DeepSeek has released software it developed with Huawei to program Huawei's Ascend chips, as Chinese tech firms continue building alternatives to Nvidia.

In an announcement on its official WeChat account on Wednesday, DeepSeek said the tools include the TileLang high-level programming language, compute libraries, and distributed communication libraries.

"Each component corresponds one-to-one with the open-source components we previously released for the Nvidia platform," said DeepSeek.

The Chinese AI lab said that Huawei's team "provided strong and unreserved support" during development for the Ascend platform. The two companies have also jointly advanced a "supernode" project based on 128 Ascend 950 chips, according to the post.

When it comes to building a new generation of independent GPU software ecosystems, DeepSeek said that the priority is to establish a high-level language that's universal, simple to program, and capable of reaching the hardware's performance potential.

Estimated AI Compute Shipped by Chip Designer

Estimated AI Compute Shipped by Chip Designer

  • AMD
  • Amazon
  • Cambricon
  • Google
  • Huawei
  • Nvidia
SOURCE: Epoch AI

The deepening ties between DeepSeek and Huawei come as Beijing pushes to build domestic alternatives to advanced U.S. technology.

Earlier this month, Huawei announced plans to launch two AI chips in 2027, the 960DT and 960PR, ahead of schedule. Alibaba also unveiled a new AI chip last week that is expected to deliver three times the performance of its predecessor.