At most mid-ocean ridges, a wide region of decompression melting must be reconciled with a narrow neovolcanic zone and the establishment of full oceanic crustal thickness close to the rift axis. Two competing paradigms have been proposed to explain melt focusing: narrow mantle upwelling due to dynamic effects related to in situ melt or wide mantle upwelling with lateral melt transport in inclined channels. Measurements of seismic attenuation provide a tool for identifying and characterizing the presence of melt and thermal heterogeneity in the upper mantle. We use a unique data set of teleseismic body waves recorded on the Cascadia Initiative's Amphibious Array to simultaneously measure seismic attenuation and velocity across an entire oceanic microplate. We observe maximal differential attenuation and the largest delays ([Formula: see text] s and δTS ~ 2 s) in a narrow zone <50 km from the Juan de Fuca and Gorda ridge axes, with values that are not consistent with laboratory estimates of temperature or water effects. The implied seismic quality factor (Qs ≤ 25) is among the lowest observed worldwide. Models harnessing experimentally derived anelastic scaling relationships require a 150-km-deep subridge region containing up to 2% in situ melt. The low viscosity and low density associated with this deep, narrow melt column provide the conditions for dynamic mantle upwelling, explaining a suite of geophysical observations at ridges, including electrical conductivity and shear velocity anomalies.