The numbers are getting awkward. Bryan Catanzaro, Nvidia's VP of applied deep learning, told Axios that compute costs for his team are "far beyond the costs of the employees." Uber's CTO already burned through the company's entire 2026 AI budget on token costs alone, according to The Information. And Swan AI CEO Amos Bar-Joseph bragged on LinkedIn about his massive Anthropic bill, calling it proof he's "scaling with intelligence, not headcount." AI now costs more than the humans it replaces. Gartner projects worldwide IT spending will hit $6.31 trillion in 2026, up 13.5% from 2025, driven by AI infrastructure and cloud services.
The sticker price on API tokens barely scratches the surface. The real costs include orchestration layers, evaluation pipelines, supervision, and engineering time spent rebuilding agent stacks every time a model version changes. Engineers are reportedly burning tokens on tasks a basic Python script could handle, because management wants to see AI usage. One Hacker News commenter made a sharp observation: human labor still wins on flexibility per dollar. Workers adapt to new requirements in weeks. Agent stacks don't move that fast.
Brad Owens, VP of digital labor strategy at Asymbl, told Axios the tone is changing. "The tone is shifting a bit more into what is the true value of a worker... human or digital?" Companies face shareholders on earnings calls now, and those shareholders want proof that AI spending produces returns. Nobody has yet produced convincing evidence of overall productivity gains. Some enterprises are routing around the problem by self-hosting open-source models like Meta's Llama 3 or Mistral's Mixtral, using inference frameworks like vLLM to keep per-token costs down. It requires upfront engineering work, but it insulates companies from the price hikes closed-source providers keep pushing through.