Debt for AI data centers moves off public markets into private hands

The buyers said no, and the borrowing got bigger anyway. The debt did not vanish when bond investors balked; it moved into deals struck off public markets where prices are invisible.

What matters now is who ends up holding hundreds of billions in data center debt without exchange pricing or daily mark tests. The public markets confronted a mismatch: borrowers burning cash want more financing, buyers demand higher yields, so the financing found a home away from scrutiny. Money keeps flowing precisely because visibility ended.

Penn Mutual Asset Management described the picture on August 6, noting that high-yield bonds tied to data centers now trade at such tight spreads over ordinary corporate debt that investors earn almost nothing extra for the trouble. Meanwhile, investment-grade bonds from the big cloud companies have lagged the broader market all year due to relentless new supply. Morgan Stanley, cited by Dataconomy’s report on August 19, projects free cash flow across five main data center operators will swing from positive $187 billion in 2025 to negative $2.8 billion in 2026.

The venue changed, not the terms. Banks hit concentration limits for single-project exposure and so migrated financing into Rule 144A bond sales and private credit funds, placing debt in pension fund and insurance company portfolios instead of traded markets — TechTimes reported this transition on August 10. A 144A deal is sold to a handful of institutions without public registration, ticker, or daily mark. An insurer holds it at whatever the model says until maturity or default.

Researchers estimate $662 billion of these shadowy obligations at five major tech companies, equal to 113 percent of their combined borrowings as counted by rating agencies. Axis Intelligence published that tally on August 1. Total US corporate bond issuance hit a record $1.68 trillion for the year, with much of it technology borrowing, according to Mezha Media figures from August 2026.

When you examine a single deal, the structure and its risks become clear. Meta’s Hyperion campus in Louisiana sits under Blue Owl Capital’s ownership through a shell registered as Laidley LLC, financed with about $27 billion, with BlackRock and Pimco as lead investors. The New York Times laid out this deal on July 27. Blue Owl committed to hold Hyperion for 24 years, but Meta can exit after four as long as it pays the gap between leftover debt and what a replacement tenant would pay.

Standard & Poor’s rated that shell’s debt A+, one notch below Meta’s own rating, because of that exit guarantee, as its August 2026 action shows. The rating entirely depends on a promise from the tenant whose aggressive spending made the extra yield necessary.

S&P grades this debt near the top because the tenant promises to pay, yet bond traders buying insurance against company defaults charge prices that suggest real doubt those promises will survive a downturn. Penn Mutual Asset Management’s August 6 note carried both figures: models fed by guarantees and lease contracts say one thing, but real-money bets say another. Both cannot be right.

Where the debt landed

Insurance company balance sheets are the slow force underneath this story. Insurers were already the chief lenders in private placements, buying long-dated assets to match policy promises, as Insurance Journal reported February 3.

The National Association of Insurance Commissioners (NAIC), which sets capital requirements for insurers, recently began reviewing whether credit ratings on private credit and infrastructure securities held by US insurers are justified, with authority to overrule ratings — Insurance Business Magazine reported this on June 12. The regulator also published updated monitoring guidance on July 24, flagging valuation practices as the key risk.

If NAIC requires lower ratings or conservative marks, insurers will need to raise capital or stop buying. Either way, the quiet buyer of last resort begins to reconsider.

Before 2008, home loans that public markets refused at quoted prices were repackaged into securities rated AAA by models trusting the housing guarantee, and insurers like American International Group wrote the guarantees. The process. Refuse the price, relocate the risk, keep building. The difference today is collateral. A data center leased for fifteen years to a cash-rich company with an exit penalty attached is not a house bought with nothing down.

The counter-case holds. If Meta, Microsoft, Alphabet, and Oracle keep paying rents, these deals work as rated, and the episode looks like prudent risk-sharing rather than hidden rot. The bear case requires tenants to stumble, not just projects. Whether that happens is the question, and the next year of hyperscaler borrowing will answer it in public.

Who pays if it breaks? Policyholders of insurers holding unlisted paper carried at full value, and pension fund investors who bought the same 144A bonds. Who profits? Blue Owl, BlackRock, and Pimco collect fees from structures that pay regardless of price write-downs, and hyperscalers get campuses built without showing the debt on their books. Valuations are set by parties paid for volume, not accuracy.

Watch NAIC’s review for moves to haircut ratings on private placements held by insurers; that would confirm real pressure. Then watch the next hyperscaler bond sales. Fewer bidders, wider yields show public buyers are finished absorbing supply. The read breaks if spreads on securitized data center debt tighten again while insurers keep buying without regulatory objections — that would mean the risk relocation found willing holders.

Buildings don’t default. Promises do, and these towers rest on promises from four companies whose rent checks depend on continued access to borrowed money.