Even if they were to use a single cloud for the managed instances, this is not at all like the centralization of platform ownership. Here’s the critical difference.

If something happens to Twitter, say a methhead buys it and turns it into a propaganda machine, its users can only stop using it and/or move elsewhere. For this to have a significant effect, the a large part of the network of people has to move. Every individual has to do non-trivial amount of labour to do so. That’s hard.

If something happens to the cloud provider hosting some sizeable Mastodon server, the owner of the server can migrate (copy) the instance to another cloud provider, or their own hardware, switch the DNS records and shutdown the old one. Their users would only notice a brief interruptin. There’s no significant labour needed by the users, apart from perhaps logging in again. Only a much smaller amount of labour by the instance owner, compared to the labour needed for mass user migration performed by individual users.

And that’s the major difference that fundamentally changes the dynamics.