DeepSeek Plans Wider Use of Huawei AI Chips for Model Training

Chinese artificial intelligence startup DeepSeek is reportedly preparing to expand its use of Huawei’s Ascend AI chips for training large-scale AI models, as the company seeks to reduce its reliance on US-based semiconductor technology.

According to a report by The Information, DeepSeek CEO Liang Wenfeng said during an investor meeting that Huawei’s Ascend processors are among the company’s major strategic bets. He emphasized the importance of successfully training AI models using domestically produced chips, while noting that Huawei continues to improve the performance of its processors.

Also read: Four Private Bank Officers Booked for Alleged Harassment of Woman in Faisalabad

DeepSeek Plans Wider Use of Huawei AI Chips for Large-Scale Model Training

DeepSeek has previously used both Nvidia and Huawei processors for AI model training and inference, although it has relied heavily on Nvidia’s advanced chips for training. Liang reportedly expects Huawei to deliver additional AI training chips as early as the fourth quarter of 2026 or during the first quarter of 2027.

The DeepSeek chief also said Huawei could narrow the performance gap with Nvidia within the next few years and become an important alternative for AI computing in China. DeepSeek is currently developing models with up to 2 trillion parameters and has plans for an 8-trillion-parameter model, which would require significantly greater computing capacity.

Liang estimated that training an AI model comparable in scale to OpenAI’s largest models could require around 50,000 Nvidia GB300 chips or at least 200,000 Huawei Ascend 950 processors. DeepSeek is reportedly investing about 30 billion yuan to expand its computing infrastructure as it prepares for increasingly demanding AI workloads.

Read more: Digital Transformation to Help Pakistan Achieve Faster, More Independent Growth