DeepSeek open-sources toolkit for Huawei chips to rival Nvidia’s CUDA

Chinese AI startup DeepSeek announced on Wednesday that it is open-sourcing a full programming toolkit for Huawei's Ascend AI chips, in one of the most concrete steps yet toward building a domestic alternative to Nvidia's ↗ 0.14% dominant CUDA software ecosystem.

The release, announced via DeepSeek's WeChat account and reported by Reuters and Bloomberg, includes TileLang — a high-level programming language positioned as a simpler alternative to CUDA — along with compute and communication libraries such as DeepGEMM, FlashMLA, TileKernel, DeepSelect, and DeepEP. The entire package is open-source and free to download.scmp+2

A Layer-by-Layer Challenge to CUDA

"To build a new generation of independent, self-controlled GPU software ecosystem, the first priority is establishing a high-level language that is universal, easy to program, and still capable of reaching the hardware's full performance potential," DeepSeek said in its WeChat post. "TileLang was created precisely to meet this need."reuters

The toolkit is optimized for Huawei's Ascend 950 chips and supports a "supernode" configuration linking up to 128 accelerators, a scale designed for large-batch AI training workloads. According to the South China Morning Post, TileLang now officially supports native code generation, automatic scheduling, and synchronization for the Ascend 950, as confirmed on the project's GitHub page.scmp+1

Huawei provided what DeepSeek described as "unreserved and vigorous support" throughout the development process. The announcement comes roughly two weeks after Huawei unveiled its next-generation AI processors and supernode computing systems, with the company saying it expects its chips to see wider use for model training by 2027.finance.yahoo+2

Reducing Dependence on Nvidia

The partnership reflects a broader push among Chinese technology firms to reduce reliance on Nvidia hardware and software amid ongoing U.S. export controls on advanced semiconductors. DeepSeek, which initially relied heavily on Nvidia's H800 chips, has been shifting increasingly toward domestic alternatives in 2026. Earlier this year, the company modified its V4 AI model to operate on Huawei hardware.cryptonomist+1

DeepSeek also plans to deploy at least 160,000 Huawei Ascend accelerators in a data center it is building in Inner Mongolia, according to Bloomberg. The company is separately finalizing a fundraising round at a valuation close to 500 billion yuan, or roughly $74 billion, ahead of a potential public listing.theedgemalaysia

The Software Ecosystem Question

CUDA's dominance stems not from technical superiority alone but from more than a decade of ecosystem development. By making its toolkit free and modeling TileLang on familiar CUDA-style workflows, DeepSeek is attempting to lower the switching costs that have kept developers locked into Nvidia's platform. Whether it can attract the developer adoption needed to make Huawei's chips a viable alternative at scale remains the central question.wionews+1