[anamax]

The second doesn't (necessarily) have to scale.

It has to handle the projected load, with some safety margin.

"projected until when?" you ask?

Well, your projections will be wrong. And, your needs will change, which means that you're going to be changing the system anyway.

That's where judgment comes in.

Suppose that each customer is worth $10/month and that the "not scalable" version can handle 1M customers. That's at least $5M MRR when the system is at 50%.

If the scalable version takes significantly longer to develop, you might choose to put it off until you have more money/resources, especially since there will probably be other changes. (And, you don't know where the actual bottlenecks are.)