CoreWeave reported third-quarter results after the close on November 10, and on the surface they were what a battered stock needed: revenue of $3.66 billion, up 168% from $1.37 billion a year earlier, a revenue backlog of about $132.5 billion, and a raised full-year outlook. Shares rose between nearly 13% and 16.93% in premarket trading on November 11, depending on which paragraph of the same market wrap you trust, after losing more than 40% in a month. The less cheerful number sits a few rows lower: net interest expense of $846 million, growing faster than revenue, on a balance sheet now carrying roughly $47 billion of debt.
The quarter, in the company's own numbers
Chief executive Michael Intrator called it "a breakout quarter for CoreWeave", and the top half of the income statement obliges. Adjusted EBITDA rose to $2.26 billion, a 62% margin. Operating income was $207 million against $52 million a year ago, and operating cash flow reached $1.98 billion. Physical scale kept pace: active power passed 1.7 gigawatts after more than 200 megawatts were added in the quarter, and contracted power rose by more than 900 megawatts to over 4.6 gigawatts.
The bottom half reads differently. The net loss widened to $709 million, or $1.27 a share, from $110 million, or $0.22. Depreciation and amortization tripled to $1.89 billion. Adjusted operating margin fell to 10% from 16%. Capital spending (cash purchases of property and equipment) was $9.44 billion, about two and a half times the quarter's revenue, up from $6.42 billion in the second quarter.
Guidance moved up across the board. Full-year 2026 revenue is now $13.0 to $13.3 billion and adjusted operating income $1.275 to $1.375 billion, the latter about 26% higher at the midpoint than the previous range. Both clear consensus: LSEG analysts had expected $12.86 billion of revenue and Visible Alpha analysts $1.07 billion of adjusted operating income. Subtracting the $8.31 billion booked in the first nine months, the range implies fourth-quarter revenue of roughly $4.7 to $5.0 billion. Capex guidance is $37 to $39 billion, the low end raised from $35 billion, and the company expects to leave the year at an annualized run rate of $20 to $21 billion.
Why the stock had been falling
The slide was not a verdict on CoreWeave alone. On November 5, the S&P 500 fell 4.4% and the Nasdaq 5.5%, the index's worst day since April 2025, and CoreWeave dropped 9.8%. The Associated Press tied the rout to months of worry that expensive borrowing would slow data-center construction, to bubble talk, and to the midterms: Democratic wins nationwide raised the possibility that governments could pause new data-center construction, and voters across parties have grown wary of AI and of the power bills attached to it. Whether such pauses would be wise is a political argument. For CoreWeave the relevant arithmetic is that, by subtraction, about 2.9 gigawatts of contracted power is not yet active, and new construction is exactly where a pause would bite.
Then there is credit. Higher rates have raised borrowing costs, and some CoreWeave bond yields recently topped 16%. For scale, the 10-year Treasury yield stood at 4.84% on November 5, after trading above 5.30% in October. A borrower paying more than three times the Treasury rate is being told something by its lenders, whatever its customers are saying.
Bull and bear, reading the same filing
The bull case is the backlog. At $132.5 billion it is about ten times the top of this year's revenue guidance, up from about $104 billion at June 30, a figure that excluded more than $25 billion of commitments signed early in the third quarter. Customers are increasingly paying for the build: CoreWeave expects about $9 billion of customer prepayments and deposits over the next three months, and secured a $2.6 billion delayed-draw facility sized for shorter contracts. The release lists OpenAI among customers that expanded and claims nine of the ten leading foundation-model providers. A 62% EBITDA margin says the machines earn cash once they are switched on.
The bear case is everything between EBITDA and net income. Net interest went from $640 million in the second quarter to $846 million in the third, and the company guides $1.05 to $1.15 billion for the fourth. Recourse and non-recourse debt add up to about $46.9 billion, against $4.1 billion of cash and $4.0 billion of stockholders' equity. The quarter also brought about $4.2 billion of convertible notes and a first at-the-market program for up to 35 million shares: dilution on an installment plan. And the release discloses no customer concentration; Microsoft and Meta are not mentioned, so how much of that backlog rests on a handful of buyers is a question for the 10-Q.
What a GPU hour actually pays for
For engineers who rent this capacity, the income statement is a price breakdown. In the third quarter, depreciation and amortization equaled 52 cents of every revenue dollar, and net interest another 23 cents. Three quarters of the price, in other words, covers hardware wearing out and the money borrowed to buy it, before a single kilowatt-hour is paid for.
That sets a floor under GPU pricing that competition alone cannot push through. If accelerators earn their keep for fewer years than the depreciation schedules assume, or if cheap debt is refinanced near 16%, the floor rises. Prepayments and long commitments are how a provider hands part of that risk to the customer, so expect more of them in the contracts that land on your desk, and price reserved capacity as what it is: a share of someone else's loan.
The 30,000-foot read
CoreWeave is a wager that demand for compute outruns the cost of the money used to build it. This quarter demand won the top line and money won the bottom line, and both can stay true for some time. The backlog is real, signed and large. So is the interest bill, and it compounds while data halls wait for power.
What management controls: delivery against the 1.85-gigawatt year-end target, the prepayment mix, the pace of dilution. What it does not: the rate path, the politics of data-center power, and how long a GPU stays worth renting. Watch the fourth-quarter interest line against its guide, whether the $9 billion of prepayments arrives, and where those 16% bonds trade after a beat. The stock bounced on a good quarter. The bonds will say whether it was a good year.
Sources: CoreWeave third-quarter 2026 earnings release, SEC 8-K exhibit (revenue, margins, net loss, EBITDA, depreciation, interest, capex, backlog, power metrics, balance sheet, financings, prepayments, customers, guidance, CEO quote); CoreWeave second-quarter 2026 release, Business Wire text reprinted by 01net (Q2 backlog, capex and interest); TheStreet, Nov. 11 market wrap (premarket move, one-month decline, bond yields, consensus figures attributed to LSEG and Visible Alpha via CNBC); Associated Press via Boston.com (November 5 sell-off, index and CoreWeave declines, election context, Treasury yields). All company figures are company-reported; adjusted EBITDA and adjusted operating income are non-GAAP. The roughly $47 billion debt total, the 2.9 GW of not-yet-active contracted power, the implied fourth-quarter revenue and the cents-per-dollar shares are my arithmetic from the releases; debt excludes about $0.65 billion of finance leases. CAVEAT: the 16% bond yield is single-sourced to TheStreet, which names no specific issue. TheStreet gives two premarket figures (nearly 13% and 16.93% to $56.98) without reconciling them, and both predate the regular session. Customer concentration, including any Microsoft or Meta share of revenue, is not disclosed in the release; the 10-Q was not reviewed.