> Reading the article, I didn’t see this answered: why not scale to more nodes if your workload is CPU bound?

It's an SDK that runs in users' apps. So userland code blocks the event loop, preventing outgoing heartbeats from the SDK