In short: Oracle closed a $16.3 billion financing for a single data centre campus in Saline Township, Michigan, the largest single-facility technology debt package ever assembled. PIMCO anchored roughly $10 billion of the bond tranche because US banks retreated from the deal, citing doubts about AI infrastructure demand sustainability. The financing is part of $72 billion in total data centre partner debt Oracle has assembled across Michigan, Texas, Wisconsin, and New Mexico for the Stargate joint venture, all backed by a company with BBB-negative credit outlook and $553 billion in performance obligations concentrated heavily in a single counterparty, OpenAI.

Oracle’s $16.3 billion data centre financing closed this week, the largest single-facility technology debt package ever assembled, and it required a bond fund to anchor it because America’s banks were not willing to. PIMCO, the world’s largest active fixed-income manager, purchased approximately $10 billion of the roughly $14 billion bond tranche, according to Bloomberg. The remainder of the capital stack includes approximately $2 billion in equity from Related Digital Infrastructure and Blackstone. Bank of America structured the deal. Goldman Sachs and Wells Fargo advised Related Digital. The financing covers a single data centre campus in Saline Township, Michigan, designed to exceed one gigawatt of capacity and serve as infrastructure for Oracle and OpenAI’s Stargate project, the joint venture announced in January 2025 with SoftBank that has become the gravitational centre of American AI infrastructure spending.

The structure

The bonds carry a 7.5% coupon with a 19.5-year maturity, structured with six years of interest-only payments followed by 13 years of amortisation. This is not cheap debt. Oracle’s own corporate bonds trade at significantly tighter spreads. The premium reflects the project finance nature of the deal: the bonds are secured against the Michigan campus itself, not against Oracle’s corporate balance sheet, which means investors are lending against a single facility’s projected cash flows rather than the creditworthiness of a $400 billion enterprise. The structure isolates the risk but prices it accordingly. At 7.5% over nearly two decades, the total interest cost will exceed the principal, a financing burden that only makes economic sense if the facility generates revenue from day one and maintains occupancy for the duration of the loan. Oracle’s $553 billion in remaining performance obligations, reported in Q3 of fiscal 2026, provides the demand signal that underwrites that assumption, but the concentration risk is considerable. A significant share of those obligations traces back to a single counterparty: OpenAI.

TD Cowen reported earlier this month that US banks have been retreating from Oracle data centre financing, citing doubts about the sustainability of AI infrastructure demand at the scale Oracle is projecting. That retreat is why PIMCO ended up anchoring $10 billion of the bond offering. When traditional lenders step back from a deal, it does not necessarily mean the deal is bad. It means the risk-reward calculation has shifted to a point where banks, constrained by regulatory capital requirements and concentrated exposure limits, cannot justify the position. Asset managers like PIMCO, which manage discretionary capital with longer time horizons, can. But the signal matters. The largest data centre financing in history required the bond market to do what the banking system would not.

The scale

The Michigan campus is not Oracle’s only data centre megaproject. It is not even the largest. Oracle has assembled at least $72 billion in total data centre partner debt across three major financing packages: the $16.3 billion Michigan deal, approximately $38 billion for campuses in Texas and Wisconsin, and roughly $18 billion for a facility in New Mexico. All of these are being built to serve the Stargate joint venture and Oracle’s broader cloud infrastructure expansion. Oracle’s own capital expenditure for fiscal 2026 is expected to reach approximately $50 billion, more than double the prior year. Oracle’s $50 billion capex programme, outlined by CFO Hilary Maxson, represents the company’s largest-ever infrastructure investment cycle and has strained the balance sheet to a degree that prompted both S&P and Moody’s to assign negative outlooks to Oracle’s BBB and Baa2 credit ratings respectively. UBS analysts have said a downgrade to junk is unlikely, but the trajectory is clear: Oracle is leveraging its investment-grade rating to the maximum extent the market will tolerate.

The sheer volume of capital flowing into AI data centres has created a parallel financial infrastructure. Jane Street’s involvement in CoreWeave’s $6 billion debt facility and $1 billion equity raise demonstrated that Wall Street’s most sophisticated trading firms see AI cloud infrastructure as a fixed-income opportunity. Blackstone’s $10 billion data centre debt financing in Australia through Firmus showed the same pattern repeating on another continent. Bain Capital’s $5 billion stake sale in Bridge Data Centres confirmed that private equity sees data centre infrastructure as a liquid, tradeable asset class. Oracle’s Michigan deal is the largest single transaction in this emerging market, but it is part of a pattern: AI infrastructure is being financed like real estate, with project-level debt secured against physical assets and long-term lease obligations, because the capital requirements have exceeded what corporate balance sheets alone can support.

The bet

Oracle’s transformation from a database and enterprise software company into an AI infrastructure provider has been the most dramatic strategic pivot in enterprise technology since Microsoft’s cloud transition under Satya Nadella. Oracle’s cloud revenue has been growing at more than 20% annually. Its remaining performance obligations of $553 billion in Q3 fiscal 2026 represent contracted future revenue that provides visibility into demand. But contracted revenue is not the same as collected revenue, and the gap between signing a cloud contract and building the infrastructure to fulfil it is where Oracle’s financial risk concentrates. The company announced 30,000 layoffs earlier this year, a workforce reduction that executives framed as a reallocation of resources from legacy operations to cloud and AI infrastructure. The human cost of the pivot is being borne by employees whose skills were built for a business Oracle is leaving behind.

The $1.4 trillion in utility capital expenditure projected through 2030 to support AI data centre power demand illustrates the infrastructure chain that must function for Oracle’s bet to pay off. A one-gigawatt data centre campus requires dedicated power generation, transmission upgrades, water cooling systems, and grid interconnection agreements that take years to complete. Saline Township is in Michigan, where DTE Energy provides electricity and where the state has been courting data centre development with tax incentives and expedited permitting. But power availability is the binding constraint on data centre expansion nationally, and every gigawatt committed to Oracle is a gigawatt unavailable to competitors.

The question

The fundamental question the Oracle financing raises is not whether AI infrastructure demand is real. It is whether the demand is durable enough to justify $72 billion in project-level debt with 19.5-year maturities. The history of technology infrastructure is littered with overbuild cycles: the fibre-optic boom of the late 1990s, the data centre construction surge of 2006 to 2008, the first wave of cloud buildouts that left Amazon Web Services with years of excess capacity before demand caught up. In each case, the technology ultimately justified the investment, but the companies and creditors who financed the first wave often did not survive to benefit. The fibre laid in 1999 carries today’s internet traffic, but the companies that laid it went through bankruptcy first.

Oracle’s counterargument is that this cycle is different because the demand is contractually committed, not speculative. The Stargate partnership with OpenAI and SoftBank provides a named customer with binding obligations. But OpenAI itself is a company that has never generated an annual profit, is burning cash at an accelerating rate, and is in the process of converting from a nonprofit to a for-profit structure in a transaction that remains legally contested. The durability of Oracle’s $553 billion in performance obligations depends on the financial health of the customers who signed them, and the largest of those customers is itself dependent on continued venture capital and strategic investment to fund its operations. PIMCO’s willingness to anchor $10 billion of the Michigan financing suggests that the world’s most sophisticated bond investors believe the risk is manageable. The retreat of US banks from the same deal suggests that not everyone agrees. The $16.3 billion got over the line. The question is whether, 19.5 years from now, the campus it financed will look like a prescient bet on the defining technology of the century or like another monument to a cycle that peaked before the debt was repaid.