Is there a limit as to how far you can go with subdivisions?

I would argue that there is a computational limit, but not a theoretical one. At some point, there won't be enough precision to split an edge.