NetGPT is an AI-native network architecture designed to enhance personalized generative services through a collaborative cloud-edge methodology. The authors propose a framework that synergizes appropriate large language models (LLMs) at the edge and cloud based on their computing capacity, enabling efficient orchestration of heterogeneous distributed communication and computing resources. Edge LLMs leverage location-based information to complete personalized prompts, while cloud LLMs handle more complex tasks. The feasibility of NetGPT is demonstrated using low-rank adaptation-based fine-tuning of open-source LLMs like GPT-2-base and LLaMA models, showing superior performance compared to alternative cloud-edge collaboration or cloud-only techniques. The architecture emphasizes deep integration of communications and computing resources and a logical AI workflow for efficient service provisioning. Additionally, edge LLMs contribute to intelligent network management and orchestration, such as popularity prediction and intent inference. The authors discuss the architectural changes required for an AI-native network, including converged C\&C resource management, data processing and privacy protection, personalized profiling, and a logical AI workflow. NetGPT offers a promising solution for provisioning personalized generative services and enhancing network efficiency.NetGPT is an AI-native network architecture designed to enhance personalized generative services through a collaborative cloud-edge methodology. The authors propose a framework that synergizes appropriate large language models (LLMs) at the edge and cloud based on their computing capacity, enabling efficient orchestration of heterogeneous distributed communication and computing resources. Edge LLMs leverage location-based information to complete personalized prompts, while cloud LLMs handle more complex tasks. The feasibility of NetGPT is demonstrated using low-rank adaptation-based fine-tuning of open-source LLMs like GPT-2-base and LLaMA models, showing superior performance compared to alternative cloud-edge collaboration or cloud-only techniques. The architecture emphasizes deep integration of communications and computing resources and a logical AI workflow for efficient service provisioning. Additionally, edge LLMs contribute to intelligent network management and orchestration, such as popularity prediction and intent inference. The authors discuss the architectural changes required for an AI-native network, including converged C\&C resource management, data processing and privacy protection, personalized profiling, and a logical AI workflow. NetGPT offers a promising solution for provisioning personalized generative services and enhancing network efficiency.