Huawei to open-source AI chip toolkit to challenge Nvidia

Huawei will open-source its Compute Architecture for Neural Networks (CANN), the software toolkit used with its Ascend AI processors, the company announced at its developer conference in Beijing on August 5, 2025.

Huawei, based in Shenzhen, said the move is intended to make it easier for developers to use Ascend chips.

The company has discussed building an open-source Ascend ecosystem with major Chinese AI companies, business partners, universities, and research institutions.

This initiative comes as China prioritizes technology self-sufficiency, and follows recent security concerns raised by Beijing over Nvidia’s H20 chip.

Huawei’s open-source approach provides public access to CANN’s source code, allowing third-party developers to modify, share, or expand its functionality.

Nvidia’s competing CUDA toolkit remains proprietary, and last year Nvidia updated its license to prevent use of CUDA on non-Nvidia hardware through code translation layers.

🔗 Source: South China Morning Post

🧠 Food for thought

1️⃣ Open-sourcing can help overcome entrenched software ecosystem disadvantages

Huawei’s decision to open-source CANN reflects a strategic response to persistent software challenges that have limited adoption of its AI chips.

Despite launching the Ascend 910 chip in 2019 with performance metrics that reportedly outperform Nvidia’s Tesla V100, delivering up to 256 teraFLOPS under FP16 compared to competitors, Huawei has struggled with software usability issues.

Users have reported that Huawei’s CANN software makes the hardware “difficult to use and prone to crashes,” while developers cite difficulties effectively using Ascend chips.

The lack of a robust developer community has further hampered the onboarding process for new users of Huawei’s AI technologies.

By open-sourcing CANN, Huawei is inviting the global developer community to help solve these fundamental usability problems that have prevented its technically capable hardware from gaining wider adoption.

This approach reflects how other technology companies have used open-source strategies to accelerate ecosystem development when competing against established proprietary platforms.