At what scale does serverless stop being the right answer?

I keep telling teams that serverless is good for bursty, unpredictable workloads. But where is the real inflection point? At 10,000 RPS? At sustained constant load? I have heard very different answers from very credible people and I still do not have mine.