The Reckoning Arrives
Across the industry, companies are starting to balk at the price of AI. The story is consistent: adoption was easy, enthusiasm was high, and the bills were ignored — until they couldn't be ignored anymore.
Uber blew through its entire 2026 AI coding budget by April. Microsoft revoked its developers' Claude Code licenses months after enabling them. A Priceline employee told TechCrunch that a routine Cursor contract renewal came back 4–5x more expensive. One company reportedly found itself with a $500 million Claude bill after forgetting to set usage limits for employees.
Per-token prices have fallen dramatically. But consumption has risen far faster than prices have dropped. Autonomous agents, multi-step workflows, and an industry-wide push to "move fast" have driven token usage up by 18.6x per developer in just nine months, according to Jellyfish's engineering management data.
Why Costs Exploded Past the Savings
Six months ago, OpenAI's head of enterprise told TechCrunch that customer conversations were all about capability: "Is it good enough?" Now, every conversation starts with the same question: "We're spending so much. What visibility do you have? What token controls do you have?"
The shift is stark. New models released in late 2025 — Claude Opus 4.5, GPT-5.1, Gemini 3 Pro — brought real improvements to agentic tools. But those improvements came with a hidden cost: agents consume tokens at rates that subscriptions never did.
Chris Reed, senior director of IT finance at Priceline, put it bluntly: "It's like the crack-cocaine epidemic. They let you try it to get you hooked on it, and now you're kind of beholden to it."
The Productivity Paradox
Jellyfish's study of 20,000 developers found something uncomfortable: engineers who used the most AI tokens were about twice as productive as those who used less — but they spent 10x the number of tokens to get there. Nicholas Arcolano, head of research at Jellyfish, noted that whether extreme spend pays off "comes down to the ultimate business value of shipped code, which most companies still can't measure."
Faros AI's CEO, Vitaly Gordon, shared a conversation he had with a CTO: "One of my engineers spent $40,000 on tokens last month, and I genuinely don't know whether I should stop him or should I go and tell everyone else to be like him."
That's the $40,000 question — and it's one most engineering leaders are now being forced to answer.
Enter FinOps for AI
The cries heard round the tech world followed CEO-driven mandates to use the best models and move fast. The result: existential budget crises by Q2 2026.
The Linux Foundation unveiled plans for the Tokenomics Foundation — a new standards body that aims to instill the same cost discipline around AI tokens that FinOps brought to cloud spend. The FinOps Foundation's executive director, J.R. Storment, said the organization started hearing from companies in April: "Oh my god, we are 3x over our entire 2026 token budget and it's only April."
What Builders and Founders Need to Do Now
This isn't just an enterprise problem. Independent developers and small teams are burning through OpenRouter, Anthropic, and OpenAI credits faster than expected. Here's the practical wake-up call:
- Set per-user or per-project token limits before rolling out AI tools company-wide. Treat AI budgets like cloud budgets.
- Track token spend per engineer — the same way you'd track compute costs. You can't optimize what you can't measure.
- Audit your agentic workflows — multi-turn agents that loop can generate thousands of tokens per task. Add checkpoints.
- Consider model efficiency, not just capability — a 70B model that costs 5x less may ship code that's 90% as good. For many tasks, that's the right trade.
- Push vendors on billing visibility — if your AI vendor doesn't give you per-user or per-project breakdowns, demand them.