—

Lab · Agentic Demand

AI Adoption — Usage, Breadth and Price, Measured

Is the use of AI growing, how widely, and at what price? Four readings, each from a source this site already collects, each with its date. Tokens are the unit of AI work — the pieces of text a model reads and writes — so their volume is the closest public measure of how much AI is being used.

Intensity

+35.8%

OpenRouter token volume, last four weeks against the four before (24.4T against 18.0T a day).

Ramp token volume: +52.4%, week to 11 Oct 2026 against the same week a month earlier.

Breadth

55.9%

US businesses paying for AI, Sep 2026 — +0.62 pp on the month, +12.1 pp on the year.

Monetisation

+23.0%

Token spend on Ramp, same comparison as the volume. Implied change in the price per token: −19.3% — our calculation from the spend and volume growth rates.

Market

+22.9%

The AI Users implementers index since 30 September, against −17.5% for the Rubin build-out index — what the market pays for the companies that put AI to work, against the ones that build it.

OpenRouter token volume, every day since July

Whole-catalog tokens per data day in trillions (thin line) and the 28-day average (thick). From 9.3T a day in the first four weeks to 24.4T in the last four — 2.6 times.

0T 7T 14T 21T 28T 35T AugSepOct

OpenRouter public rankings via the Closelooknet token worker · one point per data day since 20 Jul 2026 · last point 23 Oct 2026.

How much of each demand gauge is measured

The eight Agentic Demand pulses combine several modules each. A module without live data drops out, so a pulse reading rests only on its live share — read the two together.

PulseReadingLive modulesLive shareLast data
Enterprise Workflow 85.5 6 of 6 100% 23 Oct 2026
Consumer Cloud 78.6 2 of 5 55% 24 Oct 2026
Enterprise Infrastructure 65.1 5 of 6 80% 24 Oct 2026
Enterprise Application 78.3 3 of 3 100% 24 Oct 2026
Industrial / Physical 49.3 2 of 3 67% 19 Oct 2026
Agentic Security 42.0 1 of 3 33% 19 Oct 2026
Tollbooth 31.8 1 of 3 33% 19 Oct 2026
Consumer Endpoint 20.5 0 of 3 0% 19 Oct 2026

What the readings can and cannot say

Tokens are not finished work. Longer prompts, extra reasoning steps and retries raise the token count without more being done; the best measure would be tasks completed or agent runs that succeeded, and no market-wide series of that exists. Token volume is the closest public proxy — read it as a measure of use, not of value.

Both sources are slices of the market. OpenRouter counts only the traffic routed through it, and can gain or lose share itself; free models and model switches move its volume. Ramp counts the businesses on its platform, and a change in that sample moves its numbers. That is why the board shows both, and shows changes rather than levels.

Adoption counts a payment, not intensity. A business that pays for one AI subscription counts as adopting. Breadth and intensity are separate readings for that reason.

Sources: OpenRouter public rankings (via the Closelooknet token worker, daily); Ramp AI Index (adoption monthly, token spend and volume weekly — shown here as changes); Closelooknet index workers. Research diary, not investment advice.

Related: Agentic Demand · Token Split Pulse · Consumer Cloud Pulse · Enterprise Application Pulse · AI Users