In April I used $4,000 of enterprise Claude tokens in one week.
I was not trying to. I was just working.
If I replicated that across 100 engineers, the bill would be $400,000 a week. That could bankrupt a company in days. Enterprise AI at that rate is not a productivity tool. It is an existential liability.
The SpaceX compute deal changed the economics for Pro and Max subscribers. More capacity, higher limits, reasonable pricing. The $100 Pro subscription is now a viable option for serious individual use. But organisations that need enterprise restrictions, audit trails, data governance, and security controls cannot use the Pro tier. They are stuck with enterprise pricing that was calibrated for a different era of usage.
The constraint is not the tool. It is the commercial model.
Organisations that need AI the most, the ones operating in regulated industries with real compliance obligations, are the ones for whom the economics are most broken. The ones who can use a personal Pro subscription and look the other way are fine. The ones who cannot are left choosing between compliance and competitiveness.
PromptQ is a partial answer to one part of this: a better specified prompt costs fewer tokens. The economic constraint is also a specification quality incentive. Every word counts. But that only goes so far.
The bigger question is whether the commercial model will evolve to serve the regulated enterprise, or whether the regulated enterprise will be left behind while the less constrained organisations accelerate.