60% odds a correction (>30%) happens at all before end-2027

Tests whether AI infrastructure spending is being funded internally, from the hyperscalers' own operating cash flow, or increasingly from outside debt and equity. The score tracks how many of the five major hyperscalers are free-cash-flow negative: 0 of 5 reads as fully SUPPORT, 5 of 5 as fully ELEVATED.

Measures investor demand for new AI-linked corporate debt, expressed as how many times a bond offering is oversubscribed. A cover ratio near 5x signals ample appetite; a ratio approaching 1x means a deal barely clears — a forward read on tightening credit access before it shows up elsewhere.

Oracle's five-year credit default swap spread works as a single-name canary for AI-linked credit stress, since Oracle carries some of the largest AI-driven capex commitments of any public company. The score scales from a pre-AI-boom baseline of 50bp (SUPPORT) to a near-record-stress level of 250bp (ELEVATED).

Tracks whether investors are still funding the companies driving the AI buildout, measured as the six-month return of an equal-weight hyperscaler basket against the S&P 500. In both the 1929 utility boom and the 2000 telecom boom, the companies spending the most began underperforming the broad index months before the index itself peaked. The score scales from +20pp of outperformance (SUPPORT) to 10pp of underperformance (ELEVATED). The basket tracks the buyers of AI infrastructure rather than its suppliers, which is why Nvidia is not included.

Combines the cost of money with the direction of liquidity conditions. Higher policy rates raise the financing cost of the AI buildout and pressure the valuations underpinning it; the rate component scales from 0% (fully accommodative) to 6% (highly restrictive). The second component reads the 12-month change in the 10-year minus 2-year Treasury spread, scoring flattening as rising risk and steepening as easing. The curve leads the business cycle, but the lead time has historically ranged from roughly 18 to 90 weeks and has produced false signals, so this is context rather than a trigger. This signal's weight is split evenly between the rate and curve components.

Shows how concentrated AI-linked debt has become within investment-grade credit markets, using its share of a major liquid bond index as a proxy for systemic exposure. The score scales from 5% (a normal sector weighting) to 25% (extreme concentration).

Puts the scale of AI infrastructure spending next to what AI products actually generate in consumer revenue today, as a capex-to-revenue ratio. Because that ratio can span orders of magnitude, the score uses a log scale between 5x (sustainable) and 100x (extreme).

A broad high-yield credit spread used as a market-wide baseline against which the more AI-specific signals above can be read. This methodology scores it the conventional way — wider spreads mean higher risk, from 2.5% to 10% — rather than reading tight spreads themselves as a warning sign.

Tracks whether combined infrastructure spending across the major hyperscalers is accelerating or moderating year over year. The score scales from 10% YoY growth (sustainable) to 80% YoY (explosive) — faster growth reads as more risk, since it signals compounding commitment rather than tapering.

The bull-case counterweight to the other Tier 2 signals: strong revenue and operating growth at the core hyperscalers is what would justify the spending tracked elsewhere on this page. This signal is inverted — high growth (80% YoY) reads as SUPPORT and fading growth (10% YoY) reads as ELEVATED.