We investigated the disc-halo connection in massive (M⋆ > 5 × 1010 M⊙) disc galaxies from the cosmological hydrodynamical simulations EAGLE and IllustrisTNG, and compared this connection with that inferred from the study of H I rotation curves in nearby massive spirals from the Spitzer Photometry and Accurate Rotation Curves dataset. We find that discrepancies between the simulated and observed discs arise both on global and local scales. Globally, the simulated discs inhabit halos that are a factor ∼4 (in EAGLE) and ∼2 (in IllustrisTNG) more massive than those derived from the rotation curve analysis of the observed dataset. We also used synthetic rotation curves of the simulated discs to demonstrate that the recovery of the halo masses from rotation curves are not systematically biased. We find that the simulations predict systems dominated by dark matter with stellar-to-total enclosed mass ratios that are a factor of 1.5−2 smaller than real galaxies at all radii. This is an alternative manifestation of the ‘failed feedback problem,’ since it indicates that simulated halos hosting massive discs have been too inefficient at converting their baryons into stars, possibly owing to an overly efficient stellar and/or AGN feedback implementation.