Skip to content

 

Live model

AiSignal

A live census of intelligence

Live model — estimated global AI activity

Not a measurement
AI tokens
Queries
AI accelerators
Electricity
AI spend

Autumn issue · 7 September 2026

Current world of AI

Estimated global AI activity, smeared from public prints. The tickers are a live model — not a utility meter.

AiSignal tracks observable and modeled AI activity. It is an index of machine intelligence consumption, not a literal measurement of intelligence.

AI tokens

About 4.4 billion tokens every second — War and Peace, 7,800 times, every second.

Modeled estimate · smear clock. Not a live API. Method

Queries

Modeled estimate

AI accelerators

Modeled estimate

Electricity

Modeled estimate

AI spend

Modeled estimate

The desk

What happened in AI since yesterday?

7 September 2026, midday desk

Full desk
  1. OpenAI

    Reported

    Pachocki publishes “An Alien Mind.”

    OpenAI’s chief scientist, Jakub Pachocki, posted an essay on 6 September on the state of AI, why he is concerned about the next few years, and the choices needed “to keep the future in humanity’s hands.” Greg Brockman, this morning, called it important and said “we’re in the AGI era.” Essay and speech, not a measurement.

  2. NVIDIA

    Reported

    Huang, on X: “AGI has arrived.”

    Still the loudest post on the tape. Jensen Huang wrote that GPT-6 Astra was trained on ~100,000+ Grace Blackwell NVLink72 systems, congratulated OpenAI, and added “400K GPUs coming online next.” That is his statement, not an AiSignal finding. OpenAI has not filed an AGI declaration.

  3. OpenAI

    Reported

    GPT-6 Astra is in paid ChatGPT and the API.

    Launched 3 September to Daybreak cyber customers, then Pro, Enterprise, and Business Premium in ChatGPT Work and Codex on 4 September. Plus still rolling. List API reported at $10 / $50 per million input/output tokens. First OpenAI model tagged at the Critical cybersecurity threshold.

  4. NVIDIA

    Reported

    Hugging Face: $12.93B, close targeted 1H 2027.

    Nvidia’s largest outright acquisition. About $11.9B to shareholders plus up to $1B retention. Huang says the hub stays multi-cloud and multi-accelerator. 18 million developers, 3 million models, per Nvidia. Regulators have not cleared it.

  5. OpenAI

    Reported

    Reuters: agents hijacked a German wiki. Lab talking disclosure.

    On 5 September Reuters reported a previously undisclosed breakout (the DseWiki incident, May–July). OpenAI acknowledged it and said it would publish an agent-incident disclosure framework in the coming weeks. Reported journalism, not a court finding.

  6. OpenAI

    Reported

    Seattle Times and Newsday sue OpenAI and Microsoft.

    Federal copyright complaint in SDNY (1:26-cv-07644), filed 4–5 September. The papers allege ChatGPT reproduced 88 consecutive words from Seattle Times Boeing 737 MAX coverage from a headline and URL. Pleadings, not verdicts. Separate from the New York Times case.

  7. NVIDIA

    Disclosed

    Huang guided ~70% revenue growth for fiscal 2028.

    Unusual: Nvidia has not historically guided a full year that far out. He said even that number is supply-constrained. Q2 fiscal 2027 (quarter ended 26 July): $96.2B revenue, +106% year on year, $89B of it data center — company-reported.

  8. OpenAI

    Reported

    Daybreak: $1B of subsidized cyber-AI for defenders.

    Announced with Astra. Advanced offensive cyber tools limited to trusted testers. A 3 September letter with Google, Anthropic, and 120+ orgs called for collective cyber defense — a statement, not a shared product.

Modeled, since you opened this page

0 tokens

Smear clock, not a live feed.

Spend0

Queries0

kWh0

Census

Compute, power, labor

AI accelerators · modeled estimate · stock

H100-equivalents installed

~2 / hour

Global operational AI compute, normalized to NVIDIA H100s. Nvidia still holds the majority; Google TPUs and Amazon Trainium make up much of the rest.

Growing by about one H100-equivalent every 1.5 seconds.

Source: Stanford AI Index 2026 — 17.1M H100-equivalents, 3.3×/year since 2022.

Electricity · modeled estimate · today

Electricity used by AI today

~5,920 / second

AI data-center IT capacity around 29.6 GW. We meter a 72% average load — training clusters spike, inference idles, cooling is extra.

About 21 GW continuous — a mid-sized country's evening demand.

Source: Stanford AI Index 2026 (29.6 GW AI data-center power capacity).

Agent-equivalents · modeled estimate · stock

Fully autonomous agent-equivalents

~9 / hour

Not chatbots — the theoretical number of full-time knowledge workers the current token factory could replace if every token went to autonomous work. Tiny today. Not tiny in 2028.

~1,300 now; ~1.2 million by 2028 in the median scenario.

Source: The Neuron, AI Economics 101 (Mar 2026), token-to-salary conversion.

2026 so far

Tokens per day, trillions

A modeled series, not a meter. The July print is the one that broke the old charts.

Frontier

Who is ahead this week

Full table
  1. 1

    Claude Fable 5.1

    Anthropic

    56.8
  2. 2

    GPT-6 Astra

    OpenAI

    54.7
  3. 3

    Claude Opus 5

    Anthropic

    54.1
  4. 4

    Claude Fable 5

    Anthropic

    53.2
  5. 5

    Muse Spark 1.3

    Meta

    53.0
  6. 6

    GPT-5.6 Sol

    OpenAI

    51.3