Huawei said at Huawei Connect 2026 in Shanghai on September 19 that it is opening access to computing resources on a 10,000-neural-processing-unit scale for partners and developers.

It also launched a 100 NPU-Hour Program for developers and pledged 5 billion yuan (US$746 million) over the next three years to support universities, research institutions, industry partners, and developers.

Huawei said its agentic computing development plan covers the Kunpeng and Ascend ecosystems.

The company also highlighted SuperPoD and SuperCluster development, ThinkPro in openEuler, and expanded programming support for the new Ascend 950 generation.

The announcement comes as Chinese AI developers are seeking alternatives to Nvidia-based computing capacity. Demand for AI compute in China is driving interest in domestic platforms such as Ascend.

Huawei has also been expanding Ascend software tools to compete with Nvidia’s CUDA ecosystem, as US export controls continue to constrain access to advanced US semiconductor technology and equipment.

🔗 Source: Huawei

Recent Huawei developments