I appreciate the flexibility of keeping them decoupled more now though. Your illustrations helped. More than scalability, I like the opportunity to have a different UI solution (as in your last diagram above).
Re keeping the UI solution on e.g. Vercel, though, I don’t see the point if the UI to render is still waiting for data from Strapi hosted on e.g. an EC2 instance. Even if Vercel hosts the FE solution on “edge nodes” / closest to user’s location, that edge node needs to communicate with an EC2 instance and wait for its response. I doubt whether it’s worth the extra expense and fragmentation of process running location in “startup” cases.
Cheers