Microservice architectures promise independent evolution through loose coupling, yet large systems often exhibit strong dependency concentration around a small set of services. In an exploratory industrial case study of a product composed of 267 microservices, we triangulated multiple dependency signals - compile-time, run-time, and task dependencies - and iteratively validated our interpretations with practitioners. We observed a recurring macro-structure in the dependency network that resembles a black hole: a dense core of dependency magnets, a transitional region of services increasingly entangled with the core, and an outer region of lightly connected services. Based on these observations, we propose the dependency black hole theory, mapping the network to the black hole anatomy of a singularity, an event horizon, and an accretion disk, and formulating three hypotheses about how dependency concentration emerges and evolves at scale. The theory provides an explanatory lens for reasoning about dependency growth, identifying services at risk of becoming dependency magnets, and motivating governance interventions. We outline practical implications and directions for longitudinal and multi-case validation.