A service that answers every request as fast as possible can be knocked over by a flood of them, whether from a bug, a scraper or an attack. Rate limiting caps how many requests a given client may make in a window, smoothing spikes and keeping capacity for everyone.
Well-designed limits return a clear signal when exceeded, so callers know to slow down and retry later rather than hammering harder. The goal is fairness and stability, not punishment, so limits are usually generous for normal use and bite only on abuse.