TON618 CapitalEquity Valuation · Update
As of 10 November 2026
v1.0
Equity Valuation · AI Infrastructure · CRWV · 10 November 2026

CoreWeave (CRWV) — The Floor Moved to the Fleet

The stock fell a third from July while CoreWeave raised prices twice, lifted its operating margin from 1% to 10% and re-contracted older GPUs above their first prices. The signed book no longer covers the debt plus the build it still needs, so the value now rests on the fleet: new capacity earning near today's quoted rates, old GPUs re-contracting and asset-level lenders staying open. We value the stock at $48.57, from $60 in July, against $48.73 at the close of 10 November.

TON618 Capital Research · As of 10 November 2026 · Download PDF · Download model

$40$60$80$100July value $60This note $48.57$107.73$43.66$48.73 (10 Nov)$72.91JulAugSepOctNov

In July we called CoreWeave the neocloud thesis in a single security: revenue more contracted than any peer's, a floor under the equity worth a seventh of the price, and everything above it a levered bet on how long a GPU earns its keep. Four months on, the price has done what that framing feared, falling from $72.91 to $48.73, within a dollar of the $40 IPO price at its 5 November intraday low ($40.58; it closed that day at $43.66), with the company's unsecured bonds trading near 15% and its credit default swaps past 1,000 basis points in mid-October. The business did the opposite. Revenue rose 42% in the quarter, prices rose about 25% in July and 10% more later in the quarter, older GPUs re-priced above their original contracts, and both guides went up. This note asks which of those two moves is better argued, and finds that each is right about a different thing: the credit market about the signed book, the business about the fleet. At $48.73, the price holds both.

The questions

What this note answers

  1. What did CoreWeave report, and why did its margin turn?
  2. What is the debt now, what does it cost, and was October a credit event?
  3. What is the contracted book worth after the debt and the build it still needs?
  4. How long does a GPU earn its keep, and what does the new pricing do to that question?
  5. What is CoreWeave worth, what moved the number from $60, and what does $48.73 require?
  6. What would change our view?
$49fair value a share · distress $0 / bear $7 / base $58 / bull $132 at 20 / 25 / 35 / 20
The view. We value CoreWeave at $48.57 a share against the $48.73 close of 10 November, the session before the results traded: the price now sits on our value, after sitting well above July's $60. After hours the stock traded $52.96 at 01:04 UTC on 11 November; that is one data point, not an input. The cases are unusually wide: the distress case leaves nothing for the equity, the bear little ($6.91), the base $58.34 and the bull $132.11. The base is above the price and the weighted value is not, because 45% of the weight sits at or near zero: of the $48.57, the base contributes $20.42, the bull $26.42 and the bear $1.73, so the bull's fifth of the weight carries more than half the value. The quarter's data alone, run through July's method, would give $25.66: that method holds interest at 8% of revenue (it ran 23% in Q3) and charges capex at 1.4 times revenue (the 2026 build cost 2.6 times revenue). We value the equity on its financing path instead, with the equity in 2036 valued as what it is, a call on an enterprise that owes more than the equity is worth. The signed book alone no longer covers the debt plus the build it still needs (−$29.30 a share); with the paid-off fleet re-contracting it is worth about $7.68. In our base case, the price requires new contracts to earn $16.9 million a megawatt a year, with backlog capacity $2 million less (we assume $18 million; the 2026 build realized $13.8 million, and new four-to-five-year contracts are quoted at $20–25 million). Across the cases, it implies a 20% chance of distress, the bottom of our 20–30% reading of the CDS, and a 20% bull weight, both where we put them.
10%adjusted operating margin in Q3, from 1% in Q1 and 5% in Q2; the Q4 guide implies 15.9%–17.0%
23%interest as a share of revenue in Q3 ($846 million); it has run 23–26% for five quarters, against the 8% July's model charged
$7.68contracted-book floor a share once the fleet re-contracts; −$29.30 on the signed book alone, after the debt and the 2027 build
$16.9Mrevenue a megawatt a year that $48.73 requires on new capacity in our base case, against our $18M (the 2026 build realized $13.8M; new contracts quoted at $20–25M)

Terms used throughout, defined once. RPO (remaining performance obligations): contracted revenue not yet recognized; CoreWeave's revenue backlog is RPO plus a small amount of other committed revenue ($132.3 billion and $0.2 billion at 30 September). Take-or-pay: a contract under which the customer pays for reserved capacity whether or not it uses it. Prepayment: cash a customer pays before the service starts, booked as deferred revenue. DDTL (delayed-draw term loan): a loan drawn in pieces as GPUs are bought, secured on them and on the customer contracts they serve, held in a subsidiary; most are non-recourse or guaranteed but asset-level. Parent-level debt: CoreWeave, Inc.'s own unsecured notes and convertibles, serviced from the company's free cash flow. Convertible note: debt the holder can swap for shares at a set conversion price; a capped call is an option the company buys at issue that offsets dilution between the conversion price and a cap. ATM (at-the-market) program: shares sold into the market over time. ARR (annualized run-rate revenue): the last month of a quarter times twelve. Active power: the megawatts (MW) or gigawatts (GW; 1,000 MW) of data-center capacity in service; contracted power includes capacity leased but not yet built. Adjusted EBITDA: earnings before interest, tax, depreciation and amortization, also before stock pay; adjusted operating income: after depreciation, before stock pay. CIP (construction in progress): assets bought but not yet in service, so not yet depreciated. CDS (credit default swap): insurance against a borrower's default, quoted in basis points (bp) a year. FCFE: free cash flow to equity, cash left for shareholders after debt service and investment.

The two cases

The cases on either side of the price, in their own terms

The credit market's case. The case against paying more than $48.73 is not that CoreWeave lacks demand. It is that the equity is the last claim on an asset base that loses value fast, funded by debt that just repriced to distress levels, at a moment that may be the top of a rental cycle.

  1. The credit market has already called it. In mid-October CoreWeave's 9.625% 2032 notes traded below 80 cents, about a 15% yield (16–16.5% at the intraday low), and its five-year CDS passed 1,000bp, roughly a 60% risk-neutral chance of default over five years. On 10 November the 9.75% 2031 notes still yielded about 15.5%. High-yield spreads widened only 9bp on the day the stock fell 15%, so the market priced CoreWeave's own risk, not the market's. The company itself now says it will not issue unsecured debt "at current yields."
  2. The signed book no longer covers the debt. Total debt is $46.9 billion ($10.6 billion due within a year) against $4.1 billion of unrestricted cash. Run the $132.3 billion book off at its contract margin, pay for the 2027 build the company says that book needs, and repay the net debt: the equity is worth −$29.30 a share. Everything above zero depends on revenue CoreWeave has not yet signed.
  3. The margin turn is pricing at a cyclical peak. Prices rose about 25% in July and another 10% later in the quarter; Ampere and Hopper GPUs re-priced 30–50% in a quarter. That is what a supply squeeze looks like. Hyperscaler, sovereign and neocloud capacity ordered in 2025–26 lands in 2027–28, the same years 300,000 of CoreWeave's GPUs come off contract and the 4.6 GW of contracted power has to be filled.
  4. The reported economics do not support the equity. On July's own formula with today's measured inputs (62% adjusted EBITDA margin, capex of 2.6 times a year's revenue, interest at 18% of revenue), the business breaks even only if a GPU earns its full rate for six years, the book life. Return on capital has been about zero for the four quarters the history allows.
  5. Dilution is the release valve. Officers, directors and a large holder (Magnetar, 43% of the shares) filed 96 notices of proposed sale and sold 10.1 million shares ($0.97 billion) between the July note and this print; a 35-million-share ATM program is in place; and equity sold at $48.73 costs nearly twice the dilution it would have at the $90.32 of 11 August.
  6. Free cash flow is not in sight. It was −$7.5 billion in Q3 and −$17.9 billion for nine months, capex is guided to $37–39 billion this year, the CFO said 2027 capex would stay in line with 2026 if the company pursued only its committed backlog, and no free-cash-flow date was given.
  7. The customers are few. At 30 June one customer was 36% of revenue and two were 32% each of receivables (Q2 10-Q); OpenAI is named among expanded relationships; and $35.5 billion of leases had been signed but not yet begun. A lost anchor would strand both the capacity and the rent.

The case for the stock. CoreWeave's case is that the scarce input in AI compute is not the GPU but the powered, cooled site. It has more than 4.6 GW of contracted power, on data-center leases with initial terms of five to fifteen years, some extendable for up to ten more, in a market where new grid connections take years. On that view the margin turn (1% to 10% in three quarters, prices up about 25% and then 10%, Ampere and Hopper GPUs re-contracted above their first prices) is that scarcity being priced, and it lasts. Its scale and software keep customers who have built on them; a backlog of $132.5 billion and an exit-2027 run-rate target above $30 billion (February outlook presentation, p. 4) say demand is not the question. Asset-level debt about to be priced at investment grade, 7–7.5%, funds the build without the parent's unsecured market, so October's repricing of the notes is a cost of capital on a shrinking slice, not a verdict on the business.

This note tests the credit market's claims 1, 2 and 4, claim 3 as far as what the price requires of pricing, and claim 5's arithmetic; it states claim 6 and leaves claim 7 for the Q3 10-Q's customer and lease notes. Of the case for the stock, it tests the pricing, the renewals and the funding (§4–§5). It cannot test, inside a ten-year model, whether GPU rental prices hold through the 2027–28 supply wave, whether the next investment-grade facility prices where the company says, how much of the renewal revenue the company describes is realized, or whether sites stay scarce past 2036. Our answers are the tripwires in §7 and, for the last, the terminal judgment argued in §5.

The print

The quarter: the margin turned, and four boring checks explain it

Revenue was $3,661 million, up 168% on the year and 42% on the quarter, an increase of $1,086 million. Adjusted EBITDA was $2,256 million (62%), adjusted operating income $379 million (10%) against a guide of $200–260 million, and GAAP operating income $207 million. Net interest expense was $846 million, so the net loss was $709 million, or $1.27 a share. The company booked an $83 million tax expense on a $626 million pre-tax loss, $74 million of it deferred; the −$0.95 a share some feeds show is the adjusted net loss ($529 million) over 559 million weighted shares, not GAAP. The backlog rose to $132.5 billion from $104.2 billion: the early-Q3 signings of more than $25 billion, less the quarter's $3.7 billion of revenue, account for all but about $7 billion of that.

Both guides rose. Full-year revenue is now $13.0–13.3 billion and adjusted operating income $1.275–1.375 billion, which implies fourth-quarter revenue of $4.69 billion–$4.99 billion (up 28–36% on the quarter) at an adjusted operating margin of 15.9%–17.0%. Capex is guided to $37–39 billion, the low end raised. Exit-2026 ARR is guided to $20–21 billion, from $18.5–19.5 billion.

July's thesis was that CoreWeave's margin is a function of how long its GPUs earn, and that depreciation was catching up with it: adjusted operating margin had fallen from 17% to 1%. That reversed: 1% in Q1, 5% in Q2, 10% in Q3 and about 16–17% guided for Q4. Before crediting it, the boring checks.

The useful life was not changed: GPUs are still depreciated over six years, extended from five in 2023, and the Q3 depreciation share fell to 51.6% of revenue from 54–55% because revenue grew faster than the in-service base, not because of a policy change.

Exhibit 1 — Margin, depreciation and interest as a share of revenue, five quarters and the Q4 guide

0%20%40%60%16%6%1%5%10%16%Adjusted operating margin46%52%55%54%52%Depreciation and amortization23%25%26%25%23%23%Net interest expenseQ3-25Q4-25Q1-26Q2-26Q3-26Q4-26 guide

How to read it. Each line is a percent of quarterly revenue, Q3 2025 to Q3 2026, with the fourth-quarter guide midpoint as an open marker where the company guides it. Gold is adjusted operating margin; grey is depreciation and amortization; black is net interest expense. Read the gap between the gold line and the black line: when interest sits above the operating margin, the company loses money before tax. The chart shows the margin turning up from 1% while interest stays near a quarter of revenue.

The debt

The debt: twice July's, priced in two markets

Debt was $46.9 billion at 30 September, up $11.8 billion in the quarter and from $24.9 billion at July's anchor; $10.6 billion of it is due within twelve months, against $4.1 billion of unrestricted cash and $2.3 billion restricted. Net debt is $42.8 billion, from $22.6 billion in July. The stack has two layers that the market now prices very differently.

October, verified. On 14 October the stock fell 15.5% to $65.78 on 111 million shares. Reuters, relaying Bloomberg, reported the 2032 notes below 80 cents, about a 15% yield, and on 15 October reported CDS around 1,000bp, roughly 63% implied odds of default over five years. A company memo dated 14 October, reported by CNBC and Reuters, said the market can "misunderstand and act irrationally." Nebius fell 10% and IREN 12% the same day; Oracle fell 8.8%. High-yield spreads widened 9bp and the ten-year Treasury yield fell, so the move was specific to CoreWeave and its peers. The wires gave no single cause; the strands they reported were the debt load, comments that the company was exploring investment-grade borrowing, and peer financing strain. Nothing was missed or defaulted: this was the unsecured market refusing to fund the parent at old prices, not a payment event.

Was the market shut? In late October Elon Musk wrote that debt markets had "almost entirely shut off" for companies without revenue independent of AI data centers. The prints say otherwise for secured credit: Lambda closed $1.0 billion of secured, investment-grade-rated GPU debt at 6.78% (1 October), Nebius a $3.5 billion secured facility at SOFR+275 (22 October), and smaller borrowers priced wider (WhiteFiber at SOFR+600, 9 November). Unsecured and high-yield take-outs did stall. The market split by security, not by sector, which is the split our July thematic note predicted.

What the stack costs. At 30 September the company's weighted average cost of debt was 7.9%. On the asset-level debt it is about 9.5% contractual; on the notes, their coupons; on the convertibles, under 3%. Q4 interest is guided to $1.05–1.15 billion. The cost that matters for value is the cost of refinancing: in our base case, new asset-level money at 8.0% (above the company's 7–7.5% guide for its next facility) and the parent notes and convertibles, as they mature in 2030–33, at 11.5%. When the convertibles roll, their sub-3% coupons become market coupons; that is about $1 billion a year of extra interest from 2031 that July's percent-of-revenue line could not see.

Leverage, on a matched basis. July's "about 7 times" was net debt over 2025 adjusted EBITDA (7.3 times). Over trailing EBITDA it is 7.3 times now; on the latest quarter annualized it is 4.7 times, against 4.9 times for the quarter July annualized. Leverage is unchanged on either basis: debt and EBITDA both roughly doubled.

The funding table. The CFO said that if the company pursued only its committed backlog, 2027 capex would stay in line with 2026's ~$38 billion. Against it: about $9 billion of prepayments and deposits due in the next three months (more than $3 billion of it for re-contracted, paid-off GPUs), "nearly $5 billion available to draw under already raised DDTLs," more than $9 billion already in CIP, EBITDA less interest that the company expects to "more than double in 2027," and new DDTLs. The company expects to raise "less than $5 billion in incremental parent-level capital through 2027." Our base case funds 2027 with $45 billion of external capital, almost all of it new asset-level debt with the 35-million-share ATM, ending 2027 at $72 billion of net debt, 4.0 times its adjusted EBITDA. By case:

2027, $ billionDistressBearBaseBull
Adjusted EBITDArestructured16.618.220.4
Stock-based pay, counted as a cost—-1.1-1.2-1.3
Capex—-29.7-37.4-39.9
Working capital, net of prepayments—-10.8-7.3-3.1
JV and VIE funding—-1.0-1.0-1.0
Interest—-5.0-4.7-4.4
Cash tax—-0.0-0.0-0.0
Free cash flow to equity—-30.9-33.3-29.4
Asset-level debt maturing—-12.1-12.1-12.1
External capital needed—43.145.541.5
of which new debt—42.843.839.8
of which equity sold—0.21.71.7
at a net price a share of—$6.78$47.76$47.76
Net debt / adjusted EBITDA, end-2027—4.3×4.0×3.4×

How to read it. One column per case for calendar 2027, in billions of dollars. The rows down to free cash flow to equity add up (each figure is rounded, so a total can differ by 0.1): adjusted EBITDA, less stock-based pay (which we count as a cost rather than as new shares), capex, the change in working capital net of customer prepayments, funding for joint ventures and consolidated affiliates, interest and cash tax. Adding the asset-level debt that matures in the year gives the external capital needed, split into new debt and equity sold, with the net price a share at which the equity is sold: the 35-million-share ATM, at the market in the base and bull and, in the bear, at the case's own value, a sale after the market has seen the bad news. The last row is net debt over adjusted EBITDA at the end of 2027. The distress case restructures in 2027 and has no column.

Exhibit 2 — The debt by layer: what is owed, what it pays and what it yields now

principal, $ billionrate it pays · yield nowAsset-level secured (DDTLs, equipment)$26.5B9.5%9.0%Senior unsecured notes, 2030–32$10.0B9.2%15.5%Convertible notes, 2031–33$10.8B2.2%15.5%0%6%12%18%

How to read it. Each bar is one layer of CoreWeave's debt at 30 September 2026, its length the principal in billions of dollars. The grey tick on each bar is the coupon or contract rate the company pays; the gold dot is the yield the market priced it at around 10 November (asset-level loans from their 22 October marks, unsecured notes from 10 November, convertibles valued as unsecured debt). The distance from tick to dot is how much dearer that layer would be to refinance today. It shows cheap secured debt and a parent layer repriced to distress.

The floor

The floor: the signed book no longer covers the debt; the fleet does

July valued the contracted book at about $10 a share, 14% of the price: $98.8 billion of RPO run off at a 56% margin and an 11% discount, less $9.6 billion of construction in progress and $22.6 billion of net debt. That subtraction charged only the assets already being built. Delivering a book that size needs more capacity than CIP represents, and the CFO has now sized it: if the company pursued only its committed backlog, 2027 capex would stay in line with 2026's, about $38 billion.

We rerun it on the 30 September book: RPO of $132.3 billion recognized 36% within 24 months, 38% in months 25–48 and 26% after (deck, read from the page image), at the Q4 implied adjusted EBITDA margin of 66%; less the prepayments already received ($10.0 billion of the RPO is cash in hand and spent), plus the ~$9 billion arriving in Q4; less the 2027 build; less net debt; discounted at the company's own 7.9% cost of debt, since the contracts repay the lenders first. The $35.5 billion of leases signed but not yet begun (Q2 10-Q) are not subtracted again: they are the sites the backlog will run in, and we assume their rent costs what the fleet's does, so it sits inside the 66% margin we apply to that revenue. If half of those payments were stranded instead, rent paid on sites the book does not fill, the floor would be −$46.28.

−$29.30
the signed book's value a share after the debt and the build it needs

The book alone no longer covers the debt plus its build. That is a statement about the signed contracts, not about solvency: the GPUs bought to deliver them outlive the contracts, and the company's paid-off fleet is re-contracting now. Credit it in two steps. The ~300,000 Ampere and Hopper GPUs coming off contract in 2027–28 (outside the backlog, fully paid, no new capex) at 60% of the ~$10 billion of annualized revenue the company says they would earn at current pricing: −$9.10. Add one two-year renewal at 55% of contract pricing on the GPUs bought for the backlog: $7.68, 16% of the price. At the company's full renewal figure it is $21.15.

So the floor is about where July put it, a seventh of the price, but it now rests on the fleet re-contracting rather than on contracts already signed. The race between the backlog and the debt can now be scored: in Q3 the backlog grew 27% and net debt 45%. The debt won the quarter again, and the business won the margin.

Exhibit 3 — The contracted-book floor, from July's $10 to today

−$30−$20−$10$0$10$20$10.19July, published−$29.30Signed book, afterdebt and 2027 build−$9.10+ paid-off fleetre-contracting (60%)$7.68+ one renewal on thenew GPUs$21.15At the company'sfull renewal figure

How to read it. Value a share of the contracted book in dollars, each bar one variant: July's published floor, the 30 September book after the debt and the 2027 build, then the book with the paid-off fleet's renewals added at 60% of the company's figure, with one renewal on the new GPUs as well, and at the company's full figure. Bars below the zero line are books that do not cover the debt. Compare the second bar with the fourth: the gap is how much of the floor now rests on renewals.

The crux

The crux, corrected: how long a GPU earns, and at what price

July's table made the steady-state net margin a function of economic GPU life: 21% at six years, 14% at 4.5, 1% at three. Its levels rested on one input we can now measure. July charged capex of 1.4 times a year's revenue, derived from a management claim of a 2.5-year EBITDA payback. The 2026 build cost about $36 million per MW of active power added (nine-month capex of $30.65 billion on the company definition, construction in progress roughly flat, more than 850 MW added), and the capacity added in 2026 earns about $13.8 million per MW a year (the rise in annualized revenue from the end of 2025 to the end of 2026 over the gigawatt added). That is 2.61 times revenue, close to twice July's figure. The in-service asset base gives the same answer: depreciation was 52–55% of revenue all year at a six-year life.

Economic GPU lifeJulyMeasured nowNew contracts, base case
7.0 years23.8%5.5%26.0%
6.0 years21.0%0.2%21.1%
5.0 years17.0%−7.2%14.2%
4.5 years14.4%−12.2%9.5%
4.0 years11.1%−18.3%3.8%
3.5 years6.8%−26.3%−3.6%
3.0 years1.1%−36.8%−13.5%
Inputs: adj. EBITDA margin / capex ÷ revenue / interest ÷ revenue56% / 1.40× / 8%62% / 2.61× / 18%76% / 2.44× / 11%
Life at which net margin is zero2.9 years6.0 years3.7 years

How to read it. Each cell is the steady-state net margin a GPU earns over the economic life in its row: adjusted EBITDA margin, less capex spread over that life, less interest, after a 15% tax, all as a share of revenue. The columns are three sets of inputs, shown in the second-to-last row: July's, the 2026 build as measured, and new contracts in our base case. Negative cells lose money. The last row is the life at which the margin is zero; compare it down each column with the six-year depreciable life.

The first column is July's table. The second runs July's own formula on today's measured economics: at the 2026 build's cost and pricing, with interest at the 18% of revenue the current debt implies, the business breaks even only if a GPU earns its full rate for six years. The third is new capacity at the base case's $18 million per MW and an estimated $44 million per MW of capex for the next generation, with interest at the 11% of revenue the base case reaches by 2030: break-even at about 3.7 years.

That is the real crux now. It is less the life of a GPU than the rate new capacity earns, because a high rate pays back the GPU inside its contract. The evidence on rate since July:

The base case takes $18 million per MW on new contracts, below the quoted range, and $16 million on backlog capacity; GPU contracts of five years and one two-year renewal at 55% of contract pricing (an economic life of 6.1 years); legacy renewals at 60% of the company's figure; and pricing that holds through 2029, when the contracts signed in 2025–26 run out, then fades in a straight line to 2036, when a new GPU earns only the fleet's 10.5% after-tax cost of capital. The bear takes $15 million, a little above the 2026 build's $13.8 million, because part of the capacity coming online in 2027–28 is sold under contracts signed in early Q3, the quarter of the increase; renewals of two years at 40%; and legacy renewals at 35% of the company's figure. The distress case takes $14 million on backlog capacity and $12 million on new. Because a new GPU earns more than an old one renewed when pricing is strong, the model refreshes GPUs early in the base and bull, so a shorter book life is read through renewal pricing, not refresh timing (tripwire 7).

The value

What it is worth

Method. A levered equity whose secured debt is cheap and whose unsecured debt yields 15% is valued on its financing path, not on a multiple of a distant year. We build free cash flow to equity for 2027–2036 from the fleet: active power by vintage, each vintage earning its contract rate, then a renewal rate, then refreshed with new GPUs only if a new slot clears its cost of debt (the company calls this success-based capex), with stranded rent on power left idle. The debt runs off its ladder at its own coupons, and maturities and funding gaps are refinanced at the scenario's rates; customer prepayments enter as deferred revenue. Above a leverage cap, new shares are sold at the market price ($48.73 compounding at the cost of equity) in the base and bull, and at the case's own value in the bear and distress, a raise after the market has seen the bad news; distributions start only once net debt is at or below twice EBITDA. Stock pay is expensed, not also counted as dilution. July's two legs are retired: its P/E on fiscal 2028 earnings holds interest at a fixed share of revenue, so it cannot see the debt mature or the convertibles reprice, and its DCF at one 12% rate ignores how differently each layer of the funding is priced. The change of weighting is shown on its own line.

The terminal. The 2036 enterprise is valued at the fleet's unlevered return of 10.5%, and in every case that survives, the price of new capacity fades by then to where a new GPU earns about that return (9.9%–10.9% on capital across the cases), so growth adds or destroys little. That is a judgment about the business, applied to both tails alike: GPU capacity is rented, any operator with power and chips can add it, and no operator holds a cost or contract edge that outlasts a GPU cycle, so in the long run capacity is added until it earns its cost and withheld when it does not. The strongest objection is that the lasting asset is the powered site, not the chip. CoreWeave mostly leases its sites: operating lease liabilities were $18.7 billion at 30 September, up from $8.2 billion at the end of 2025, with more signed but not yet begun, and the 10-K gives data-center leases initial terms of five to fifteen years and a weighted remaining term of 11 years at the end of 2025, a figure that will move as the new leases begin. A scarcity rent on power goes to whoever controls the site at renewal: CoreWeave keeps some of it past 2036 where it holds options to extend at set rents, and the landlord takes it where leases reprice. We judge that most of it has repriced by the terminal. If CoreWeave kept it, the bull's 2.5-point edge forever would be worth $55.34 (1 point: $51.25); relative to keeping the edge, our rule takes $6.77 from the weighted value through the bull and adds $0.56 through the bear, whose equity is a call far out of the money. The cases differ in the decade to 2036 and in the debt they carry out of it, not in a permanent edge.

The equity in 2036. It is not simply enterprise less debt. With debt more than twice the equity in the base and above the enterprise in the bear, the equity is a call on the enterprise with the debt as its strike, and we value it that way. The volatility inside a case is 20% a year to 2036, the level of a mature contracted-compute or data-center lessor, less the spread the four cases already carry (a lognormal spread of 0.56 over the decade). CoreWeave's own equity has moved about 97% a year, about 40% for the enterprise at market weights, but that reflects the October funding shock, which the bear and distress cases already carry, and a 2036 call ignores the earlier point at which maturities and lender caps would hand a sinking enterprise to its creditors; we use the mature figure and show the higher one as a sensitivity.

Four cases. The distress case is the credit market's: pricing breaks to $12 million per MW on new capacity ($14 million on backlog capacity) in 2027–28 while debt peaks, renewals realize 20% of the company's figure, the parent pays 18% to refinance, and the 2027 build cannot be funded without selling shares worth almost nothing; we treat that as a restructuring and value the equity at zero. The bear avoids restructuring, because asset lenders keep funding the committed build at 10%, but pricing stays near the 2026 level, new capacity earns below its cost of capital through the decade, and by 2036 the enterprise is worth less than its debt; its $6.91 is entirely the option value of that enterprise recovering, which is why it is all terminal. The reason we hold volatility at the mature level (creditors would take control before 2036 if the enterprise kept sinking) makes $6.91 an upper bound; it is $1.73 of the weighted value. The same holds, less strongly, for the base, whose debt is about twice its equity in 2036: with no option value in any case, the weighted value is $44.33, below the price. The base is §4's rates. The bull takes the new-contract quotes nearer their midpoint, renewals at the company's full figure, investment-grade asset debt at 7.25% and the parent market reopening at 9.5%.

Distress (20%)Bear (25%)Base (35%)Bull (20%)
New capacity, $M per MW a year (backlog / new contracts)14 / 1215 / 1516 / 1818 / 20
Capex per MW, 2027$46M$44M$44M$42M
Contract years + renewal (years at share of price)4 + 1 at 30%5 + 2 at 40%5 + 2 at 55%5 + 3 at 75%
Legacy renewals, share of the company's figure20%35%60%100%
New asset-level debt / parent refinancing11% / 18.0%10% / 15.5%8% / 11.5%7.25% / 9.5%
Leverage cap (net debt / adjusted EBITDA)4.5×4.5×5.0×5.5×
Revenue 2027 / 2030— / —$25B / $50B$27B / $63B$30B / $81B
Net debt / adjusted EBITDA, end-2027—4.3×4.0×3.4×
Equity sold 2027–36$0.0B$0.2B$1.7B$1.7B
Restructured2027nonono
Terminal: competitive pricing / return on capital—$19.5M / 9.9%$18.7M / 10.1%$16.5M / 10.9%
Value a share$0.00$6.91$58.34$132.11

How to read it. One column per case, its weight in the heading. The top six rows are the inputs we set: pricing in millions of dollars per MW a year on backlog and on new capacity, capex per MW, contract and renewal terms, debt rates and the leverage cap. The next four are what the model produces from them: revenue, leverage at the end of 2027, equity sold over the decade and the year of any restructuring. The terminal row is the 2036 price of new capacity and the return on capital it earns. The bottom row is the equity value a share; weighted by the headings, the four give the fair value.

Stages. Each stage's length is argued from what drives it and checked against base rates of growth persistence.

BusinessStageYearsWhat drives itWhat sets the length
CoreWeave fleetContracted build-out44.6 GW of contracted power coming online (1.85 GW active at end-2026, guided), filled from the $132.3bn RPO and new contracts2027-30: the leases behind the contracted power commence between 2026 and 2029 (Q2 10-Q, 'Leases Not Yet Commenced'); 36% of RPO recognizes within 24 months and 38% in months 25-48 (deck p13); the bear slips a year to 2031
CoreWeave fleetContract-pricing era3New-capacity pricing held at contract levels ($16M/MW on backlog capacity, $18M/MW on new contracts in the base; CFO quotes $20-25M on 4-5 year contracts)2027-29: the 4-5 year contracts signed in 2025-26 lock pricing on the capacity they fill; from 2030 pricing fades in a straight line to the competitive rate by 2036
CoreWeave fleetRenewal wave4~300k Ampere/Hopper GPUs (130k in 2027, 170k in 2028) come off contract fully paid; re-contracted with no new capex2027-30 in the base (renewal terms of ~3 years after the 2027-28 roll-off); 2 years in the bear, 4 in the bull
CoreWeave fleetFade to competition6Contracting beyond the 4.6 GW (0.30 GW a year in the base through 2033) while a new slot clears its cost of debt; pricing fades until a new slot earns only its after-tax hurdle (about the 10.5% unlevered return in every surviving case: near zero-NPV growth)2031-36: one full GPU cycle (contract 5 + renewal 2 years) after the build-out, so every slot has been re-priced at least once by the terminal year; terminal growth 4.0% (Fed SEP 16 Sep 2026: 2.0% real + 2.0% PCE), bear 3%

How to read it. One row per stage of the fleet's growth in the base case, in order: its length in years, what drives growth in that stage, and the observable that sets when it ends.

Weights: distress 20%, bear 25%, base 35%, bull 20%, against July's 0/28/44/28, each argued on its own. CDS above 1,000bp on every reading since 16 October (about 1,150–1,280bp in early November, one source) imply risk-neutral default odds near 60% over five years. Market-implied odds include a premium for bearing default risk, so the real-world chance is lower; we judge it at a third to a half of the implied figure, which puts a default at 20–30%. That ratio is our judgment, not a measured one. We put the distress case at 20%, the bottom of that range, because the secured loans stayed close to par through October (marked at 97.75–99.5 on 22 October, one source, though DDTL 5.5 slipped about three points that month) and the company says it does not intend to issue unsecured debt through 2027 and will not use that market at current yields. The 20% also assumes the unsecured market reopens within 2027; yields still above 15% at mid-2027 move distress to 30% (tripwire 5). The bear gets 25%: it avoids a restructuring through 2027 but leaves the equity little, so 45% of the weight sits at or near zero. No market price anchors the bull the way the CDS anchors distress, so its 20% is our judgment, argued from its inputs: the company's own quotes and figures (new contracts near $20 million a megawatt, renewals at the full $10 billion, investment-grade asset debt, the unsecured market reopening), each with evidence behind it (signed price increases, two re-contracts above original prices, an investment-grade facility in process) and none yet in reported revenue. We set it equal to distress because the evidence for each tail is of the same kind: real, but not yet in the numbers. Between the tails, the evidence since July moved the body of the distribution toward the base, while leverage did not fall on any matched basis, so the left tail stays wide. July's weights, which had no zero case, would give $64.60; distress at 30% (base 30%, bull 15%) gives $39.05.

Cost of equity: 16.0% through 2030, then each case's own rate. The statistical reads are mostly noise. CoreWeave has 19 monthly returns since its IPO, and its own beta to the S&P 500 is 5.9 with a standard error of 2.1, which would imply a cost of equity above 30%. A peer prior gives 2.83 (a 17.6% rate), but two of its four peers carry bitcoin-mining histories and a third a two-year trading halt. We argue the rate from the capital structure instead. An asset beta of 1.2–1.3 for contracted compute, relevered at a market debt-to-equity of about 1.4 with a debt beta of 0.3–0.4 and no tax shield (NOLs and the interest-deduction cap mean an extra dollar of interest saves no cash tax for years), gives an equity beta of 2.3–2.7: 15.3–17.0% on a 4.88% ten-year yield and a 4.5% equity premium. We take 16.0%, the middle, just above the 15.5% promised yield on the company's unsecured notes, which rank ahead of the equity but whose promised yield exceeds their expected return. The same asset beta gives the fleet its 10.5% unlevered return. Each case's terminal rate relevers that beta at the case's own 2036 leverage, faded in from 2030: 18.3% in the base, where debt is still 1.9 times the equity, and 13.0% in the bull. At 14% to 2030 the value is $53.43; at the 17.6% peer-prior read, $45.07.

Exhibit 4 — From July's $60 to $48.57, one change at a time

$0$20$40$60$80$59.51July−18.19Net debt−1.92Shares+31.58EBITDA margin−3.50FY26 guide−18.76Interest burden−23.07Capex intensity+41.41Method+6.70Option value−9.16Cost of equity−16.03Weights$48.57This noteprice $48.73the quarter's data, on July's method

How to read it. Fair value in dollars a share. The black bar is July's value as its script computed it ($59.51, published as $60); each bar after it applies one more change to the running total, gold where it adds and grey where it subtracts, ending at this note's value in black. The first six changes are the quarter's data run through July's method; the last four are the change of method, the 2036 equity valued as a call on the enterprise, the discount rate and the weights. The order changes how the total splits, not the total.

ChangeKind$ a shareFair value after
Data: net debt $22.6bn -> $42.75bn (30 Sep 2026)data-18.1941.32
Data: shares 545.6M -> 575M (scenario paths shifted by the same 29M)data-1.9239.41
Data: adjusted EBITDA margin 56% -> 62% (Q3)data+31.5870.99
Data: FY26 revenue $12.5bn -> $13.15bn and capex $33bn -> $38bn (guide midpoints)data-3.5067.49
Data: interest 8% of revenue -> 18% (debt 2.3x exit-2026 ARR at the 7.9% weighted cost)data-18.7648.73
Data: capex intensity 1.4x -> 2.61x revenue (2026 vintage: $36bn per GW / $13.8bn per GW-year)data-23.0725.66
Method: fleet and financing FCFE model replaces the P/E and DCF legs (July's 28/44/28 weights, July's 14% base rate flat; equity at 2036 = enterprise less debt)method+41.4167.06
Method: equity at 2036 valued as a call on the enterprise with the debt as strike (within-case volatility 0.56)method+6.7073.76
Cost of equity: 14% flat -> 16.0% to 2030, then to each case's rate relevered at its 2036 leverage (base 18.3%)rate-9.1664.60
Weights: 28/44/28 -> distress 20 / bear 25 / base 35 / bull 20weights-16.0348.57
Fair value 59.51 → 48.57-10.9448.57

How to read it. Each row applies one more change to the running value, in the order shown, starting from July's $59.51 as its script computed it (published as $60). Kind says whether the change is data, method, rate or weights; the third column is its effect in dollars a share and the fourth the value after it. The first six rows are the quarter's data run through July's method (each scenario floored at zero); a different order would change the split, not the total.

What moved the number. The quarter's data, on July's method, take $60 to $25.66: net debt nearly doubled (−$18), the measured interest burden replaces July's 8% (−$19) and the measured capex intensity replaces 1.4 times (−$23), against a better EBITDA margin (+$32). The method adds +$41.41, because the fleet model sees what a fixed share of revenue cannot: revenue per MW rising as new capacity comes in at higher prices, the paid-off fleet re-contracting, and interest falling as a share of revenue as EBITDA grows faster than debt (from 23% now to about 11% in the base by 2030). Valuing the 2036 equity as a call on the enterprise adds +$6.70. The rate path, including the higher terminal rates the remaining leverage implies, changes it by −$9.16, and the weights, with 45% at or near zero, take off $16.03.

What the price requires. At $48.73, holding the base's other inputs: new capacity at $16.9 million per MW a year (base $18 million; the 2026 build realized $13.8 million); a cost of equity of 20.0% to 2030; asset-level debt at 10.8% every year; a GPU economic life of 5.5 years instead of the base's 6.1 (life here is contract years plus renewal years at a share of the contract price, so a shorter life is a lower renewal share); or fiscal 2028 revenue of $38.9 billion instead of $40.3 billion. Across the cases, it requires a distress weight of 20% (the other cases in proportion) or a bull weight of 20% (the base taking the difference). On the reverse grid, the required pricing is 91% of the base's at a 12% cost of equity and 100% at 20%. Two levers matter less than they look. Dilution: the base reaches the price only at 695 million shares outstanding today, 21% more than 575 million, because it sells just $1.7 billion of equity over the decade; selling shares near their own value is what costs the equity in the bear and distress cases. And the length of the expansion: ending the contracting beyond 4.6 GW in any year from 2027 to 2036 gives the base $57.62 to $58.52, because new capacity there earns close to its cost, so the price is not a bet on how long the expansion lasts. The price, in short, asks new contracts to earn about $1.15 million a megawatt a year less than we assume, or GPUs to earn for about half a year less: still well above what the 2026 build earned.

Sensitivities (weighted fair value): new asset-level money at 10% instead of 8% in the base, $46.20; parent refinancing at today's 15.5%, $48.12; base pricing of $15 million new / $14 million backlog, $37.70, or $20 million new / $18 million backlog, $54.49; renewals at 30% of the company's figure, $44.63, or 100%, $51.16; capex of $50 million per MW, $43.76; prepayments on the bear's path, $47.65. The rules that decide when the company must sell shares or may pay out matter too: a bear leverage cap of 4.0× instead of 4.5× forces a 2027 restructuring and gives $46.84; every cap 0.5× looser changes nothing, because the looser caps never bind; distributions allowed at 3.0× EBITDA instead of 2.0× give $50.91, at 1.5× $45.99. The option value turns on the volatility inside a case: none (equity at enterprise less debt) gives $44.33; today's regime fading to 20% gives $54.57. A permanent edge in the terminal, which we do not assume, gives $54.78 (the bull above and the bear below its cost of capital) or $55.34 (the bull's edge alone). About 44% of the base's value is terminal; the bear's is all terminal.

Cross-check: what the peers trade at. A relative method is shown at zero weight. On the 10 November closes, CoreWeave's enterprise (equity plus net debt, leases excluded) is $70.8 billion, 3.45 times its exit-2026 ARR guide, with debt 60% of it.

CompanyEnterprise valueNet debt shareForward run-rate usedEV / run-rateCoreWeave a share at that multiple
CoreWeave$70.8B60%exit-2026 ARR guide3.45×—
Nebius (converts at carrying value)$41.2B-1%year-end 2026 ARR guide $7-9bn (midpoint)5.14×$109.08
Nebius (converts at fair value)$53.5B23%year-end 2026 ARR guide $7-9bn (midpoint)6.68×$163.90
IREN$15.3B45%>$4bn ARR targeted for 31 Jan 20273.82×$61.84
Oracle$465.9B21%FY27 revenue guide >=$90bn5.18×$110.20

How to read it. One row per company, at 10 November closes. Enterprise value is equity plus debt less unrestricted cash, in billions, with operating leases left out; net debt share is net debt over enterprise value. EV / run-rate divides enterprise value by the forward run-rate named in its column, which is defined differently by each company. The last column applies that multiple to CoreWeave's exit-2026 run-rate guide, subtracts CoreWeave's net debt and divides by its shares. Nebius appears twice, with its earlier convertibles at carrying value and at fair value.

Three things keep it a cross-check rather than a method. Nebius is the only GPU cloud in the set, and its balance sheet is from June: we add the $0.8 billion secured facility and $5.75 billion of convertibles it raised since and the cash they brought, but not its unreported spending since June, so its net debt is understated and its multiple, and the CoreWeave value read from it, is if anything too low; its earlier convertibles are so far in the money that its value moves by $12 billion between carrying and fair value. The multiples price growth stage more than economics: Nebius's ARR is guided to roughly 2.7 times in a year from a small base, while CoreWeave's exit-2027 target of more than $30 billion, set in the February outlook presentation and not restated on this call beyond a promise to exceed it meaningfully, is about 1.5 times its exit-2026 guide. And the definitions differ: IREN's ARR is contracted capacity, not revenue, so its $61.84 is no support for our value, and Oracle's guide is whole-company revenue. What the cross-check does say is that the market prices CoreWeave's run-rate at the low end of the group, about half Nebius's multiple, consistent with debt at 60% of its enterprise value against −1%–23% at Nebius. Applied Digital, CoreWeave's landlord at Polaris Forge, is a lessor rather than a peer and is left out.

Exhibit 5 — Return on capital against its cost, four quarters

−4%0%4%8%12%16%−0.2%Dec '25−0.7%Mar '26−0.8%Jun '26−0.2%Sep '26WACC 12.7% (test)cost of equity 16.0% (reference)new capital earned −0.2%model assumes 10.1%

How to read it. Trailing-twelve-month operating income after the tax it would bear (none while losses run), over average capital (equity, debt and customer prepayments, less unrestricted cash), at each quarter-end, December 2025 to September 2026. The solid line is the 12.7% WACC, the test; the dashed line is the 16.0% cost of equity, for reference. The marker at the right edge sets the return earned on new capital over the window beside the 10.1% the model assumes on new capital in its terminal value. Read the gap between the bars and the solid line: below it, the business earns less than its capital costs.

Return on capital. On a trailing basis CoreWeave has earned about zero on capital for as long as the history allows (−0.2% at September against a WACC of 12.7%: weakness), because capital rose from $21 billion to $57 billion in twelve months while trailing operating income stayed between −$230 million and −$45 million. Capital here includes the $10.0 billion of customer prepayments, which fund the build and on which the company itself books interest, and excludes operating leases, whose rent sits inside EBITDA. The WACC is built on the same capital, at market weights: equity $28.0 billion at 16.0% (36%), asset-level debt marked near par at about 9% (34%), unsecured notes at 80 cents yielding 15.5% (10%), convertibles valued as unsecured debt at the same yield (7%), and the prepayments at book (13%), charged the 9% secured yield because they fund the build in place of secured draws on the same contracts. The interest the company books on them implies a lower rate, which would lower the WACC. It overstates the expected cost of capital, because promised yields on distressed paper exceed expected returns; it is the hurdle at today's prices. Q3 is the first sign of a turn: operating income annualized is a 1.1% return, and the Q4 guide about 3.5%. Our base needs the fleet to earn 10.1% on capital in the terminal value, about its 10.5% unlevered cost of capital.

The receipt

The receipt: July's tests and claims, scored

fired, passed or holds reversed, corrected, superseded or widened not fired, not scored or pending
FIRED (toward the bull case)First-generation contracts renew at or near original rates vs at steep discounts

Re-contracts priced at or above their original rates, in three filings and on the call: an A100 (2020 silicon) contract renewed for three years into 2029 at pricing 'in line with typical 1-year terms' and an H200 three-year renewal at a 'premium to original contract' (convert deck, 8-K 17 Sep, slide 10, read from the page image); Ampere and Hopper pricing up 30-50% quarter over quarter and above original contracts (Q3 call, three transcriptions); more than $3bn of the ~$9bn of Q4 prepayments is for re-contracted, fully paid-off infrastructure 'sold at meaningful premiums to their original price'. Sample size: two named contracts plus the pricing statement; ~300k Ampere and Hopper GPUs (130k in 2027, 170k in 2028) come off contract (the company gave GPU counts, not power, so the share of the fleet is not stated) and none of that renewal revenue is reported yet. The company sizes it at ~$4bn of annualized revenue (2027) and ~$10bn including 2028; the base case realizes 60%. Fired on price; the volume is the new tripwire (cwq3-renewal-realized).

NOT FIREDCompany guides to, or reports, a positive free-cash-flow year

No crossover date: 'free cash flow' is not said on the call. Q3 free cash flow $-7.5bn (operating cash flow $1.98bn, of which deferred revenue +$2.13bn, less capex $9.44bn), nine months $-17.9bn; FY26 capex guide raised at the low end to $37-39bn and 2027 capex 'in line with' 2026 on committed backlog alone. What was said instead: EBITDA less interest 'to more than double in 2027', adjusted net loss to 'reduce meaningfully in 2027' and adjusted net income positive 'in the first half of 2028 and for the full year 2028'. The base case has free cash flow to equity turning positive in 2029. Carried as cwq3-adj-net-income.

NOT FIRED (PENDING-10Q)Depreciable life of computing equipment shortened from six years toward four

The release is silent and the call never states the policy; the CEO said contracts now extend 'beyond the GPU's depreciable lives' (company transcript). The six-year life (extended from five in 2023) stands in the FY25 10-K. The Q3 10-Q property note is the reading [pending]; the FY26 10-K is the next scheduled one (cwq3-useful-life). Q3 D&A fell to 51.6% of revenue from 54-55% because revenue grew faster than the in-service base, not because of a policy change.

ACCESS PASSED; PRICE FAILED FOR UNSECUREDThe $7.5bn current maturity is refinanced; terms cheap+secured vs expensive or equity-heavy (due 2027-03-31)

Access: the stack grew from $24.9bn (Mar) to $46.9bn (Sep) through DDTL 4.0 ($8.5bn, SOFR+2.25%, Mar), DDTL 5.0 ($3.1bn, SOFR+4.50%, May), DDTL 5.5 ($2.6bn, SOFR+5.50%, Aug), 9.75% 2031s ($2.75bn, Apr), 9.625% 2032s ($1.25bn, Jun), 8.50% EUR 2032s (EUR 2bn, Jun), 1.75% 2032 converts ($4bn, Apr) and 2.875% 2033 converts ($4.2bn, Sep); no common stock sold for cash in Q3 (ATM established, 35M shares). Price: secured spreads widened 2.25 -> 4.50 -> 5.50 over SOFR within the year; the unsecured notes, issued at 9.6-9.75%, traded near 15% (2032s below 80c, 14 Oct; 15.5% on the 2031s on 10 Nov) and CDS passed 1,000bp on 15 Oct. The company's answer: no unsecured issuance 'at current yields', an investment-grade DDTL 'in the coming weeks' at a 7-7.5% yield financing >90% of capex, and ~$9bn of prepayments. Current debt is now $10.6bn against $4.1bn of cash. Retired: access is settled, price splits by seniority. Replaced by cwq3-ig-ddtl (secured) and cwq3-unsecured-yield (parent).

PENDING-10QMaterial expansion or contraction of the Microsoft revenue concentration

Q2 10-Q: Customer A 36% of revenue (71% a year earlier), Customer B 26%, Customer C 10%. Microsoft is not named on the call or in the deck. The Q3 10-Q customer table is the reading [pending].

NO PRIMARY PAIR; SECONDARY GAP WIDENEDCRWV secured-vs-unsecured differential at the next issuance vs ~334bp

No straight secured and unsecured pair has priced since July: the September deal was a 2.875% convertible (an option-adjusted coupon, not a clean unsecured read). In the secondary market the gap widened: secured loans marked 97.75-99.5 (DDTL 5.5, a SOFR+5.50% loan, ~10% at its mark; PitchBook 22 Oct, single source) against unsecured notes near 15.5%, about 550bp; the investment-grade DDTL the CFO describes (7-7.5%) would put the primary gap near 800bp. The note's reading rule (widening = the market hardening on the operating company) is met on the secondary read; the primary read comes with the IG DDTL. Score in neocloud-credit-2026 only when that facility prices.

PENDING-10QCoreWeave discloses OpenAI above 10% of revenue

Q2: three customers at or above 10% (36/26/10%), anonymized. OpenAI is named in the release among 'expanded relationships' and in the deck; the Q3 10-Q customer table is the reading [pending].

NOT SCORED (no comparable print)A levered builder refinances at or below its prior comparable coupon/spread

CoreWeave's only print since the hike was the 2.875% convertible: a convert's coupon is lowered by the conversion option and is not comparable to the 9.625% straight coupon, so it does not score. Other builders: Lambda's $1.0bn secured DDTL at 6.78% fixed (Baa1, 1 Oct) and Nebius's $3.5bn secured facility at SOFR+275 (22 Oct) priced inside CoreWeave's DDTL 5.5 (SOFR+5.50%) but on different collateral and tenor; WhiteFiber's term loan at SOFR+600 (9 Nov) priced wider. CoreWeave's own investment-grade DDTL (7-7.5%, weeks away) would fire it on the secured leg only if its collateral is comparable to the DDTL 5.5's shorter-duration contracts. Carried.

CITED (due 2027-02-15)Any company other than Amazon shortens its server or network-equipment useful life

CoreWeave has not shortened (six years, FY25 10-K; Q3 10-Q pending). Read again in the FY26 10-K cycle.

SUPERSEDEDJuly: the contracted book is worth ~$10 a share, 14% of the price

Rerun on the $132.3bn book (36/38/26 schedule), net debt $42.75bn, and the capex the company says the committed backlog needs in 2027 (in line with 2026, ~$38bn): the book alone is worth $-29.30 a share. July charged only the $9.6bn of construction in progress against a book that also needed new capacity. Crediting the paid-off fleet's renewals (company figure at 60%) gives $-9.10; adding one renewal term on the GPUs bought for the backlog gives $7.68 (16% of the price). The floor no longer rests on the signed book; it rests on the fleet re-contracting.

RESTATEDJuly: ~86% of the price is a levered growth option

At $48.73 with a $7.68 floor that already needs renewals, about 84% of the price rests on business beyond the signed book; the share is similar, the price is a third lower.

CORRECTED (input error)July: steady-state net margin by GPU life, 6.0yr ~21% ... 3.0yr ~1%

The table charged capex of 1.4x revenue, derived from a 2.5-year EBITDA payback claim. The reported asset base shows 2.61x for the 2026 vintage ($36bn per GW against $13.8bn of revenue per GW-year), so at every life the table overstated the margin by roughly 15-20 points. July's direction (margin is a function of life) holds; its levels did not. The corrected arithmetic is §4's table.

REVERSEDJuly: adjusted operating margin fell from 17% (Q1-25) to ~1% (Q1-26) as depreciation caught up

1% (Q1) -> 5% (Q2) -> 10% (Q3) -> ~16-17% guided for Q4. Depreciation fell to 51.6% of revenue from 54-55% as revenue per active MW rose (pricing +25% in July and +10% later in the quarter).

HOLDS ON A MATCHED BASISJuly: $24.9bn of debt at ~7x adjusted EBITDA

Debt $46.9bn; net debt over trailing adjusted EBITDA 7.3x (July's 7x was net debt over FY25 EBITDA, 7.3x), and 4.7x on Q3 annualized (July's quarter annualized: 4.9x). Leverage is unchanged on either matched basis; debt and EBITDA both roughly doubled.

HOLDS (PENDING-10Q)July: GPUs depreciated over six years, extended from five in 2023

Unchanged in the FY25 10-K; Q3 10-Q pending.

WIDENEDNeocloud Fuse: ~334bp wedge between CoreWeave secured and unsecured

Secondary ~550bp at 22 Oct marks; ~800bp against the CFO's IG DDTL guide. See nc-wedge-next-print.

CONFIRMED IN FORMNeocloud Fuse: 'The fuse is a refinancing, not a detonation'

The October break was a funding-market event with no missed payment: unsecured yields ~15% and CDS >1,000bp (14-15 Oct), HY OAS +9bp the same day, so specific to CoreWeave and the neoclouds. Secured, contract-backed credit kept clearing (Lambda 6.78%, Nebius SOFR+275), unsecured did not. The company's answer is to stop needing the unsecured market: prepayments and investment-grade asset-level debt.

The tripwires

What carries the story to the next update

7–7.5%the yield the company guides for its next, investment-grade DDTL; above 9% or not priced by the 10-K, the base's new-money rate moves to 10%

Each tripwire has its reading rule set now, so the next update scores it rather than reinterprets it. Effects are on fair value at today's other inputs.

  1. Pricing of the next DDTL (CFO: investment grade, 7-7.5% yield, >90% of capex financed, 'in the coming weeks'). At or below 7.5% with an investment-grade rating: the base's new-money rate 8% -> 7.5% ($49.11); 7.25% throughout would give $49.37. Between 7.5% and 9%: no change. Above 9% (the base case reaches the price at 10.8%), below investment grade, or not priced by the 10-K: base asset-level rate 8% -> 10% ($46.2). Reading: 8-K / 10-K debt note: facility size, spread over SOFR, advance rate, ratings.
  2. Collection of the ~$9bn of prepayments and deposits guided for Q4 (CEO). Every case opens 2027 with ~$18.1bn of deferred revenue ($10.0bn at 30 Sep plus the ~$9bn guided for Q4, less ~$0.9bn recognized in the quarter). $13-20bn at year-end: no change. At or above $20bn, read as customers prepaying more of each contract: the base's prepayment path (deferred revenue at 50% of the year's revenue in 2027, falling to 20%) moves to the bull's (60% to 24%) ($49.27). Below $13bn: the bear's (40% to 15%) ($47.65). Reading: FY26 10-K: deferred revenue (current + non-current) at 31 Dec 2026 vs $10.0bn at 30 Sep; Q4 operating cash flow.
  3. Revenue actually realized from the 2027 roll-off (130k GPUs; company: ~$4bn annualized at current pricing). Annualized renewal revenue at or above $2.4bn (60% of the company figure): base holds. At or above $4bn: base realization 60% -> 100% ($51.16). Below $1.2bn: 30% ($44.63). Reading: Q1/Q2-27 calls and 10-Qs: renewal revenue, re-contract pricing vs original, utilization of paid-off fleet.
  4. Revenue per active MW at the exit of 2027 (ARR / active power). Exit-2026 is ~$11.1M (blended fleet). Exit-2027 at or above $13M: consistent with the base's $16M on new capacity. Above $15M: the base's new-contract rate $18M -> $20M ($54.49). Below $11.5M: $15M ($37.7). Reading: Q4-27 release: exit ARR and active power.
  5. Yield on CoreWeave's senior unsecured notes (9.75% 2031s / 9.625% 2032s). Below 11% on 30 Jun 2027: the parent market has reopened; base parent rate 11.5% -> 9.5% ($48.8), and distress 20% -> 10% / bull 20% -> 25% / base 35% -> 40% ($58.09). Above 15% (about today's 15.5%): the 20% distress weight assumes the parent market reopens within the 2027 funding year (the base refinances at 11.5%), so a market still shut at mid-2027 is new evidence; base parent rate -> 15.5% ($48.12) and distress 20% -> 30% / bull 20% -> 15% / base 35% -> 30% ($39.05). Between: no change. Reading: TRACE or Bloomberg/Reuters quotes on the 9.75% 2031s and 9.625% 2032s, 30 Jun 2027.
  6. Adjusted net income positive in H1-2028 and FY2028 (CFO, Q3 call). Positive in H1-2028: no change. Negative in H1-2028: distress 20% -> 30% / bull 20% -> 15% / base 35% -> 30% ($39.05). Positive in FY2027 already: distress 20% -> 10% / bull 20% -> 25% / base 35% -> 40% ($58.09). Either weight tripwire acts alone; both firing the same way does not stack. Reading: Q2-28 release (H1) and Q4-28 release (FY): adjusted net income.
  7. Depreciable life of technology equipment (six years since 2023). Six years held: no change. Shortened to five years or less: read as the company expecting less from aging GPUs, so the base renewal rate falls from 55% to 40% of contract pricing with refresh timing unchanged ($46.92). Lengthened to seven years or more: read the same way in reverse, renewal rate 55% -> 65% ($49.73). Neither is scored through refresh timing: the model would otherwise refresh earlier into new slots, which is a pricing assumption, not a life finding Reading: FY26 10-K, Note 1 / property and equipment.
  8. Capex per MW of capacity brought online in 2027. At or below $40M per MW: base capex per MW $44M -> $40M ($50.44). $40-50M: no change. Above $50M: $44M -> $50M ($43.76). Reading: FY27 10-K and Q4-27 deck: capex (company definition) less change in construction in progress, over MW added.
  9. Price against the case values. A close 30% or more below or above fair value ($34.00 / $63.14) for 20 sessions: re-underwrite the cases before the next print. The price itself is not an input Reading: Settled close (Robinhood close.price / Schwab closePrice).

What would change our view

We would move toward the bull case if the evidence tracks it: the next DDTL at or below 7.5% with an investment-grade rating; deferred revenue of $20 billion or more at the end of 2026; annualized renewal revenue from the 2027 roll-off reaching $4 billion; exit-2027 revenue per MW of active power above $15 million; 2027 capex at or below $40 million per MW; the depreciable life lengthened to seven years or more; and the unsecured notes below 11% by mid-2027. That path is worth $132.11. The mirror moves us toward the distress case: a DDTL priced above 9% or not at all, renewal revenue below $1.2 billion annualized, exit-2027 revenue per MW below $11.5 million, deferred revenue below $13 billion, capex above $50 million per MW, the life cut to five years or less, the notes above 15% at mid-2027, or an adjusted net loss in the first half of 2028, when the company has said it will be positive. Most of these move the base's inputs; two move the weights, each on its own. The notes above 15% at mid-2027, or a loss in the first half of 2028, takes the distress weight to 30% and the bull to 15%, and fair value to $39.05. The mirror: the notes below 11% at mid-2027, or adjusted net income positive in 2027, takes the distress weight to 10% and the bull to 25%, and fair value to $58.09. Because our value sits at the price, any of these moves it off the price. Two alone take our base case down to the price: new contracts on capacity added in 2027 earning $16.9 million a megawatt a year or less (backlog capacity $2 million less), read as the rise in annualized revenue over 2027 divided by the megawatts added in 2027 (the measure that gave $13.8 million for 2026), or new asset-level money at 10.8% or more.

The call

The call

CoreWeave's price fell through July's fair value while its business improved, and now sits on ours: $48.57 against $48.73, with our base case at $58.34. Each market is right about a different thing. The credit market is right about the signed book, which no longer covers the debt plus the build it needs, and about the parent's unsecured claim, priced at 15%; that is a case inside our distribution, and with the bear, 45% of our weight sits at or near zero. The business is right about the fleet. The equity's value rests on three things the October selloff priced as unlikely and the quarter made more likely: new capacity earning near the rates the company now quotes, the paid-off fleet re-contracting, and asset-level lenders funding the build at single-digit rates. More than half of the $48.57 comes from the bull's fifth of the weight; the history is short and the renewal revenue unreported, which is why the cases run from zero to $132.11.

Sources

Sources & Method

Filings. CoreWeave Form 8-K Exhibit 99.1, Q3 2026 earnings release (acc. 0001769628-26-000461, filed 10 November 2026): income statement, balance sheet, cash flows, non-GAAP reconciliations, backlog and guidance. Earlier releases for Q3 2025 to Q2 2026 (acc. 0001769628-25-000059, -26-000094, -26-000220, -26-000362). Q2 2026 Form 10-Q (acc. 0001769628-26-000366): debt table and maturity ladder, RPO schedule, property and equipment, leases not yet commenced ($35.5 billion), significant customers, prepayments (15–25% of contract value). FY2025 Form 10-K (acc. 0001769628-26-000104): six-year useful life, federal NOLs of $4.3 billion. Convertible notes 8-K (acc. 0001769628-26-000432, 22 September) with the indenture and capped-call confirmations; convertible marketing 8-K (acc. 0001769628-26-000429) and its investor presentation, read from the slide images; ATM prospectus supplement (424B5, acc. 0001628280-26-062362); DDTL, notes and Meta, Jane Street 8-Ks from March to August 2026. Form 4 and Form 144 filings 17 July – 10 November 2026: 80 Form 4 filings, 96 Form 144 notices, 10.1 million shares sold in open-market sales ($0.97 billion) by 10 reporting persons: officers, directors and one large holder (Magnetar, 4.4 million shares). The Q3 2026 Form 10-Q, which the release says will be filed on or about 10 November, had not posted to EDGAR when this note's figures were set; Q2 10-Q figures are labeled with their date.

Peers. Nebius Q2 2026 6-K and later 6-Ks; IREN fiscal Q1-27 10-Q and earnings 8-K; Applied Digital fiscal Q1-27 10-Q and release; Oracle fiscal Q1-27 10-Q and 8-Ks (cloud infrastructure revenue, guidance); closes for 10 November from the brokerage feed, matched to a second vendor's tape (archived in data/relative/).

Company materials. Q3 2026 earnings presentation (backlog schedule p. 13, RPO p. 15 and capex calculation p. 16, read from the page images) and outlook presentation; fourth-quarter 2025 outlook presentation (26 February 2026, p. 4: exit-2026 ARR $17–19 billion, exit-2027 ARR above $30 billion).

The call. 10 November 2026, 5:00 PM ET. Quotes are checked word for word against the company's proofread transcript, posted on its investor site on 11 November, and against three independent transcriptions (Investing.com, Gloom, Benzinga); all are archived in data/tells/.

Credit and market data. Bond prices and yields from Reuters (relaying Bloomberg and Tradeweb), CNBC and PitchBook LCD, October–November 2026; no TRACE print was available to us, and the November yields rest on one outlet. CDS from Reuters (S&P Global Market Intelligence) and CNBC. Peer financings from their own 8-K and 6-K filings and releases. FRED: high-yield option-adjusted spread (BAMLH0A0HYM2) and Treasury yields. CRWV settled close of 10 November ($48.73) and daily and monthly history from a brokerage feed; the after-hours trade ($52.96 at 01:04 UTC on 11 November) is reported, not an input. Fed Summary of Economic Projections, 16 September 2026, for long-run growth.

Method. Free cash flow to equity by scenario, weighted 20/25/35/20, mid-year from 10 November 2026: the fleet by vintage (end-2026 fleet at its realized rates; capacity added from 2027 at the scenario's rates), slot costs of $2.5 million per MW a year plus 10–12% of revenue, stock pay expensed, cash tax at 23% after NOLs and the 30%-of-EBITDA interest cap, the debt stack by layer refinanced at scenario rates, a leverage cap above which shares are sold (at the market price in the base and bull, at the case's own value in the bear and distress), distributions only at twice EBITDA or below, and a terminal enterprise at the 10.5% unlevered return, with pricing at a competitive return; the 2036 equity is valued as a call on that enterprise with the debt as strike, and its rate is relevered at each case's 2036 leverage. Contracted-book floor at the company's 7.9% cost of debt. The model and every input are in the downloadable workbook.

Limitations. Revenue per MW by vintage, capex per MW for the next GPU generation, renewal realization and the competitive return are house judgments, shown with their effects. The model rolls the whole end-2025 fleet's contract revenue off in 2027–28 and replaces it with the company's renewal figure; if the 300,000 GPUs are only part of that fleet, revenue is understated; the company's renewal and pricing figures are its statements, not reported results. Bond yields for November rest on a single outlet. The share count, debt ladder and customer table are from the Q2 10-Q and the Q3 release until the Q3 10-Q is read.

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 1.0.