[munchbunny]

I'm guessing Twitter runs their own infra for the really heavy throughput stuff. They're big enough that I'd imagine economies of scale kicked in a long time ago.

It sounds like they moved stuff like logging/metrics and query systems onto GCP, which makes sense because as another poster said utilization is probably bursty.

[derekalyon]

Indeed, these workloads are bursty. They also tend to involve running lots of different processing frameworks over the same data, which makes the value proposition for separating compute and storage stronger.