Chime famously (but maybe not externally) ran almost all infra off two co-located beefy servers all the way up to their millionth customer. Maybe that is dumb, but we never had any downtime or slowness that was a result of work we did. The APIs we would rely on sometimes though did.