- Huawei makes its cann ai source GPU toolbox to challenge the Cuda owner owner of Nvidia
- The domination of Cuda almost 20 years has locked developers in the Nvidia material ecosystem exclusively
- Cann provides multilayer programming interfaces for AI applications on Huawei GPUs Ascende
Huawei has announced its intention to make its Cann software toolbox for Ascend Ai Gpus Open Source, a decision aimed squarely to challenge the domination of Cuda for a long time of Nvidia.
Cuda, often described as a “ditch” or closed “swamp”, was considered a barrier for developers in search of multiplatform compatibility by some for years.
Its tight integration with Nvidia Hardware has locked developers in a single supplier ecosystem for almost two decades, with all efforts to provide CUDA functionality to other GPU architectures through translation layers blocked by the company.
Open Cann to developers
Cann, abbreviation of the calculation architecture for neural networks, is the heterogeneous IT framework of Huawei designed to help developers create AI applications for its GPU Ascens IA.
The architecture offers several programming layers, offering developers options to build high -level and high performance intensity applications.
In many ways, this is the equivalent of Huawei to Cuda, but the decision to open its source code indicates an intention to develop an alternative ecosystem without the restrictions of a owner model.
Huawei would have already started discussions with the main players in Chinese AI, universities, research institutions and trading partners on the contribution to a community of development of open source ancestry.
This awareness could help accelerate the creation of optimized tools, libraries and AI executives for Huawei GPUs, which makes them potentially more attractive for developers who are currently counting on Nvidia equipment.
The material performance of Huawei AI have improved regularly, with statements that certain Ascend chips can surpass Nvidia processors under specific conditions.
Reports such as the reference results of Cloudmatrix 384 against NVIDIA performing Deepseek R1 suggest that the performance trajectory of Huawei fills the gap.
However, raw performance alone will not guarantee the migration of developers without stability and equivalent software support.
While the Open-Source Cann could be exciting for developers, its ecosystem is at its beginnings and can not be close to Cuda, which has been refined for almost 20 years.
Even with open-source status, adoption may depend on how the SANN supports existing AI frameworks, in particular for emerging workloads in large language models (LLM) and the writer of AA tools.
Huawei’s decision could have broader implications beyond the convenience of developers, because the Open-Source Cann aligns with the wider push of China to technological self-sufficiency in AI computer science, reducing dependence on Western chipmakers.
In the current environment, where American restrictions target Huawei hardware exports, the creation of a robust domestic software battery for AI tools as critical as it is to improve flea performance.
If Huawei can succeed in promoting a dynamic open source community around Cann, it could present the first serious alternative to Cuda for years.
However, the challenge does not only reside in the availability of the code, but in the construction of confidence, documentation and compatibility on the scale that Nvidia has obtained.
Via Toms equipment
