The demand for storage and computing of massive amounts of data has led to the emergence of multiple private clouds. However, there are data isolation and knowledge isolation between cloud services offered by different cloud service providers as a result of the need for privacy preservation. Existing solutions have introduced federated learning (FL) to train a global model collaboratively while ensuring the sensitive data remains within the private cloud. Nevertheless, the sharing of well-trained FL models remains inflexible, limiting the models’ potential value. And there still exists the lack of full lifecycle FL service frameworks that are designed to be adaptable for multi-cloud environments. Motivated by this, in this paper, we propose a serverless federated learning service ecosystem for multi-cloud collaborative environments to provide customized model training and fine-grained model sharing. Specifically, we improve Attribute-based Encryption (ABE) by designing a distributed secret key generation and management scheme to accomplish a fine-grained sharing control in a peer-to-peer multi-cloud environment. The evaluation results demonstrate the adaptability of our ecosystem in a multi-cloud environment.