While a vast body of work has documented the importance of face-to-face interactions in science and innovation, the share of distributed teams in scientific research has steadily risen since ICT revolution of the 1990s. Building on the simple intuition that some ideas are more important than others, we explore how the upsurge of remote collaboration has shaped innovation in scientific discovery. To distinguish between breakthroughs and incremental improvements, we implement a new method of classifying scientific publications according to their novelty or disruptiveness. Using a difference-in-difference design, we exploit instances in which colocated teams become distributed and compare the disruptiveness of their publications before and after the split, controlling for the characteristics of their research field. We document a robust and significant negative impact of switching to a distributed model on breakthrough innovation. However, beginning in the 2010s, the negative impact tapers off and even becomes positive. We hypothesize that this reversal relates to improvements in key technologies needed for effective remote collaboration, including video conferencing and the cloud. Consistent with this view, we provide evidence that colocation matters less when the split involves team members in places with better broadband infrastructure.