Still your point is very valid and I guess only time can tell!
For websocket, you can move it to another "not-serverless" hosting as well.
Splitting then combine is far more easier than combine then splitting.
At that point, I'd rather just go fully in on Kubernetes and keep the benefits of containers, rather than using two cloud providers.
Me, setting that up from scratch on a single VPS, that might get hacked if I mess it up, and then my customer's availably is compromised? I wouldn't want to be in that situation.
Currently we're starting out by looking at feasibility with Apollo. If we can manage to get that to work, then hopefully we can move on to other frameworks.
Pusher does Websockets. It can only be so opinionated, since it's a fairly focused, well-defined problem: Websockets. I'ts less likely to be subject to breaking changes versus Now. I'd say they have both roughly the same chance of randomly shutting down, with only a few months notice.
It killed the category in my mind, it isn't even your fault. I now feel far more comfortable running on bread-and-butter tools, like AWS/GCP/Azure with Kubernetes. The overhead sucks, but at least I can plan for things better than I can when hosting with fast-moving PaaS providers.
Hosting is too important of a problem to wholly outsource entirely to a PaaS platform if it's subject to shuffling at a rate I can't control (exception: personal projects with no real availability expectations). That's the double edged sword with platforms like these. They let you get your application online rapidly, but you could be subject to unforeseen technical debt at the hosting provider's will. I learned (the hard way) that I'm not a fan on managing risk under those terms.