Huawei's OceanStor A600 is a next-generation high-performance distributed file storage solution for AI, accelerating the entire training and inference process of large models with extreme performance. A single chassis can provide 160GB/s bandwidth and 3.6 million IOPS. Based on an innovative data controller separation architecture, the product supports flexible expansion to 32 controllers per cluster, matching the smooth evolution requirements of multimodal large models with trillions/tens of parameters. It supports new data paradigms such as vectors, tensors, and KV Cache; it has the capability to accelerate large model inference, reducing the first token response latency by 78% and increasing inference throughput by 60% through a multi-level KV Cache architecture, enabling long sequence inference with super-context windows and optimizing inference cost-effectiveness.