Faster startup -> lower serverless bills for a lot of shops. Shaving 25ms off a request might not seem like it's important, but if that saves you a million serverless seconds in aggregate, you just earned a bonus.
Millions auf serverless seconds? Better get a cheap server for all basic executions and only serverless for randomly huge traffic.
But what if you're Azure or Amazon? Then you definitely care a lot about this kind of thing
I’m fairly sure they have their own servers where they can just run their processes continuously in the background.
For any well trafficked site, the vast majority of their lambda invocations will be from a warm start.