CoreWeave (CRWV) Q3 10-Q Addendum — Who Gets Paid First
CoreWeave's quarterly report, filed the day after our update, answers what the release could not: who holds which claim on the fleet, what falls due before 2028, and what the customer money costs. With three parts of our method rebuilt to fit the company's capital structure, our value moves from $48.57 to $49.55, against $44.91 at the 19 November close.
Our update of 11 November, The Floor Moved to the Fleet, valued CoreWeave at $48.57 on the earnings release; the Form 10-Q reached EDGAR on 12 November. The filing's four new inputs are small, and rebuilding three parts of our method moves the value more. What the filing changes is the picture of who is paid first: the lenders secured on each fleet, the customers who prepaid for capacity, the landlords of sites not yet built and, last, the shareholders. Nothing due in the next 15 months is a bond; by our estimate about half of it is vendor credit rather than contract-funded loans. $1.9 billion of customer prepayments sat on contracts that became cancellable in a single quarter, without explanation. And the parent's guarantees, its revolver and the older loans' own default clauses are the routes by which one subsidiary's default, or the parent's, could become the group's. We trace that path quarter by quarter and list the events that would show it starting.
What this addendum answers
- What did the 10-Q change in the value, and what did we change in the method?
- Of the ~$9 billion of prepayments and deposits due over the next three months, how much is fixed, how much waits on delivery or power, and what does that money cost?
- How do prepayments sit beside the delayed-draw loans, and who is paid first if a customer defaults?
- Why did $1.9 billion move from deferred revenue to customer liabilities, and is it refundable in cash?
- Why did recourse debt rise $10.2 billion in a quarter under an asset-level model, and how much of the $13.4 billion due in the next 15 months is amortization paid from customer cash?
- How much of the pending leases and commitments reaches the balance sheet within 18 months, and should prepayments on undelivered capacity count as debt?
- How do the GPUs' book life and the loans' lives compare, what renewal realization does the price need, and what does the capital cost?
Terms used throughout, defined once. Recourse debt: debt CoreWeave, Inc. owes directly or has guaranteed; non-recourse debt is owed by a subsidiary whose lenders can reach only that subsidiary's assets. DDTL (delayed-draw term loan): a loan drawn as GPUs are bought, held in a special-purpose subsidiary and secured on the GPUs and the customer contracts they serve. Vendor (OEM) financing: credit from equipment makers on the gear they sell. Deferred revenue: cash a customer paid before the service, owed back as service. Customer liabilities: prepayments on contracts that became cancellable, held as a current liability. RPO (remaining performance obligations): contracted revenue not yet recognized. Ladder: principal due by year. Treasury method: options and warrants counted as the net shares they would add after their exercise price buys some back at the market price. Realization: the share of the company's own renewal-revenue figure that we credit. Event of default: a breach named in a loan agreement that lets its lenders act. Acceleration: lenders demanding immediate repayment of the whole loan after an event of default. Cross-default and cross-acceleration: clauses that make another debt's default, or its acceleration, count as a default under this one. Bankruptcy-remote subsidiary: a company set up so that its assets serve its own lenders and stay apart from its affiliates' creditors. Restricted and significant subsidiaries: the subsidiaries a given document counts; the revolver counts every one, while the bond indentures leave out the financing subsidiaries. Cash trap: a lender's right to hold back a subsidiary's spare cash instead of releasing it to the parent; a sweep uses that cash to repay the loan. Letters of credit: payment guarantees the revolver's banks issue for the company; they use up revolver capacity without being drawn cash. Covenant: a financial test the borrower must pass, here each quarter. Indenture: the contract that governs a bond or convertible issue. SPV (special-purpose vehicle): the financing subsidiary that borrows a delayed-draw loan. Step-up: the point at which a loan has paid down to 30% of its balance at the end of drawing, after which DDTL 1.0's and 2.0's thresholds on the parent's debt rise.
The cases on either side of the price, in their own terms
The case for caution, as the filing extends it. Recourse debt rose $10.2 billion in one quarter to $41.6 billion, and the parent guarantees up to $13.1 billion of the delayed-draw loans called asset-level, so the claim ahead of the equity is larger than the label suggests. The fastest-growing recourse layer is vendor financing: the recourse part rose $2.7 billion in the quarter, vendor financing totals $7.7 billion, and it is short-dated (one to three years in the 2025 annual report; maturities from October 2026 to July 2030 in the Q3 table). About a fifth of the opening prepayment balance sat on contracts that became cancellable in a quarter. Leases signed but not begun rose to $47.4 billion from $35.5 billion, before a 453 MW site capped at $16.9 billion that sits outside that figure. The risk factors add covenants to the debt heading, drop June's sentence "we expect our cost of capital to continue declining", and add a factor for the October "negative publicity and market rumors." The chief executive set up a plan in August to sell up to about $500 million of stock by 31 March, subject to price thresholds. The parent's 9.75% 2031 notes yielded about 15.5% on 10 November (as reported, in our update), above the line at which we would call its market shut. And the value still rests on GPUs earning past their contracts, an economic life the filing cannot show.
The company's case, as the filing extends it. The contracted book is $132.3 billion of RPO, with Meta (up to about $21.0 billion) and OpenAI (up to about $6.5 billion) named as significant customers to come. No bond or convertible matures before June 2030; everything due in the next 15 months is scheduled amortization, and by our estimate about half of it is paid from customer contracts pledged to the loans, as $1.06 billion of the three oldest facilities was in Q3. The secured subsidiaries hold $25.9 billion of assets against their loans ($22.2 billion long-lived, $3.7 billion current), up from $14.5 billion at the end of 2025. Prepayments cost about 3% a year as booked, below any debt the company can issue. GPUs are depreciated over six years while the newest guaranteed loans are repaid within about five, so the fleet comes off contract owned outright, and more than $3 billion of the prepayments due in the next three months relate to re-contracting existing infrastructure. And prices are rising: by our arithmetic the 2026 vintage earns about $13.8 million a megawatt-year against $7.8 million on the legacy fleet, new contracts are quoted at $20–25 million, and an investment-grade facility is in process.
This addendum tests the first case's claims on recourse, vendor credit, the prepayment reclassification and the leases, and the second case's on maturities, amortization, pledged assets and the cost of prepayments, each against the filing. It cannot test whether the $1.9 billion is refunded or served, which the year-end report will show, or how much of the ~$9 billion arrives on schedule. Nor can it test renewal realization or GPU economic life; both are carried from the update, at 60% of the company's renewal figure and 6.1 years, and §6 shows what the price needs of each.
What moved the value: the filing, then the method
Four inputs were pending the 10-Q. Each is applied on its own, in this order, holding everything else in the published model; then the price anchor moves to 19 November; then three changes to the method, each argued below.
| Change | $ a share | Fair value after |
|---|---|---|
| Shares: 575M placeholder -> 582.0M (30 Oct cover 556.1M + options and warrants by treasury method) | -0.55 | 48.02 |
| Debt at principal owed ($47.5bn ladder) instead of carrying value ($46.9bn) | -0.29 | 47.73 |
| Asset-level maturities on the 10-Q ladder instead of the Q2 ladder spread | +0.02 | 47.75 |
| $1.9bn customer liabilities: cash already received, owed back as service or refund | -0.83 | 46.92 |
| Price anchor 10 Nov $48.73 -> 19 Nov $44.91 (equity issue price in the base and bull) | -0.04 | 46.88 |
| Method: each case's cost of equity follows its own leverage year by year (was 16% for every case to 2030) | -1.58 | 45.30 |
| Method: convertibles settled case by case, net of capped calls (bull share values at the maturities sit between conversion prices and caps) | +0.00 | 45.30 |
| Consistency: terminal timed like the explicit flows (mid-year) and struck at end-2036 net debt | +4.25 | 49.55 |
| Fair value 48.57 → 49.55 | +0.98 | 49.55 |
How to read it. Each row applies one more change to the running value, starting from the $48.57 published on 11 November. The middle column is the row's effect in dollars a share, the right column the value after it; the fifth row moves the price anchor from the 10 November close to the 19 November close, and the last three rows are the method changes argued below the table. A different order would change the split, not the total.
- Shares. The update used a placeholder of 575 million. The cover gives 556.1 million Class A and B shares at 30 October; options (23 million at $1.85) and warrants (4 million at $1.5495) add 25.9 million by the treasury method: 582.0 million. The 30 million unvested restricted stock units stay out, because we expense stock pay as a cost and would otherwise count it twice. The convertibles, which convert at $97.85–$119.60, about twice the price, are settled case by case below.
- Principal, not carrying value. Debt was taken at the balance sheet's $46.9 billion ($42.75 billion net of cash), after $632 million of unamortized issue costs. The ladder gives the principal actually owed, $47.5 billion ($43.4 billion net of cash).
- The ladder. The model had spread Q3's new asset-level debt evenly over 2027–31; the filing's own schedule front-loads it. The effect is small because the refinancing rate is close to the rate on the debt it replaces.
- Customer liabilities. The release showed $2,015 million of "other current liabilities"; the filing shows $1,915 million of it is prepayments on contracts that became cancellable. The model counted that cash but not the obligation. It now enters the opening deferred-revenue balance; served (revenue with no new cash) or refunded (cash out), the equity loses about the same: −$0.83 as served, $49.51 in value if refunded in the fourth quarter.
Exhibit 1 — From $48.57 to $49.55, one change at a time
How to read it. Fair value in dollars a share. The black bar on the left is the published value of 11 November; each bar after it applies one more change to the running total, in the order of the table above, grey where it subtracts and gold where it adds, ending at this addendum's value in black. The first five changes are the filing's facts and the price anchor; the last three are the method. Read the bar lengths: the terminal-timing correction is the largest single change, and the filing's facts together take off less than it adds.
The method, rebuilt where it did not fit. Tested against CoreWeave's own capital structure, three parts of the update's method did not fit.
- Each case's cost of equity follows its own leverage. CoreWeave's equity risk is mostly leverage: net debt is 4.8 times annualized EBITDA, and the cases diverge sharply from here. By 2030 the bull is at 1.5 times and falling, the base borrows for the build and ends 2036 with debt 1.7 times its equity, and the bear stays above 3 times. The update held 16% for every case to 2030. We now relever one asset beta, 1.25 (1.10 gives $55.33, 1.40 $44.17), at each case's own net debt over its own equity value each year: 16.6% in the first year at today's market leverage, then about 13–14% in the bull, 18–20% in the base and the 30% ceiling in the bear. The rate is the equity's expected return for its market risk, not the notes' promised yield, so the distress case is not a second charge for the same default risk. Effect: −$1.58.
- The terminal value is timed like everything else. The model discounts each year's cash flow from mid-year, but the update valued the 2036 enterprise as if its first cash flow, the 2037 fleet's, came a full year later, and set the equity against net debt grown by a further year. At the end of 2036 the equity is the enterprise (its cash flows from mid-2037) less the debt owed then. The equity is levered at that point, so both errors fall on it several times over: +$4.25, the largest line in the bridge, and a correction of an inconsistency rather than a change of judgment.
- The convertibles convert where they would. The update kept them as debt in every case. Settled case by case, the bull's own share value in the maturity years, $165.03–$188.10, is above the conversion prices ($97.85–$119.60) and below the capped calls' caps ($199.70–$230.00), so the capped calls absorb the conversion and the equity gives up nothing beyond repaying principal; the 2033 notes come closest, $188.10 against a $199.70 cap. Effect: none.
What we kept, and why.
- Weights: distress 20%, bear 25%, base 35%, bull 20%. Credit-default swaps imply about a 60% chance of default over five years at risk-neutral odds; we judge real-world odds at a third to a half of that, 20–30% (our judgment, not a measured ratio). Distress sits at the bottom of that range because no bond falls due before June 2030 and the 10-Q shows $9.8 billion of liquidity (cash, marketable securities and $5.7 billion undrawn under its facilities) with ~$9 billion of prepayments due; against that, the parent guarantees up to $13.1 billion of the asset-level loans and about a fifth of the opening prepayments became cancellable. §7 tests the 20% against the parent's own runway. The bull's 20% rests on signed price increases, renewals above original prices and an investment-grade facility in process, none of it yet in reported revenue; it carries 59% of the value on a fifth of the weight and credits 100% of the company's renewal figure, and at a bull weight of 15% the value meets the price.
- Base pricing: $18 million a megawatt-year on new contracts, $16 million on backlog capacity. The step from the legacy fleet's $7.8 million to the 2026 vintage's $13.8 million came with one GPU generation and part of this year's price rises; the capacity arriving in 2027–28 is the next generation, and new contracts are quoted at $20–25 million. So $13.8 million is the floor of the evidence and the quotes its ceiling; at the quotes' low end, $20 million on new contracts and $18 million on backlog, the value would be $57.37.
- Option value: 20% within-case volatility on the 2036 equity, a mature contracted lessor's level. The company's enterprise is near 40% today, but the cases already carry the spread between outcomes (no option value gives $45.75; today's level throughout, $55.60).
- The tails. The bear's $4.15 is all terminal value: its enterprise ends 2036 below its debt, so its equity is only the option on a recovery, worth nothing with no volatility, $3.42 at 2% terminal growth and $5.16 at 4%, with its rate at the 30% ceiling in every year after the first. The distress case restructures in 2027, so its equity is worth zero from then and has no terminal value.
- Growth beyond the contracted 4.6 GW continues while a new site covers its financing, which is how management says it underwrites; ending it in any year moves the base only between $51.61 and $54.89.
The filing confirms the RPO schedule the update took from the deck: $132.3 billion, 36% within 24 months, 38% in months 25–48 and the rest by month 86. The contracted-book floor, which values only signed contracts after the debt and the 2027 build, moves with the same inputs: −$33.32 a share on the signed book alone (from −$29.30), $3.21 with the paid-off fleet and one renewal on new GPUs (from $7.68), $16.52 at the company's full renewal figure, and −$55.71 with half of the pending lease payments stranded, now on $47.4 billion rather than $35.5 billion.
The prepayments: what is fixed, what waits, what it costs
The ~$9 billion. On the call of 10 November the company said that "in the next three months alone, we will collect approximately $9 billion in prepayments and deposits from signed customer contracts", of which "more than $3 billion" relates to "recontracting existing cloud infrastructure". The 10-Q does not split the figure by payment terms, so how much is fixed in amount and date cannot be read from it; what it fixes is the boundary. The $3 billion-plus for installed GPUs does not wait on construction or power. The call does not say what the other ~$6 billion is for; if it is for capacity still to be delivered, it waits: committed contracts "typically include a prepayment from our customers prior to them receiving access to our services", RPO is reduced for "amounts that may not be recognized as revenue due to delivery delays", and Q3 showed that prepaid contracts can become cancellable. So at most about $6 billion of the $9 billion depends on delivery or power. Prepayments averaged 15–25% of contract value across active contracts at the end of 2023, 2024 and 2025; this year the company says it collected them on 70% of deals, some covering up to 100% of contract value.
What it costs. The company books interest on prepayments it judges to contain a financing component: $204 million of non-cash interest in the nine months, about 3.0% a year on the average deferred-revenue balance of $9.1 billion. That is the accounting cost, set when each contract is signed. The economic cost is the price the company gives up for cash up front, which it says it weighs "against the cost of financing" and does not disclose. In our cost of capital we charge prepayments 9%, the secured rate they replace; at the booked 3% the weighted cost of capital would be about 11.9% rather than 12.7%. That rate is the hurdle for the return on capital, not the discount rate in our equity model, so the choice changes the test, not the value.
The $1.9 billion. The filing says $1.9 billion of "the deferred revenue balance from the beginning of the period was reclassified from deferred revenue to customer liabilities, which are included in other current liabilities, as contracts became cancellable during the quarter." It does not say which period: the amount is 20% of the $9.7 billion at 30 June and 23% of the $8.2 billion at 31 December; "during the quarter" points to June, the base we use. Customer liabilities were $137 million at the end of 2025. The filing gives no cause, names no customer and states no refund terms. The classification itself says two things: the amounts are no longer service owed under a binding contract, and they are current, so they could leave within a year. A delivery milestone that gave customers a right to cancel and a renegotiation lead to different places, and the year-end report will separate them: a balance that returns to deferred revenue or becomes revenue was served; one that is paid out was refunded. Our value takes either path at about the same cost; the events in §8 read it.
Who is paid first
CoreWeave borrows in two places. Six delayed-draw loans sit in bankruptcy-remote subsidiaries, each repaid first from its own fleet and contracts; the parent owes the revolver, the bonds, the convertibles and most of the vendor credit, and guarantees most of the subsidiaries' loans. Who is paid first therefore depends on where a claim sits and on the few links that let a default in one place reach the other. We read every filed credit agreement, guarantee and indenture to map both.
Inside each subsidiary. The delayed-draw loans are borrowed by bankruptcy-remote subsidiaries "formed to hold the financed infrastructure and the related customer contracts", each secured by a first-priority pledge of its equity and substantially all of its assets: the GPUs, the contracts, the receivables and the reserve cash ($22.2 billion of long-lived and $3.7 billion of current assets at 30 September, which "may be used only to settle the obligations of those entities"). If a customer defaults, the lenders to that subsidiary have first claim on its contract cash and its GPUs. Cash runs down a fixed waterfall (operating costs, interest, principal, reserves, a cash trap if one is running) before anything reaches the parent. The six facilities differ in how much they lend, what they test and how they treat a lost customer:
| Facility (30 Sep balance) | Parent's guarantee | Lends, share of capex | Test | A customer terminates | Default inside the subsidiary above | Default from the parent's debt |
|---|---|---|---|---|---|---|
| DDTL 1.0 ($1.1bn) | Full | Redacted | Minimum liquidity only | Event of default; cured by a replacement within 90 days | $5M | Bank lines above $50M called due or unpaid |
| DDTL 2.0 / 2.1 ($5.3bn) | Full | 2.0: 90% for investment-grade customers, 60% otherwise; 2.1: 90% (Microsoft), 75% (OpenAI) | Coverage over the loan's life, 1.40x | Event of default; cured by a replacement within 90 days | $15M | Bank lines above $15M called due or unpaid |
| DDTL 3.0 ($2.3bn) | In part; can fall away | 80% | Debt service, 1.40x from Oct 2027 | Event of default; cured by a replacement within 90 days | $20M | Guarantee trigger: drawn revolver above ~$1.16bn |
| DDTL 4.0 ($4.6bn) | Bad acts only | 90% | Debt service, 1.15x | Event of default, no cure; a late site forces repayment | $100M | None on the loan; the parent's own voluntary bankruptcy filing makes it liable under the limited guarantee |
| DDTL 5.0 ($2.7bn) | Full | 71% | Debt service, 1.35x | Event of default, no cure | $35M | Guarantee trigger: drawn revolver above ~$1.16bn |
| DDTL 5.5 ($1.6bn) | Full | 70%; sweeps all excess cash until the loan is $1.3bn | Debt service, 1.35x | Event of default, no cure | $35M | Guarantee trigger: drawn revolver above ~$1.16bn |
How to read it. One row per delayed-draw loan, oldest first, from its credit agreement and parent guarantee. "Test" is the financial covenant the subsidiary must pass; "default from the parent's debt" is what the parent must do wrong, on its own borrowing, for the loan to default. The DDTL 1.0 and 2.0 thresholds rise (to the greater of $50 million and 10% of EBITDA, and of $150 million and 30%) once each loan has paid down to 30% of its balance at the end of drawing. Read down the columns: the older three are looser on a lost customer, and DDTL 1.0 and 2.0 are tied far more tightly to the parent's own borrowing.
So the older three can cure a lost customer that the newer three cannot, but DDTL 1.0 and 2.0 fall with the parent's bank lines. Lenders have waived before: in December 2024 DDTL 2.0's lenders waived four events of default the company acknowledged, among them an overstated collateral value used to size five draws.
Who holds the contracts, and the prepayments. CoreWeave, Inc. signs most customer contracts. Under DDTL 4.0 it assigned the contract to the borrowing subsidiary; under DDTL 5.0 one contract was assigned and the other signed by the subsidiary; DDTL 5.5's five contracts name CoreWeave, Inc., and whether they were assigned is redacted. None of these three credit agreements mentions customer prepayments, deposits or refunds; the three older facilities (DDTL 1.0, 2.0/2.1 and 3.0) mention upfront payments only to leave them out of the cash flows that size the loans, and none says who holds one. The documents therefore do not settle which entity holds a prepayment or would pay a refund: on an assigned contract a refund would compete with that subsidiary's lenders for its cash, and on a contract still with the parent it would come out of the parent's own cash.
Do prepayments net down the loans? The agreements describe no netting. DDTL 4.0's draws are subject to "debt-sizing limitations tied to the purchase price of eligible assets" and "projected debt service coverage"; DDTL 5.0 and 5.5 are sized on the depreciable cost of the GPU servers, at 71% and 70% of it. A contract partly paid up front leaves less future cash to cover debt service, so on a coverage test it supports a smaller loan; that is our reading of the sizing terms, not a figure the filing gives.
What the parent guarantees. By the company's figure, CoreWeave, Inc. guarantees $13.1 billion of the loans in full; for DDTL 4.0's $4.6 billion it answers only for bad acts: fraud, misappropriating collateral, wilful misconduct that materially damages the project, or a voluntary bankruptcy filing. DDTL 3.0's guarantee is narrower than that figure: it covers only the borrower that does not hold the OpenAI contract, the two borrowers are not jointly liable for each other's loans, how the $2.3 billion splits between them is not filed, and the guarantee falls away when a set of conditions is met (OpenAI rated investment grade by both S&P and Moody's, a manager and servicer in place, coverage tests met, and a date no earlier than 30 April 2027), which its lenders may also waive.
How a default could cross: three bridges. The bond and convertible indentures, all eight of which we read, fence the financing subsidiaries off: they are excluded from the significant subsidiaries whose defaults count. The notes' cross-acceleration clause is word for word the same in all five note issues, and the three convertibles' narrower version (debt of the company or a significant subsidiary, with no guarantee language) is the same in substance in each. Three bridges link the two sides. The first two cross the fence and count a default above the greater of $200 million and 5% of trailing EBITDA (about $291 million on reported figures; more on the indentures' own EBITDA, which adds back the year's rise in deferred revenue and the run-rate profit of new contracts, by an amount the filings do not let us size); the third runs the other way, from the parent back into the older loans.
- Guaranteed loans, straight into the notes. The senior notes cross-accelerate on accelerated debt "the payment of which is Guaranteed by the Company", so an acceleration of DDTL 5.0, 5.5 or any other guaranteed loan (for DDTL 3.0, the guaranteed borrower's share) reaches the notes directly, and the notes' acceleration reaches the convertibles. Guaranteed debt is the parent's debt, so this is no surprise.
- Any subsidiary, through the revolver. Since its November 2025 amendment the revolver treats every subsidiary as a restricted subsidiary, so a default on any subsidiary's debt above the threshold, DDTL 4.0's included, is a revolver event of default; an accelerated revolver is parent debt, and the notes and convertibles cross-accelerate on it. A default on more than the greater of $200 million and 20% of EBITDA of drawn revolver debt (about $1.16 billion on reported figures, letters of credit excluded, against $1.2 billion drawn; more on the covenant's larger EBITDA, so likely above today's drawings) is also a trigger event under the DDTL 3.0, 5.0 and 5.5 guarantees, giving those lenders remedies against the parent.
- From the parent back into the older loans. DDTL 1.0 and 2.0/2.1 are in default if the parent's bank revolving or working-capital debt above $50 million (DDTL 1.0) or $15 million (DDTL 2.0/2.1) is called due or left unpaid at maturity, until each loan's step-up (the table's guide), and the documents do not say whether either has passed it. A voluntary parent bankruptcy filing accelerates both automatically, an involuntary one after 60 days undismissed, and both default on a parent judgment left unpaid or unstayed for 60 days above the same thresholds. So, while the low thresholds apply, an accelerated revolver also puts $6.4 billion of guaranteed loans in default, with no bar like the $1.16 billion the other guarantees require; past DDTL 2.0/2.1's step-up (about $1.7 billion at 30% of reported EBITDA, above the $1.2 billion drawn) it may reach only DDTL 1.0's $1.1 billion.
One bridge the documents cannot settle. Each indenture names the financing subsidiaries it excludes as of its own date. Four issues ($10.3 billion: the 2030 and 9.00% 2031 notes and the 2031 and 2032 convertibles, which borrow the 9.00% notes' list) name none of the DDTL 4.0, 5.0 and 5.5 borrowers; the 9.75% 2031 notes name DDTL 4.0's but not 5.0's; and no indenture names DDTL 5.5's, signed in August 2026. A borrower left off a list is outside the significant subsidiaries only if the company has designated it a "Future SPV", which the indentures let it do, or revoke, at its sole discretion and at any time, with no certificate or notice to the trustee required, or an unrestricted subsidiary, which does need a certificate. No designation is public. If none was made and a borrower also passes the significant-subsidiary size test, a default on its loan above the threshold reaches those issues directly, without the revolver; for the guaranteed DDTL 5.0 and 5.5 the straight notes already count it through the guarantee, so the open question matters most for non-recourse DDTL 4.0 and for the convertibles.
In plain steps. Through the revolver, for any financing subsidiary:
- A subsidiary defaults on more than about $291 million of its own debt, for example because a customer terminates a financed contract.
- Under the revolver's terms that is automatically a revolver default, even for DDTL 4.0, which the parent never guaranteed.
- The revolver's lenders call the $1.2 billion drawn due at once, and the older guaranteed loans fall into default with it: DDTL 1.0 and 2.0/2.1's lenders can call up to $6.4 billion due, which the parent owes under its guarantees.
- That is the parent's own debt called due above the threshold, which is what the bonds' and convertibles' clauses look for; they can then be called due too.
Through the guarantees, for the guaranteed loans, step 2 is not needed: an accelerated guaranteed loan counts under the senior notes directly, and the notes' acceleration reaches the convertibles. The revolver is the only bridge the parent can cut: repaying it removes step 2, DDTL 4.0's path through it and the older loans' trigger on the parent's bank lines, though not a direct path from the newer borrowers into the issues whose lists omit them, if no designation was made. It can do so while it holds the cash; how likely step 1 is, is §7's question.
What would then fall due. Either path ends with the parent's bonds and convertibles, $20.8 billion ($10.0 billion of notes and $10.8 billion of convertibles, none otherwise due before June 2030), and the revolver due at the same moment, with up to $13.1 billion of guaranteed subsidiary loans behind them that the parent would owe if their lenders called on the guarantees. Refinancing all of that at once, in the conditions that caused the default, is in practice a restructuring of the parent's debts together; a parent bankruptcy is itself a trigger event under the DDTL 3.0 (for its guaranteed borrower), 5.0 and 5.5 guarantees and accelerates DDTL 1.0 and 2.0/2.1 automatically. That is the path our distress case takes, with the equity at zero. Two brakes sit inside the chain: the bonds need an actual acceleration, not a default alone, and a note acceleration is annulled if the underlying default is cured within 30 days.
Covenants. The revolver tests net leverage at 6.0 times and contracted revenue at $1 billion each quarter; nothing tests the parent's liquidity. The 10-Q says only that the company was in compliance with all covenants, without the ratio. On reported trailing adjusted EBITDA of $5.8 billion (the last four quarterly releases), net debt is about 7.5 times, so compliance points to the covenant's EBITDA running well above reported: its definition lets the company add, at its option, the projected run-rate profit of contracts signed within the past 24 months (and the debt needed to serve them). A terminated contract loses its add-back, so a customer failure tests the subsidiary and the revolver covenant at once.
Exhibit 2 — Who holds which claim, and in what order
How to read it. Boxes are claimants on CoreWeave at 30 September 2026, in billions of dollars. Across the top are the parent's own creditors, numbered by their claim on the parent; the parent sits in the middle, customers on the left, landlords and vendors on the right, and the three groups of borrowing subsidiaries along the bottom. Grey arrows are cash; gold dashed lines are the parent's guarantees of subsidiary debt; the red dotted lines are the two cross-default paths, both at about $0.3 billion: an accelerated guaranteed loan crosses into the senior notes, and any subsidiary's default becomes the revolver's, whose acceleration crosses into the notes and convertibles. A third path runs the other way and is written in the older loans' box: DDTL 1.0 and 2.0/2.1 default if the parent's bank lines above $15–50 million are called (higher once each loan steps up). The strip at the bottom is the order in which each claim is paid. Read it for the red paths: the subsidiaries are fenced off from the notes except through the parent's guarantees and the revolver, unless the company has not designated the newer borrowers as excluded under the issues whose lists omit them (§3).
The debt: why recourse grew, and what falls due
Recourse debt rose from $31.4 billion at 30 June to $41.6 billion. The filing's debt table shows where it came from:
Exhibit 3 — Where Q3's $10.2 billion of new recourse debt came from
How to read it. Each bar is one source of the change in CoreWeave's recourse debt between 30 June and 30 September 2026, in billions of dollars, from the 10-Q's debt table; the bars add to the total in the title. Gold bars are borrowed or guaranteed by the parent outside any customer-contract subsidiary; grey bars are the guaranteed delayed-draw loans, net of what the three oldest repaid. Compare the gold total with the grey: most of the quarter's recourse debt is not contract-matched.
Of the increase, $5.4 billion is parent-level borrowing (the $4.2 billion of 2033 convertibles and $1.2 billion drawn on the revolver), $2.7 billion is vendor financing classed as recourse to the parent, and $2.25 billion is guaranteed delayed-draw loans, net. The asset-level model is real but narrower than its name: of $17.6 billion of delayed-draw loans, $4.6 billion is non-recourse. Non-recourse debt rose a further $1.6 billion net: DDTL 4.0 rose $1.7 billion, offset by other movements. The risk factors repeat June's warning that "the majority of the debt under our Credit Facilities bears interest at variable rates, the majority of which is unhedged", now prefaced by "Market interest rates have increased". The filing puts the effect of a 100-basis-point move at about $41 million of interest a quarter, which we read as roughly $16 billion of floating-rate debt; it does not say whether that is before or after its $6.8 billion of swaps.
The next 15 months. $13.4 billion of principal falls due from October 2026 through December 2027 ($10.6 billion within twelve months of 30 September). None of it is a bond: the first note matures in June 2030 ($2.0 billion) and the revolver ($1.2 billion drawn) in November 2029. The filing does not split the ladder by facility, so we estimate it. The three oldest delayed-draw loans repaid $1.06 billion in Q3 with no new draws, $5.3 billion over five quarters at that pace. DDTL 5.0 and 5.5 begin monthly amortization in November 2026 and February 2027, and DDTL 4.0 by July 2027 at the latest (earlier at sites that missed delivery dates): about $1.7 billion through December 2027 on level payments. That puts contract-funded amortization at about $7.0 billion. The residual, about $6.4 billion, we attribute to vendor financing ($7.7 billion outstanding), though it also holds any DDTL 3.0 amortization and the $195 million Magnetar loan. The first is paid from the contracts pledged to it; the second from general cash flow, or rolled into new facilities. Our model refinances every asset-level maturity at the new-money rate, so the split moves timing, not value.
The leases and commitments still to land
Leases signed but not yet begun total $47.4 billion of undiscounted payments, from $35.5 billion at 30 June, starting between 2026 and 2029 for seven to seventeen years. Outside that figure: a 453 MW single-site lease with rent capped at $16.9 billion over sixteen years, and 354 MW at sites where rent follows construction cost. Other commitments: $2.1 billion of data-center construction, about $4 billion of purchase commitments, $1.0 billion of unfunded commitments to data-center partnerships, up to $120 million for the New Jersey joint venture plus a $95 million guarantee, and $0.2–0.8 billion of equipment the company must install in leased sites.
What reaches the balance sheet. Construction and purchase commitments become property, paid in cash or vendor credit, and are inside our capex. The leases arrive as liabilities as each site is handed over. Discounted the way the commenced leases are (their liability is 57% of their undiscounted payments, at a 10% average rate over twelve years), the $47.4 billion is about $26.9 billion of liability. Spread over the contracted megawatts not yet active, outside the two sites above, that is an estimated $12.8 million a megawatt, close to the lease assets added per megawatt in 2026 (more than 850 MW, so that figure is an upper bound). Treating every megawatt added on our base path for active power as leased, about $10 billion arrives by September 2027 and $16 billion by March 2028, taking lease liabilities from $19.3 billion toward $30–35 billion; these are estimates, because the filing does not say when each lease begins.
Is undelivered-capacity prepayment debt? Only the part a customer can take back. The $1.9 billion of customer liabilities sits on contracts that became cancellable, so the cash may have to be returned; we count it as debt. Deferred revenue is settled by delivering service, at the cost of running the capacity, not by repaying principal; counting all of it as debt treats every customer as able to walk away. Q3 says that is not hypothetical for a fifth of a quarter's balance, so we show it as a stress line rather than ignore it. Our cost of capital already charges the full $10.0 billion as capital.
| Claim counted as debt, 30 September 2026 | Times Q3 EBITDA, annualized |
|---|---|
| Parent view: net debt less the $5.4 billion of non-recourse debt (DDTL 4.0 $4.6 billion, vendor financing $0.8 billion) | 4.2× |
| Net debt at principal ($43.4 billion) | 4.8× |
| + customer liabilities ($1.9 billion) | 5.0× |
| + lease liabilities ($19.3 billion), over EBITDA before rent | 5.5× |
| + all deferred revenue ($10.0 billion): stress, every prepayment refundable | 6.3× |
How to read it. The first row is the parent's view, leaving out debt only a subsidiary owes but keeping all EBITDA, including the non-recourse sites', so it flatters the parent; from the second row on, each row adds one more claim to net debt and divides by the third quarter's adjusted EBITDA annualized ($9.0 billion); the lease rows add the quarter's rent back to EBITDA ($11.8 billion), so leased and owned capacity compare on one basis. Read down the column: each step is a judgment about what counts as debt, and the last is a stress, not our view.
Lives, realization and what the capital costs
Book life against loan life. GPUs are depreciated over six years (2025 annual report); the 10-Q reports no change to its accounting policies and does not restate the life. The newest guaranteed loans close and finish within about five years: DDTL 5.0 from May 2026 to May 2031, DDTL 5.5 from August 2026 to July 2031. Their lenders are repaid with roughly a year of book life left, and whatever the GPUs earn after that belongs to the equity: this is the fleet the company says comes off contract "fully paid", and why the renewal question matters more than the depreciation schedule. DDTL 4.0 runs six years, March 2026 to March 2032, about as long as the book life; vendor credit runs one to three years.
Realization. We credit 60% of the company's renewal figure (about $10 billion a year of revenue from GPUs coming off contract in 2027–28) in the base, 35% in the bear, 20% in distress and 100% in the bull. At $44.91, our weighted value reaches the price with base renewals realized at 24%. The filing reports no renewal revenue; the evidence remains the call's "more than $3 billion" of prepayments for re-contracted capacity.
The cost of capital. On its books, CoreWeave pays 7.9% on its debt (Q3 contractual interest and issue-cost amortization over average principal, matching the company's 7.9%) and about 3.0% on prepayments, about 7.0% on the $50.6 billion the two fund together. At the margin it pays more: the next secured facility is guided at 7–7.5%, the 9.75% 2031 notes yielded about 15.5% on 10 November (as reported, in our update), and the equity's own rate starts at 16.6%. Weighted at market values, that is about the 12.7% our update used. The gap between the 7.0% the company books and the 12.7% the market charges is the cost of the equity cushion under the cheap layers, and why the business must earn well above its booked cost of debt to create value.
What the price requires, at $44.91 and our weighted value, holding everything else: base new contracts at $17.2 million a megawatt a year (we assume $18 million); or an asset beta of 1.38 in every case (we use 1.25); or base new asset-level money at 10.5% (we assume 8%, the company guides 7–7.5%); or base renewals at 24% of the company's figure (we credit 60%); or a distress weight of 27% (ours 20%), or a bull weight of 15% (ours 20%). The pricing and asset-beta steps are small; the funding-rate, renewal and weight steps are larger, though the price's distress weight lies inside our own 20–30% range. That is what a value 10% above the price means.
The distress path: how a funding problem becomes a default
This is a mechanism, not a new valuation case: it traces how the distress case in our weights would happen, and when. The engine is a quarter-by-quarter liquidity runway for CoreWeave, Inc. alone. The subsidiaries' cash is fenced off; only what their waterfalls release reaches the parent, less DDTL 5.5's share, which it sweeps to its lenders from early 2027 until its loan is down to $1.3 billion. From its cash, revolver headroom, the ATM, the prepayments and those distributions, the parent pays coupons, vendor credit, refunds, guarantee calls, rent, joint-venture funding and its share of each build: the part of capex the delayed-draw loans do not lend. That share is the lever. DDTL 4.0 lends 90% of capex, DDTL 5.0 and 5.5 about 70%, and more than three-quarters of the remaining committed 2027 build is tied to investment-grade and enterprise customers, so we take 85%. Vendor credit is assumed to roll, new credit funding the next equipment as each vendor loan amortizes, as the recourse balance's rise in Q3 shows; if only half of the ~$6.4 billion due through 2027 rolls, the parent needs $4.4 billion beyond the ATM by the end of 2027, and if none does, $7.6 billion. With the parent's market shut, the ATM sells a third below the 19 November price.
The base path. Parent liquidity is about $12.2 billion at the end of 2026, the ~$9 billion of prepayments included, then $7.1 billion at June 2027 and $3.7 billion at September as the build is paid for. The parent sells the ATM and needs about $1.2 billion more by the end of 2027, about $2.8 billion of parent capital in 2027, inside the company's "less than $5 billion", and $6.1 billion beyond the ATM by the end of 2028. At a 90% advance rate throughout, it needs nothing new through 2027; at 70%, $6.9 billion. The path credits all of the prepayments to the parent, which signs most contracts; if half of them land in the subsidiaries' accounts instead, the parent needs $5.7 billion beyond the ATM by the end of 2027, above the company's figure. So our runway matches the company's figure only if most of that cash reaches the parent and vendor credit keeps rolling; the year-end report will show where the cash went.
Five shock channels, each on its own. With the parent able to raise capital: (a) a quarter of the 2027–28 build slips and the $1.9 billion already cancellable is refunded, so parent capital needed by the end of 2027 rises to $2.0 billion; (b) a customer that is 30% of revenue stops paying from the second quarter of 2027, its subsidiaries trap their cash and the guaranteed loans behind it call the parent for the shortfall: $5.7 billion by the end of 2027 and $17.7 billion by the end of 2028; (c) both funding markets shut from January (no new delayed-draw facilities, vendor credit not rolled, no parent raise): the parent reaches its buffer in the first quarter of 2027 and runs out in the third; (d) renewals in the 2027–28 roll-off realize 20% of the company's figure rather than 60%: $2.6 billion and $11.4 billion; (e) floating rates rise 2 points on the unhedged debt: $1.6 billion and $6.8 billion. Only $1.9 billion of prepayments is classified as cancellable today; the ~$10 billion of deferred revenue becomes refundable only through cancellability, so the refunds in these paths are a conditional exposure, not a forecast.
The turning point. The loop starts when the parent cannot pay its share of a build: deliveries slip, slipping deliveries make contracts cancellable, refunds drain the parent, and the next build is harder to pay for. With the loan market open but the parent's own market shut, the parent reaches its buffer of about $2.5 billion in the fourth quarter of 2027, cuts capex from there and runs out of cash in the second quarter of 2028. With both markets shut and a large customer failing, it reaches the buffer in the first quarter of 2027 and runs out in the second. The cross-default thresholds then decide how far it spreads: inside the subsidiaries a default counts above $5–100 million depending on the facility (the table in §3); into the notes and convertibles, above about $291 million; and from the parent's bank lines back into DDTL 1.0 and 2.0/2.1, above $15–50 million until each loan steps up.
Exhibit 4 — The parent's own liquidity, by quarter, under the stress paths
How to read it. Each line is CoreWeave, Inc.'s liquidity at the end of each quarter from Q4 2026 to Q4 2028, in billions of dollars: unrestricted cash plus revolver headroom, after the ATM but before any other new parent capital. The dashed grey line is the buffer, about two quarters of the parent's fixed charges plus $1 billion of operating float; open circles mark where a path first cuts capex because it is below the buffer with no market to raise in. Compare the lines' end points with zero and the circles' dates: in the base the parent needs new capital from late 2027; it runs out within 2027 only if both its own market and the loan market shut (by the third quarter, or the second with a large customer failing too), and by mid-2028 if its own market alone stays shut.
The brakes. Six things slow or stop the sequence. Each buys time or cuts a link, and each has a point where it gives out.
- About $9 billion of prepayments due in the next three months. Customers pay up front, and the cash arrives before the parent's large 2027 bills; on our runway it carries the parent through most of 2027, though even with it the base still needs about $2.8 billion of parent capital that year. Where it gives out: it has not been collected yet; about $6 billion of it may depend on capacity being delivered, and slipping deliveries make contracts cancellable, as the $1.9 billion already is; and if half of it lands in the subsidiaries, the parent needs $5.7 billion beyond the ATM by the end of 2027.
- No bond or convertible matures before June 2030. Until then the parent's only hard dates are interest payments, so no maturity can force a default. Where it gives out: the $13.4 billion due through 2027 is amortization on the subsidiary loans and vendor credit; by our estimate about half is paid from pledged contracts, but the vendor-credit half, about $6.4 billion, needs a working funding market.
- DDTL 4.0 is non-recourse. Its lenders can reach only that subsidiary's assets; the parent answers only for bad acts. Where it gives out: through the revolver, a default there can still reach the bonds, and directly into the issues whose lists omit its borrower, if the company has not designated it as excluded (§3).
- Repaying the revolver. Repaying the $1.2 billion drawn and cash-collateralizing the $0.5 billion of letters of credit, against $4.1 billion of unrestricted cash at 30 September, removes the revolver bridge, DDTL 4.0's path through it, and the older loans' trigger on the parent's bank lines. Where it gives out: it spends cash and gives up borrowing capacity just when both are scarce, so it works best early; it does not cut the guaranteed loans' path, which is the parent's own debt in all but name, nor a direct path from the newer borrowers into the issues whose lists omit them, if there is one.
- The 30-day cure. A note acceleration is annulled if the underlying default is cured within 30 days, so a refinancing or a waiver can unwind it. Where it gives out: the cure needs money or lenders' cooperation within a month, in a stressed market.
- The ATM and the capped calls. The parent can sell up to 35 million shares, about $1.5 billion at the 19 November close after commission, before it needs a new bond or convertible, and the capped calls absorb conversion up to their caps. Where it gives out: the ATM raises less, at more dilution, as the price falls (our shut-market paths sell it a third lower), and the capped calls add no cash.
Built into the chain. A subsidiary's default must exceed about $291 million to reach the revolver or the bonds (more on the indentures' own EBITDA); the older loans' trigger on the parent's bank lines sits far lower, at $15–50 million until each loan steps up. The subsidiaries hold $25.9 billion of assets against $17.6 billion of loans, each with the tests in §3's table, so the realistic first failure is a large customer terminating a financed contract, which the three older facilities let the borrower cure by replacing it within 90 days. A default gives lenders the right to call their loan, which they may waive instead, as DDTL 2.0's did for four acknowledged events of default in December 2024; the bonds need an actual acceleration.
The first three brakes buy time; the last three need cash, an open market or lenders' cooperation. They are strongest now and weaken with time and funding stress, and they strengthen as the prepayments are collected, the guaranteed loans amortize and the parent raises at a workable price, which the events in §8 read; within 2027 a default needs two failures together, but by mid-2028 one is enough.
Exhibit 5 — The distress sequence, its two inflection points and the brakes
How to read it. Read the eight boxes left to right along the top row and back along the bottom: each is a step that the one before it can set off, with the date or threshold that marks it. The two gold boxes are the inflection points, where a funding problem becomes a delivery problem (3) and a delivery problem becomes a default (6); the dashed red curve is the loop, refunds draining the parent back at step 2. The six boxes underneath are the brakes that slow or stop the sequence. It shows that within 2027 no single step makes a default: it takes a funding failure and a delivery failure together, carried to the group by the guarantees or the revolver; by mid-2028 the parent's own market staying shut is enough.
What it says about the 20% distress weight. It supports 20%, on one condition. In the base the parent needs about $2.8 billion of parent capital in 2027, which the ATM and a modest raise cover, so within 2027 a default needs two failures together (the parent's market and the loan market shut, or a financed contract terminated while the parent is short) and a bridge to carry it to the group. By mid-2028 one failure is enough: if the parent's own market stays shut while the loan market stays open, it reaches its buffer at the end of 2027 and runs out in the second quarter of 2028. That market is open today at a price: the company placed $4.2 billion of 2033 convertibles in the quarter, but its 9.75% 2031 notes yielded about 15.5% on 10 November (as reported, in our update), past the 15% line our events table reads as shut. So 20% rather than 25–30% rests on the parent raising at a workable price by mid-2027, and the test arrives within 2027 if half of the prepayments land in the subsidiaries or only half of the vendor credit due through 2027 rolls. The notes above 15% at 30 June 2027 take the weight to 30%; below 11%, to 10%.
What the full read of the credit documents changes. The bridges, and what reading every credit document added to them, change how far and how fast a default spreads, which the distress case already takes to zero for the equity, not how likely the first failure is. The older loans' $15–50 million trigger fires only when the parent's bank lines are actually called due, the same event on which the notes and convertibles already cross-accelerate, so, while today's drawings sit above their threshold, a revolver covenant breach becomes a group event no sooner than before. It does bring up to $6.4 billion of guaranteed loans with it, and it keeps that path open if the parent pays the revolver down rather than off. Whether a breach is waived or called stays with the revolver's lenders. Against that, the older loans' 90-day replacement cure, DDTL 3.0's partial guarantee and the indentures' larger EBITDA narrow the spread, and a possible direct path from the newer borrowers widens it.
Events to watch
Each event is observable on a date or in a filing, with the reading that starts the sequence and the reading that stops it. All are armed tripwires with a stated check.
| Event | When | Starts the sequence | Stops it | Effect (tripwire) |
|---|---|---|---|---|
| Next delayed-draw facility prices | by Feb 2027 (8-K, 10-K) | above 9%, below investment grade, or not priced (step 1) | at or below 7.5% with an investment-grade rating (both: the company has guided to both, and the rate alone would price one facility, not the next ones) | new-money rate 8% to 10%, or to 7.5% (cwq3-ig-ddtl) |
| Q4 prepayments collected | 10-K, Mar 2027 | deferred revenue plus customer liabilities below $13bn (step 2) | $20bn or more | prepayment path to the bear's or the bull's (cwq3-prepayments) |
| The $1.9bn cancellable balance | 10-K, Mar 2027 | more than $1bn refunded, or customer liabilities above $3bn (step 5) | back in deferred revenue, below $0.5bn | bear 25% to 30%: $47.04; served: none, the base already counts the $1.9bn as service, not refund (cwq3-custliab) |
| Delivery against plan | Q4 release, 10-K | active power below 1.75 GW, or a delayed-site repayment (step 4) | 1.85 GW or more, no delayed sites | parent liquidity tested a quarter early; on plan: none, the base already assumes delivery on schedule (cwq3-delivery) |
| Revolver covenants | each 10-Q, 10-K; 8-K | an amendment loosening the covenants, or a waiver (step 7) | reported net debt at or below 6.0 times trailing adjusted EBITDA (inside the covenant without its add-back), or an investment-grade rating | distress 25%: $44.51; 15%: $54.59 (cwq3-revolver-covenant) |
| Parent liquidity | Q2-27 10-Q, Aug 2027 | below $3bn at 30 June with no parent raise (step 2) | at or above the $7.1 billion path, or a parent raise of $2.8 billion or more | distress 25%: $44.51; on the path: none, it is the base runway (cwq3-parent-liquidity) |
| Unsecured notes yield | 30 Jun 2027 | above 15% (step 2: the parent's market still shut) | below 11% | distress 30%: $39.46; 10%: $59.64 (cwq3-unsecured-yield) |
| Parent capital raised through 2027 | 10-Q, 10-K, 8-Ks | a raise at 13% or more, or equity sold below $30 (step 3) | under $5bn with the ATM at $40 or more, or none needed | distress 25%: $44.51; 15%: $54.59 (cwq3-parent-capital) |
| A financed contract terminated or trapped | any time (8-K, 10-Q) | termination in full, a cash trap or a sweep (step 6) | a breach cured before its third monthly payment date (before a trap becomes a sweep), or a maturing contract renewed or replaced by its deadline on the facility's coverage tests | distress 30%: $39.46; cured or renewed: none, the base assumes no termination (cwq3-customer-termination) |
| Renewal revenue from the 2027 roll-off | Q2-27 10-Q | below $1.2bn annualized (shock d) | $4bn or more | renewal realization 30% or 100% (cwq3-renewal-realized) |
How to read it. One row per event, in the order the sequence would meet them. The second column is when it can be read; the third, the reading that starts or advances the sequence, with the step in Exhibit 5 it belongs to; the fourth, the reading that stops it; the last, what each reading does to our weights or value. Read across a row: every event has a favorable reading as specific as its adverse one. Where the favorable reading's effect is "none", it confirms what the base already assumes. The weight moves hold the bear at 25% and shift the rest between base and bull, as do the 30% and 10% bounds; the verdict's price-implied weight and its 25% midpoint scale all three in proportion.
What else the filing says
- Customers. Two customers were 30% and 24% of Q3 revenue; no other reached 10%, and four hold 77% of receivables. The filing names neither; OpenAI (up to about $6.5 billion) and Meta (up to about $21.0 billion) are named as expected significant customers. If the 24% customer is the one that was 67% a year earlier, which July's note took to be Microsoft (the filing warns its customer letters may not denote the same customers across periods), it bought about the same dollars ($0.88 billion against $0.91 billion), so its share fell by dilution rather than loss.
- Risk factors, against June. Added: restrictive covenants in the debt heading; storage and memory suppliers alongside GPUs; local community opposition as a permitting delay; the October "market rumors" factor; open-weight models as a risk to its business and pricing models; the ATM program's collared forwards. Expanded: New York's July moratorium, now described as pausing, for up to a year, discretionary state environmental permits for data centers of 50 MW or more; the warning that the convertibles become current liabilities if their conversion trigger is met, extended to the 2033 notes. Dropped: June's expectation that the cost of capital would keep falling through "asset-backed securitizations and rated parent-level debt."
- Insiders and shares. The chief executive's plan covers up to about $500 million of sales from 14 November 2026 to 31 March 2027, subject to price thresholds. No shares were sold under the ATM through 30 September. The filing has no subsequent-events note and reports no financing after 30 September.
- Controls and litigation. The material weaknesses in IT controls, segregation of duties and accounting staff continue. The securities class action is pending and the derivative suits are stayed. The company bought $275 million of surety bonds in August for an electric-service agreement ($154 million of cash collateral). No change to the six-year depreciable life is reported.
What would change our view
The tripwires in our update stand, with their reading rules; this filing moves none of them, and §8 adds six, each with a favorable reading as specific as its adverse one. We would move toward the bull case on the next delayed-draw loan at or below 7.5% with an investment-grade rating, deferred revenue of $20 billion or more at the end of 2026 with customer liabilities back below $0.5 billion, annualized renewal revenue from the 2027 roll-off reaching $4 billion, or the unsecured notes below 11% by mid-2027. We would move toward the distress case on a delayed-draw loan priced above 9% or not at all, customer liabilities above $3 billion or a refund of the $1.9 billion disclosed in the year-end report, renewal revenue below $1.2 billion annualized, the depreciable life cut to five years or less, or the notes above 15% at mid-2027. At about 15.5% on 10 November (as reported, the last reading in our update) the notes are already past that line: the move toward distress needs the level to hold to mid-2027, the move toward the bull case a fall of four and a half points. The notes above 15% at mid-2027, or a loss in the first half of 2028, take the distress weight to 30% and fair value to $39.46; the notes below 11%, or adjusted net income positive in 2027, take it to 10% and fair value to $59.64.
Sources & Method
Filings. CoreWeave Form 10-Q for the quarter ended 30 September 2026 (acc. 0001769628-26-000465, filed 12 November 2026; period confirmed on EDGAR): cover share count; Notes 2 (customers, contract balances, RPO), 3 (investments, VIEs, joint venture), 5 (property, construction commitments, consolidated VIEs), 7 (accrued and other current liabilities), 8 (leases, leases not yet commenced), 9 (commitments, litigation, surety bonds), 10 (debt table, ladder, interest, facility terms), 11 (equity, ATM, options, RSUs, warrants) and 13 (anti-dilutive securities); MD&A liquidity and contractual obligations; Item 3 rate sensitivity; Item 1A risk factors; Item 5 trading plans and severance. Q2 2026 Form 10-Q (acc. 0001769628-26-000366) for the June comparatives and risk factors; FY2025 Form 10-K (acc. 0001769628-26-000104) for vendor-financing terms and the six-year life. Q3 release (acc. 0001769628-26-000461).
The call. 10 November 2026; prepayment and renewal figures from the company's proofread transcript and two independent transcriptions, archived with the update.
Prices. CRWV settled close of 19 November 2026 ($44.91) from a brokerage feed's official close; the update's 10 November close ($48.73) for the published value.
Method. The update's model (fleet-and-financing free cash flow to equity in four cases, weighted 20/25/35/20, the 2036 equity valued as a call on the enterprise) with the four filing inputs, the price anchor and three method changes, each applied on its own and shown in Exhibit 1: each case's cost of equity relevered every year at its own net debt over its own equity value (asset beta 1.25, debt beta 0.35, no tax shield, ceiling 30%; year one at the market debt-to-equity of 1.51 at the 19 November close); the terminal enterprise valued at mid-2037 cash flows to match the mid-year explicit years and the 2036 equity struck at end-2036 net debt; and the convertibles settled at maturity in each case at that case's own share value, net of the capped calls. Return on new capital is measured as the after-tax unlevered return on a new site at that year's price (10.5% in the last explicit year and in the terminal of every surviving case); the accounting ratio is not used because GPU replacement is lumpy. Customer liabilities enter as opening deferred revenue; the refund path is shown as a sensitivity. The split of the 15-month principal is our estimate: the three oldest delayed-draw loans at their Q3 repayment pace, DDTL 5.0 and 4.0 on level monthly payments on their drawn balances and DDTL 5.5 on its full commitment, to their stated final dates, DDTL 3.0 at zero (its amortization began in April 2026 on a schedule that is not filed, and it still drew net in Q3), and the residual attributed to vendor financing, which therefore also holds any DDTL 3.0 amortization and the Magnetar loan. Pending leases are discounted at the ratio of liability to undiscounted payments on the commenced leases, spread over the contracted megawatts not yet active outside the two separately disclosed sites, and timed to the base case's active-power path as if every added megawatt were leased. Leverage uses Q3 adjusted EBITDA annualized; the covenant comparison and the cross-default thresholds use trailing adjusted EBITDA of $5.8 billion, the sum of the company's last four quarterly releases. The $7.8 million and $13.8 million a megawatt-year are our arithmetic: Q4 2025 revenue annualized and grossed up 6% for growth within the quarter (our estimate), over the 0.85 GW active at the end of 2025; and the rise from there to the company's end-2026 exit run-rate over the 1.0 GW added in 2026 (to 1.85 GW). Without the gross-up they would be about $7.4 million and $14.2 million. The cost of prepayments is the filing's non-cash interest on significant financing components over the whole average deferred-revenue balance, an accounting rate that leaves out any price given up for the cash. The 15.5% note yield and the credit-default-swap levels are the reported figures our update used, not new readings. All arithmetic is in the archived scripts, and the model and every input are in the downloadable workbook.
Credit documents. Every filed credit agreement, amendment, parent guarantee and indenture behind the 10-Q's debt table, read in full: DDTL 1.0 (credit agreement of July 2023 and its first amendment; parent guarantee and four amendments; S-1 and 2025 10-K exhibits), DDTL 2.0 and 2.1 (credit agreement of May 2024 and five amendments, the fifth creating DDTL 2.1; parent guarantee and two amendments), DDTL 3.0 (credit agreement, parent guarantee and first amendment; 8-K of 31 July 2025 and 2025 10-K), DDTL 4.0, 5.0 and 5.5 (Exhibits 10.1 and 10.2 to the 8-Ks of 30 March, 15 May and 7 August 2026), the revolving credit agreement as amended by Amendment No. 4 (Exhibit 10.7 to the Q3 2025 10-Q), and the indentures for all five note issues and all three convertibles, with the April 2026 supplemental indenture. Amendments filed as marked copies were read with their deletions marked. Not public: the vendor (OEM and software) financing agreements, the Magnetar loan, a second amendment to the DDTL 1.0 credit agreement dated 10 November 2025 (cited in a later amendment, not filed), DDTL 3.0's amortization schedule and the agreement governing how contract revenue moves between entities, and DDTL 2.0's commitment schedule. Customer names, some percentages and schedules are redacted.
Distress runway. CoreWeave, Inc. alone, quarterly from Q4 2026 to Q4 2028, on the base case's operating plan; subsidiaries hold contracts that cover their debt service 1.25 times and release the excess, except DDTL 5.5's share, swept to its lenders from 2027 until its loan is $1.3 billion; delayed-draw loans fund 85% of capex; vendor credit is refinanced as it amortizes (shown at half and none rolled); all prepayments reach the parent (shown at half); fixed parent charges from the coupons, vendor interest and joint-venture funding; the buffer is two quarters of those charges plus $1 billion; with the parent's market shut, the ATM sells a third below the 19 November price. All parameters and paths are in the archived runway script.
Limitations. The filing does not split the ~$9 billion of prepayments by payment terms, the ladder by facility or the pending leases by year; those figures are estimates or bounds, labeled as such. The cause of the $1.9 billion reclassification is not disclosed.
Disclosures
Information only. TON618 Capital. This report is for information purposes only. Nothing here is an offer to sell or a solicitation of an offer to buy any security, fund interest, or digital asset, and nothing here is personalized investment advice or a recommendation regarding any instrument.
Publisher's exclusion. All research is published solely as general, impersonal information of regular circulation. It is not tailored to the objectives or circumstances of any individual and is not issued in connection with compensation from any client. The Fund has no clients and distributes all research free of charge. On that basis it publishes in reliance on the publisher's exclusion from the definition of "investment adviser" under the Investment Advisers Act of 1940 (§202(a)(11)(D); cf. Lowe v. SEC, 472 U.S. 181 (1985)).
Registration & conflicts. TON618 Capital is not registered as an investment adviser or broker-dealer in any capacity. The Fund is a Bitcoin fund and may hold or transact in the securities or digital assets it discusses; material conflicts are disclosed where they exist. Ownership: the Fund holds no position, long or short, in CRWV or in any other security named in this report as of the report date, and has no economic interest in their price. Compensation: the Fund received no compensation from any party in connection with this report and charges nothing for it.
Use of AI. Artificial intelligence is used in the creation of this research. All methodology and data integrity are reviewed and approved before publication by TON618 Capital's Chief Investment Officer, Keyth Beck; errors may nonetheless occur, and readers should verify independently.
CFA. This report was prepared to align with CFA Institute analytical standards (methodology only). CFA® and Chartered Financial Analyst® are registered trademarks owned by CFA Institute. That reference describes the analytical framework applied; it does not imply the report was prepared, reviewed, or authored by a CFA charterholder, and the report is not issued, reviewed, endorsed, certified, or approved by — nor affiliated with — CFA Institute.
Risk & feedback. Past performance is not indicative of future results. Digital assets and equities are volatile and may result in total loss of capital. Corrections and feedback are welcome — please direct them to CIO Keyth Beck at keyth@ton618capital.com. Version 1.3.