DeepSeek and Huawei unveil open-source AI toolkit

6.7 Tech

Sep 30, 2026

DeepSeek and Huawei have launched open-source tools for Ascend AI chips, offering a domestic alternative to Nvidia’s CUDA and supporting a major AI data center rollout in China.

DeepSeek, in collaboration with Huawei, has launched an open-source software toolkit tailored for Huawei’s Ascend AI chips. This suite features the high-level programming language TileLang and libraries such as DeepGEMM, FlashMLA, TileKernel, DeepSelect, and DeepEP. It is positioned as a domestic alternative to Nvidia’s CUDA ecosystem and is optimized for Ascend 950 chips, supporting configurations with up to 128 accelerators. DeepSeek plans to deploy over 160,000 Ascend chips at a new data center in Inner Mongolia, signaling a significant investment in China’s AI infrastructure. The software is freely available, aiming to lower barriers for developers and reduce reliance on Nvidia’s solutions. Additionally, DeepSeek is approaching a funding round that could value the company at $74 billion ahead of a potential IPO. Huawei has also introduced next-generation Ascend processors, expected to become mainstream for AI model development by 2027. This initiative marks a major step in China’s drive for technological self-sufficiency in AI hardware and software.