CoreWeave Shares Close Up About 19% After Q2 Revenue Doubles to $2.58 Billion and Backlog Reaches $104.2 Billion
The AI cloud provider beat sales estimates and raised full-year guidance, while its net loss widened to $626 million on rising interest costs.
CoreWeave's Big Quarter Doubles Revenue and Doubles Down on Debt
CoreWeave spent $2.624 billion running its business in the second quarter of 2026. It brought in $2.575 billion. That gap is small — about $49 million — but it means the company lost money just keeping the lights on, before counting a dime of interest[4].
And yet CoreWeave's stock closed up about 19% the next day[3]. Both of those things are true. They just answer different questions.
CoreWeave rents out Nvidia chips so companies can run AI workloads without building their own data centers. On Tuesday, August 11, it reported quarterly revenue of $2.58 billion, more than double what it made a year earlier[1][5]. That beat what Wall Street expected, and the company raised its full-year revenue forecast to $12.4 billion-$13.2 billion[1][2]. Shares rose about 14% in after-hours trading that night and kept climbing, closing up roughly 19% on Wednesday[2][3].
The Number Everyone Quoted, and What It Doesn't Mean
The figure that got the most attention wasn't revenue. It was backlog: $104.2 billion, up 246% from $30.1 billion a year earlier[1][10].
Backlog is the total value of contracts customers have signed for future work. It is not cash CoreWeave has. It's a promise of money to come, sometimes years out. CoreWeave says the $104.2 billion figure doesn't even include more than $25 billion in new deals signed in the first weeks of the current quarter[2]. Behind those contracts are some of the biggest names in tech and finance: Anthropic, OpenAI, Meta and Jane Street[1]. One deal alone — a Meta agreement running through 2032 — is worth $21 billion, on top of a prior $14.2 billion commitment between the two companies[4].
For a company trying to convince investors and lenders that it's worth backing, a number like that is powerful. It says: the demand is real, it's locked in, and it's coming from customers who can pay.
Why the Loss Got Bigger Even as Sales Doubled
Here's the collision. Revenue doubled. The loss did too.
CoreWeave lost $626 million in the quarter, more than double the $290 million it lost a year earlier[4][5]. The main driver wasn't the cost of running data centers. It was interest — $640 million paid on debt in the quarter, up from $267 million a year before[4][5]. Total debt now stands at about $35.6 billion, after CoreWeave raised more than $10 billion in new unsecured debt and convertible bonds during the quarter alone[4].
This is the arithmetic underneath the whole story. Building AI data centers costs enormous amounts of money upfront — CoreWeave is guiding to $35-$39 billion in capital spending this year, against projected revenue of just $12.4-$13.2 billion[1]. That gap has to be filled somehow, every quarter, and CoreWeave is filling it with borrowed money. Whether that's smart or reckless depends entirely on how long it takes those $104.2 billion in contracts to turn into actual cash — and whether the chips being financed are still useful by the time the bills come due.
There's a second number worth sitting with: three customers account for 36%, 26% and 10% of CoreWeave's revenue — 72% combined[4]. That's the flip side of landing giant contracts with giant companies. It means CoreWeave's future depends heavily on just a few renewal decisions made by a few executives elsewhere.
Two Ways to Read the Same Set of Numbers
Nobody disputes the figures. What people disagree on is what they mean, and that split showed up clearly in how outlets covered the results.
CoreWeave's own case is about time horizons. Building a data center takes years, and the contracts that pay for it get signed years ahead of when the cash arrives. Judged that way, adjusted EBITDA — a measure of operating profit before interest, taxes and certain other costs — came in at $1.5 billion, a 59% margin, which the company points to as proof the underlying business is healthy before financing costs are added in[1][10]. Analysts at Citi called it one of the cleanest quarters since CoreWeave's IPO, and some pointed out that older Nvidia A100 chips are still generating revenue under contracts running to 2029 — pushing back on the idea that GPUs become worthless within a few years[9]. Fox Business framed the quarter as evidence that AI capital spending generally is paying off[8].
Credit investors and short sellers read the same numbers and see a timing mismatch that gets worse, not better. Their case doesn't rely on opinion — it relies on price. In late July, credit-default swaps on CoreWeave's debt — contracts that function like insurance against a company failing to pay what it owes — traded at levels implying roughly 50% cumulative default odds[12]. Loan investors, spooked by widening spreads, pushed for tighter covenants, the contract terms that restrict what a borrower can do[12]. The Register put the debt and the growth in the same sentence, calling out both the $35.6 billion debt pile and the 72% customer concentration in one breath[4]. Forbes contributor Peter Cohan called CoreWeave a "risky growth machine," conceding the growth while questioning whether the funding model holds[6].
Meanwhile, the customers signing these giant contracts — Meta, OpenAI, Anthropic and others — have their own logic. Building their own data centers takes years they don't have. Renting from CoreWeave gets them computing power now, and locks in supply in a market where AI chips are scarce, without putting the construction costs on their own books[4]. From where they sit, CoreWeave carrying the debt is a feature of the deal, not a warning sign.
What the Coverage Reveals About the Story Itself
Coverage split largely along one line: does the headline lead with the stock price, or with the balance sheet?
CNBC and Bloomberg led with the market reaction — "surges," "pops," "booming demand" — market-desk language that treats investor approval as the story[2][3][7]. The Register, a UK trade publication, led instead with the debt figure sitting right next to the revenue figure, refusing to let one be read without the other[4]. Tech Times went further, combining backlog with new signings into a $129 billion figure CoreWeave itself never reported, and describing the CDS pricing as flat "50% default odds" rather than what it actually is — a market-implied probability, not a certainty[12].
Two of the more pointed pieces, from Fox Business and Forbes, were opinion columns rather than straight news reporting, something not always obvious from the headline alone[6][8].
What Nobody Can Answer Yet
The one thing that decides whether this quarter was CoreWeave's turning point or its warning sign is something no earnings release can settle: how long AI compute demand stays this strong.
If it holds, a $104.2 billion backlog outlives the debt used to build toward it, and the interest payments become a temporary cost of an infrastructure buildout[1][4]. If it slows, or if just one of those three major customers scales back, the debt doesn't shrink along with the revenue. Both the bulls and the bears are reading the same $35.6 billion and the same $104.2 billion. The next few quarters will decide which one of those numbers was telling the real story.
Summary
CoreWeave, a company that rents out Nvidia chips for artificial-intelligence work, reported second-quarter results after the close on Tuesday, August 11, 2026. Revenue was $2.58 billion, up about 112% from $1.21 billion a year earlier, and slightly ahead of the roughly $2.56 billion analysts expected[1][2]. The company raised its full-year 2026 revenue forecast to between $12.4 billion and $13.2 billion[1]. Shares rose about 14% in after-hours trading and closed up about 19% on Wednesday, August 12[2][3].
The number that drew the most attention was the backlog: $104.2 billion as of June 30, up 246% from $30.1 billion a year earlier[1][10]. Backlog is the total value of work customers have committed to buy in the future. It is not money in the bank. CoreWeave said the figure leaves out more than $25 billion in new commitments signed early in the third quarter[2]. Big names behind it include Anthropic, OpenAI, Meta and Jane Street[1].
The same quarter also showed the other side of the ledger. CoreWeave lost $626 million, more than double the $290 million it lost a year earlier[4][5]. The main reason was interest: net interest expense hit $640 million, up from $267 million[4][5]. Total debt reached about $35.6 billion[4]. And three customers made up 36%, 26% and 10% of revenue — 72% between them[4].
That is the real dispute. Bulls say a $104.2 billion book of signed, long-dated contracts from the most creditworthy buyers in tech makes the borrowing rational[8][9]. Bears say the contracts do not become cash fast enough to cover the debt, and point to credit-market pricing: in late July, default-swap contracts on CoreWeave debt traded at levels implying roughly 50% cumulative default odds[12]. Both sides accept the same reported figures. They disagree about what those figures mean.
The Event
CoreWeave released second-quarter 2026 results after the U.S. market close on Tuesday, August 11, 2026[2]. It reported revenue of $2.58 billion, an adjusted loss of $1.03 per share, adjusted EBITDA of about $1.5 billion, a net loss of $626 million, and a revenue backlog of $104.2 billion as of June 30[1][5]. The company raised full-year 2026 revenue guidance to $12.4 billion-$13.2 billion and guided third-quarter revenue to $3.45 billion-$3.6 billion[1][2]. Shares rose about 14% in extended trading and closed up roughly 19% on Wednesday, August 12; rival Nebius rose about 34% the same day[2][3].
Undisputed Facts
- Q2 2026 revenue was $2.58 billion, up about 112% from $1.21 billion a year earlier[1][5].
- Revenue backlog was $104.2 billion as of June 30, 2026, up 246% from $30.1 billion a year earlier[1][10].
- Net loss widened to $626 million, from $290 million in Q2 2025[4][5].
- Net interest expense was $640 million in the quarter, up from $267 million a year earlier[4][5].
- Operating expenses of $2.624 billion exceeded revenue of $2.575 billion, producing an operating loss of about $49 million[4].
- Total debt stood at about $35.6 billion, after more than $10 billion in unsecured debt and convertible bonds raised during the quarter[4].
- Three customers accounted for 36%, 26% and 10% of quarterly revenue — 72% combined[4].
- Full-year 2026 guidance was raised to $12.4B-$13.2B in revenue, with capital spending guided to $35-$39 billion[1].
- CoreWeave signed a $21 billion agreement to supply AI cloud capacity to Meta through 2032, on top of a prior $14.2 billion commitment[4].
The Pressure
Strip away the moralizing and blame. What structural realities persist regardless of which narrative wins?
- Build ahead of demand or lose the contract
- Customers buy compute years forward. To win a deal like Meta's $21 billion through 2032, CoreWeave must commit capital before the revenue exists[4]. That structurally forces borrowing, whatever management's preference[1].
- The funding gap is arithmetic, not sentiment
- Guided 2026 capex of $35-$39 billion sits far above guided revenue of $12.4-$13.2 billion[1]. The difference has to come from debt or equity markets every quarter. That is why credit spreads matter as much as bookings[12].
- Few buyers, enormous checks
- Three customers are 72% of revenue[4]. Concentration cuts both ways: it delivers backlog at a scale a diversified vendor could never sign, and it means renewal risk sits with a handful of decision-makers.
- Depreciation schedules are the hidden argument
- How long a GPU earns money decides whether this is a good business. CoreWeave points to A100 chips still under contract into 2029 as evidence of long useful life[9]. Skeptics assume faster obsolescence. Neither side can prove it yet; the answer arrives in future quarters.
Material realityThe verifiable facts are not in dispute. CoreWeave took in $2.58 billion in the quarter, spent $2.624 billion to operate, and lost $626 million after $640 million of interest[4][5]. It holds about $35.6 billion of debt and $104.2 billion of contracted future work[1][4]. It plans to spend $35-$39 billion this year while collecting $12.4-$13.2 billion[1]. Physically, that means more buildings, more power hookups and more chips going in now, paid for with borrowed money, against payments due later from a small number of very large, very well-capitalized customers. Whether that works turns on one thing no earnings release can settle: whether AI compute demand stays strong long enough for the contracts to outlive the loans.
Narrative as a weaponCoreWeave is shaping this story most actively. Its release leads with revenue growth, backlog and adjusted EBITDA — the measures that look best — and wants readers to judge an infrastructure builder by contracted future revenue rather than current net income[1][10]. Bullish market outlets amplify that by leading with the share price, which makes market approval the story[2][3][7]. Fox Business extends it further, treating one company's bookings as evidence that AI capital spending as a whole is working[8]. Pushing the other way are credit investors and short sellers, whose case is carried by trade and columnist coverage foregrounding the $35.6 billion debt load and 72% customer concentration[4][6]. The most aggressive bear framing comes from lower-tier aggregators that convert a credit-default-swap price into a flat claim of '50% default odds' and add words like 'spiral'[12] — a real market signal, stated with more certainty than the instrument supports. One housekeeping note for readers: the wave of AI-infrastructure earnings this week that moved alongside CoreWeave was Nebius and Supermicro[3][11]; Leidos and Solidigm did not figure in this reporting.
How Each Side Sees It
Each major actor’s view — how it frames things, its underlying incentive, and how it’s materially affected. Tap a side to read it.
Frames it asThe company's case is that it is building infrastructure, not flipping chips. A data center takes years and billions to build; the contracts that pay for it are signed years ahead. So the right way to judge the business is contracted future revenue against the cost of building, not this quarter's bottom line. On that measure the quarter was strong: $104.2 billion booked, plus more than $25 billion signed since[1][2]. Management also argues the assets last longer than critics assume — older Nvidia A100 chips are still under contract generating revenue into 2029, which pushes back the claim that GPUs go worthless in three years[9]. Adjusted EBITDA of $1.5 billion at a 59% margin is offered as evidence the underlying operation is profitable before financing costs[1][10].
WhyKeep the cost of capital down. CoreWeave must raise tens of billions more to hit $35-$39 billion in 2026 capex[1]. A higher stock price and tighter credit spreads make that cheaper, so every quarter is partly a pitch to lenders[12].
Impact on themShares closed up about 19% on August 12[3]. But interest expense of $640 million a quarter is already larger than the operating result, so financing terms directly determine whether growth turns into profit[4][5].
Frames it asBears do not dispute the backlog. They dispute the timing. Their argument is about the gap between when money goes out and when it comes in. CoreWeave spends now — $35-$39 billion this year — against revenue of $12.4-$13.2 billion[1]. The difference is borrowed. Interest is already $640 million a quarter and rising[4]. Meanwhile 72% of revenue comes from three customers[4]. If one renegotiates or slows down, the debt payments do not shrink with it. Their strongest evidence is not opinion but a price: credit-default swaps — insurance contracts against a company failing to pay its debt — traded in late July at levels implying roughly 50% cumulative default odds[12]. Loan investors also pushed for tighter covenants after CoreWeave's spread widened about 125 basis points[12].
WhyGet paid, or get paid to be right. Lenders want covenants and higher yields; short sellers profit if the equity falls. Both benefit from making the debt story visible.
Impact on themWider spreads raise CoreWeave's borrowing cost directly, which feeds back into the interest line that is already driving the loss[4][12].
Frames it asTheir case is capacity now versus capacity later. Building your own data centers takes years; renting from CoreWeave buys compute today. Signing long, large contracts — Meta's $21 billion deal running to 2032 is the clearest example — is how they lock in supply and price in a market where chips are scarce[4]. From their side, CoreWeave's leverage is a feature: someone else is carrying the construction risk.
WhySecure compute at predictable prices without putting the capital on their own balance sheets.
Impact on themThey hold the leverage. Because so few customers make up so much of the revenue, their renewal decisions set CoreWeave's future more than CoreWeave's own execution does[4].
Frames it asInvestors in this group read CoreWeave as a demand signal for the whole chain. The argument is that a rented-compute provider sees demand before the chipmakers report it, because customers book capacity ahead. Strong bookings therefore mean the spending cycle has not turned. That view showed up in prices: Nebius rose about 34% and AI server maker Supermicro also reported strong results the same week[3][11].
WhyConfirm that AI capital spending is still expanding, which supports valuations across chips, servers, power and data centers.
Impact on themThe S&P 500 and Nasdaq rose on August 12 as AI names climbed[11]. The same linkage works in reverse if bookings ever stall.
Like this article?
The Bias Ledger average rating 4.8
The same story, as framed by outlets across the spectrum, ordered least to most biased. The bias score (1 = straight, 10 = heavily spun) is an AI assessment of that framing — click an outlet to see its track record. The tell is the word choice or omission that reveals the angle.
| Outlet | Vantage | Bias | How they frame it | The tell |
|---|---|---|---|---|
| CNBC | U.S. center / business | 3 | "CoreWeave gains 19%, Nebius surges 34% in post-earnings neocloud rally" and "CoreWeave stock pops 14% as revenue doubles"[2][3]. | Leads with the stock move rather than the results. 'Pops', 'surges', 'rally' are market-desk verbs; the widening loss and debt appear but below the price action. |
| Bloomberg | U.S. center / business | 3 | "CoreWeave Revenue Surges on Booming Demand for AI Computing"[7]. | 'Booming demand' is the company's own frame. Demand is the subject; the financing cost that produced the loss is not in the headline. |
| The Register | UK tech trade press, skeptical house style | 4 | "CoreWeave revenue doubles as debt pile reaches $35.6B"[4]. | Puts growth and leverage in one sentence, so neither can be read alone. Also the only piece to foreground the 36/26/10 customer concentration — an emphasis choice that steers toward fragility. |
| Fox Business (Opinion) | U.S. right / business; Varney & Co. commentary segment, not a news desk report | 5 | "AI spending is working for CoreWeave" — a Varney segment treating the quarter as proof the capex wave pays off[8]. | Frames a single company's bookings as a verdict on private AI investment generally. Profitability is raised as a future question, not a present problem. |
| Forbes (Opinion) | U.S. center-right business platform; signed contributor column | 6 | "CoreWeave Stock Up 18% — Inside The Risky Growth Machine"[6]. | 'Risky growth machine' is a characterization, not a finding. The stock gain is granted in the headline and then reframed as the setup for a warning. |
| Tech Times | U.S. tech aggregator, high-alarm framing | 8 | "CoreWeave Q2 2026: $129B Backlog Grew $29.6B in Six Weeks Despite 50% Default Odds" and earlier "Bond Market Calls Time on GPU Debt Spiral"[12]. | Combines backlog and third-quarter signings into a $129B figure the company did not report as backlog, then presents a CDS-implied probability as settled 'default odds'. 'Debt spiral' and 'calls time' are verdicts. |
References
- CoreWeave Reports Strong Second Quarter 2026 Results — CoreWeave · The company itself — primary source, promotional by design
- CoreWeave stock pops 14% as revenue doubles on accelerating AI infrastructure demand — CNBC · U.S. business news, owned by Comcast/NBCUniversal; market-desk framing
- CoreWeave gains 19%, Nebius surges 34% in post-earnings neocloud rally — CNBC · U.S. business news, owned by Comcast/NBCUniversal
- CoreWeave revenue doubles as debt pile reaches $35.6B — The Register · UK enterprise-tech trade publication; skeptical, vendor-critical house style
- CoreWeave Q2 2026 earnings: revenue beats, losses widen — Quartz · U.S. business site, center-left editorial lean
- CoreWeave Stock Up 18% — Inside The Risky Growth Machine — Forbes (Opinion) · U.S. center-right business platform; signed contributor opinion column
- CoreWeave Revenue Surges on Booming Demand for AI Computing — Bloomberg · U.S. financial wire owned by Bloomberg LP; institutional-investor audience
- AI spending is working for CoreWeave: Kenny Polcari — Fox Business (Opinion) · U.S. right-leaning business network, Fox Corporation; Varney & Co. commentary segment
- CoreWeave Stock Surges Nearly 16% After Earnings, Analyst Says Old Nvidia GPUs Can 'Still Generate Revenue' — Benzinga · U.S. retail-investor financial media; trading-oriented, bullish tilt
- CoreWeave Q2 2026 slides: revenue doubles, backlog surges 246% — Investing.com · Commercial financial data portal; summarizes company-supplied investor decks
- AI infrastructure stocks surge after strong earnings from CoreWeave, Supermicro — Yahoo Finance · U.S. aggregator; market-performance framing
- CoreWeave CDS Hits 50% Default Odds: Bond Market Calls Time on GPU Debt Spiral — Tech Times · U.S. tech aggregator; traffic-driven, high-alarm headline style