Nvidia's AI hardware lead questioned by Chinese supernodes and AMD

DeepSeek founder and CEO Liang Wenfeng said Huawei's upcoming Atlas 950 SuperPod could match or replace Nvidia's GB200 and GB300 AI accelerator racks. He noted that four Huawei chips deliver the performance of one Nvidia GPU and that the Chinese supernode is slated for launch in the third or fourth quarter of 2026, despite a higher price.

At the same time, AMD corporate vice‑president Andrew Dieckman stated that the importance of Nvidia's CUDA software stack is fading. Customers are moving to higher‑level abstraction frameworks and alternative serving platforms, reducing CUDA's competitive moat. AMD highlighted its own ROCm platform as an open alternative.

Both executives signal a potential shift in the AI accelerator market, challenging Nvidia's longstanding dominance.