Huawei announces new AI infrastructure as Nvidia gets locked out of China (techcrunch.com)

Huawei unveiled SuperPoD Interconnect at its Huawei Connect keynote — a high‑scale interconnect that can link up to 15,000 graphics cards, including Huawei’s Ascend AI chips, to form massive compute clusters. The technology is positioned as a competitor to Nvidia’s NVLink, enabling high‑bandwidth, low‑latency communication across many accelerators so users can aggregate raw compute for large‑scale training and inference. The announcement follows China’s ban on domestic firms buying Nvidia hardware (including the RTX Pro 600D servers), making Huawei’s push into cluster interconnects both timely and strategic. For the AI/ML community, SuperPoD matters because it addresses a core bottleneck: access to large, tightly coupled compute pools needed for training modern models. Even if individual Ascend chips trail Nvidia GPUs on single‑device performance, an effective interconnect and software stack can close gaps by enabling greater horizontal scale. The move also amplifies the geopolitical and supply‑chain shift toward domestic alternatives in China, raising questions about interoperability (CUDA vs. Huawei’s stack), real‑world bandwidth/latency performance versus NVLink, and software ecosystem maturity — all key factors for researchers and enterprises evaluating non‑Nvidia infrastructure for model scaling.