With the robust growth in the social environment, millions of families have vast medical plans nowadays. Each client requires a personalized medical rescue decision, making an intelligent recommendation system highly important. In practice, wealthy families tend to purchase tailored medical services, while others tend to seek medical services from local community hospitals. Noticeably, trillions-scale of medical service transactions occurs daily, making the development of a fully automatic intelligent recommendation system to match the suitable medical services to clients urgent. In this work, we propose a novel Internet-scale deep architecture that automatically recommends personalized and tailored medical rescue services based on the previously discovered clients with different financial tolerance. Specifically, given a massive number of client families, first, we create a set of features that represents the multiple medically related attributes of each family. These features capture the income, health status, type of diseases, occupation, and so on. The features are then combined into a descriptive feature by multi-view learning, which can dynamically assign the importance of each feature channels. Based on the combined feature, affinity graphs are created on a large scale to represent the relations between the client families. Also, an efficient dense subgraph mining is used to categorize the million-scale clients into 10 different medical groups. Finally, various customized medical services can be tailored for each client in a fully automatic and efficient manner. Comprehensive experiments on our collected data sets demonstrated the competitiveness of our proposed intelligent recommendation system. Moreover, the results of visualized dense subgraph mining showed that the client families with different wealth levels can be distinguished accurately.