Huawei accelerators have become more in demand in China after the implementation of support for their model AI DeepSeek V4

Chinese AI startup DeepSeek has adapted its new V4 model for Huawei accelerators—sparking industry buzz

For companies developing and deploying artificial intelligence, the ability to run models on familiar software without large investments in new hardware is critically important. Switching to another platform often requires both financial outlays and significant time commitments.

DeepSeek—a Chinese startup known for its competitive AI models—has decided to adapt the latest version of its model (V4) for Huawei accelerators. This makes the model more attractive to users of Chinese chips, which in turn has drawn attention from major market players.

The gist of the event

* Reuters, citing The Information, reports that DeepSeek V4’s compatibility with Huawei accelerators has already prompted giants such as Alibaba, ByteDance, and Tencent to place orders for a new series of Ascend chips in quantities of several hundred thousand units.

* The model launch is planned for the coming weeks. The developer has actively collaborated with Huawei and competing Cambricon Technologies to optimize V4 for the nuances of both platforms.

Technical details

* DeepSeek V4 will be released in at least two configurations that account for the specifics of Chinese accelerators.

* Unlike typical industry practice, DeepSeek did not share preliminary results with American chip suppliers. Instead, early access to the source code was granted to Chinese developers, including Huawei Technologies.

Why it matters

The new model has already attracted global attention. Previously, DeepSeek’s solutions significantly shifted the balance of power in the AI market, becoming a serious competitor to U.S. closed models. Thus, adapting V4 for Huawei accelerators strengthens the position of the Chinese player and opens new opportunities for companies seeking cost‑effective AI solutions.