Abstract We study mass ejection from accretion disks around newly formed hypermassive neutron stars (HMNSs). Standard kilonova model fits to GW170817 require at least a lanthanide-poor (“blue”) and lanthanide-rich (“red”) component. The existence of a blue component has been used as evidence for an HMNS remnant of finite lifetime, but average disk outflow velocities from existing long-term HMNS simulations fall short of the inferred value (∼0.25c) by a factor of ∼2. Here we use time-dependent, axisymmetric hydrodynamic simulations of HMNS disks to explore the limits of the model and its ability to account for observations. For physically plausible parameter choices compatible with GW170817, we find that hydrodynamic models that use shear viscosity to transport angular momentum cannot eject matter with mass-averaged velocities larger than ∼0.15c. While outflow velocities in our simulations can exceed the asymptotic value for a steady-state neutrino-driven wind, the increase in the average velocity due to viscosity is not sufficient. Therefore, viscous HMNS disk winds cannot reproduce by themselves the ejecta properties inferred from multi-component fits to kilonova light curves from GW170817. Three possible resolutions remain feasible within standard merger ejecta channels: more sophisticated radiative transfer models that allow for photon reprocessing between ejecta components, inclusion of magnetic stresses, or enhancement of the dynamical ejecta. We provide fits to our disk outflow models once they reach homologous expansion.