Mammalian genomes are folded in a hierarchy of topologically associating domains (TADs), subTADs and looping interactions. The nested nature of chromatin domains has rendered it challenging to identify a sensitive and specific metric for detecting subTADs and quantifying their dynamic reconfiguration across cellular states. Here, we apply graph theoretic principles to quantify hierarchical folding patterns in high-resolution chromatin topology maps. We discover that TADs can be accurately detected using a Louvain-like locally greedy algorithm to maximize network modularity. By varying a resolution parameter in the modularity quality function, we accurately partition the mouse genome across length scales into a hierarchical nested structure of network communities exhibiting a wide range of sizes. To distinguish high probability subTADs from the full detected set, we developed and applied a new hierarchical spatial variance minimization method. Moreover, we identified a large number of dynamically altered communities between pluripotent embryonic stem cells and multipotent neural progenitor cells. Cell type specific boundaries correlate with trends in dynamic occupancy of the architectural protein CTCF, thereby validating their biological relevance. Together, these data demonstrate the utility of metrics from network science in quantifying a nested hierarchy of dynamic 3D chromatin communities across length scales. Our findings are significant toward unraveling the link between higher-order genome folding and gene expression during healthy development and the deregulation of molecular pathways linked to disease.