OpenAI’s infrastructure partners have built up a massive, highly leveraged financial bet on AI—close to 96–100 billion dollars in debt—because they are racing to provide the data centers, chips, and power needed to satisfy OpenAI’s long‑term compute commitments while keeping most of the borrowing off OpenAI’s own balance sheet.
How the $96B–$100B debt pile is structured
The headline figure comes from an analysis of loans, bonds, and private credit facilities raised by data‑center builders, GPU cloud providers, and energy‑intensive infrastructure firms whose projects are tied directly to OpenAI’s compute needs. These partners are financing multi‑billion‑dollar campuses with long construction timelines, using a mix of investment‑grade bonds, high‑yield debt, bank syndications, and private credit arrangements that push leverage well above what their historical cash flows would normally support.
A rough breakdown shows three big buckets: around 30 billion dollars already borrowed by major names like SoftBank, Oracle , and CoreWeave to either invest in OpenAI or build dedicated capacity; roughly 28 billion dollars in loans tied to structures arranged by Blue Owl Capital and Crusoe; and about 38 billion dollars more in bank debt now being negotiated to fund additional AI data‑center sites via Oracle and Vantage Data Centers. When you add in other loans that are OpenAI‑linked but not always labeled as such, the total exposure around OpenAI’s compute ecosystem approaches the 100 billion dollar mark, comparable to the net debt levels of legacy industrial giants like major telecom or auto groups.
Why the debt sits with partners, not on OpenAI’s books
OpenAI’s strategy is to sign enormous, long‑dated contracts for compute and energy rather than owning most of the physical infrastructure itself, effectively “renting” capacity at scale while its partners borrow to build the underlying assets. The company has committed to about 1.4 trillion dollars of spending on compute and power over roughly eight years, even though it expects around 20 billion dollars in revenue this year and, in one bank scenario, something over 200 billion dollars of annual revenue only by 2030—leaving a very large funding gap that must be bridged by external capital.
To keep its own balance sheet relatively light, OpenAI relies on supplier financing: the partners raise debt against long‑term contracts and leases, while OpenAI maintains an undrawn credit facility of about 4 billion dollars and avoids taking on comparable levels of borrowing itself. This approach shifts much of the financial risk to the companies building and owning the data centers and GPU farms, even though those assets are effectively locked into serving OpenAI’s AI workloads for years.
Key players and their individual exposures
Several names stand out in this ecosystem. Oracle has reportedly raised on the order of 18 billion dollars in bonds to support AI data‑center expansion, and some analysts estimate it could ultimately carry around 100 billion dollars of total debt over the next few years if it fully builds out capacity to honor its OpenAI‑linked contracts. SoftBank, which has positioned itself as a major AI capital provider, has raised roughly 20 billion dollars in 2025 for AI‑related deals, with a meaningful portion connected to OpenAI‑related infrastructure and investments.
CoreWeave, a specialized GPU cloud provider that supplies compute to Microsoft and, indirectly, to OpenAI, has borrowed more than 10 billion dollars and has tens of billions in future lease commitments for data‑center space, despite expecting only a single‑digit‑billion revenue figure this year. Blue Owl has structured large special‑purpose vehicles (SPVs) that borrowed around 18 billion dollars from mostly Japanese banks to fund OpenAI‑linked sites—projects then leased to Oracle—which concentrates project risk inside these vehicles while still tying their economics to OpenAI’s success.
How this fits into the broader AI credit boom
The debt surge around OpenAI is part of a wider wave of borrowing by so‑called “hyperscalers” and AI‑focused infrastructure players. The big cloud platforms—Amazon , Google , Meta , Microsoft , and Oracle—have together taken on roughly 121 billion dollars of new debt in 2025 to fund AI and cloud projects, more than four times their typical annual issuance over the last five years. Bank research notes show that a large share of the increase in investment‑grade corporate bond supply this year can be attributed to AI‑driven capital expenditure and related mergers and acquisitions.1
This flood of issuance is moving credit markets: investment‑grade supply has run well above seasonal norms, spreads on some AI‑exposed names have widened, and credit‑default‑swap prices for borrowers like Oracle and CoreWeave have risen, signalling that investors see higher default risk as leverage climbs. At the same time, private credit funds and non‑bank lenders have become central in structuring the more complex, bespoke financings behind many of the largest GPU and data‑center projects, blurring the line between traditional corporate debt and project finance.
The economic logic and strategic bet
The core bet behind this borrowing spree is that demand for AI services—chatbots, copilots, enterprise agents, and AI‑enhanced cloud services—will grow fast enough, and stay profitable enough, to justify multi‑decade investments in power‑hungry, capital‑intensive infrastructure. For operators, the logic is that once a site is built and filled with GPUs, it can generate high‑margin recurring revenue over years if utilization remains strong and customers like OpenAI, Microsoft, and other AI firms keep committing to contracts.
However, the timing is tight. Many of these projects must be financed upfront with debt, while the associated AI revenue may ramp more slowly and can be sensitive to pricing pressure, competition, and technological shifts, such as more efficient models or new chip architectures that could reduce compute needs. If demand under‑delivers or if interest rates stay elevated, operators with thin equity cushions could find themselves squeezed between high debt service costs and lower‑than‑expected cash flow.
Main risks for partners, OpenAI, and the system
For OpenAI’s partners, the most immediate risk is concentration: a large fraction of their new capacity is economically tied to one customer’s AI roadmap, even if contracts are legally structured with intermediaries like Microsoft or SPVs. Should OpenAI’s growth slow, its product mix change, or customers migrate to rival models, the infrastructure owners might be left with overbuilt or mis‑configured assets, especially in locations optimized specifically around OpenAI‑driven workloads.
For OpenAI, the model reduces on‑balance‑sheet leverage but does not eliminate risk. Long‑term minimum‑spend and take‑or‑pay commitments mean that, if its own revenue projections fall short, it could end up locked into costly compute and energy contracts that are hard to renegotiate, forcing further fundraising or strategic shifts. Systemically, the concern is that a cluster of highly leveraged AI‑infrastructure projects financed by overlapping banks and private credit funds could amplify stress if a downturn or AI‑specific shock hits, particularly in high‑yield and private credit markets that are already absorbing a lot of this risk.
What this means for the future AI build‑out
In the near term, this financing model likely accelerates AI infrastructure deployment, because it lets OpenAI and similar firms scale faster than their own balance sheets would otherwise permit. It also sets a template other AI companies can follow: lock in long‑term compute and power contracts, shift capex and debt to specialized partners, and use those partners’ balance sheets and access to bond and loan markets to fund expansion.
Over the medium to long term, though, the sustainability of this approach will depend on whether AI revenues broadly—across consumer, enterprise, and platform use cases—actually validate the trillions of dollars in planned spending commitments. If the growth story plays out, today’s 96–100 billion dollar debt pile around OpenAI could look like the early foundation of a durable AI infrastructure layer; if not, it may instead be remembered as one of the most aggressive leveraged bets of the current tech cycle.
