What Is the Tokenomics Foundation? The New Plan to Control AI Costs

The AI Token Bill Comes Due: Inside the Industry Scramble to Control Runaway AI Costs:

How enterprises are reckoning with exploding AI token spend — and the emerging market of tools, standards, and strategies built to bring it under control:

When the AI Spending Honeymoon Ends:

The bill has arrived, and for many of the world's largest enterprises, it is far larger than anyone anticipated. Uber exhausted its entire 2026 AI coding budget before April was over. Microsoft quietly revoked Claude Code licenses it had issued to developers just months earlier. A Priceline employee reported to TechCrunch that a routine Cursor contract renewal came back four to five times more expensive than the year before. The era of unchecked AI adoption is colliding with a very real financial reckoning — and the industry is scrambling to respond.

Per-token AI pricing has fallen significantly over the past two years, but that cost reduction has been more than offset by a surge in consumption. The push for greater AI adoption, combined with the rise of increasingly autonomous AI agents, has driven token usage to levels that few finance teams modelled or prepared for. Companies that signed up for all-you-can-eat AI subscriptions in early 2025 are now urgently asking a new set of questions: Where is our money going? Can we audit this spend? And is any of it actually delivering ROI?

It is a question that even the most sophisticated technology organisations are struggling to answer. One company reportedly racked up a $500 million Claude bill after failing to set usage limits for its employees — a figure that illustrates how rapidly AI token costs can spiral without proper governance in place.

From Token Maximalism to Token Governance:

The mood shift inside enterprise technology teams has been dramatic and swift. Just six months ago, the conversations between AI vendors and enterprise clients were almost entirely focused on capability: What can the model do? Is it good enough for our use case? Today, those conversations have changed entirely. Alexander Embiricos, OpenAI's Head of Enterprise, described the shift plainly at a recent New York event: the questions now are about visibility, auditability, token controls, and model efficiency — not raw performance.

The catalyst for this shift was a combination of executive pressure and model releases that dramatically expanded what AI agents could do. New models released in late 2025 — including Anthropic's Claude Opus 4.5, OpenAI's GPT-5.1, and Google's Gemini 3 Pro — delivered significant improvements in agentic capabilities, unlocking more complex, multi-step workflows. Those workflows, by their nature, consume far more tokens than a single-turn chat interaction. CEOs had pushed their teams to adopt the best models and move fast. The token bills that followed reflected exactly that mandate.

J.R. Storment, Executive Director of the FinOps Foundation, captured the atmosphere that began emerging in spring 2026: "In April and May, I started hearing from companies: 'Oh my god, we are 3x over our entire 2026 token budget and it's only April.' We started hearing existential crises, and the whole conversation shifted from tokenmaxxing and 'go fast' to 'we need guardrails, how do we control this?'"

The $40,000 Engineer and the Productivity Paradox:

At the heart of the AI cost management crisis lies a genuinely difficult question: does extreme AI token consumption actually pay off? Vitaly Gordon, CEO of engineering operations platform Faros AI, recently spoke to a CTO who found that one of his engineers had spent $40,000 on AI tokens in a single month. The CTO's dilemma was revealing — he was not sure whether to stop the engineer or tell the rest of the team to follow his lead.

The data on AI developer productivity tells a similarly complicated story. A two-year study of 20,000 developers by Faros AI found that output was rising alongside token consumption — but so were bugs and rewrites. Research from Jellyfish, an engineering management platform, found that engineers who used the most AI tokens were roughly twice as productive as lighter users, but they consumed ten times as many tokens to achieve that productivity uplift.

Nicholas Arcolano, Head of Research at Jellyfish, noted that per-developer token consumption has risen approximately 18.6 times in nine months — driven largely by the adoption of agentic AI features. But whether that spend translates into genuine business value remains an open question. "Whether extreme spend pays off comes down to the ultimate business value of shipped code (e.g. revenue), which most companies still can't measure," Arcolano said. Until that measurement capability matures, the ROI case for AI investment will remain murky at best.

Why AI Token Cost Tracking Is Harder Than Cloud Cost Management:

The challenge of managing AI token spend is not simply a matter of willpower or policy — it is a data infrastructure problem of extraordinary scale. Storment put the contrast with cloud cost management in stark terms: tracking cloud costs is a hundreds-of-millions-of-rows-per-month data problem. Tracking AI token costs is a trillions-of-rows-per-month data problem. The tooling, specifications, and accounting systems required to handle that volume do not yet exist in mature form.

The billing accuracy problem adds another layer of complexity. Chris Reed, Senior Director of IT Finance at Priceline — who began his career in telecom expense management — is already finding discrepancies between vendor-reported token usage and Priceline's own internal data. He draws a direct parallel to the early days of cloud billing: "Anytime you introduce something new, it's ripe for billing errors and audit and optimization opportunities.

" The analogy to telecom and cloud is instructive: both industries went through extended periods of opacity, overcharging, and eventual standardisation. AI is now entering that same cycle.

Reed's description of the dynamic — likening AI adoption to a dependency cycle where enterprises are "hooked" before the true cost becomes visible — resonates across the industry. Priceline has responded by placing token limits on certain employee groups, a move that reflects a broader trend toward AI spend governance rather than unconstrained access.

The Emerging Market for AI Token Management and Observability:

Where enterprises see a problem, the market sees an opportunity — and the market for AI cost management and token observability tools is forming quickly. Pure-play startups are leading the charge. Pay-i tracks, measures, and optimises the costs and performance of generative AI investments. Paid enables developers to track costs, measure usage patterns, and bill users based on actual delivered value rather than flat subscription fees. Both represent a new category of financial tooling purpose-built for the AI era.

Engineering intelligence platforms are also pivoting to address the ROI measurement gap. Jellyfish, Waydev, and Faros AI all now offer AI agent monitoring capabilities designed to connect token consumption data to engineering output and business outcomes. The goal is to give technology leaders the visibility they need to make defensible investment decisions — and to identify which AI usage patterns are genuinely value-generative versus which represent waste.

Established vendors are moving fast to capture this opportunity as well. Ramp has entered the AI spend management space. Datadog and New Relic have expanded their offerings to include token-level observability and GPU monitoring alongside their existing cloud cost management capabilities. At the enterprise application layer, Factory — a startup building AI agents for enterprises — launched a model router this week that automatically selects the most cost-appropriate model for every individual task, dynamically optimising spend without requiring manual configuration.

Gordon anticipates that frontier AI labs and model providers will increasingly adopt OpenRouter-style optimisation, automatically routing queries to lower-cost models when the task does not require top-tier capability. "The financial report for how much you spend on Anthropic, even if you call the Opus model, some of the spend will be on Sonnet or Haiku, because they are smart enough to do it," he noted — a trend that is already appearing on enterprise AI bills.

The Tokenomics Foundation — Building a Common Language for AI Economics:

The market for AI cost management tools is growing rapidly, but it is being built on a foundation of incompatible definitions, inconsistent metrics, and vendor-specific billing frameworks. The Linux Foundation is moving to address that structural problem with the launch of the Tokenomics Foundation — a new standards body explicitly modelled on what the FinOps Foundation did for cloud spend discipline.

The Tokenomics Foundation's mandate is ambitious and foundational. It aims to establish a canonical definition and framework for tokenomics; create open standards, specifications, and shared metrics for AI token usage and billing; and develop new economic metrics tailored to the AI era — including cost-per-intelligence and tokens-per-watt.

The Foundation also plans to define standard metrics for token factory effectiveness and consumption efficiency, giving enterprises and vendors a shared vocabulary for measuring and comparing AI spend.

Nishant Gupta, Chief Availability Officer at Salesforce, articulated the core problem the Foundation is trying to solve: "Token economics is fundamentally more abstract and opaque than anything we've managed at this scale before. It requires a different operational muscle than the one the industry built for cloud." The Foundation is planning a formal launch in July, with additional member announcements expected at the upcoming FinOps X conference.

What the Numbers Say About Where This Is Headed:

The scale of the challenge ahead is difficult to overstate. Goldman Sachs projects that global AI token usage will multiply by 24 times by 2030 — meaning the cost management problems enterprises are experiencing today represent the very early stages of a much larger structural challenge.

The companies that are already over budget on AI token spend do not have the luxury of waiting for July standards launches or next-generation observability platforms. They need governance frameworks, usage limits, and audit capabilities now. The market is responding with urgency, but the tooling is still maturing, the standards are still being written, and the ROI models are still being developed.

Tiffany Luck, a partner at NEA, expects token efficiency and observability capabilities to be embedded at the application and harness layer — becoming a standard feature of enterprise AI infrastructure rather than a standalone add-on. That trajectory points toward a future where AI spend management is as routine and disciplined as cloud cost management is today — but getting there will require the same years of iteration, standardisation, and hard-won institutional knowledge that FinOps required.

The New Discipline Every Enterprise AI Team Needs:

The AI industry is entering its FinOps moment — the point at which the conversation shifts from capability maximisation to cost discipline, ROI accountability, and sustainable consumption. For enterprise technology leaders, that shift demands a new operational capability: the ability to track token spend at scale, attribute it to specific teams and workflows, measure it against business outcomes, and govern it with the same rigour applied to any other major cost centre.

The tools to support that capability are being built in real time, by startups, established vendors, and a new standards body working in parallel. The market signal is clear.

The direction of travel is set. The question for every enterprise AI team is not whether to build this capability — it is how quickly they can do it before the next token bill arrives.