CoreWeave Pops On Beat And Raise... Then Guides To A $1.1 Billion Quarterly Interest Bill As Debt Hits $47 Billion

The bar could hardly have been lower.

CoreWeave (CRWV) walked into tonight's third-quarter report with its stock down more than 73% from its June 2025 high of $183, its 9.75% 2031 bonds yielding north of 15%, its CDS above 1,100bps, and a freshly elected class of governors who would rather not see another data center built in their states, thank you very much. Options were pricing a 15.5% move in either direction, and after a week in which exactly one of 45 data center ballot measures went the industry's way, most of the nervous money was looking down.

Well, the neocloud delivered: revenue, margins and guidance all came in above expectations, and the stock immediately spiked as much as 10% after hours to $54, a welcome change for a name that had fallen after five of its six reports as a public company.

But as has been the case lately, it didn’t last: within minutes most of the gain was gone.

What did the market see once it got past the first page of the release?

Probably the two sentences that will matter most to anyone who owns the company's bonds (and after the past month, maybe its stock too) were tucked into the outlook: CoreWeave expects to pay between $1.05 billion and $1.15 billion in interest in the fourth quarter alone, and it is pulling capex forward, lifting the low end of its full-year range to $37 billion. In other words, the best quarter in the company's history still ended with $11.8 billion more debt than it started with.

First, The Good News (And There Is Plenty)

Here is what CoreWeave reported for the third quarter, with estimates from Bloomberg consensus:

• Revenue $3.66 billion, up 168% y/y and 42% q/q, beating the estimate of $3.56 billion and the top of the company's own $3.45-$3.60 billion guide
• Adjusted operating income $379 million, beating the $245.8 million estimate by 54% and blowing through the $200-$260 million guide
• Adjusted EBITDA $2.26 billion, beating the estimate of $2.05 billion; margin 62% vs. 57.6% expected
• GAAP operating income $207 million vs. estimate of $63.7 million
• Net loss $709 million, smaller than the $850.4 million estimate (but 6.4x the $110 million loss a year ago)
• Loss per share $1.27 vs. estimated loss of $1.46
• Interest expense, net $846 million, below the $860-$940 million guide, and up 173% y/y
• Revenue backlog $132.5 billion, up from $104.2 billion at June 30
• Active power over 1.7 GW; contracted power over 4.6 GW
• Cash and equivalents $4.12 billion, down from $5.52 billion at June 30
• Total liabilities $87.46 billion

Sequentially, revenue rose by more than $1 billion in a single quarter, a first for the company, and the adjusted operating margin climbed back to 10% from 5% in Q2 and just 1% in Q1. CEO Michael Intrator said margins reached "double digits a quarter ahead of plan", which is fair: the July price increases and a wave of new capacity showed up exactly where the bulls said they would.

The guidance was even better:

• Full-year revenue raised to $13.0-$13.3 billion from $12.4-$13.2 billion, which implies Q4 revenue of roughly $4.7-$5.0 billion against a Street sitting near the bottom of that range
• Full-year adjusted operating income raised to $1.275-$1.375 billion, a 26% increase at the midpoint, implying about $800 million in Q4, or more than double tonight's number
• Exit annualized run-rate revenue raised to $20-$21 billion from $18.5-$19.5 billion
• Year-end active power target unchanged at more than 1.85 GW

So far, so good. If this were a software company, we could probably stop here.

It is not a software company.

Now For The Bill: $1.1 Billion A Quarter, Payable To The Lenders

Here are management's own words on the two numbers that matter:

We now expect full year 2026 capital expenditures of $37 billion to $39 billion, raising the low end of our prior guidance range of $35 billion to $39 billion. This is as a result of our expectation to deploy infrastructure and deliver it to customers sooner than planned.

We expect interest expense for Q4 2026 in the range of $1.05 to $1.15 billion.

Start with interest. The midpoint of the Q4 range, $1.1 billion, is 30% higher than the $846 million CoreWeave just paid, almost three times the $388 million of a year ago, and annualizes to $4.4 billion. That is against exit run-rate revenue of $20-$21 billion, which means more than one in every five dollars the company collects at year-end is already spoken for by creditors before a single GPU is depreciated (and depreciation, at $1.89 billion in Q3, ate another 52% of revenue).

Put differently: even on the raised guide, and even assuming the midpoint of an operating income number that has to double in three months, Q4 interest expense will still be about 1.4x adjusted operating income. That is a big improvement from 2.2x in Q3 and the frankly comical 25x in Q1. It is also still upside down.

To be fair, Q3 interest came in below the guided range, and the operating line is finally moving faster than the financing line. The trouble is what sits underneath the interest line.

$47 Billion And Counting

CoreWeave ended September with $46.9 billion of balance sheet debt, up from $35.1 billion at the end of June, $21.4 billion at year-end and $14.0 billion a year ago. That is $11.8 billion of net new debt in 92 days, or roughly $128 million a day, weekends included.

A few more balance sheet items worth a look:

• $10.6 billion of that debt is current, due within twelve months, against $4.1 billion of unrestricted cash (plus $2.3 billion restricted)
• Operating lease liabilities, which sit outside the debt number, rose to $18.7 billion from $8.2 billion at year-end. Add them and the fixed obligations come to $65.6 billion
• Total liabilities of $87.5 billion sit on top of $4.0 billion of equity. CNBC noted this morning that debt-to-equity was 702% at the end of June; by our math it is now about 1,165%

Why does the debt keep growing faster than the revenue? Because the business still burns cash at a spectacular rate. Operating cash flow was a respectable $1.98 billion in the quarter. Cash spent on property and equipment was $9.44 billion. The difference, $7.5 billion, is the largest quarterly cash burn in the company's history, bringing the total to $17.9 billion for the first nine months of the year and $25.2 billion since the start of 2025. The hole was filled the usual way: $10.3 billion of new debt in the quarter, $27 billion year-to-date.

And that was before the capex pull-forward. Q3 capex came in at $14.5 billion, above the top of the company’s own $11.5-$13.5 billion guide and almost 8x the $1.85 billion of a year ago. With $16.2 billion spent in the first half, that leaves just $6-$8 billion for Q4 to land inside the new $37-$39 billion range. Note also that only $9.4 billion of that $14.5 billion was paid in cash during the quarter, even as payables and accruals fell, which means roughly $5 billion was financed some other way. One more for the 5pm call. Deploying sooner is good for revenue. It is also, by definition, borrowing sooner.

Here is what that looks like: after blowing through its own Q3 guide by $1 billion, CoreWeave’s full-year range implies capex gets cut in half in Q4. Either the spending was simply pulled forward, or the $39 billion ceiling is about to be raised again.

Which brings us to the funding question that Wall Street has been circling for a month. Regular readers will recall that Bernstein put the hole at $76 billion of incremental financing just to deliver the existing backlog: (see "Core-We've Got Bills To Pay", available to pro subs), while Goldman, which had just upgraded the stock to Buy, got to a far friendlier $22.9 billion in its base scenario, a number that rested on customers prepaying 20% of contract value. Strip out the prepayments and Goldman's own figure jumped to $36.7 billion.

Tonight CoreWeave leaned hard into that same assumption. In the company's telling: "Our growth is supported by multiple sources of funding, including financing backed by customer contracts, operating cash flow, and approximately $9 billion in customer prepayments and deposits expected in the next three months alone."

Roughly $9 billion of customer cash in three months would be a big deal: it is almost as much as the entire $10 billion of deferred revenue on the balance sheet today, and it goes some way toward what Wolfe Research asked for on Monday: 

“Investors need neoclouds to quantify how much financing they expect to raise to execute on existing backlog. We think outlooks that communicate when and which avenues of funding will be tapped will increase investor confidence.”

It is worth remembering who those customers are, though. A prepayment from a hyperscaler with $100 billion of free cash flow is one thing. A prepayment from an AI lab that is itself raising debt to fund its compute bill is the same circular dollar taking a different route. And the rest of the toolkit is not exactly cheap: the quarter's financing list includes $4.2 billion of converts and an at-the-market program for up to 35 million shares, which at tonight's price raises maybe $1.8 billion. Bob Lang of Explosive Options was blunter:

“They don’t have enough cash to service all that debt,” he said. “Unless some big orders come in, unless some big money flows come in for them in the next 17, 18 months, I think this is going to be the start of the end for CoreWeave, unless somebody comes in and wants to buy them. But who wants to buy all that debt?” 

Wells Fargo's Michael Turrin, who has an Overweight rating and a $115 target, is on the other side: he sees this year's capex as largely funded already, but even he pencils in 2027 capex of as much as $66 billion and at least $25 billion of additional debt to pay for it. When the bull case needs another $25 billion of borrowing at a 15% bond yield, the bear case more or less writes itself.

The Check Was In The Mail After All

One number did move the right way, and since it was the chart we were watching most closely going in, credit where due. Accounts payable, which had tripled in three quarters to $3.63 billion as CoreWeave took its time paying suppliers, fell by $705 million to $2.93 billion.

Accrued liabilities fell too, by $783 million to $5.64 billion, so the combined IOU pile shrank by $1.5 billion to $8.6 billion. That is still double the level of a year ago, but it is the first decline since the buildout went vertical, and it probably explains a good part of why cash capex ran so hot: some of last quarter's bills finally got paid. With borrowed money, naturally.

Suppliers looking for the next slow payer may want to glance one ticker over. IREN's payables hit $2.1 billion in its September quarter, up from $97 million in March.

About That $132.5 Billion Backlog

The headline backlog number, $132.5 billion, is up $28 billion in a quarter and 2.4x in a year. Impressive, but not exactly news: back in August the company had already disclosed more than $25 billion of new commitments signed in the first weeks of Q3. That suggests that in the remaining two-plus months of the quarter, CRWV added only about $3 billion on a net basis, roughly the same as the revenue that rolled off. Lumpy? Sure. But for a company whose equity story is the second derivative of bookings, a quiet August and September right before an election that turned on data centers is at least worth a raised eyebrow.

Then there is the matter of turning signatures into electrons. Contracted power jumped by more than 900 MW to 4.6 GW, but active power rose just 200 MW to 1.7 GW, after a 500 MW surge in Q2. Only 37% of what CoreWeave has signed is switched on, down from 41%.

And here is the odd part. The year-end target of more than 1.85 GW was left unchanged, which requires just 150 MW in Q4, the slowest pace in a year. Yet capex is being pulled forward because infrastructure is arriving early, and revenue is supposed to grow by more than $1 billion again. Either the power target is a layup that management chose not to raise, or the Q4 revenue step-up leans more on price than on megawatts. One for the call.

What about the politics? More than 1.3 GW of the contracted total is now outside the US (India, Indonesia and Spain got a mention), which is one way to deal with moratoria that Goldman thinks could freeze about 45% of planned US capacity. As for Texas, where Governor-elect Gina Hinojosa wants an immediate moratorium, CoreWeave has already sent a letter offering to work with her. No word yet on a reply.

A Billion More In Revenue, Twelve Billion More In Debt

So where does that leave the company that imploded a month ago and then became the market's beta barometer for the entire AI trade?

The operating story is better than the bears (HSBC, which tore apart Goldman's bull case last month, chief among them) expected: pricing is firm, margins are recovering, suppliers are getting paid and the Q4 guide is well ahead of the Street. CreditSights' Jordan Chalfin, who told CNBC before the print that fundamentals and pricing were strong and the next few quarters should be supportive, looks right about the near term.

The capital structure is another matter. Coming into tonight the unsecured bonds were trading at yields the market reserves for companies it expects to restructure, and a long list of junk bonds sits one CoreWeave accident away from the same fate.

50 Park's Adam Sarhan put the bear case this way before the print:

“If AI revenue dries up, then fair value is almost certainly lower than where they’re trading now. Where will their growth come from if not AI, and how will they service their debt if they can’t grow their way out of it?”

Tonight's answer is that AI revenue is very much not drying up. But the arithmetic of the quarter is hard to get around: revenue grew by $1.1 billion, debt grew by $11.8 billion, and the interest bill is about to grow by another $250 million. We have been warning since last October that the AI buildout is a debt story first and a technology story second, and nothing in this report changes that. CoreWeave can grow into its balance sheet only if the bond market keeps lending at prices that make the growth worth having, and at 15%, well, that is a narrow window. The beat buys time and probably a relief rally in the bonds. It does not buy $25 billion of new 2027 financing.

Then again, when Gartner expects more than 80% of neoclouds to disappear, being the one that can still borrow counts as a moat.

We'll update after the 5pm call, where the only questions that matter are the 2027 capex number and who is lending against it.