Huawei Releases AI Model Without Nvidia Chips

Huawei's openPangu release has sparked attention in Washington and within the AI community. On June 30, Huawei announced that its openPangu-2.0-Flash model had gone live with 92 billion total parameters, a 512K context window, and training and inference operators published through the Ascend Tribe community on GitCode.

The model's size is significant, but what's more notable is the fact that it was built without using Nvidia chips. Huawei's announcement comes as part of its larger open-source roadmap for the second half of 2026.

While Huawei has announced the 505 billion parameter Pro model, it has yet to release full weights for this model, with only a promise of availability in July. The Pro model is expected to have 18 billion active parameters and achieve twice the single-card throughput of other mainstream open-source models on Ascend compute.

Experts caution that Nvidia's position remains strong due to its years-long head start and established ecosystem. However, Huawei's move could indicate a shift in China's approach to AI development, with a focus on domestic stacks rather than reliance on Nvidia hardware.