If binaries consisting of two ∼ 100 M ⊙ black holes exist they would serve as extraordinarily powerful gravitational-wave sources, detectable to redshifts of z ∼ 2 with the advanced LIGO/Virgo groundbased detectors. Large uncertainties about the evolution of massive stars preclude definitive rate predictions for mergers of these massive black holes. We show that rates as high as hundreds of detections per year, or as low as no detections whatsoever, are both possible. It was thought that the only way to produce these massive binaries was via dynamical interactions in dense stellar systems. This view has been challenged by the recent discovery of several 150 M ⊙ stars in the R136 region of the Large Magellanic Cloud. Current models predict that when stars of this mass leave the main sequence, their expansion is insufficient to allow common envelope evolution to efficiently reduce the orbital separation. The resulting black-hole-black-hole binary remains too wide to be able to coalesce within a Hubble time. If this assessment is correct, isolated very massive binaries do not evolve to be gravitational-wave sources. However, other formation channels exist. For example, the high multiplicity of massive stars, and their common formation in relatively dense stellar associations, opens up dynamical channels for massive black hole mergers (e.g., via Kozai cycles or repeated binary-single interactions). We identify key physical factors that shape the population of very massive black-holeblack-hole binaries. Advanced gravitational-wave detectors will provide important constraints on the formation and evolution of very massive stars.