I wouldn't say "don't worry about performance". It's good to at least have an idea of where you're heading, so that you can build in anticipation where it makes sense. However, the other end of the spectrum, where you are definitely heading, is premature optimization. As Geon says, a given modern, non-virtualized server with a well-design codebase, can easily handle thousands of connections for casual games, and hundreds to low thousands for more real-time games.
First, get the game going on one server, but keep your code very modular. For example, can you cut out the authentication/authorization subsystem without affecting too much other code? If so, it becomes easier to pull parts out when 1) it is coded well, and 2) you have cashflow to spend on a dev to focus on this for the company.