scaling real time data processing when things get really busy?

Hey everyone, I'm struggling with a system that has to take in and process tons of real-time events. Things get slow when the event rate really jumps up. We've thought about sharding our processing part, but I'm curious if there are other good ways to deal with these huge, temporary load spikes without slowing down. What approaches or tech have worked for you when real-time data pipelines get super busy?