AI Burn Clock
live · sources refreshed 2026-09-09

AI works like 2026. It searches like 1975. The bill runs every second.

An agent that needs one function opens whole files to find it, and pays for every byte it loads. This furnace prices that habit for a team like yours, from public model prices and measured ratios, with every knob on the table. The national figures below are scale, not cause.

This team burns on reading it never needed
$0
per year at the knobs below · per second
The whole reading bill, as it runs today
$0
agents searching by reading: tokens a year
Kept, retrieval-first
$0
index answers "where is it"; the agent reads the passage

For scale

Official numbers, shown as published. Context for the size of the habit, not its cause.

US total public debt · Treasury, Debt to the Penny
$40.095T
as of , ticking at the last 30-day rate
Avoidable reading, scenario · world AI spend
$0
since you opened this page · per second · four-factor estimate, knobs below
Kept with retrieval-first · same scenario
$0
what stays with the payer when the agent reads the passage, not the file

The machine as it runs today

$0 / second

The agent searches by reading. Every file it opens to find one function is billed in full. Rate for the team selected above.

The same machine, retrieval-first

$0 / second

A local index answers "where is it" first. The agent reads the passage it needs. Open source does this today; XERJ is one example.

The formula, with the knobs

No hidden model. Four numbers multiply. Every default has a source; every knob is yours. Set them where you believe they belong and the clock recomputes.

A coding agent working through a task issues dozens of tool calls; 40 is an ordinary day. Move it.
Measured on a Next.js codebase: the whole files behind one answer were 103 KB, 242 KB and 244 KB. Tokens ≈ bytes / 4.
Public list prices in 2026 run from under a dollar for small models to fifteen for frontier models. Set yours.
Agents that run unattended overnight push this to 365.

One year, five years, ten years

Two curves. Do nothing, or read the passage instead of the file. The gap is the story. Growth starts at the Gartner 2026 rate and tapers to the second knob.

Federal AI contracts by state

Place of performance, FY2026 to date, contracts whose descriptions name artificial intelligence, machine learning or language models. USAspending.gov, live. This is the floor: most AI inside larger contracts is not labelled.

Receipts

Who is awarding and who is receiving, FY2026 to date. Same filter, same source.

Awarding agencyObligated
RecipientObligated

Your company

Teams tell us the same thing in different words: the monthly limit melts in days, and most of it went to the agent reading the tree. Put your numbers in. Names appear here only with permission.

Avoided share uses the overhead knob above. Post your card with the plugin and we will add your number, anonymised, to the corporate ledger.

Measure yours in one session

Claude Code plugin: local index on port 9200, retrieval before reading, a score card at the end of every session. Nothing leaves your machine.

/plugin marketplace add nikolaichuk7/xerj-plugins
/plugin install xerj-memory@xerj-plugins
The plugin

For newsrooms and offices

Embed the live clock in one line. Pull the numbers from the open endpoint. Cite the methodology page. Ask for a state memo.

<script src="https://aiburnclock.org/embed.js" data-scope="world"></script>

Renders a 300×120 live counter with a link back. Data: /data.json, refreshed daily. Method: /methodology.html. Press kit: /press.html.

For a state CIO or budget office we prepare a one-page memo with your state's numbers, the sources, and a pilot that runs on one laptop with no procurement. Write to hello@aiburnclock.org with the state in the subject line.