A $27 billion bond issuance has moved through private placement channels over the past eighteen months to fund data-center construction tied to AI compute infrastructure, according to capital structure filings reviewed by multiple credit desks. The notes never touched public exchanges. The buyers are insurance balance sheets, sovereign wealth vehicles, and a handful of private credit funds with minimum $500 million check sizes. The borrowers are the usual names—hyperscalers and their joint-venture subsidiaries—but the documentation sits behind non-disclosure agreements and the pricing remains opaque outside participant circles.

The structure is simple. Special-purpose vehicles issue senior secured notes against physical assets: land, power contracts, cooling systems, fiber runs. Coupon rates are reportedly in the 5.8% to 7.2% range for seven- to ten-year paper, depending on credit enhancement and whether the facility has a signed capacity agreement with a named tenant. Construction loans convert to term debt once the facility reaches mechanical completion and passes commissioning. The collateral is hard. The risk is adoption lag—whether the AI workloads materialize at the scale these facilities assume, and whether the hyperscalers renew capacity contracts when the initial terms expire in 2029 through 2031.

This matters because the private bond market has become the primary funding vehicle for physical infrastructure that public equity markets no longer want to capitalize directly. Hyperscalers are keeping data-center construction off their balance sheets by routing capital through these SPVs, which keeps reported capex figures lower and shifts construction risk to bondholders. The trade works as long as utilization stays high and power costs remain predictable. If either assumption breaks—if model training demand plateaus or if electricity price volatility spikes in key markets like Virginia, Texas, or Dublin—the bondholders absorb the loss, not the tech operating companies. The public never saw the issuance, so the public has no forward visibility into how much leverage is now stacked against AI infrastructure build-out.

Allocators should watch for three follow-on events. First, whether any of these SPVs attempt to securitize their note tranches and move them into CLO structures, which would signal either liquidity stress or an attempt to re-price risk. Second, whether utilization data from the largest facilities starts appearing in quarterly earnings calls, which would indicate the hyperscalers are being asked harder questions about capacity absorption. Third, whether insurance commissioners in New York, California, or Connecticut begin requesting exposure reports on private data-center debt, which would suggest regulatory attention is turning toward concentration risk in illiquid credit portfolios. All three are likely within the next eight to twelve months.

The private bond market now holds more AI infrastructure debt than the public market ever priced.