It has recently been noted that many discs around T Tauri stars appear to comprise only a few Jupiter masses of gas and dust. Using millimetre surveys of discs within six local star formation regions, we confirm this result, and find that only a few per cent of young stars have enough circumstellar material to build gas giant planets, in standard core accretion models. Since the frequency of observed exoplanets is greater than this, there is a `missing-mass' problem. As alternatives to simply adjusting the conversion of dust flux to disc mass, we investigate three other classes of solution. Migration of planets could hypothetically sweep up the disc mass reservoir more efficiently, but trends in multiplanet systems do not support such a model, and theoretical models suggest that the gas accretion time-scale is too short for migration to sweep the disc. Enhanced inner-disc mass reservoirs are possible, agreeing with predictions of disc evolution through self-gravity, but not adding to millimetre dust flux as the inner disc is optically thick. Finally, the incidence of massive discs is shown to be higher at the protostellar stages, Classes 0 and I, where discs substantial enough to form planets via core accretion are abundant enough to match the frequency of exoplanets. Gravitational instability may also operate in the Class 0 epoch, where half the objects have potentially unstable discs of >~30 per cent of the stellar mass.However, recent calculations indicate that forming gas giants inside 50au by instability is unlikely, even in such massive discs. Overall, the results presented suggest that the canonically `protoplanetary' discs of Class II T Tauri stars have globally low masses in dust observable at millimetre wavelengths, and conversion to larger bodies (anywhere from small rocks up to planetary cores) must already have occurred.