Jag's AI Frothiness Index
A daily 0–100 reading of how stretched the AI boom is, built only from free public data: stock prices, borrowing, hyperscaler spending, speculation, and real demand. 50 is normal for the last few decades; the dot-com peak is the yardstick for 100.
0 · Calm40 · Warm60 · Frothy80 · Bubble100
The index over time
Weekly before 2018, daily since. Shaded bands mark Calm, Warm, Frothy and Bubble.
The five pillars today
Each pillar averages its indicators' percentiles. Real demand counts against froth: fast-growing usage makes heavy spending more justified.
Every indicator
Percentile = where today's reading sits against that indicator's whole history up to today (100 = most stretched ever). Click a source to check the raw data.
Bubble talk: the noise meter
Not part of the index. How loudly people are talking and betting about an AI bubble over the last 28 days, compared with every 28-day stretch since 2022. Talk tends to peak before the top, not after.
Noise meter (0–100)
Hacker News: "AI bubble" stories + comments per week
Wikipedia: "AI bubble" article views per week
Betting markets: odds the AI bubble bursts
All AI-themed betting, last 24 hours
How it works
- Inputs: 19 indicators from free primary sources: FRED (St. Louis Fed), SEC EDGAR filings and full-text search, Yahoo Finance prices, the npm registry, Y Combinator's company list, and the Ramp AI Index. No hand-entered numbers.
- Scoring: each reading becomes a percentile against that indicator's own history up to that day, so the past is scored without hindsight. Indicators flagged "inverted" (VIX, credit spreads) score higher when they are low, because calm markets mean complacency.
- Pillars and weights: Prices 30%, Borrowing & risk appetite 20%, Spending strain 20%, Speculation 15%. Real demand takes the remaining 15% as a counterweight: index = 0.85 × (weighted pillars) + 0.15 × (100 − demand).
- History: inputs join as their data begins, so the 1990s use only prices and borrowing; AI-specific inputs (SEC capex, npm, YC, Ramp) arrive after 2009–2023. Slow data (quarterly filings) carries forward until the next release.
- Bands: under 40 Calm, 40–60 Warm, 60–80 Frothy, 80+ Bubble.
- Bubble talk is shown separately because it is contrarian: Hacker News mentions (Algolia API), Wikipedia views (Wikimedia API), and volume on Polymarket, Kalshi and Manifold.
Caveats
- A high reading says things are stretched, not when they will turn. Stretched markets can stay stretched for years.
- Official data is revised; history is rebuilt from today's vintage, not what was known at the time.
- Hyperscaler figures come from XBRL tags that differ between companies; a quarter counts when at least three of the five report.
- Kalshi has no AI-bubble market, so only its overall AI-themed betting volume is counted. Manifold uses play money.
- This is a personal project. Nothing here is investment advice.