Most businesses are measuring AI wrong, and it’s costing them

Amazon just shut down its AI leaderboard tracking internal token usage. The gamification was driving more AI-powered tasks but fewer useful results. “Please don’t use AI just for the sake of using AI,” the Amazon SVP instructed his staff. Amazon is not alone. Uber blew past its 2026 artificial intelligence coding budget in just four months. Google’s CEO, Sundar Pichai, revealed that the company’s token usage has grown sevenfold in a year. Many companies including Meta, Microsoft and Salesforce are reportedly pushing to limit token usage. It’s unsurprising what happens when you set the wrong incentives: You get the wrong results. You get what you measure. Hype, these days, is invariably accompanied by jargon. In an attempt to demonstrate corporate progressiveness, boardrooms and C-suites across America scramble to keep up with modern phraseology. They’re throwing around terms like tokens per query, cost per inference, GPU hours, and even model utilization. “Tokens are the new oil for the enterprise” is the latest slogan; tokens are apparently the new measure for AI adoption and productivity, the proof of AI discipline, the real unit for driving AI ROI. “Tokenmaxxing” has topped the charts for weeks. It all sounds smart, but it’s the wrong conversation. While companies get better at measuring AI spend, many still have no idea whether their AI investments are driving increased revenue, creating faster decisions, reducing friction, or creating any tangible and measurable advantage at all. With all the token math, they know what the intelligence costs, but not whether the intelligence is useful. And, very quickly, the AI ecosystem is running into unsustainable economics. The dashboards are getting better; returns are not. Word counts and lines of code are proliferating, but none of that trickles down to the bottom line. Cost Obsession is Short-sighted After a decade of cloud overspending, finance leaders are now homing in on everything AI. T