Recent observations using the Atacama Cosmology Telescope (ACT) have provided ground-based cosmic microwave background (CMB) maps with higher angular resolution than the Planck satellite. These have the potential to put interesting constraints on models resolving the 'Hubble tension'. In this paper we fit two models of Early Dark Energy (EDE) (an increase in the expansion rate around matter/radiation equality) to the combination of ACT data with large-scale measurements of the CMB either from the WMAP or the Planck satellite (including lensing), along with measurements of the baryon acoustic oscillations and uncalibrated supernovae luminosity distance. We study a phenomenological axion-like potential ('axEDE') and a scalar field experiencing a first-order phasetransition ('NEDE'). We find that for both models the 'Planck-free' analysis yields non-zero EDE at 2σ and an increased value for H0 ∼ 70 − 74 km/s/Mpc, compatible with local measurements, without the inclusion of any prior on H0. On the other hand, the inclusion of Planck data restricts the EDE contribution to an upper-limit only at 95% C.L. For axEDE, the combination of Planck and ACT leads to constraints 30% weaker than with Planck alone, and there is no residual Hubble tension. On the other hand, NEDE is more strongly constrained in a Planck+ACT analysis, and the Hubble tension remains at ∼ 3σ, illustrating the ability for CMB data to distinguish between different EDE models. We further explore the apparent inconsistency between the Planck and ACT data and find that it comes (mostly) from a slight tension between the temperature power spectrum at multipoles around ∼ 1000 and ∼ 1500. Finally, through a mock analysis of ACT data, we demonstrate that the preference for EDE is not driven by a lack of information at high-when removing Planck data, and that a ΛCDM fit to the fiducial EDE cosmology results in a significant bias on {H0, ω cdm }. More accurate measurements of the TT CMB power spectra above ∼ 2500 and EE between ∼ 300 − 500 will play a crucial role in differentiating between EDE models.