AI's Hidden Cost: Companies Overlook Token Usage Bills Amid Layoffs

AMD CIO Hasmukh Ranjan flagged that employee AI token usage could cost a company up to $900 million a year—a hidden bill many CEOs cutting jobs over AI are overlooking.

Companies have spent this year cutting staff and pointing at AI as the reason. Almost nobody in the corner office, though, has done the simple multiplication on what the AI itself costs. AMD's chief information officer, Hasmukh Ranjan, did, and the figure he landed on should make any executive trading workers for tokens sit up. Train an employee to lean on AI agents, let them burn through roughly $200 a week in tokens, and you are looking at about $10,000 a year per person. Scale that across 40,000 employees and the bill hits $400 million. Push it to 90,000 and you are staring down $900 million—a line item that, as Ranjan put it, simply did not exist a few years ago. That math is the catch in the layoff story. Employers announced almost 40% of May's job cuts with AI as the stated reason, the highest such share this year, even as the AI meant to replace those workers is quietly running up a bill of its own.

When the replacement costs more than the replaced

The awkward part is that AI is not obviously cheaper than the people it is displacing. An Nvidia executive admitted compute now costs his team more than the employees using it. A 2024 MIT study found humans were still the cheaper option for 77% of vision-related roles. Some analysts argue companies citing AI for layoffs may be reaching for an acceptable excuse—workforce cuts being one of the few levers available to offset ballooning AI invoices. Tokens, the units models charge by, add up fast once agentic tools that run for hours with little human input enter the picture.

Push them to use it, then tell them to stop

For most of the past year, the message to staff was the opposite of restraint. Companies celebrated heavy AI users on internal leaderboards—Meta handed top users "Token Legend" titles, and the practice earned a name: tokenmaxxing. Then the bills caught up. Uber blew through its entire 2026 AI coding budget by April. One firm reportedly spent $500 million on tokens in a single month. As SFGate has reported, the sticker shock is now landing across corporate America, not just the AI-native startups.

From open bar to spending caps

Now the same companies are reining it in. Meta projects internal AI usage could cost it billions in 2026 and is rolling out budgets and a tracking dashboard. Amazon and Microsoft have pulled back too, with Microsoft cancelling most of its Claude Code licenses. Uber capped employees at $1,500 of tokens a month per coding tool. Disney is pushing staff to move faster with AI while warning them off wasteful usage in the same breath. The result is a strange whiplash for workers—told to embrace the tool that may cost them their job, then told to go easy on it because it costs too much. Ranjan's number is the one that ties the knot: for now, AI is both the reason for the cuts and a bill no one fully budgeted for.