Using testing data from 2.1 million students in 10,000 schools in 49 states (plus D.C.), we investigate the role of remote and hybrid instruction in widening gaps in achievement by race and school poverty. We find that remote instruction was a primary driver of widening achievement gaps. Math gaps did not widen in areas that remained in-person (although there was some widening in reading gaps in those areas). We estimate that high-poverty districts that went remote in 2020-21 will need to spend nearly all of their federal aid on academic recovery to help students recover from pandemic-related achievement losses.