OpenAI’s partners are carrying $96 billion in debt, highlighting growing risks around the loss-making AI company

Suppliers of data center capacity, chips and compute for OpenAI have taken on roughly $96 billion of debt (breakdown: $30bn already borrowed by SoftBank/Oracle/CoreWeave; $28bn loans via Blue Owl/Crusoe; $38bn in ongoing talks) while CoreWeave alone reports $3.7bn current debt, $10.3bn non-current debt and $39.1bn of future data‑center leases, expecting only $5bn of revenue this year with a $56bn revenue backlog. OpenAI has committed $1.4 trillion of future energy/compute spend but forecasts ~$20bn revenue this year and faces large funding shortfalls per HSBC; meanwhile the five hyperscalers added ~$121bn of new debt this year (BofA), driving heavy IG supply and notable CDS spread widening (Oracle ~+60bps to 104bps; CoreWeave ~+280bps to ~640bps), signaling rising credit risk and increased market hedging demand. Investors should treat this as a material credit-market development that amplifies refinancing and default risk in AI‑related capital structures and could pressure corporate bond spreads and CDS liquidity.

Analysis

Market structure: The debt burden (~$96bn among OpenAI partners; hyperscalers ~$121bn new debt this year) shifts marginal funding from cash-rich Big Tech to credit markets, boosting short-term funding demand and pricing power for banks/lenders while exposing smaller GPU/data‑center providers (CRWV, CoreWeave) to liquidity and covenant risk. Hyperscalers (MSFT, GOOGL, AMZN, META, ORCL) gain preferential access to capacity and can compress vendor margins; independent builders face a two‑tier market and higher funding costs (CoreWeave 5y CDS ~640bps).

Risk assessment: Tail risks include a funding shortfall at OpenAI (HSBC’s $207bn gap to 2030) producing cascade defaults or lease-strain (CoreWeave $39.1bn leases), regulated capital controls, or a sharp rise in rates that blows out interest expenses. Near term (days–weeks) expect volatility as IG supply impacts spreads; medium term (3–12 months) covenant tests and refinancing windows matter; long term (1–5 years) demand vs. oversupply in GPU capacity will determine EBITDA realization.

Trade implications: Tactical trades: buy single‑name CDS protection on CRWV and ORCL or long deep OTM put spreads (6–12m) sized 1–3% book to guard tail risk; rotate fixed‑income exposure from long IG paper into floating‑rate loans and short 7–10y IG duration to protect against spread widening. Equities: overweight MSFT/GOOGL (2–3% each) financed with a small hedge (buy ORCL/CRWV puts) and go long utilities/power producers and copper for 6–18 months to capture rising energy/capex input costs.

Contrarian angles: The market underestimates two outcomes—(1) hyperscalers will reabsorb much AI demand, leaving independents overlevered and prime M&A targets; (2) if AI monetization accelerates, credit spreads could compress sharply in 12–24 months. Positioning should therefore be asymmetric: buy cheap, short-tail protection now (CDS/puts) and selectively scale into long equity exposure if spreads normalize below pre‑Sept levels (Oracle 5y CDS <80bps).

Key Decisions for Investors

• Establish a 2–3% notional long in MSFT and GOOGL equity (equal weight) over 3–12 months to capture secular AI upside; fund by reducing long-duration IG bond exposure by an equivalent amount and buy ORCL 6–12 month 5–10% OTM put spreads as a hedge.
• Buy credit protection on CRWV (single‑name CDS or equivalent) sized 1–2% notional if 5y CDS >600bps; alternatively buy CRWV 6–9 month put spreads (30–40% OTM) to protect against a >30% equity drawdown. Take profits if CDS drops below 400bps or equity recovers >50%.
• Reduce exposure to long-dated corporate IG bonds by rotating 30–50% of duration into floating-rate loan vehicles (e.g., BKLN or similar) within 2–6 weeks to hedge rising spread/ rate risk from tech capex issuance.
• Go long US utilities/power producers (2% position) and copper exposure (via physical-equivalent ETF) for 6–18 months to capture higher electricity and copper demand driven by data center buildouts; size up if energy contracts for hyperscalers show >10% price pass‑through.
• Set automated alerts: purchase ORCL protection if 5y CDS >125bps or CRWV CDS >750bps; sell protection / reduce shorts if ORCL CDS <80bps or CRWV CDS <400bps, review positions at each quarterly earnings or major OpenAI funding announcement (within 30–90 days).