At the AI Summit, Huang Jin, Vice President of Huawei Cloud, emphasized that the soaring demand for computational resources in large model training and inference has rendered traditional computing architectures insufficient. To tackle this challenge, the super-node architecture has emerged as a groundbreaking solution. Huawei Cloud has responded by introducing the CloudMatrix 384 super-node, which addresses issues such as communication efficiency, memory bottlenecks, and reliability. Offering up to 300 Pflops of computing power, a 67% enhancement over NVIDIA's offerings, this super-node showcases six significant technological advancements. Notably, the EMS elastic memory storage technology significantly boosts resource utilization and performance, while reducing the first token latency by up to 80%.