- Huawei manufactures its Cann Open Open Gpu Toolkit Kit Kit to challenge Nvidia’s patented CUDA platform
- The domain of about 20 years of CUDA has locked developers in the NVIDIA hardware ecosystem exclusively
- Cann provides multiple layer programming interfaces for AI applications in Huawei GPUs
Huawei has announced plans to make its Cann software tool kit for open source of GPU Ascend AI, a movement directly directed to challenging the CUDA domain of long data from NVIDIA.
CUDA, often described as a “pit” or “closed swamp”, has been seen as a barrier to developers looking for multiplatform compatibility for some for years.
Its close integration with Nvidia hardware has locked developers in a unique supplier ecosystem for almost two decades, with all efforts to bring CUDA functionality to other GPU architectures through translation layers blocked by the company.
Opening Cann to developers
Cann, abbreviation for computing architecture for neural networks, is Huawei’s heterogeneous computer framework designed to help developers create AI applications for their GPU Ascend AI.
The architecture offers multiple programming layers, providing the developers options to build high -level applications and performance intensive.
In many ways, it is Huawei’s equivalent to Cuda, but the decision to open its source code indicates an intention to grow an alternative ecosystem without the restrictions of a patented model.
According to the reports, Huawei has already begun discussions with the main Chinese players of AI, universities, research institutions and commercial partners to contribute to an open source Asce development community.
This scope could help accelerate the creation of optimized tools, libraries and AI frameworks for Huawei GPUs, which potentially makes them more attractive to developers who currently depend on NVIDIA hardware.
Huawei’s hardware performance has constantly improved, with statements that certain Ascend chips can overcome NVIDIA processors in specific conditions.
Reports such as the reference results of Cloudmatrix 384 against NVIDIA that execute Deepseek R1 suggest that Huawei’s performance trajectory is closing the gap.
However, unprocessed performance will not guarantee the migration of the developer without stability and equivalent software support.
While the open emission cannon could be exciting for developers, its ecosystem is in their early stages and may not be close to CUDA, which has been refined for almost 20 years.
Even with the open source status, the adoption can depend on how well Cann admits the existing frames, particularly for emerging workloads in large language models (LLM) and the writer of AI and AI tools.
Huawei’s decision could have broader implications beyond the convenience of the developer, since the open emission cannon is aligned with the broader impulse of China for technological self -sufficiency in AI computing, reducing the dependence of western chip manufacturers.
In the current environment, where US restrictions are directed to Huawei hardware exports, build a robust national software battery for AI tools becomes as critical as improving chip performance.
If Huawei can successfully encourage an open source vibrant community around Cann, it could present the first alternative would be CUDA in years.
Even so, the challenge lies not only in the availability of code, but also in the creation of trust, documentation and compatibility at the scale that Nvidia has achieved.
Via Hardware Toms
