It depends on how it's implemented. If it's a fixed window, then your absolute ceiling is tokens/windows in a month. If it's a function of other usage, like a timeshare, you're still paying for some price for a month and you get what you get without paying more per token. There's an intrinsic limit based on how many tokens the model can process on that gpu in a month anyway, even if it's only you.