In Summary
- Equity captures the upside of the AI build-out, while credit absorbs the loss if it fails – yet creditors are compensated with under 1% p.a. on 5-year bonds and around 2% on 10-year. The asymmetry has grown as total long-term debt at the eight dominant US tech companies rose by 86% in the 12 months to mid-2026. If AI investment pays off, equity holders benefit; if it does not, the added debt, increasingly compounded by off-balance-sheet commitments, leaves creditors exposed to a skewed distribution of outcomes that current spread levels may not reflect. This paper tests whether spreads price that skewed risk using three structural credit models – Merton, and KMV-inspired, and CreditGrades – with financial-filing and equity data for eight mega-caps.
- On recognized debt, the market’s calm is justified but we see the first drifts. Our modelling agrees with market pricing and ratings on most names. As of mid-2026, recognized leverage across the panel is low, at 0.2% to 6.8%, implying credit risk broadly consistent with Aaa–A ratings. Three of the eight screen weaker than their formal rating due to high leverage (Meta) or equity volatility (Oracle, SpaceX).
- The real risk lurks below the waterline: off-balance-sheet debt lifts the debt burden by nearly 150% on average, and pulls the model-implied credit quality down a notch or two. The deterioration surfaces as quality migration long before it surfaces in price, because distance-to-default remains too large for spreads to react. The market has noted the direction, but not yet the size: CDS spreads more than doubled over the past year, while the average model-implied credit spread rose six-fold once off-balance-sheet obligations are included. Off-balance-sheet debt keeps growing as new commitments are signed, and AI profit potential, even realized, offers little spread-tightening against it.
- Ranked against the broad corporate credit universe, the panel is bifurcated: model-implied credit quality spans from the 97th percentile (strongest) to the 2nd (weakest), with the average sitting near the 62nd — barely better than a median credit. In effect, the rank bought by rising market capitalizations and compressed volatility is being spent on debt growth, both recognized and hidden. The two legs are not symmetric – debt is contractual and accumulates on schedule, while equity relies on the future.
- A severe market repricing – equity down 10%, volatility up 10pp, debt up 50% – would cost the panel two standard deviations of distance-to-default, translating into ~ 20bps of model spread. Debt sensitivity is actually lower: recognizing leases – equivalent to 150% of debt – moves spreads by only ~10bps, with wide dispersion across names. Mid-2022 is the natural stress test: the same balance sheets, carrying far less debt, still showed vulnerability – volatility and time horizon matter more than rates or equity levels.
- As tech allocation in credit portfolios grows, spreads do not fully compensate for skewed risk. Default risk is still remote, but spread risk will increasingly surface. The liability leg is contractual and arrives on schedule, while the equity leg can reverse in a quarter. Tech’s curve steepness already prices the long horizon as the key risk, and that may extend to the front-end. Tech has avoided the iceberg so far, but the navigation is getting trickier, and the metric to watch is spread vs. quality ranking within the corporate market, including by financial commitments and equity volatility. With technology accounting for an ever-larger share of credit, integrating this into position sizing and maintaining the flexibility to adjust course if the (A)Iceberg appears will be key.