Check with finance
This month I hit my AI token limit at work. I wasn't doing anything exotic. I'm not running a fleet of agents chewing through tasks in the background, just using AI across a handful of projects. It had always felt effectively limitless, so hitting a wall stopped me cold with a question I'd been happy to ignore: who's actually paying for this?
The industry is loudly of two minds. On one end, Nvidia's Jensen Huang says a $500K engineer should be burning something like $250K a year in tokens, half their salary on top of their pay, because it amplifies them 10x. On the other end, Uber blew through its entire 2026 AI budget in four months and ended up capping employees at $1,500 a month per tool. Spend like it's free, or ration it. Same technology, opposite conclusions.
It took me back to earlier roles. When a team ran over budget, or a sales forecast came in wrong, leadership always wanted a throat to choke. And the cleanest answer in the room was almost always, "go check with finance." Not a dodge, exactly. Finance is where the real accounting of a decision lives.
At a company like Amazon, I know the answer to "who's paying," and I trust how we think about it. But the smaller, personal version is harder. As the person running a program, how should I think about its token cost? How do I account for someone on my team using AI to get smarter on one hard task and deliver more value because of it? That's not waste, it's leverage, but it still shows up on a bill somewhere.
Some days I think about friends back home who use AI zero percent at work. The gap between what their day looks like and what mine looks like is night and day. Something that valuable isn't free, and pretending the cost is zero is how you end up like Uber in April.
I don't have all the answers here, but in the meantime, if you want to know who's paying for all these tokens, go check with finance.
