⚠ Contracts Are Only as Good as the CounterpartiesHigh threat
CoreWeave (CRWV) — threat to the moat
Take-or-pay from a loss-making lab is a credit exposure wearing a contract's clothes.
The take-or-pay contracts underpinning CoreWeave's backlog and financing are only as valuable as the customers' ability and willingness to pay, and CoreWeave's concentration in a few enormous, in places financially fragile, counterparties makes this a real risk rather than a formality. The most striking example is OpenAI, which has contracted multi-billion-dollar commitments to CoreWeave but is itself a deeply loss-making company whose ability to honor tens of billions in obligations depends on continually raising vast sums of new capital. A take-or-pay contract is a promise to pay, and a promise from a counterparty that is burning cash and dependent on external funding carries credit risk that the headline backlog figure does not reveal — if such a customer's funding falters, the contract's value can evaporate regardless of its legal terms.
Even the strong counterparties introduce renegotiation risk. The hyperscalers and large labs that dominate the backlog have the leverage and the incentive to press for better terms if GPU prices fall, if their own capacity comes online, or if the competitive landscape shifts, and a few large customers renegotiating can move CoreWeave's economics materially. The concentration means CoreWeave's contracted revenue is not a diversified, reliable stream but a set of large bets on a handful of specific counterparties, each of whose circumstances can change. The take-or-pay structure is genuinely valuable and does shift real risk onto customers, and the largest counterparties are among the most important companies in technology. But an investor should not treat the backlog as risk-free contracted revenue: it is concentrated in a few hands, exposed to the credit fragility of at least one major counterparty and the renegotiation power of the rest, and its true worth depends on customers whose ability and willingness to pay across a multi-year term is far less certain than a $104 billion headline suggests — the biggest counterparties, OpenAI at ~$22.4B among them, are themselves unprofitable ventures1.
- ReportedOpenAI at ~$22.4B is itself an unprofitable venture.CoreWeave customer-commitment announcements — OpenAI up to ~$22.4B; Meta ~$21B — 2025-2026 · publ. 2025-2026 · source ↗