OpenAI's growth is running on partner debt. Here's where the risk sits.
OpenAI's expansion is being bankrolled largely by its partners' balance sheets, not its own. Reporting indicates suppliers and data center operators tied to OpenAI now carry close to $100 billion in debt, while OpenAI itself remains mostly debt-free.
SoftBank, Oracle, and CoreWeave have taken on at least $30 billion to either invest in or build for OpenAI. Another $28 billion has gone to groups like Blue Owl Capital and infrastructure firms such as Crusoe-entities whose ability to repay depends on OpenAI-related contracts. Banks are in talks to arrange an additional $38 billion for Oracle and Vantage Data Centers to fund more facilities.
This is happening as OpenAI has committed to an estimated $1.4 trillion of procurement over eight years, compared with roughly $20 billion in expected annualized revenue this year. As one senior executive put it, the company is "using other people's balance sheets."
The structure (and why it matters)
- OpenAI has a $4 billion credit line agreed last year but hasn't drawn on it. Most financing is off its balance sheet.
- Many deals run through special purpose vehicles and include non-recourse features, so lenders-not OpenAI or some investors-eat losses if cash flows fall short.
- Oracle appears the most exposed. Since announcing a large OpenAI-related deal, it has seen a sharp market value drop, and some analysts expect it to borrow up to $100 billion over four years to meet commitments.
- Debt tied to OpenAI-linked buildouts reportedly now matches the combined net debt of several top global corporate borrowers, including Volkswagen and Toyota.
What finance teams should extract from this
This is vendor financing at hyperscale. Capital is flowing from banks and private credit into SPVs and operators, then coming back as revenue from OpenAI and its ecosystem. The circularity creates concentration risk: a handful of AI demand sources are underwriting huge capex decisions.
Non-recourse structures shift default losses toward lenders and project vehicles, but that doesn't erase system risk. If utilization lags or pricing resets, equity cushions and covenant protections will be stress-tested quickly.
How the capital stack is being engineered
Think of this as layered project finance built around AI compute and data center capacity. Sponsors use SPVs to isolate asset risk, ring-fence cash flows, and enable higher leverage than corporate balance sheets might tolerate.
Non-recourse terms place repayment on the project's economics rather than the sponsor's full faith and credit. That can be smart risk allocation-until demand, power costs, or chip supply cycles shift unexpectedly. For a refresher on how non-recourse debt works, see this overview from Investopedia: What is non-recourse debt?
Key numbers to track
- Debt pipeline: Watch how the rumored $38 billion for Oracle and Vantage is syndicated, priced, and tranched.
- Utilization and pre-commits: Capacity take-or-pay coverage, churn, and renewal prices across data center partners.
- Covenant packages: DSCR thresholds, minimum liquidity, and cure rights in SPVs and operating companies.
- Refinancing risk: Maturity walls versus expected AI demand ramps and chip refresh timelines.
- Spread behavior: Credit spreads and CDS for heavily exposed operators and suppliers.
- Counterparty mix: Revenue concentration in OpenAI or a small set of AI clients.
Scenario checks for your models
- Base case: Demand growth fills capacity at modeled prices; projects refinance at tolerable spreads; sponsors earn equity IRRs within underwriting targets.
- Rate shock: 150-250 bps higher-for-longer; DSCR headroom tightens; equity top-ups or partial de-levering required at refi.
- Demand slip: Delayed deployments or price concessions; watch for covenant trips, waivers, or amortization accelerants.
- Tech shift: Architecture or vendor change reduces required footprint; stranded asset risk rises without multi-tenant backfill.
Who bears what risk
- OpenAI: Keeps optionality with limited direct borrowing and uses purchase commitments to mobilize external capital. Cash burn is pointed at compute and services often provided by the same partners financing the build.
- Operators and suppliers: Carry leverage and execution risk. Returns hinge on on-time delivery, energy pricing, and sustained AI workloads at contracted rates.
- Banks and private credit: Face asset-level risk via SPVs and non-recourse loans. Covenant strength, collateral packages, and step-in rights matter more than ever.
- Large strategic partners (e.g., Oracle): Deepest exposure through both capex and balance sheet leverage. Equity outcomes swing on utilization, pricing power, and refinancing windows.
Practical actions for CFOs, treasurers, and credit teams
- Map exposure: Counterparty, facility, SPV, and maturity wall views across your portfolio.
- Stress test DSCR/ICR: Run rate, demand, and power-cost scenarios; pre-wire waivers and liquidity backstops.
- Tighten covenants: Springing cash sweeps, minimum liquidity, capex pacing gates, and information rights.
- Stage capital: Tranche commitments to deployment milestones; align draw schedules with pre-committed workloads.
- Diversify offtake: Reduce single-buyer risk; secure multi-tenant demand where possible.
- Lock risk where you can: Hedge rates and power; align contract tenors with debt maturities.
Why the circularity isn't automatically bad
Yes, capital flows from partners to OpenAI and back to those same partners through compute spend. That can still work if the flywheel spins: more usage, more revenue, more capacity, better unit economics.
The fragility appears if growth underdelivers or pricing compresses. In that case, the structure will test who actually owns the risk: equity sponsors, SPV lenders, or strategic partners with implied support expectations.
Bottom line
OpenAI is scaling via other people's balance sheets. The leverage is piling up where the assets live, not at the platform commissioning them.
If you underwrite or hold this risk, assume the cycle will wobble at least once before it settles. Paper your covenants, pace your capital, and model the exit before you wire the first dollar.
Related for finance teams
- Concept refresher on non-recourse structures: Investopedia: Non-Recourse Debt
- If you're evaluating AI vendors and tools from a finance perspective, this curated list may help: AI tools for finance
Your membership also unlocks:
