NGPU is a Web3 decentralized AI computing power network, and a DePin dApp in tech architecture. It is dedicated to providing permissionless-access of idle GPU nodes, offering cost-effective, user-friendly, and stable GPU computing resources for various AI applications, thereby delivering enterprise-level AI inference services. Its main innovative technologies include pay-per-task GPU real usage billing, intelligent allocation, and efficient HARQ network transmission. This allows NGPU to achieve auto-scaling, high reliability, and low latency on permissionless-access GPU nodes.