The paper introduces a novel collaborative learning framework called Diffusion-Based Cloud-Edge-Device Collaborative Learning for Next POI Recommendations (DCPR), which leverages the diffusion model to provide region-specific and highly personalized Point-of-Interest (POI) recommendations on devices. DCPR operates with a cloud-edge-device architecture to reduce on-device computational burdens while maintaining recommendation accuracy and efficiency. The framework consists of three layers: cloud server, edge server, and device. The cloud server trains a global diffusion model, which is then sent to region-specific edge servers to capture regional preferences. Each edge server distributes a region-specific model to users within the region, which is further refined locally using personal data. The evaluation with two real-world datasets demonstrates DCPR's superior performance in recommendation accuracy, efficiency, and adaptability to new users and regions. Key contributions include the integration of the diffusion model into on-device POI recommendations, the development of a fast-adapting framework, and the introduction of an acceleration strategy to speed up inference.The paper introduces a novel collaborative learning framework called Diffusion-Based Cloud-Edge-Device Collaborative Learning for Next POI Recommendations (DCPR), which leverages the diffusion model to provide region-specific and highly personalized Point-of-Interest (POI) recommendations on devices. DCPR operates with a cloud-edge-device architecture to reduce on-device computational burdens while maintaining recommendation accuracy and efficiency. The framework consists of three layers: cloud server, edge server, and device. The cloud server trains a global diffusion model, which is then sent to region-specific edge servers to capture regional preferences. Each edge server distributes a region-specific model to users within the region, which is further refined locally using personal data. The evaluation with two real-world datasets demonstrates DCPR's superior performance in recommendation accuracy, efficiency, and adaptability to new users and regions. Key contributions include the integration of the diffusion model into on-device POI recommendations, the development of a fast-adapting framework, and the introduction of an acceleration strategy to speed up inference.