Crowding Index

The number ships after the method, not before it.

The index will measure divergence between an AI-exposed input basket and matched controls, with January 2024 = 100. It is a pressure gauge—not a claim that every divergence is caused by AI.

StatusCalibration
Headline valueWithheld
Target scored releaseAug 4, 2026

A calendar target is not permission to publish. If the checklist fails, the release remains “withheld” and the failure is logged on Changes.

Observed components

Signals visible before a composite exists

These official PPIs are reproducible inputs, not an index preview. The aggregate semiconductor series falling while storage rises is exactly why an intuitive composite can mislead.

Storage-device PPI65.056
+30.4% · Jan 2024 to Jun 2026
Transformer PPI365.306
+5.9% · Oct 2024 to Jun 2026
Semiconductor PPI29.695
−6.3% · Jan 2024 to Jun 2026

Provisional exposed basket

Markets AI plausibly touches

  • HBM / advanced-memory capacity
  • Enterprise storage devices
  • Power and distribution transformers
  • Switchgear and power-control equipment
  • Gas-turbine manufacturing slots
  • Electrician and mission-critical trade labor
  • Large-load electricity and grid capacity
  • Data-center corridor construction inputs

Control rules

A matched market, or an honest blank

  • Same measurement family and release cadence
  • Similar input intensity before January 2024
  • Low direct AI exposure
  • No controls selected for a convenient result
  • Control changes logged before recomputation
  • No headline when a credible match does not exist

Publication gate

Six tests before “100” becomes a moving number

  1. 1
    Series coverage

    At least 80% of exposed and control basket weights have current observations.

  2. 2
    Economic sign

    Each series states whether scarcity should raise or lower the observation.

  3. 3
    Control match

    Every exposed series has a documented non-AI comparison or an explicit no-control flag.

  4. 4
    Weighting

    The launch release publishes equal-weight and sensitivity results before selecting a headline.

  5. 5
    Vintage

    Raw values, transformations and retrieval dates are downloadable and revision-aware.

  6. 6
    External read

    One reviewer outside the project can reproduce the number from the file.

Open data

Download the launch evidence file

One row per published signal with unit, date, grade, caveat and source URL.

Download CSV