China's most prominent AI lab is preparing to deploy Huawei's Ascend processors at massive scale, accelerating the country's push away from Nvidia dependence
DeepSeek, the Hangzhou-based AI lab that rattled Silicon Valley earlier this year, is getting ready to swap out a critical piece of its hardware stack. CEO Liang Wenfeng revealed in investor discussions that Huawei will begin delivering AI training chips to DeepSeek around Q4 2026 or Q1 2027.
The move signals that Huawei’s homegrown AI accelerators have matured enough to satisfy one of China’s most demanding AI developers, and represents one of the clearest examples yet of Chinese tech firms building an alternative supply chain that routes entirely around US export controls on advanced semiconductors.
The hardware play #
The chip at the center of this transition is Huawei’s Ascend 950DT, a processor that has reportedly shown stable results in testing. DeepSeek is reportedly contemplating a deployment of at least 160,000 of these chips at a new data center in Ulanqab, Inner Mongolia, with the facility expected to draw roughly 1 gigawatt of power.
160,000 chips dedicated to inference tasks would make this one of the largest known implementations of Huawei’s AI accelerators anywhere.
DeepSeek already completed full-parameter post-training of its V4 Pro model using around 1,000 Ascend 910C chips back in June 2026. That earlier deployment served as a proving ground, and apparently the results were encouraging enough to justify a much larger commitment.
Huawei has pulled forward the launch of its next-generation Ascend 960DT training chip to Q1 2027, roughly three quarters ahead of the original schedule, due to unusually high demand from Chinese AI developers.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Why Nvidia’s grip is loosening in China #
For years, Nvidia’s A100 and H100 GPUs were the gold standard for AI training worldwide, including in China. That started changing when the US government began restricting exports of advanced AI chips to Chinese companies in late 2022, with subsequent rounds of controls tightening the screws further. Huawei stepped into that gap with its Ascend line. The Ascend 950DT is expected to see widespread adoption among Chinese AI developers starting in 2027.
DeepSeek’s willingness to go all-in on Huawei hardware carries particular weight because of the lab’s reputation for delivering robust open-source AI models at lower training costs. The company has engaged in significant collaboration with Huawei, fine-tuning software optimizations such as CANN adaptations that enhance performance on domestic hardware. Currently, DeepSeek still relies on Nvidia for large-scale frontier pre-training tasks.
What this means for the AI chip landscape #
The semiconductor supply chain implications run deep. China has been investing heavily in domestic chip manufacturing capacity, and guaranteed demand from major AI labs provides exactly the kind of market signal that justifies continued capital expenditure.
The timing of Huawei’s accelerated Ascend 960DT launch is also worth watching closely. Bringing a next-generation training chip to market three quarters early suggests Huawei is confident enough in its manufacturing pipeline to compress development cycles, or that competitive pressure to deliver is simply too high to stick with the original timeline.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our