Would you rather double your users or halve your AI cost?

You have one quarter and enough engineering capacity to do only one:

A. Double active users, with the same unit economics

B. Cut AI cost per successful outcome by 50%, with no user growth

The startup answer is supposed to be growth.

But if every new user increases an invisible loss, growth just helps you reach the wall faster.

Which would you choose—and at what stage would your answer change?