CoreWeave (CRWV) credit breakdown: $51.4B in debt and leases, 15% yields, and the circular Nvidia backstop

I spent some time looking into Barclays credit research and UBS initiation files on CoreWeave (CRWV). Beyond the headline capacity expansions, the debt mechanics and hardware depreciation profile look structurally fragile.

A few balance sheet data points worth tracking:

1. $51.4B in Total Commitments

The liability structure consists of $35.1B in funded debt obligations plus $16.3B in operating leases. Funded debt is heavily skewed toward delayed draw term loans ($21.5B in DDTLs) alongside roughly $13.6B in non-recourse project financing and senior notes. Cash on hand sits at roughly $6.4B against an estimated $260B in infrastructure CapEx required through 2030 (roughly 90% debt-funded).

2. Borrowing Against 3-Year Depreciating Collateral

The core vulnerability is the collateral underpinning the DDTLs. These loans are directly secured by GPU clusters. Unlike traditional real estate or telecom infrastructure that amortizes over 15 to 30 years, GPUs face a hard 3-year economic depreciation cliff. As newer silicon compresses secondary market rental yields, collateral resale values decay faster than debt amortizes, creating severe LTV covenant pressure.

3. Credit Markets Are Pricing High-Yield Distress

Barclays credit research data shows debt pricing that public equity markets are largely ignoring:

Equipment loans / DDTLs: SOFR + 450 to 550 bps (roughly 9.8% to 10.8% all-in cost)

Senior secured notes: 11.25% fixed coupon (trading around 850 bps spreads)

Subordinated HoldCo tranches: yielding between 13.50% and 15.00%

Servicing 15% cost of capital on physical hardware with multi-billion-dollar annual rollover needs leaves very little margin for error.

4. The Closed-Loop Nvidia Financing Structure

CoreWeave does not operate at arm's length. Nvidia sits across the entire capital stack: minority equity holder, OEM financier allocating hardware to secure borrowing bases, anchor customer with a contractual $6.3B capacity purchase commitment through 2030, and residual capacity backstop for unsold clusters. Nvidia books immediate hardware revenue and gross margin upfront, while long-term debt and silicon obsolescence sit in an off-balance-sheet vehicle.

5. Counterparty Concentration

For 2Q26, recognized revenue is heavily clustered:

Microsoft: 36%

OpenAI: 26%

Meta: 10%

Top 3 make up 72% of total revenue. If hyperscalers pull back lease renewals or shift workloads to proprietary ASICs, the cash flow profile will compress long before the $16.3B lease obligations amortize.

Question for allocators: When rapid silicon obsolescence forces margin compression, who takes the haircut between private credit syndicates and Nvidia?