Credit Markets Are Pricing AI Debt Like It’s Ordinary Debt. It Isn’t.
Quantumvest Macro Research • 3 min read
|
THEME ID MT-26 |
CONVICTION Medium–High |
TIME HORIZON 2–4 quarters |
TOPIC Credit / Fixed Income |
Investment-grade corporate bond spreads recently touched roughly 71 basis points - their lowest level since 1998 and high-yield spreads have compressed below 300 basis points, a level reached only about 5% of the time since 2000. By historical standards, credit markets are pricing corporate America's risk of default at levels associated with only the calmest periods on record. We think that pricing is increasingly disconnected from what is actually being financed.
The disconnect is one of composition, not just level. A rising share of new investment-grade issuance is being driven by hyperscalers and data-center-linked borrowers funding the AI buildout: 2026 high-grade issuance is forecast to reach roughly $2.25 trillion, a 35% increase from the prior year, with AI-related borrowing a primary driver. Under normal market dynamics, a sudden, concentrated surge in supply from one thematic vertical would be expected to cheapen — widen — spreads for that segment as the market absorbs the new debt. Instead, these deals have been oversubscribed by as much as five times, and aggregate spreads have stayed compressed to some of the tightest levels investors have ever seen.
|
Spreads are near 30-year lows at the exact moment the composition of new supply is shifting toward a concentrated, capital-intensive, monetization-unproven business model. The market is pricing the label “investment grade,” not the balance sheet behind it. |
We think this is a technical phenomenon being mistaken for a fundamental one. Persistent demand from insurance companies, pension funds, and yield-seeking retail and institutional buyers — alongside reinvestment needs from existing bond funds - has overwhelmed the increase in supply, keeping spreads tight regardless of what that supply is actually financing. That is a demand-technical story, not a verdict on credit quality. As we've written elsewhere, a meaningful and growing share of this same issuance sits behind a capital-intensive AI infrastructure build whose revenue payback timeline remains genuinely unresolved, and whose accounting (see our companion piece on GPU depreciation) leaves real room for reported earnings to move against issuers. Pricing that risk at the same spread as a diversified, seasoned corporate universe assumes those two categories of debt carry equivalent risk. We don't think they do.
|
WHY WE THINK THIS IS MISPRICED Credit spreads are supposed to compensate investors for the risk that a specific issuer's business model or financing structure deteriorates. When one of the largest and fastest-growing sources of new supply is tied to a single, relatively young, high-capex business model with an unresolved monetization timeline, a rational market would demand a differentiated risk premium for that segment. Instead, oversubscription and technical demand have compressed spreads market-wide, blending AI-related issuers into the same historically tight pricing as the rest of the investment-grade universe. |
|
WHAT WOULD CHANGE OUR MIND A visible “AI premium” beginning to open up - data-center and hyperscaler-linked issuers trading meaningfully wider than the broad investment-grade index - would suggest the market is starting to price this distinction, and would reduce our concern. We would also revisit this view if elevated 2026 issuance volumes cause spreads to widen broadly, as several credit strategists now forecast, since that would suggest the technical supply-demand imbalance is correcting on its own. |