Tokenmaxxing isn't an AI strategy
Tokenmaxxing—scaling model size and inference cost in lockstep—is an operational treadmill, not a strategy.
The Register examines whether AI ROI calculations account for actual deployment economics, not just capability benchmarks. The gap between lab performance and profitable production reveals whether a model's cost-per-token serves a real market demand or subsidises investor narratives.
Oliver's take: Bigger models, bigger tokens, bigger inference bills. Eventually someone has to buy the thing and actually use it. That date keeps moving.
Original source
