AI token values plummeted by 98%, while costs for enterprises surged threefold.

      TL;DR: Enterprise AI costs are soaring despite a 98% decrease in per-token prices, as agentic tools lead to consumption increasing 18.6 times per developer. The Linux Foundation is establishing the Tokenomics Foundation to bring financial discipline to AI expenditures.

      Uber has already used its entire AI coding budget for 2026 by April. Microsoft revoked Claude Code licenses from its developers six months after they were granted. One company reportedly accrued a $500 million Claude bill in just one month due to forgetting to set usage limits. A Priceline employee informed TechCrunch that a standard Cursor contract renewal returned with a price four to five times higher than expected.

      This trend is consistent across the board. While per-token costs have plummeted, the demand for autonomous AI agents has skyrocketed usage. Companies that indulged in unlimited subscription plans in early 2025 are now trying to track where the funds went and if there was any return on investment.

      The numerical paradox

      GPT-4-equivalent performance now costs about $0.40 per million tokens, down from $20 per million at the end of 2022, a 98% drop. However, enterprise AI bills have surged by around 320%, based on various industry analyses. The average enterprise AI budget rose from $1.2 million annually in 2024 to $7 million in 2026.

      The driving force behind this is volume. Agentic AI tools introduced since November 2025, such as Anthropic’s Claude Opus 4.5, OpenAI’s GPT-5.1, and Google’s Gemini 3 Pro, have significantly increased token consumption per task. A basic linear workflow in 2023 cost about $0.04 per interaction, while a coordinated agentic system in 2026 costs approximately $1.20, nearly 30 times more. Individual engineers at Microsoft were reportedly spending between $500 and $2,000 monthly on tokens before licenses were retracted.

      Nicholas Arcolano, head of research at engineering management platform Jellyfish, told TechCrunch that token use per developer has surged about 18.6 times in nine months. Engineers who utilized the most tokens had about double the productivity of those who used fewer, yet they consumed 10 times the tokens to achieve that. “The question of whether high spending is worthwhile hinges on the overall business value of deployed code, which most organizations still struggle to quantify,” Arcolano stated.

      From tokenmaxxing to guardrails

      “Six months ago, discussions with customers revolved around ‘What can it do? Is it sufficient?’” said Alexander Embiricos, OpenAI’s head of enterprise. “Now, conversations focus on ‘We’re spending so much. What insight do you have? What controls are in place for token usage?’”

      J.R. Storment, executive director of the FinOps Foundation, explained the shift clearly. “In April and May, companies started saying: ‘Oh my goodness, we are three times over our entire 2026 token budget and it’s only April.’ The narrative changed from tokenmaxxing and ‘going fast’ to ‘we need restrictions, how do we manage this?’”

      Chris Reed, Priceline’s senior director of IT finance, compared the situation to the telecom billing period. “It’s reminiscent of the crack-cocaine crisis. They lure you in to get you addicted, and now you find yourself dependent on it.” The company has started to impose token limits on specific groups. Reed noted discrepancies he is already observing between vendor-reported usage and Priceline’s internal figures.

      The Tokenomics Foundation

      Amid this situation, the Linux Foundation has announced the creation of the Tokenomics Foundation, intended to impose the same financial rigor on AI tokens that FinOps has provided for cloud spending.

      The Foundation aims to establish a definitive definition of “tokenomics,” develop open standards for AI token utilization and billing, and create new metrics such as cost-per-intelligence and tokens-per-watt. A formal launch is scheduled for July. Nishant Gupta, Salesforce's chief availability officer, remarked that “token economics is inherently more abstract and less transparent than anything we’ve managed at this scale previously.”

      The challenge is immense. “Monitoring cloud expenses involves managing hundreds of millions of rows of data monthly,” said Storment. “In contrast, tracking token expenses will entail trillions of rows of data each month.”

      A market emerges to address the issue

      Both startups and established companies are attempting to bridge this gap. Pay-i helps track and optimize AI expenditures. Paid allows developers to charge based on actual value rather than subscription fees. Companies such as Jellyfish, Waydev, and Faros AI offer agent monitoring to demonstrate the ROI of developer tools, while Ramp has ventured into AI spend management. Datadog and New Relic have introduced token-level observability features.

      Model routing is becoming a key cost-control strategy. Factory, an