There is a class of business metric that is technically true and strategically useless. Seat count at SaaS companies that measure active users as licensed seats. Page views at publishers who never track whether anyone read past the headline. And now: tokens generated, where every token is profitable as long as someone is paying for compute.
The problem isn't the math. The problem is what the metric leaves out. Profitable tokens and valuable tokens are not the same set. You can run a system that generates unlimited profitable tokens for no business purpose whatsoever. The economics of production tell you nothing about the economics of value creation.
This is the activity-equals-productivity trap in a new coat. Engineering teams learned this the hard way with story points: a team can close 40 points per sprint while shipping nothing users need. The token equivalent is already visible in the gap between AI investment and AI-driven revenue at most organizations.
"Every token is profitable" is a supply-side metric dressed up as a business case. The demand side - whether any of those tokens should have been generated - is the question it doesn't answer.
Thoughts? Find me on Bluesky.