CoreWeave: contracts, leverage and the limits of protection

01Committed revenue

Committed contracts generated 98% of Q2 revenue.

CoreWeave's Q2 2026 filing describes committed contracts as take-or-pay. They generated almost all reported revenue. This is evidence of contracted demand, not a measure of protection across future construction or outstanding debt.[1]

A strong customer can improve the quality of a receivable. It does not eliminate delivery, acceptance, service-performance or contract-exit conditions. Nor does buying compute make that customer a guarantor of the equipment loan.

Credit-market stress is not a verified payment default.

On October 15, Bloomberg reported five-year default-insurance spreads reaching 1,194.5 basis points, with a model-implied default probability above 60%. Those are dated market signals, not observed default rates or an objective forecast.[16] This review did not establish rising actual missed payments. CoreWeave reported DDTL covenant compliance at June 30; that does not certify October compliance.[1]

02Repayment case

DDTL 4.0 has a credible contract-funded repayment case.

Meta capacity payments underpin a financing designed to repay equipment debt during the contracted arrangement. DBRS's April 3 assessment expected full amortization without refinancing or re-leasing. It assigned A (low) to the borrower and facility, using internal documents as well as public evidence.[2]

That supports a serious economic case: equipment debt need not depend on speculative resale if contracted operating cash repays it. It does not prove the parent's solvency, equity value or every project's economics.

DDTL 4.0: from Meta service payments to conditional parent cash

Meta compute paymentsService contract; performance and acceptance conditions remain.Filed Meta framework [5]
CCAC VIII / controlled accountsAssigned contract cash supports the borrower’s equipment financing.Credit agreement [3]
Costs, reserves and debt serviceCash controls and repayment requirements constrain available distributions.Cash controls [3]
Conditional parent distributionsOnly cash permitted by the financing can reach the parent.Distribution restrictions [3]

Equipment and lender security. Financed assets and assigned receipts support the secured borrowing. Exact installment matching is unresolved because amortization schedules are omitted. [3]

Separate parent support. CoreWeave guarantees specified carveout obligations; separate lease support concerns specified Polaris premises. Meta’s service payments are not an evidenced guarantee of this debt. [4] [6]

Schematic relationships, not an exhaustive legal payment waterfall. April 3, 2026 rating assessment and filed terms retain their original dates. Exact Dalton/Chester asset assignments remain unresolved; the diagram does not assign the entire pool to a campus. [2] [18] [23]

Read the relationships and their evidence limits
Relationship	Supported by reviewed evidence	Limit
Customer → borrower	Meta-backed capacity payments support CCAC VIII’s repayment case. [2] [3]	Meta is a service customer, not an evidenced debt guarantor. Specific orders govern performance obligations. [5]
Borrower → lenders	Equipment financing, controlled cash and repayment restrictions. [3]	Omitted amortization schedules prevent a full receipts-to-installments reconstruction.
Borrower → parent	Distributions are subject to financing restrictions. [3]	Scheduled lender coverage does not establish cash available to the parent.
Parent → specified obligations	Carveout debt support and separately scoped premises lease support. [4] [6]	Neither support arrangement establishes an unconditional guarantee of the whole campus or all group debt.
Financing → physical sites	The review identifies related facilities and project records. [18] [23]	Exact Dalton/Chester scope is not established.

What the structure protects

Take-or-pay capacity payments reduce volume risk. Draw conditions, controlled accounts and distribution restrictions support repayment. DBRS nevertheless identified completion, acceptance, uptime, reliability and power-cost risks.[2][3]

What it does not prove

The 1.15× maintenance test excludes specified payments, including a maturity bullet. The published Meta framework allows conditional exits and gives orders precedence. Omitted schedules prevent an independent installment-by-installment cash match.[3][5]

Keep the guarantees separate. Meta is the customer, not an evidenced debt guarantor. CoreWeave's DDTL 4.0 parent debt guarantee covers specified carveouts. Separate parent lease support applies to specified Polaris premises. Neither should be generalized to every obligation or the whole campus.[4][6]

03Remaining exposure

Other structures retain customer, renewal and parent exposure.

There is no single “box” model. DDTL means delayed draw term loan. A financing can span sites; several tranches can share a borrower. DDTL 2.0 and 2.1 share CCAC IV and allow reciprocal excess-cash support. They are not fully isolated legal boxes.[7]

USD billions of outstanding principal, June 30, 2026 unless stated otherwise. Dated balances, not commitments or current October debt.[1]
Financing	Principal	Known customer backing	Decisive qualification
DDTL 1.0	$1.300B	Assigned contracts; names redacted.	Permission for uncontracted activity does not establish actual usage.[15]
DDTL 2.0	$3.190B	Microsoft orders 4/5 identified.	Customer-grade principal split unresolved.[7]
DDTL 2.1	$3.000B	Microsoft order 7; OpenAI order 2.	Shares 2.0's borrower; advance rates do not reveal customer shares.[7]
DDTL 3.0	$2.215B	OpenAI/Phoenix assignment identified.	Parent guarantee covers CCAC VII, not CCAC V. Do not infer an unguaranteed dollar slice.[8][9]
DDTL 4.0	$2.837B	Meta take-or-pay capacity.	Non-recourse classification, with parent carveouts and separate lease support.[4][6]
DDTL 5.0	$1.101B	Two disclosed non-investment-grade customers.	Names and customer-specific repayment matching unresolved.[11][12]
June subset	$13.643B	Six rows; not total company debt.
DDTL 5.5	$1.200BAfter August closing; exact day unspecified	Average contract term shorter than debt.	Renewal/resizing exposure. Excluded from June subtotal.[13][14]

Read the denominator correctly: the issuer classifies $10.806B, or 79.2% of the June subset, as recourse. Exact guarantee scopes vary. DDTL 5.0's disclosed non-investment-grade customer backing represents 8.1% of this subset; the remaining 91.9% is not established as investment-grade-backed. Excludes bonds, revolvers, OEM financing, leases and later DDTL 5.5. Recourse and customer grade overlap.[1][8]

DDTL 5.5 explicitly underwrites renewal risk.

Its debt lasts about five years; customer contracts average about three. The $2.6B commitment is capacity, not the later $1.2B drawing.[13][1] Residual cash sweeps toward a $1.3B debt threshold, with distribution limits and renewal or resizing tests. That threshold is not a promised year-three balance, balloon or loss cap. Insufficient sweep cash does not itself constitute default. Missing inputs prevent sizing principal at contract expiry.[14]

Covenant compliance is not proof of current operating coverage.

DDTL 3.0's amended 1.40× testing starts on the first Monthly Date after October 1, 2027; realization tests permit equity cures. Guarantee fallaway is conditional, subject to lender waiver, not automatic. These provisions require more care than reading a headline coverage ratio.[10][9]

Debt can stay covered while residual cash shrinks.

In the illustration below, a 10% fall in cash available after operating costs reduces the residual from 20 to 8 units, a 60% drop. A lender can retain scheduled coverage while the parent loses most of its expected cash.

Illustration only · arbitrary cash units

Change available operating cash

Base cash 120; scheduled debt service fixed at 100. Move the slider or use the arrow keys.

Cash change versus base0%

−30%0%+20%

Cash / scheduled debt service1.20×

Residual before other uses20.0

Scheduled debt service is covered. The residual is 20.0 units before other uses.

Not a CoreWeave forecast or covenant model. Cash starts at 120; scheduled debt service stays at 100. Reserves, taxes, new capital needs and maturity payments are not modeled. Actual covenant definitions differ.

04Evidence limits

Public gaps prevent a portfolio-wide repayment verdict.

AI Buildout's research inventory distinguishes risk mechanisms. It does not establish debt left at each customer's expiry or classify individual sites as distressed. The missing bridge is assigned-order receipts, operating costs, full repayment schedules and cash available to the parent.

Question	Supported today	Still unresolved
Is demand contracted?	Yes by reported revenue; identified loans have contract support.	Strong-customer share of total debt and future construction.
Which pool is better protected?	DDTL 4.0 has an affirmative dated rating assessment; 5.5 has explicit renewal exposure.	Comparable actual cash coverage and principal at expiry.
Which sites lack offtake?	No reliable list established.	Missing site/customer links do not prove absent contracts.
Is capital fully mapped?	19 equipment-linked project records; $25.59B of allocations.	Complete site census and a common capital denominator.[23]

More project records do not establish capital coverage.

The inventory grew from 10 to 19 equipment-linked projects, including future programs. CoreWeave reports 51 operating data centers. The units differ; subtracting them does not count missing facilities.[17][23]

The $25.59B figure comprises lifetime project allocations, including estimates. It has not been reconciled to issuer-reported capital spending on a common period and accounting basis. It is a partial research mapping, not an audited corporate asset base or a measure of missing investment.[23]

Site questions remain specific: Crawley's building, Conapto's second campus, three unnamed Indonesian facilities, and DDTL 4.0's exact Dalton/Chester scope.[22][19][21][18][23] An anonymous Conapto customer provides a candidate, not a confirmed attribution.[20]

Separate unfinished research from withheld information.

Further lender holdings, later filings and inaccessible Crawley accounts may close research gaps. Omitted amortization schedules, redacted orders and private lender models need disclosure or direct confirmation. Public opacity does not prove weak underwriting: lenders may have information outsiders lack. Their access does not guarantee success.

The next evidence should test the parent cash bridge.

Collections and paydown: match accepted capacity and customer receipts to all principal and interest payments.
Accessible cash: separate reserves, sweeps and trapped cash from distributions the parent can use.
Funding dependence: identify principal relying on renewals and corporate needs requiring fresh debt or equity.

Timely collections, contract-matched amortization and usable distributions would strengthen the case. Repeated equity cures, service credits, resizing needs or inaccessible cash would weaken it. These are tests to investigate, not claims that adverse events have occurred.

ReferenceSources & method

Evidence and qualifications.

Sources retain their original dates. Issuer statements, executed clauses, rating opinions and analytical judgments are distinguished in the prose.

01CoreWeave, Q2 2026 Form 10-Q
Notes 2 and 10; committed revenue, June debt rows, compliance and subsequent events. June balances are not October balances.
02Morningstar DBRS, CCAC VIII assessment
April 3, 2026. Meta support, expected amortization and installation/operating risks. Historical rating opinion, not a current rating check.
03DDTL 4.0 executed credit agreement
Definitions; §§2.08, 2.20 and 6.12. Repayment, cash waterfall and coverage tests; amortization schedule omitted.
04DDTL 4.0 parent guarantee
Specified carveout obligations. Distinguishes debt support from ordinary repayment guarantees.
05Filed Meta master services agreement
§§11–12 and 15(c). Conditional exits and order precedence; not the complete assigned-order package.
06Applied Digital, Polaris lease disclosure
Specified ELN-02/03 scope and separate parent lease support. Does not establish whole-campus guarantees.
07DDTL 2.0 / 2.1 amended agreement
September 29, 2025; definitions, §§2.08, 2.24 and 6.12. Shared borrower, identified orders, cash support and pricing grid.
08DDTL 3.0 original Form 8-K
July 2025. CCAC VII guarantee expressly distinguished from CCAC V. Later amendments must also be considered.
09DDTL 3.0 original credit agreement
Borrowers, contract assignment, revenue route and conditional guarantee termination; lender-waiver qualification applies.
10DDTL 3.0 December amendment
Coverage-testing commencement and equity-cure provisions. Current coverage is not established by a deferred test.
11DDTL 5.0 issuer announcement
May 18, 2026. Two large non-investment-grade customers; identifies neither customer by name.
12DDTL 5.0 credit agreement
Security, support, payment tests and omitted amortization detail.
13DDTL 5.5 issuer announcement
August 10, 2026. $2.6B commitment; approximately five-year debt and three-year average customer contracts.
14DDTL 5.5 credit agreement
§§2.08–2.09 and 2.20; sweep threshold, renewal/resizing mechanisms and redacted inputs.
15DDTL 1.0 amended agreement
Contract-linked repayment rules and permissions for uncontracted activity; permissions do not measure usage.
16Bloomberg News, credit-risk repricing
October 15, 2026. Dated CDS level and market-implied probability; not a payment-default report.
17CoreWeave data-center overview
Company website checked October 19, 2026; 51 operating data centers, a different unit from inventory project rows.
18Core Scientific facility supplement
April 21, 2026. Distinct Dalton identities; does not itself prove the DDTL 4.0 assignment.
19CoreWeave / Conapto announcement
June 24, 2026. Two-campus relationship; initial Stockholm 4 South deployment identified.
20Conapto prospectus
May 2026. Anonymous new-customer disclosure suggests a second-campus candidate; not a verified CoreWeave attribution.
21CoreWeave Indonesia announcement
Three facilities announced; individual physical-site identities not established in this review.
22CoreWeave initial UK deployments
January 2025. Crawley and Docklands partners; exact Crawley building unresolved.
23AI Buildout research inventory (unpublished)
As of October 19, 2026. Seven financing identities linked to dated observations; 19 equipment-linked project records; $25.59B of project allocations, including estimates and future programs. Partial research data, not an issuer disclosure or an audited capital reconciliation.

Scope and methodology

This assessment uses public filings, executed agreements, issuer and landlord disclosures, rating analysis and AI Buildout's research inventory available through October 19, 2026. Rating opinions remain attributed to their authors; the cash-flow illustration is analytical, not a company forecast.

The June subset is 1.300 + 3.190 + 3.000 + 2.215 + 2.837 + 1.101 = $13.643B. Recourse rows sum to $10.806B, or 79.2%. DDTL 5.5 is a later observation, excluded from that denominator. “Classified as recourse” does not establish identical parent guarantees for every dollar; DDTL 3.0's borrower distinction matters.

The review is not an exhaustive audit of all filings and lender holdings available at that date. Private lender models were not obtained. It makes no stock-valuation or independently calibrated default-probability estimate.

CoreWeave's contract-backed model can work. Its parent is not insulated from the risks.