18
Chinese AI company DeepSeek has partnered with Huawei Technologies to develop programming tools for Huawei’s Ascend AI chips, as China
works to expand its domestic AI computing ecosystem and reduce reliance
on Nvidia technology.
DeepSeek said it is open-sourcing software for Huawei’s Ascend platform, including computing and communication libraries. It is also working with TileLang, an open-source programming language that helps developers write software for AI chips.
The partnership focuses on the software needed to make AI processors useful for developers. While powerful chips are critical for training and running AI models, programming tools and software libraries are also needed to use that hardware efficiently.
Nvidia’s CUDA platform has become a widely used software environment for AI development. It supports applications ranging from AI model training to inference, making software compatibility an important part of the competition between chipmakers.
Huawei has been expanding its Ascend AI chip business as Chinese technology companies seek domestic alternatives for advanced computing. The company is also developing systems that connect large numbers of AI processors for demanding workloads.
DeepSeek and Huawei have previously worked on a system using 128 Ascend 950 chips. The companies have focused on improving computing and communication performance for AI workloads, according to DeepSeek.
The latest collaboration could give DeepSeek additional options for developing and running its AI models on Huawei hardware, while strengthening software support for Huawei’s Ascend platform.
The partnership comes as China continues to invest in domestic AI and semiconductor technology amid US restrictions on access to some advanced computing technologies.
The development highlights the growing importance of software in AI chip competition. As companies build alternatives to Nvidia, programming languages, libraries and developer tools are becoming increasingly important alongside the processors themselves.
DeepSeek said it is open-sourcing software for Huawei’s Ascend platform, including computing and communication libraries. It is also working with TileLang, an open-source programming language that helps developers write software for AI chips.
The partnership focuses on the software needed to make AI processors useful for developers. While powerful chips are critical for training and running AI models, programming tools and software libraries are also needed to use that hardware efficiently.
Nvidia’s CUDA platform has become a widely used software environment for AI development. It supports applications ranging from AI model training to inference, making software compatibility an important part of the competition between chipmakers.
Huawei has been expanding its Ascend AI chip business as Chinese technology companies seek domestic alternatives for advanced computing. The company is also developing systems that connect large numbers of AI processors for demanding workloads.
DeepSeek and Huawei have previously worked on a system using 128 Ascend 950 chips. The companies have focused on improving computing and communication performance for AI workloads, according to DeepSeek.
The latest collaboration could give DeepSeek additional options for developing and running its AI models on Huawei hardware, while strengthening software support for Huawei’s Ascend platform.
The partnership comes as China continues to invest in domestic AI and semiconductor technology amid US restrictions on access to some advanced computing technologies.
The development highlights the growing importance of software in AI chip competition. As companies build alternatives to Nvidia, programming languages, libraries and developer tools are becoming increasingly important alongside the processors themselves.
