Why the AI token economy is the next economic frontier

Why the AI token economy is the next economic frontier

When Stripe announced its planned acquisition of AI platform OpenRouter in August, CEO Patrick Collison called tokens the new "central currency" of AI. OpenRouter, which helps companies route AI queries across hundreds of models, was processing more than 10 trillion tokens per day when Stripe agreed to buy it, Reuters reported.

The deal is the latest high-profile sign of an emerging "token economy," which essentially means tokens are moving from a technical measure of AI usage to a resource that companies increasingly need to manage, allocate and finance.

In this article, Accels explores the rise of the AI token economy and what it means for the infrastructure companies that will need to manage AI consumption at scale.

AI agents are turning tokens into a much bigger bill

Tokens, for the uninitiated, are the chunks of text that AI models process and generate when completing a task. While many consumers still pay for AI through monthly subscriptions, most enterprise and API-based AI services are priced by the number of tokens consumed.

The growing use of AI agents is dramatically increasing the number of tokens enterprises use. Agentic traffic grew 7,850% in 2025, according to cybersecurity firm Human Security.

AI agents devour tokens more quickly than their chatbot counterparts simply because they can be given an objective and left to get on with it. Cisco found that an AI agent carrying out a research task generated 450% more network traffic than a person performing the same task manually. All of that is billable in tokens.

Cheaper tokens do not mean cheaper AI

Thankfully for enterprises, the cost of tokens has been falling.

The LLM Token Expenditure Index, a measure of daily token prices compiled by intelligence firm Silicon Data, fell to 97 cents per million tokens at the start of September, its lowest level since the index was launched late last year, CNBC reported. The index has more than halved from its peak earlier this summer. The decline is due to improving model efficiency and intensifying competition among AI model providers.

However, while this is welcome news for enterprises — 93% of whom have reported AI cost overruns in the past year, according to McKinsey — a falling price per token does not necessarily mean a falling AI bill.

First, AI is handling more complex tasks, offsetting lower token prices with higher consumption. That same McKinsey study found enterprise LLM spending tripled in 12 months, despite sharply lower inference costs, as companies expanded AI across more workflows. Goldman Sachs projects token consumption will increase 24-fold from current levels by 2030.

More from this section

AI vendors can change their risk profile between reviews, and most oversight programs never notice

Why the AI token economy is the next economic frontier

+4

Coughing at night? Study links air pollution to poor sleep

Second, token use is difficult to predict, meaning businesses still often set budgets well below what they need. A 2026 study by researchers, including Stanford's Erik Brynjolfsson, tested eight frontier models on software engineering tasks and found that the models systematically underestimated their own token consumption. The same study found that runs of the same task consumed tokens differently, and by as much as 30 times.

Together, these factors are still driving AI bills sharply higher. OpenAI recently said its heaviest users of AI coding agents were consuming more than $7,000 worth of tokens a day. Most businesses could not tolerate this level of usage, so AI infrastructure providers like Google Cloud have begun introducing spending caps that allow customers to automatically halt eligible API traffic when limits are reached.

The next challenge is knowing where the tokens go

Yet, while controlling token consumption is an obvious first step, it is also a rather blunt instrument when used in isolation. For instance, a $50,000 monthly cap can stop a company from overspending, but it cannot tell a CEO whether the money is being spent wisely.

That is still a blind spot for many companies: KPMG found last year that just 15% of business leaders had established formal metrics for measuring returns on AI investments. As token consumption grows, closing that gap will require more sophisticated infrastructure to track, manage and measure AI spending.

If tokens are becoming a kind of commodity or raw input for AI, as some analysts argue, then the financial and operational infrastructure supporting them is still in its infancy. Take oil or gold, for example: Entire markets have grown around them, including futures, credit, pricing benchmarks, exchanges and risk-management tools. Tokens do not yet have anything close to that level of infrastructure.

McKinsey describes the emerging discipline as "AI FinOps," which encompasses AI spend, usage, performance and business outcomes. Only 20% to 25% of companies in its May 2026 Enterprise AI FinOps survey had mature AI FinOps capabilities, indicating a disconnect between AI adoption and the infrastructure needed to manage it.

However, that discrepancy could disappear sooner than many firms expect. The Linux Foundation launched the Tokenomics Foundation in August to develop standards for measuring, managing and monetizing AI investments.

"Every CEO is being asked to show returns on all of that without a shared way to count it," said J.R. Storment, executive director of the Tokenomics Foundation, regarding the “real total cost of AI.” Storment added, "That is why this foundation exists and is working together on pre-competitive frameworks, benchmarks and specifications."

Accenture, meanwhile, launched its own tokenomics offering in July, combining usage monitoring with model routing, budget controls and tools for measuring the cost of AI against business outcomes. Stripe's planned acquisition of OpenRouter is another example.

The ambition behind all these efforts is to make AI spending a more predictable technology expense, in other words, something closer to an economic input that can be planned, financed and optimized.

For now, the token economy is far closer to an emerging market than a mature one. But as tokens become a fundamental input into how businesses operate, the companies that can price, route, finance and measure them will become as important as the infrastructure that surrounds other major commodities.

This story was produced by Accels and reviewed and distributed by Stacker.