The rise of Cloud Computing has progressively dimmed the interest in volunteer and peer-to-peer computing, in general. However, efficient and cost-effective large scale distributed collaborative environments cannot be achieved leveraging upon the Cloud alone. In this paper, we propose a novel hybrid P2P/cloud approach where components and protocols are autonomically configured according to specific target goals, such as cost-effectiveness, reliability and availability. The proposed approach is based on the Networked Autonomic Machine (NAM) framework, which allows distributed system designers to include different kinds of cost and performance constraints. As an example, we show how the NAM-based approach can be used to design collaborative storage systems, enabling the definition of an autonomic policy to decide, according to cost minimization and data availability goals, how to part data chunks among peer nodes and Cloud, based on the local perception of the P2P network.