DeepSeek Open-Sources Software Tools for Huawei AI Chips

DeepSeek said Wednesday that it would open-source software tools that are customized for Huawei’s AI chips, as the Chinese AI firm steps up its efforts to use domestic semiconductors for training and running its models.

The software tools include a Huawei-compatible version of TileLang—a high-level programming language that DeepSeek has also used for Nvidia chips—as well as computing and communication libraries, the company said in a social media post on WeChat. DeepSeek, which previously open-sourced similar tools for Nvidia’s platform, is now offering all the equivalent tools for Huawei’s platform.

DeepSeek’s latest move shows how the company is accelerating its adoption of Huawei chips. CEO Liang Wenfeng told investors in a recent meeting that a major priority for his company is to use more domestic chips to train its models, adding that he expects Huawei to start delivering training chips to DeepSeek as early as the fourth quarter this year, The Information reported.

For DeepSeek and other Chinese AI firms, one of the biggest challenges ahead is how to secure enough compute to train and run their powerful new models in light of U.S. export restrictions on Nvidia’s advanced chips.

In the WeChat post, DeepSeek said it has been working closely with Huawei to build advanced “supernode” AI computing system using Huawei’s hardware.

During the research and development process, “Huawei’s team provided full and unwavering support,” DeepSeek said in the post.

DeepSeek first started experimenting with Huawei chips for AI model training, and Huawei only started to collaborate with DeepSeek after that, The Information previously reported.