The first neutron star-neutron star (NS-NS) merger was discovered on August 17, 2017 through gravitational waves (GW170817) and followed with electromagnetic observations. This merger was detected in an old elliptical galaxy with no recent star formation. The origin of this event is yet unknown. We perform a suite of numerical calculations to understand the formation mechanism of this merger. We probe three leading formation mechanisms of double compact objects: classical isolated binary star evolution, dynamical evolution in globular clusters, and nuclear cluster formation to test whether they are likely to produce NS-NS mergers in old host galaxies. Our simulations show current NS-NS merger rates at the level of 6x10^-2 yr^-1 from binary stars, 2x10^-4 yr^-1 from globular clusters and 6x10^-5 yr^-1 from nuclear clusters for all elliptical galaxies within 100 Mpc from Earth. These models are thus in tension with the detection of GW170817 with an observed rate of 1 yr^-1. If the detection of GW170817 is not a statistical coincidence our results imply that either (i) physics in our three models is not correct in the context of the evolution of stars that can form NS-NS mergers or that (ii) another very efficient formation channel with a long delay time between star formation and merger is at play. Recent observations show that most O/B stars are found in binary systems with a prevalence to very close orbits. Strong tidal forces may inhibit the expansion of these stars in close binaries leading to the formation of relatively wide NS-NS systems with long delay times. This so called homogeneous binary evolution may be the prime channel for the formation of GW170817, despite the fact that this mechanism was claimed so far to work only for black hole formation.