FRAMEWORK DESIGN
Why eight tests—and why equal weight?
WHY EIGHT?Broad enough to test the cycle. Small enough to avoid double-counting.
The tests cover eight observable channels through which the AI investment cycle could strengthen or come under pressure: capital returns, funding, demand, competition, monetisation, market concentration, valuation and wider economic exposure.
A test is included only when it has a clear economic relationship to the thesis, a sufficiently reliable public source, a repeatable observation method and predetermined scoring thresholds. Fewer tests would leave the reading overly dependent on one company or one part of the cycle. Adding closely related indicators could count the same underlying pressure more than once and create false precision.
WHY EQUAL WEIGHT?Transparency before unsupported precision.
Each available verified test currently receives equal weight. This is not a claim that every indicator has identical economic importance in every environment.
BoomRisk does not yet have a sufficiently long live history to defend calibrated unequal weights without introducing subjective judgement or overfitting. Any future addition, removal or weighting change would require a new published methodology version and clear disclosure; it would not be made simply because it produced a preferred score.
THE EIGHT PRESSURE TESTS
The fixed rules behind every dashboard number.
Higher scores always mean greater pressure on the economics of the AI investment cycle. The live Evidence & Method panel highlights the band containing today’s measurement.
01 · CAPEX & CASH RETURNSFree cash flow
Aggregate free-cash-flow growth for Microsoft, Alphabet, Amazon and Meta. 1: at least 10% · 2: 0–9.99% · 3: -10–0% · 4: -25–-10% · 5: below -25%. Capex growth is context, not a second score.
02 · FUNDINGCredit stress
90-day change in broad US investment-grade spreads. 1: tightened at least 10 bp · 2: tightened 0–10 bp · 3: widened 0–20 bp · 4: widened 20–50 bp · 5: widened more than 50 bp.
03 · DEMANDSemiconductor demand
NVIDIA Data Center revenue growth year over year. 1: at least 50% · 2: 25–50% · 3: 10–25% · 4: 0–10% · 5: contraction.
04 · COMPETITIONOpen-model pressure
Price-performance pressure from qualifying open-weight models. 1: capability gap above 20% · 2: gap at most 20% · 3: gap at most 15% and at least 30% cheaper · 4: gap at most 10% and at least 50% cheaper · 5: gap at most 5% and at least 50% cheaper.
05 · RETURNSAI monetisation
Amazon and Alphabet capex growth minus AWS and Google Cloud revenue growth. 1: revenue keeps pace · 2: capex leads by at most 15 points · 3: 15–30 points · 4: 30–50 points · 5: more than 50 points.
06 · MARKET FRAGILITYMarket concentration
Weight of the ten largest S&P 500 companies. 1: at most 25% · 2: 25–30% · 3: 30–35% · 4: 35–40% · 5: above 40%.
07 · EXPECTATIONSValuation
S&P 500 one-year forward P/E. 1: at most 16× · 2: 16–19× · 3: 19–22× · 4: 22–25× · 5: above 25×.
08 · MACRO EXPOSUREData-centre investment/GDP
US data-centre construction as a share of GDP, plus annual growth. 5: share at least 0.35% or growth at least 75% · 4: 0.20% or 40% · 3: 0.10% or 20% · 2: 0.05% or non-negative growth · otherwise 1.
OVERALL READING
What the number and colour mean.
- Lower risk: 1.00–2.49. The monitored economics remain broadly supported.
- Elevated risk: 2.50–3.99. Material pressure is visible, but the cycle has not broadly broken.
- High risk: 4.00–5.00. Severe pressure is present across the scored evidence.
- Signal scores remain whole numbers because each is a fixed band. The overall average uses two decimals to preserve real movement between those bands.
SOURCES & REFRESH
Primary evidence first.
Company fundamentals come from SEC filings and earnings releases. Credit comes from FRED’s ICE BofA US Corporate spread. Construction and GDP come from Census and BEA. Market concentration and valuation use documented index or fund data. Model price-performance uses published model catalogues and benchmark results.
Each signal refreshes when its source produces a comparable observation. Daily checks do not manufacture daily changes in quarterly or annual data. Every live panel shows its source, observation period, method, limitation and update time.
HISTORY
Live monitoring and reconstructed context are kept separate.
Stored live readings begin on 22 August 2026. The longer monthly view adds twelve month-end reconstructions from August 2025 through July 2026 using only evidence available by each date. Reconstructed points are analytical context, not evidence that Boom Risk operated or issued alerts at that time.
Reconstructed points use six historically recoverable tests and are labelled differently from live readings. Daily, weekly and monthly views select recorded observations. They never interpolate missing values. A flat line means the stored score did not change.
LIMITATIONS
What the dashboard cannot prove.
Cloud revenue is not purely AI revenue. The credit input is a broad-market proxy rather than AI-issuer debt. NVIDIA revenue does not capture every chip supplier or cancellations. Construction excludes servers, chips and much power investment. Open-model prices are hosted list prices, and the valuation test does not yet include a reliable earnings-revision feed.
BoomRisk is monitoring research, not an investment recommendation or a prediction of when markets will rise or fall.