How to read AiSignal

For lawyers, and for anyone who just got here

Nothing on this site is a certified measurement, a forecast to trade on, or a finding of fact for a lab, a regulator, or a court. It is a magazine with a smear clock: public prints, modeled rates, openly sourced. Estimates, not hard claims.

If a number on this site matters to a contract, a filing, a model card, or a headline, go back to the primary source listed under that meter. Treat AiSignal as journalism with a clock: we smear public prints so scale is felt in the body. We do not certify, warrant, or audit the underlying activity.

• We claim: these are estimates, built from named public sources, updated when those sources move, and labeled as a model.
• We do not claim: that any ticker is a live API, a complete census, a measurement of “intelligence,” or a number a court should treat as a finding of fact.

Three kinds of number

Disclosed. A lab or vendor said it in public — OpenAI’s more than 2.5 billion ChatGPT messages per day, Stanford’s 17.1 million H100-equivalents, Gartner’s $2.59 trillion. We repeat the print and name the source.

Modeled. We (or Epoch, or Tokens Per Day) built a stack on top of disclosed floors. Example: Epoch’s ~4 billion daily ChatGPT messages is an estimate, not the same kind of number as OpenAI’s disclosed floor. We keep them apart. OpenAI publicly reported ChatGPT at more than 2.5 billion messages per day. Epoch subsequently modeled ChatGPT closer to ~4 billion daily messages. That second figure is an estimate, not the same kind of disclosed number. AiSignal does not collapse the two.

Smear clock. A published daily or annual total, divided across time so a counter ticks in your peripheral vision. Today counters start at local midnight. Year counters start on 1 January. Stock counters begin at a dated snapshot and add a derivative when the source implies one. Reality is bursty. We would rather show an honest smear than a fake live API.

Tokens are an index, not a barrel

Tokens from different models are not perfectly comparable. Tokenizers differ, and input, output, reasoning, and multimodal accounting differ. Aggregated token counts are a directional index, not a physical unit like barrels of oil.

We still sit a token meter because the industry prices, ships, and wastes tokens. That count is the best directional read we have of consumption. It is not physics. Digit after the first two is weather.

Primary sources

• Stanford AI Index Report 2026 — compute, power, U.S. private investment.
• Gartner Worldwide AI Spending Forecast, May 2026 — $2.59T print.
• OpenAI public usage disclosures — ChatGPT more than 2.5B messages/day (not prompts).
• Epoch AI modeled ChatGPT ~4B messages/day — an estimate, labeled as such.
• Tokens Per Day; Exponential View — token-volume triangulation.
• Artificial Analysis Intelligence Index v4.1 / v4.2 — frontier table.
• Microsoft, McKinsey, PwC, Deloitte — usage, adoption, wages.
• aifunding.me — tracked venture tallies, September 2026.
• gradually.ai cluster estimates — large GPU fleets, including Colossus.

What this is for

There is a research desk (Epoch), a leaderboard (Artificial Analysis), a university yearbook (Stanford), and a swarm of newsletters. This site is the morning desk plus features: the live model to feel the scale, the desk to answer “what happened since yesterday,” the essays to say what the number means this week.

Read the masthead How to read this issue.