A growing number of major enterprises are discovering that their metrics for measuring artificial intelligence adoption are fundamentally flawed, leading to ballooning costs and diminished returns. Amazon recently shuttered its internal AI leaderboard that tracked token usage after realizing the gamification encouraged more AI-powered tasks but fewer useful outcomes. "Please don't use AI just for the sake of using AI," an Amazon SVP instructed staff, signaling a shift away from vanity metrics.
The scale of the problem extends well beyond Amazon. Uber blew past its AI coding budget for 2026 in just four months, while Google CEO Sundar Pichai revealed the company's token usage had grown sevenfold in a year. Meta, Microsoft, and Salesforce are reportedly among the firms now pushing to limit token consumption. The pattern suggests that setting the wrong incentives inevitably produces the wrong results.
At the heart of the issue is a corporate obsession with jargon-laden metrics like tokens per query, cost per inference, GPU hours, and model utilization. The phrase "tokens are the new oil for the enterprise" has emerged as a misguided slogan, treating tokens as the definitive proof of AI adoption and productivity. This "tokenmaxxing" culture prioritizes volume over value, according to the Fast Company report. The problem is that measuring AI by raw usage ignores whether the technology is actually solving business problems or generating meaningful returns.
As companies race to demonstrate AI progress to boardrooms and C-suites, the disconnect between activity and outcomes is costing them dearly. The experience at these tech giants suggests that enterprises need more sophisticated metrics that track business impact rather than computational appetites. Without a course correction, the industry risks repeating the same mistakes that plagued early cloud computing adoption, where spending exploded before optimization became a priority.
Counter-argument: Some analysts argue that in the early stages of a transformative technology, raw usage metrics provide a necessary baseline for understanding adoption patterns and infrastructure needs. Excessive spending could be a natural and acceptable cost of experimentation that yields long-term competitive advantages.
AI context: This brief is based on a single Fast Company report published hours ago. The claims about Uber, Google, Meta, Microsoft, and Salesforce rely on a single source and could not be independently verified during composition.