TL;DR
Enterprise AI bills are tripling despite a 98% drop in per-token prices, as agent tools increase consumption 18.6x per developer. The Linux Foundation is launching the Tokenomics Foundation to bring cost discipline to AI spending.
Uber flew by your entire 2026 AI coding budget by April. Microsoft revoked Claude Code licenses from its developers six months after enabling them. a company reportedly racked up a $500 million Claude bill in a single month after forgetting to set usage limits. A Priceline employee told TechCrunch that a routine Cursor contract renewal was four to five times more expensive.
The pattern is the same everywhere. Prices per token have plummeted, but the push for autonomous AI agents has skyrocketed consumption. Companies that gorged on all-you-can-eat subscriptions in early 2025 are now struggling to understand where the money went and whether any of it produced any returns.
The paradox in numbers
GPT-4 equivalent yield now costs about $0.40 per million tokens, down from $20 per million at the end of 2022. That’s a 98% reduction. However, enterprise AI bills have increased by approximately 320%, according to multiple industry analyses. The average enterprise AI budget has increased from $1.2 million per year in 2024 to $7 million in 2026.
The culprit is the volume. Agent AI tools launched from November 2025, including Claude Opus 4.5 by AnthropicOpenAI’s GPT-5.1 and Google’s Gemini 3 Pro have multiplied the consumption of tokens per task. A simple linear workflow in 2023 would cost about $0.04 per interaction. An orchestrated agent system in 2026 will cost approximately $1.20, about 30 times more. Individual Microsoft engineers were reportedly spending between $500 and $2,000 a month on tokens before licenses were withdrawn.
Nicholas Arcolano, head of research at engineering management platform Jellyfish, told TechCrunch that consumption per developer has increased approximately 18.6 times in nine months. Engineers who used the most tokens were about twice as productive as the lightest users, but spent 10 times as many tokens to get there. “Whether the extreme spending pays off comes down to the ultimate business value of the shipped code, which most companies can’t yet measure.”Arcolano said.
From tokenmaxxing to railings
“Six months ago, I had a conversation with a client and it was all about ‘What can you do? Is it good enough?‘” Alexander Embiricos, business director at OpenAI, told TechCrunch.Now the conversations are around: ‘We’re spending too much.’ What visibility do you have? What token controls do you have?‘”
JR Storment, CEO of the FinOps Foundation, described the change bluntly. “In April and May, I started hearing from companies, ‘Oh my gosh, we’ve tripled our entire token budget for 2026 and we’re only in April.’ The whole conversation went from tokenmaxxing and “go fast” to “we need guardrails, how do we control this?”‘”
Priceline senior director of IT finance Chris Reed made a comparison to the telecommunications billing era. “It’s like the crack and cocaine epidemic. They let you try it to get hooked and now you are indebted to it.“The company has begun imposing token limits on certain groups. Reed said he is already seeing discrepancies between vendor-reported usage and Priceline’s internal data.
The Tokenomics Foundation
It is against this backdrop that the Linux Foundation this week revealed plans for the Tokenomics Foundation, a new standards body that aims to bring the same cost discipline to AI tokens that FinOps brought to cloud spending.
The Foundation plans to construct a canonical definition of “tokenomics,“Open standards for AI token usage and billing, and new metrics including cost per intelligence and tokens per watt. A formal launch is planned for July. Nishant Gupta, chief availability officer at Salesforce, said in a statement that “the economics of tokens are fundamentally more abstract and opaque than anything we’ve managed at this scale before.”
The challenge is enormous. “Tracking cloud costs is a hundreds of millions of rows per month data problem,“Storment said.”Tracking token costs is a billion-row-a-month data problem.“
A market is formed around the problem
Startups and established providers are competing to fill the void. Pay-i tracks and optimizes AI spend. Paid allows developers to bill based on actual value instead of subscription fees. Jellyfish, Waydev, and Faros AI provide agent monitoring to demonstrate ROI of development tools. Ramp has moved into AI expense management. Datadog and New Relic have added token-level observability.
Model routing is emerging as the main cost lever. Factoryan enterprise AI coding startup, this week launched a router model that automatically selects the cheapest suitable model for each task. Vitaly Gordon, CEO of Faros AI, said frontier labs are already doing this internally. “The financial report of how much you spend on Anthropic, even if you call the Opus model, some of the spending will be on Sonnet or Haiku, because they’re smart enough to do that.“, said.
Goldman Sachs projects Global token usage will increase 24-fold by 2030. Companies already over budget need solutions now, and the first result from the Tokenomics Foundation is still months away. As Gordon said: “Maybe we created a steam engine, but we haven’t figured out the assembly line yet.“
