Part 2: Stress testing the AI buildout

This is the central stress. It assumes rating agencies and bond investors begin to treat AI-related commitments, leases, guarantees, and special-purpose vehicles as economically relevant leverage. The immediate move is not a collapse in AI demand. It is a repricing of the capital structure that funds the buildout.

A practical equity stress would combine a negative shock to broad market sensitivity, a larger negative shock to high-growth technology and semiconductors, a drawdown in data-center beneficiaries, and wider investment-grade and high-yield credit spreads. In factor terms, focus on Market Sensitivity, Downside Risk, Residual Volatility, Leverage, Investment, and Profit Growth.

Suggested calibration: Nasdaq down 20 percent, semiconductors down 30 percent, AI infrastructure beneficiaries down 25 percent, US 10-year yields up 100 basis points, investment-grade spreads wider by 75 basis points, high-yield spreads wider by 150 basis points, and volatility up sharply. 

This is the non-recessionary version of the stress. Compute demand remains durable, but power availability, permitting, transformers, turbines, and grid interconnection delay delivery. Investors do not conclude that AI is false. They conclude that the path from spending to revenue is longer and less linear than valuations had assumed.

The equity shock should be more targeted than in the first scenario. It should hit long-duration growth, capital-intensive technology, industrial suppliers tied to the buildout, and firms whose free-cash-flow assumptions depend on rapid capacity deployment. Utilities and selected infrastructure names could be neutral or positive if investors begin to price scarcity in delivered power.

Suggested calibration: broad US equities down 8 percent, technology down 15 percent, selected AI infrastructure and semiconductor beneficiaries down 20 percent to 25 percent, industrials tied to the buildout down 10 percent, long yields up 50 basis points, and regulated power exposure flat to modestly positive.

This scenario isolates tenant and counterparty concentration risk. One large AI customer delays payment, restructures commitments, reduces future capacity reservations, or becomes less credible as a long-term counterparty. The issue is whether a small number of private, unrated, or weakly financed customers can support the commitments that hyperscalers, data-center owners, and financing vehicles have already made.

The stress should hit companies with concentrated exposure to OpenAI, Anthropic, or similar counterparties, together with suppliers whose revenue depends on the same expansion path. It should also widen spreads for issuers whose capex has been justified by backlog rather than demonstrated cash conversion.

Suggested calibration: mega-cap AI platform exposure down 25 percent, concentrated AI infrastructure names down 35 percent, semiconductors down 30 percent, private-credit and data-center financing proxies under pressure, credit spreads wider by 200 basis points, and defensive equity sectors flat to positive.

This is the cleanest valuation stress. No major customer fails. Gross domestic product does not roll over. Earnings estimates are not immediately cut. Investors simply decide that too much future success has been capitalized into current prices. Portfolios can lose money even when the fundamental story remains directionally valid.

The stress should focus on high Profit Growth, high Momentum, high Residual Volatility, and high Market Sensitivity exposure, with less emphasis on macro rates or credit. It is a multiple-compression exercise, not a default cycle.

Suggested calibration: S&P 500 down 15 percent, AI leaders down 35 percent to 40 percent, semiconductors down 40 percent to 50 percent, long yields unchanged, credit spreads only modestly wider, and defensive quality factors outperform.

The fifth scenario is behavioral. It assumes the narrative shifts from AI can absorb any amount of capital to AI is overbuilt, overfinanced, and vulnerable to funding discipline. Nothing has to fail first. The shock is a change in risk appetite and a rapid unwind of crowded exposures.

In Axioma Risk, combine adverse shocks to Market Sensitivity, Downside Risk, Residual Volatility, Momentum, and Crowding, while allowing the correlated stress framework to propagate the shock across related factors and sectors. This is particularly useful when apparent sector exposure understates behavioral exposure to the same underlying trade.

Suggested calibration: broad equities down 12 percent to 18 percent, high-beta AI exposure down 25 percent to 35 percent, crowded winners down more than sector peers, volatility higher, and lower-volatility quality exposure outperforming.