This piece cuts through the noise of sensational headlines about corporate AI spending sprees to reveal a far more nuanced reality: the era of unchecked "tokenmaxxing" was less a market-wide frenzy and more a series of isolated incentive failures. Dylan Patel brings on-the-ground data from over 50 enterprise conversations, challenging the narrative that major companies are burning through budgets at an unsustainable rate.
The Myth of the Spending Frenzy
Patel immediately dismantles the prevailing media story. "Widely reported responses to Tokenmaxxing budgets from companies like Meta and Uber are overstated and stem from poor incentives and employee allocation we didn't find present at other organizations," he writes. This is a crucial distinction. The article suggests that while headlines focused on specific outliers, the broader enterprise market has quietly shifted toward discipline rather than panic.
The author highlights how early adoption was driven by strange internal competitions. At Meta, an employee built a dashboard ranking power users, where one individual consumed roughly 280 billion tokens in a month just to climb a leaderboard. "Employees started competing for rankings like 'Token Legend' and 'Cache Wizard' by having agents do research for hours simply to burn tokens," Patel notes. This behavior mirrors the economic concept of Goodhart's law—once a measure becomes a target, it ceases to be a good measure. The dashboard was shut down within days, but the media narrative of unbridled excess persisted long after the experiment ended.
The headlines on early 2026 tokenmaxxing were a result of poor incentives and lax oversight rather than an absence of high ROI activities.
Patel argues that these stories are outliers. His data shows that even Meta, often cited as the most profligate spender, represents only a "3-5% customer" for major AI labs like Anthropic. The distribution of spending is heavily skewed: 90th percentile customers spend around $7,300 per employee annually, while the median spends just $136. This suggests that the market is not collapsing under weight but is instead maturing into a tiered structure where heavy usage is concentrated in specific, high-value roles.
The Reality of Budgeting and Conservation
As companies move from experimentation to production, the approach has shifted to strict budgeting, though there is no consensus on the numbers. Patel observes that budgets range wildly "starting at $250 and going up to tens of thousands a month." This lack of standardization reflects the early stage of AI integration, where organizations are still calibrating their tools.
Some companies have taken aggressive steps to curb costs. Uber, for instance, imposed a "$1,500/month/employee limit" after burning through its annual budget in just four months. In contrast, a top aerospace manufacturer capped employees at $250 a month. "Management believes that handing employees larger token budgets would push them to automate tasks that shouldn't be automated at all, like writing emails," Patel writes regarding one conservative firm's strategy.
This anti-automation stance is where the argument faces its sharpest friction. While management fears waste, Patel contends this view is short-sighted. "We believe this anti-automation view by management teams is naive," he asserts. He argues that email and communication workflows will inevitably become AI-native, and restricting tokens now may stifle long-term productivity gains. Critics might note, however, that in a recessionary environment, the immediate pressure to cut costs often overrides long-term strategic bets on automation.
To stretch their allowances, employees are getting creative. Patel points out that workers with Microsoft 365 subscriptions can "game the system by using Copilot's 365 chat to draft and synthesize ideas first, before spending metered tokens on Claude or Codex." This behavior highlights a gap in how companies track usage; if the tool isn't metered, it becomes an infinite resource, creating its own form of inefficiency.
AI as a Headcount Lever
Perhaps the most insightful reframing in the piece is how successful companies view AI spend not as an expense to be minimized, but as a lever for output. "Output expectations rise to match spend, and many workers have found themselves putting in even longer hours than before," Patel explains. The goal isn't just faster work; it's more work.
The article cites Amazon as a prime example of this dynamic. Despite public narratives about layoffs, the company is hiring at a faster pace because AI tools have unlocked efficiencies that allow for expansion. "A recruiter at Amazon responsible for scouting and placing principal engineers... noted that the process from initial screening call to team placement used to take 6-9 months, but with AI tools... that timeline been cut in half," Patel reports.
This shift is transforming how budgets are allocated. At a major US airline, token usage is now tied directly to project revenue, treating AI spend like travel or contractor fees. "When a project comes in, the financing team decides what percentage of revenue is set aside for expenses... and now that same budget must cover token usage as well," Patel writes. This financial discipline suggests that the market is moving past the hype cycle into a phase of rigorous ROI calculation.
Bottom Line
Dylan Patel's analysis provides a necessary corrective to the alarmist narrative surrounding enterprise AI spending, grounding the discussion in hard data from actual users rather than press releases. The strongest part of this argument is its demonstration that high spend is concentrated and productive, not wasteful and diffuse. However, the piece may underestimate the cultural friction of imposing strict caps on tools that employees view as essential to their daily efficiency. As organizations continue to refine these budgets, the real story will be whether they prioritize cost-cutting or productivity scaling—a choice that will define the next phase of AI adoption.