Résumé : We provide a revised assessment of the number of exoplanets that should be discovered by Gaia astrometry, extending previous studies to a broader range of spectral types, distances, and magnitudes. Our assessment is based on a large representative sample of host stars from the TRILEGAL Galaxy population synthesis model, recent estimates of the exoplanet frequency distributions as a function of stellar type, and detailed simulation of the Gaia observations using the updated instrument performance and scanning law. We use two approaches to estimate detectable planetary systems: one based on the S/N of the astrometric signature per field crossing, easily reproducible and allowing comparisons with previous estimates, and a new and more robust metric based on orbit fitting to the simulated satellite data. With some plausible assumptions on planet occurrences, we find that some 21,000 (+/-6000) high-mass (1-15M_J) long-period planets should be discovered out to distances of ~500pc for the nominal 5-yr mission (including at least 1000-1500 around M dwarfs out to 100pc), rising to some 70,000 (+/-20,000) for a 10-yr mission. We indicate some of the expected features of this exoplanet population, amongst them ~25-50 intermediate-period (P~2-3yr) transiting systems.