A QTS bond tied to a Microsoft data center project has widened more than 30 basis points since April while Microsoft’s own debt barely moved, and Citi says investors are finally pricing AI data center bonds as project finance rather than borrowed hyperscaler credit.
For the past two years, buying an AI data center bond often looked like a simple bet on the tech giant whose name appeared on the lease. If Microsoft, Amazon, Meta, or another hyperscaler was the anchor customer, many investors treated the project almost as if the tenant’s balance sheet stood behind it. Citi’s credit analysts argued this week that the shortcut is breaking down, and the clearest evidence is now showing up in secondary trading.
The case study is a bond issued by QTS, the Blackstone-owned data center operator, for a facility in Fayetteville, Georgia tied to Microsoft. QTS priced $4.6 billion of debt in April in a deal that drew about $12.5 billion of peak demand, a sign of how hungry investors have been for exposure to the AI infrastructure build-out. Since then, the spread on the bond has widened by more than 30 basis points. Microsoft’s own corporate bonds, over roughly the same period, have barely moved. That gap matters because it shows investors are no longer treating the project and the tenant as the same credit.
Citi traces the underperformance to structure, not to a sudden concern about Microsoft or the broader need for data center capacity. The QTS bond uses a bullet repayment format, meaning the principal comes due at maturity rather than being paid down gradually over the life of the deal. That creates a refinancing cliff. When the bond matures, QTS has to return to the market and roll the debt. If financing conditions are worse, or if the economics around the Microsoft lease have changed, the bondholder is exposed to that moment. In an amortizing structure, that risk falls over time as the outstanding balance declines.
The significance is that discipline is arriving in a corner of credit that had been running on implied hyperscaler comfort. The presence of a big technology name on the lease was enough to pull project debt toward corporate-style pricing, even when the repayment profile carried project-level risk. Now investors are doing what project finance investors are supposed to do: reading the structure, testing the cash flow, and asking who really bears the risk if the market is less forgiving at maturity.
The timing is important because the borrowing wave is still building. Global AI-related debt issuance is on track to nearly double to roughly $570 billion in 2026, according to Morgan Stanley estimates cited in recent market reporting. About $236 billion had already been raised globally through the end of May, roughly four times the pace from the same period a year earlier. The exact totals will move as new facilities close, but the direction is clear. AI infrastructure is becoming one of the largest capital formation stories in the investment-grade market.
CoreWeave’s $8.5 billion delayed draw term loan, which closed in March, shows the other side of the same market. That financing was built around GPU infrastructure and tied to major customer commitments, with an investment-grade rating that helped prove compute-heavy assets could tap deep pools of credit. Investors responded because the deal was structured to match the risk. The QTS bond is now being judged by the same standard. The market is not rejecting AI infrastructure debt. It is separating stronger structures from weaker ones.
Amazon’s newly announced $17.5 billion delayed draw term loan, confirmed on June 10, operates on different logic. It is a direct corporate obligation from Amazon, led by Citigroup, and sits against the balance sheet of a company planning enormous AI and cloud spending. That is not the same thing as a project bond tied to a facility leased by a hyperscaler. The distinction sounds technical until spreads start moving. Then it becomes the whole story.
What Comes Next for the Build-Out
The next test is whether widening spread dispersion changes how new data center deals are built. Bullet structures should face a higher hurdle as investors price refinancing risk more explicitly. Developers may have to accept more amortization, offer wider spreads, shift more financing into private credit, or lean harder on tenant commitments that give bondholders cleaner protection. For infrastructure owners including QTS, Digital Realty, and Equinix, the financing mix is now more complicated than it looked six months ago.
The AI build-out still needs capital, and plenty of it. What is changing is the price of convenience. A hyperscaler lease can support a deal, but it does not erase the legal and financial structure beneath it. The next major data center bond sale will show whether arrangers adapt quickly or ask investors to keep accepting project risk at near-corporate prices. Either way, the easiest phase of AI infrastructure credit is ending.
Also read: CoreWeave's $8.5 billion investment-grade loan rewrites how AI infrastructure gets financed • Einride's 74% Nasdaq Debut Reveals How Much the Market Is Paying for an Autonomous Future That Barely Exists Yet • May CPI landed at 4.2% and the Fed's rate-cut window just slammed shut
