In this work, we compare galaxies from the NIHAO and HESTIA simulation suites to ultra-diffuse galaxies (UDGs) with spectroscopically measured dynamical masses. For each observed UDG, we identify the simulated dark matter halo that best matches its dynamical mass. In general, observed UDGs are matched to simulated galaxies with lower stellar masses than they are observed to have. These simulated galaxies also have halo masses much less than would be expected given the observed UDG's stellar mass and the stellar mass─halo mass relationship. We use the recently established relation between globular cluster (GC) number and halo mass, which has been shown to be applicable to UDGs, to better constrain their observed halo masses. This method indicates that observed UDGs reside in relatively massive dark matter haloes. This creates a striking discrepancy: the simulated UDGs are matched to the dynamical masses of observed ones, but not their total halo masses. In other words, simulations can produce UDGs in haloes with the correct inner dynamics, but not with the massive haloes implied by GC counts. We explore several possible explanations for this tension, from both the observational and theoretical sides. We propose that the most likely resolution is that observed UDGs may have fundamentally different dark matter halo profiles than those produced in NIHAO and HESTIA. This highlights the need for a simulation that self-consistently produces galaxies of a stellar mass of in dark matter haloes that exhibit the full range of large dark matter cores to cuspy NFW-like haloes.