There's a word circulating in enterprise AI circles right now: "tokenmaxxing." It describes what many large companies have been doing for the past 18 months - pumping as many tokens as possible through AI APIs (tokens are the chunks of text an AI model processes; more tokens means more data sent to the model, and a bigger bill) in hopes that volume alone would produce better business outcomes. Uber's COO Andrew Macdonald says that bet is getting harder to defend.
Macdonald's skepticism matters because Uber is exactly the kind of company that should be a showcase for enterprise AI ROI. The company runs millions of transactions daily, has rich behavioral data, and has the engineering budget to experiment broadly. If a company like Uber is struggling to connect AI token expenditure to measurable returns, smaller organizations spending proportionally on AI tooling should pay attention.
What Tokenmaxxing Actually Costs
The term captures a real pattern. Teams spin up AI features, integrate LLM APIs into workflows, and then - because it's easy to keep calling the API - they call it constantly. Context windows fill up. Prompts get longer. Costs compound. The assumption is that more AI input produces proportionally better output, but that relationship breaks down quickly past a certain point.
For a company running at Uber's scale, even a modest increase in per-request token usage multiplies into meaningful infrastructure spend. The ROI calculation requires knowing not just what the AI costs, but what it produced - a number many teams cannot actually measure.
The Broader Accounting Problem
Macdonald's comments reflect a shift in how mature AI adopters are starting to think about this spending. The 2024-2025 "just try everything" posture is giving way to harder questions: Which AI workflows actually reduced headcount, increased revenue, or cut support costs? Which ones are running in production but not moving any metric?
This reckoning was predictable. Early AI adoption cycles tend to reward experimentation; the second phase rewards measurement. Companies that can tie specific AI deployments to specific outcomes will know where to keep spending. Companies that cannot will be the ones cutting AI line items in Q3 and Q4.
For freelancers and small teams, the lesson is simpler: the same question applies at any budget. Whether you're paying $20/month or $200/month for AI tools, it's worth occasionally asking which ones you'd actually miss if they disappeared tomorrow. The ones you wouldn't miss are the ones doing tokenmaxxing on your credit card.