CloudMatrix 384 super-node uses 6912 x 400G OSFP silicon photonic (SiPh) Linear Drive Pluggable Optics (LPO) optical modules and 3168 fibers, connecting 384 Ascend 910C computing chips through a complete mesh interconnection architecture. In the AI era, Huawei provides a full range of GE to 800GE optical modules, featuring three major capabilities: Spanning (ultra-long transmission), Stable (ultra-high reliability), and Secure (ultra-solid security). Together, they ensure resilient data center interconnectivity and empower. Deep dive into Huawei's AI super-node: 384 Ascend 910C chips, 6912×400G OSFP SiPh LPO modules (1:18 ratio), 1. 2% failures stem from optics & how QSFPTEK cuts costs by 69. On May 14, 2025, the "2025 Chip and Optical Forum" hosted by HiSilicon and organized by. Huawei's CloudMatrix 384 Supernode outperforms Nvidia's 180 PFLOPs NVL72 with 300 PFLOPs, higher memory bandwidth, and sanction‑proof design, fueling China's AI infrastructure ambitions. Huawei announced the CloudMatrix 384 Supernode—a 16‑rack AI compute cluster achieving 300 petaflops of BF16. This paper introduces Huawei CloudMatrix, a next-generation AI datacenter architecture, realized in the production-grade CloudMatrix384 supernode. It's designed to rival NVIDIA's GB200 NVL72, and early reports suggest it may achieve ~2× the throughput in BF16 compute. Let's explore how it. Huawei has officially entered the AI supercomputing race with the launch of CloudMatrix 384, a large-scale AI training system designed to compete with Nvidia's most powerful offerings.
