Wire · ai
Huawei Pangu Pro Trains 505 Billion Parameters Without Nvidia: Supply Chain Tells Different Story
◆ Sectors
◆ Geography
◆ Source
◆ Verified
Fusion42 · 1 August 2026 · Fusion42 review
Huawei's openPangu-2.0-Pro, a 505-billion-parameter language model, completed training using Ascend 910B NPUs without Nvidia hardware, marking a frontier-scale AI training on non-Nvidia chips. However, the chips contained key components fabricated by TSMC, indicating the training was not fully reliant on a domestic Chinese hardware stack, which remains unproven at scale.
This Wire brief sits within Fusion42's coverage of AI & ML.
◆ ◆ The Wire takeaway
You must factor in lingering foreign chip dependencies despite local training claims. Founders building AI hardware or cloud services in China face a short-term bottleneck as fully domestic chip and memory supply chains struggle to match global leaders.
◆ Coverage
1 source · 1 Aug 2026
◆ Related on Wire
◆ Topics