Nvidia's VP of applied deep learning, Bryan Catanzaro, told Axios that compute costs for his team now far exceed what he pays the people on it. Uber's CTO reportedly burned through the company's entire 2026 AI budget on token costs alone, according to The Information. The bills are real and they're stacking up fast.
Why? A few reasons. AI agents don't just make one call and call it done. They loop. They retry. They stuff entire documents into context windows that charge by the token. A single coding task might involve dozens of back-and-forth exchanges, each one burning tokens. Add sloppy prompting to the mix and you've got models doing unnecessary work. It adds up.
Global IT spending is projected to hit $6.31 trillion in 2026, up 13.5% from this year, per Gartner. AI infrastructure and cloud services are driving that jump. And some founders wear massive AI bills as a badge of honor. Swan AI CEO Amos Bar-Joseph bragged on LinkedIn about his Anthropic bill, calling it proof of "scaling with intelligence, not headcount." Maybe. Or maybe it's just expensive.
Companies trying to fix this have options. Caching repeated queries helps. Running smaller models for simpler tasks cuts costs. Better prompting reduces wasted tokens. Anthropic already adjusted pricing to account for surging demand. An OpenAI investor told Axios that Codex beats Anthropic's Claude Code on token efficiency, which starts to matter a lot when your budget vanishes by April.
The math has to work eventually. Brad Owens, VP of digital labor strategy at Asymbl, told Axios the conversation is shifting toward "what is the true value of a worker... human or digital?" Companies dropping fortunes on AI will need to show actual returns to shareholders.
The irony is thick. AI was supposed to make labor cheaper. Turns out digital workers can burn through cash faster than humans ever did.