 160,000 of Huawei’s next-generation Ascend-950DT processors in a data center in Inner Mongolia. That would make it the largest known Huawei chip cluster to date — and a concrete signal that China’s push to decouple from Nvidia hardware is moving past the pilot phase.
Here’s the catch: these chips would only handle inference, the cheaper, lighter workload of running already-trained models. For the heavy lifting of training, DeepSeek still leans on Nvidia. It’s a telling division of labor. Chinese firms are comfortable letting domestic chips serve users at scale, but they’re not yet betting their frontier training runs on them.
Delivery won’t be quick. Huawei is still wrestling with production constraints and a memory chip shortage, with the full order likely taking over a year to fulfill. That bottleneck might loosen soon — China’s top memory maker, CXMT, has started producing small batches of HBM3E, the high-speed memory AI processors depend on. But CXMT remains three to five years behind Samsung, SK Hynix, and Micron, all of which are already mass-producing the next generation, HBM4.
What’s interesting here is the strategic patience. DeepSeek isn’t waiting for Huawei to match Nvidia on training performance; it’s building out inference capacity now, presumably because serving costs and supply-chain independence matter more in the short term than raw training throughput. If Huawei can actually deliver, expect other Chinese AI firms to follow the same playbook — and Nvidia to feel the squeeze in the one market it’s already restricted from selling its best chips into.
