The class of models in which a QSO is powered by a massive black hole accreting gas produced in a dense galactic nucleus is examined. Application of the standard loss-cone diffusion theory shows that, for reasonable stellar densities, tidal disruption of stars cannot provide an effective source of fuel. Tidal disruption may be effective if collective effects enhance the rate of diffusion into disruptive orbits. In the absence of enhanced relaxation, stellar collisions can provide sufficient gas at lower densities than are required for tidal disruption. This process still requires densities higher than can be associated with the isothermal core of a normal galaxy; for example, collisions in a nucleus of 10 to the 9th solar masses and density 10 to the 9th solar masses/cu pc release gas at a rate about 1 solar mass/yr as a result of coalescence followed by rapid stellar evolution. Less extreme densities are required if active nuclei store mass during quiescent periods and this mass then becomes available through an instability, feeding the hole for a shorter period of intense activity. A mechanism for such intermittent activity may be storage and release of gas in a disk.