CoreWeave Q3: Can a $132.5B Backlog Outrun Its Debt and Interest Bill?
The question this piece answers: can the money CoreWeave earns from its $132.5B backlog grow faster than its debt and interest costs, and was this quarter a turning point for the business model or just relief from a credit scare? Data as of the Nov 11, 2026 close (after-hours as of 4:46 pm ET).
1. A 24.7% one-day rebound that ended a credit scare
On Nov 11, CoreWeave (CRWV) closed at $60.77, up 24.71% on the day, and added about 1.2% after hours. That is an unusual move for a stock worth over $30B. The setup matters just as much: in the month before earnings, the stock fell almost without a break, from $80.26 on Oct 12 to a $43.66 close on Nov 5.
The trigger was the Q3 report released after the close on Nov 10. Revenue was $3.661B, up 168% year over year. Full-year revenue guidance was raised to $13.0–13.3B, and backlog (contracted revenue not yet recognized) reached $132.5B. Another number in the report stood out just as much: net interest expense of $846M for the quarter, 2.2x adjusted operating income of $379M.
So this was not simply a beat. Since mid-October, the market's worry had not been whether CoreWeave had customers. It was whether CoreWeave could keep borrowing, and at what price. According to Motley Fool (Oct 16), CoreWeave's credit default swap spread (in effect, the price of insurance against it defaulting) hit a record of about 1,195bp on Oct 15. One of its 2031 bonds yielded over 16%, and some bonds traded near 79 cents on the dollar. What the report really addressed was where the money will come from.
All three neoclouds fell more than 40% together before the report. CRWV and NBIS won back a large share of that on 11/11; IREN won back the least.
2. The business model: borrow to buy GPUs, then rent them out under contract
The simplest way to think about CoreWeave is equipment leasing backed by contracts. It signs multi-year compute contracts with AI labs, big tech companies and enterprises. It then pledges those contracts and the GPUs it will buy as collateral for delayed-draw term loans (DDTLs: project loans drawn down as construction progresses). With that money it buys chips, builds and powers data halls, delivers capacity, and collects monthly rent that pays down principal and interest.
Three features follow. First, revenue lags spending. GPUs are bought upfront, but revenue is only recognized month by month once capacity is live, so capex (spending on equipment and buildings) runs ahead of revenue. Q3 capex was about $14.5B, roughly 4x the quarter's revenue. Second, depreciation is heavy. GPUs lose value quickly, and depreciation weighs directly on operating profit. That is why the company leans on adjusted EBITDA (earnings before interest, taxes, depreciation and amortization, roughly what the business earns before those two big costs), which was $2.256B in Q3, a 62% margin. Third, interest is heavy. The equipment is almost entirely debt-funded, so interest comes straight out of profit.
So the useful way to judge CoreWeave is not revenue growth alone. It is a race between three lines: how fast contracts turn into revenue, how fast depreciation and interest grow, and whether the next loan can be raised at a tolerable cost.
3. Question one: can the backlog support guidance and a $20B+ run rate?
Bottom line: 2026 guidance looks safe. The $20–21B year-end run-rate target depends on how fast capacity is delivered in Q4. Customer concentration is improving but is still high.
Start with guidance. Revenue for the first nine months totals $8.314B, so hitting $13.0–13.3B for the year needs $4.69–4.99B in Q4. That is about 28–36% above Q3's $3.661B. It is a big step, but two things support it. Q3 itself came in above the top of guidance. And management's year-end annualized run-rate revenue target (ARR: year-end monthly revenue times 12) of $20–21B works out to about $1.7B a month, which means Q4 is a steep ramp that is already built into contracted delivery schedules. On the call, management said that of its contracted power, "more than 50% is attached to a contract where we expect customer delivery to commence by the end of this year." The company also expects more than 1.85 GW of active power by year-end, up from 1.7 GW at the end of Q3.
Next, how many years of revenue does $132.5B represent? At the $20.5B midpoint run rate, about 6.5 years. So the amount of backlog is not the constraint. The constraint is how fast it can be delivered, which depends on power, data halls and GPUs arriving on time. The gap between 4.6 GW contracted and 1.7 GW active is what capex has to fill. Management says qualified demand is about 10x deliverable capacity through end-2027. That is the company's own claim and cannot be checked independently.
Customer concentration is the strongest evidence on the other side. Microsoft alone was about 67% of FY2025 revenue. According to the Q2 10-Q (quarter ended June 30, 2026), the largest customer's share of revenue fell from 45% in Q1 to 36% in Q2, with the second- and third-largest customers at 26% and 10%. The top three still made up 72%. A secondary source says no single customer exceeds about 35% of backlog; we could not confirm that in a primary filing, so we do not rely on it. This quarter management emphasized diversification: 10% of the Fortune 500 are now customers, and more than half of 2026's new customers come from outside AI labs and tech. Even so, a large part of the $132.5B still comes from a handful of AI labs and hyperscalers. If they build their own capacity or renegotiate prices, CoreWeave feels it directly. A UBS analyst asked on the call about frontier labs building their own capacity; management's answer focused on "long term relationships" and gave no quantified protections.
One variable is easy to overlook: renewals. Management said 130k GPUs come off contract in 2027 and 170k in 2028. It estimates renewals will add close to $4B of incremental annualized revenue in 2027 and about $10B in 2028, based on prior-generation GPU pricing rising 30–50% quarter over quarter. If that holds, renewing older GPUs is profit that needs almost no new capex. But it is a forecast about future prices, and it is the first thing to fail if compute prices fall.
4. Question two: debt and interest, and whether the financing holds up
Bottom line: the financing holds up over the next two or three quarters, because the company has shifted the burden onto project loans secured by contracts and GPUs, and onto customer prepayments. But interest is still climbing fast, and operating profit won't fully cover interest until 2028 at the earliest. The "no more unsecured debt" claim is only half right.
First, the size of the debt. As of Sep 30, recourse debt (current plus non-current) was $41.6B and non-recourse debt was $5.27B, for a total of $46.87B, in line with the "about $46.9B" in the topic brief. Net of $4.12B in cash, net debt is about $42.75B. On interest, Q3 net interest expense was $846M, up 172.9% year over year, slightly faster than revenue growth of 168%. Q4 guidance is $1.05–1.15B, and the midpoint is another 30% above Q3.
The two sets of bars rise almost in step. Interest has stayed at 23–26% of revenue, and scale has not pushed it down yet.
The financing rests on three pillars.
1) DDTLs secured by GPUs and contracts. Management said most of the remaining capex will be funded with "delayed draw term loans matched to specific customer contracts." The next DDTL launches in the coming weeks, is structured to be investment grade (the higher tier of credit ratings), and is expected to yield 7–7.5%. The weighted average cost of debt is 7.9%, down 250bp from a year ago. About $4B of principal has already been repaid, nearly half of the first two DDTLs. The idea is to borrow against the customer's credit rather than CoreWeave's own; management said more than 75% of committed 2027 capex not yet attached to a DDTL is tied to investment-grade and enterprise customers.
2) Customer prepayments. Deferred revenue (cash customers have paid for services not yet delivered) was about $10.0B at the end of Q3: $2.27B current and $7.73B non-current. Management expects about $9B more in prepayments and deposits over the next three months, more than $3B of it tied to re-contracting existing capacity, and says it collected prepayments on 70% of 2026 deals. Prepayments carry no interest. The catch is that they are essentially the price customers pay to lock in capacity, so the more bargaining power customers have, the harder prepayments are to collect.
3) Parent-level capital. The "no more unsecured debt" claim needs to be checked word for word. Management said: "We will not use the unsecured debt market to fund our business at current yields." There are two qualifiers. One is "at current yields": the 2032 unsecured notes issued in April and June at 9.75% and 9.625% coupons yielded more than 16% by Nov 6. The other is that the company's own release lists an upsized $4.2B convertible senior notes offering among its financing activities; we could not confirm its pricing date in the materials we read. Management also said it expects to raise less than $5B of incremental parent-level capital through 2027. So the accurate version is this: the company has committed not to issue high-yield unsecured bonds at today's rates, and it will keep new parent-level funding small, but it has not committed to stop parent-level financing altogether.
The evidence on the other side is just as clear. Q3 operating cash flow was about $2B against $14.5B of capex, and the gap is filled by borrowing and prepayments. New borrowing in the first nine months totaled $27.05B. Full-year capex guidance of $37–39B implies another $13.4–15.4B in Q4. Management itself does not expect adjusted net income to turn positive until 1H28. Whether this model can keep going depends on credit markets continuing to lend against "GPUs plus contracts" at around 7%. October showed how quickly that door can narrow when sentiment turns.
5. Question three: is a 10% adjusted operating margin a turning point for neoclouds?
Bottom line: the quarter-over-quarter trend shows operating leverage starting to appear. But this is not yet a turn that covers the cost of capital, and there is not enough evidence to call it a business-model inflection.
Three pieces of evidence point to a turn. Adjusted operating margin rose from 1% in Q1 and 5% in Q2 to 10% in Q3. Q3 adjusted operating income of $379M beat guidance of $200–260M. Full-year adjusted operating income guidance went up to $1.275–1.375B, 26% higher at the midpoint than before. Management also laid out unit economics: about $40M per MW per year of revenue on short-term contracts and $20–25M on 4–5 year contracts, with infrastructure payback of about 6–18 months and 1–2 years respectively.
The evidence against is just as strong. First, 10% is below the 16% of a year ago. First-half margins were squeezed to 1% by the costs of bringing a large amount of new capacity online, and Q3 is a recovery from that low that has not yet returned to last year's level. Second, the profit turns into a loss as soon as interest is subtracted: Q3 GAAP operating income was $207M, the net loss was $709M, and the adjusted net loss was $529M, an adjusted net margin of -14%. Third, interest coverage is still getting worse. Q3 interest was 2.2x adjusted operating income, and the Q4 interest midpoint of $1.1B is almost equal to the low end of full-year adjusted operating income guidance.
The gap between the two lines is how far operating profit still falls short of paying the interest bill. It is narrowing, but it has not closed.
A more precise read on the neocloud model: adjusted EBITDA margins above 60% show that renting out GPUs is a healthy business at the gross level. The problem is the two capital costs, depreciation and interest. Operating profit will only cover interest consistently once revenue from renewed older GPUs (which need no new capex) becomes a bigger share and borrowing costs fall to investment-grade levels. Management's own 1H28 target for positive adjusted net income tells you the turning point is still ahead.
6. Spillover: Nebius, IREN, Oracle and upstream memory
The sympathy rally on Nov 11 is best read as one shared risk factor being repriced. Nebius (NBIS) closed up 15.44%, IREN up 5.63%, and Oracle (ORCL) up 8.82%. On Aya's US heat list, Micron (+11.7%) and SanDisk (+13.5%) ranked #1 and #5, with CRWV #3 and NBIS #4. On X, @TapeAndTicker described the move in memory stocks as a "memory bid" driven by CoreWeave's results.
There is a fundamental logic to the spillover, but its strength varies, so each group needs to be looked at separately.
- For peer neoclouds (NBIS, IREN), what spills over is mainly the outlook for financing access. All three grow through contract-plus-equipment-backed borrowing. CoreWeave's next investment-grade DDTL, at 7–7.5%, effectively resets the reference cost of funding for this kind of asset. But that is a spillover in sentiment and valuation; each company's actual funding terms depend on its own disclosures. Nebius reports before the open on Nov 12 ET. According to public reports, its earlier 2026 targets were $3.0–3.4B in revenue and $7–9B in year-end ARR (secondary sources, to be confirmed by the report). IREN's core asset is the five-year, roughly $9.7B GB300 compute contract it signed with Microsoft in Nov 2025, 20% prepaid. IREN rose only 5.63% on 11/11 and is still down about 40% since 10/12, which shows the market did not spread CoreWeave's good news evenly across the group.
- For Oracle, what spills over is how the market prices whether AI compute contracts will actually be delivered and paid for. Oracle is also defined by large remaining performance obligations and heavy capex. Its rally happened on the same day as CRWV's and the correlation is obvious, but we could not verify any Oracle-specific catalyst that day, so the gain should not simply be credited to CoreWeave.
- For upstream GPUs and memory, what spills over is whether demand will last. CoreWeave kept its $37–39B full-year capex guidance and said that if it focuses only on committed backlog, 2027 capex would be "in line with 2026." For suppliers, that means CoreWeave's orders are unlikely to fall off a cliff in 2027, but there is no promise of continued rapid growth either. Memory stocks rose far more than Nvidia (+3%) that day, which looks more like a sharp rebound after steep earlier losses than a direct read-through from new order information.
7. What the market is betting on
At the $60.77 close and about 551.5M shares (yfinance implied shares outstanding), CoreWeave's market cap is about $33.5B. Adding about $42.75B of net debt gives an enterprise value of about $76.3B, roughly 3.7x the $20.5B midpoint of the year-end run-rate target. On yfinance's definition, which includes lease liabilities, enterprise value is about $89.3B.
That structure means shareholders hold only a thin layer of the balance sheet: more than half of every dollar of enterprise value belongs to creditors. Two things follow. First, the stock is extremely sensitive to funding costs and asset values. In the month when bond yields went from about 9% to 16%, the shares fell more than 40%. Second, a report only has to show the company can still borrow for the stock to rebound sharply. As of the latest reporting date, short interest was about 74.2M shares, 23.8% of the float, which probably amplified the 11/11 move. yfinance shows a mean analyst target of $125.46 across 37 analysts, but many of those targets likely predate the report, and perhaps the credit scare, so they are of limited use.
Our reading of what the market is now pricing in: Q4 lands on guidance, the next investment-grade DDTL prices at 7–7.5%, and the ~$9B in prepayments arrive on schedule. If all three hold, the company gets through 2027 without diluting shareholders or issuing high-yield debt. That expectation rests largely on management's stated commitments. Q3 execution did beat guidance, so it is not without basis. But it leaves little room for outside variables: AI labs' ability to pay, secondary-market and renewal prices for GPUs, and credit markets' appetite for risk. If any one of these deteriorates, spreads could blow out again the way they did in October.
8. What to watch next
- Nov 12, before the open (ET): Nebius earnings. This is the first peer test of whether neocloud demand is strong across the board. If NBIS holds or raises its ARR target, CoreWeave's demand story looks representative. If it cuts, the group-wide rally on 11/11 was partly an over-extrapolation.
- The coming weeks: launch and pricing of CoreWeave's next investment-grade DDTL. Management's anchor is a 7–7.5% yield. Pricing at or below that range would show the GPU-backed lending channel is open. Pricing well above it, or a delay, would mean that even after giving up unsecured debt, the company still lacks a cheap alternative.
- Secondary bond yields and CDS. The 2032 notes yielded more than 16% on Nov 6. We do not yet have data on how far they fell after the report. If the stock keeps rising while bond yields don't come down meaningfully, the stock and bond markets are disagreeing about the risk, and that is worth taking seriously.
- The Q4 report, around Feb 2027 (date not yet announced). Four numbers matter: whether Q4 revenue lands in the $4.69–4.99B range, whether interest stays within $1.05–1.15B, how much of the ~$9B in prepayments actually arrived, and 2027 capex and revenue guidance.
- Customer concentration in the Q3 10-Q. Whether the largest and top-three customers' shares of revenue keep falling is the hardest evidence for or against the diversification story.
9. Conclusion
This report proved two things. Demand is real: $132.5B of backlog is enough to support 2026 revenue guidance of $13.0–13.3B. And management laid out a financing plan that avoids expensive unsecured debt: an investment-grade DDTL, about $9B of prepayments, and less than $5B of new parent-level capital. What it did not prove is that operating profit can cover interest yet. Q3 net interest of $846M was 2.2x adjusted operating income, and Q4 interest will rise to $1.05–1.15B. The 24.71% jump on 11/11 is best read as relief from a credit scare, not confirmation of a business-model turn. The key tests now are the pricing of the next DDTL and the pace of Q4 deliveries.
Appendix: Data and Sources
A1. Key financials and valuation ($M unless noted)
| Metric | Value | Basis / formula |
|---|---|---|
| Q3'26 revenue | 3,661 (Q3'25: 1,365) | Company release; +168% YoY |
| Adj. operating income / margin | 379 / 10% (Q3'25: 217 / 16%) | Release |
| GAAP operating income | 207 | Release |
| Net loss / adj. net loss | -709 / -529 (-14%) | Release |
| Adj. EBITDA / margin | 2,256 / 62% | Release |
| Net interest expense | 846 (Q3'25: 310); +172.9% YoY | (846-310)/310 |
| Q4 interest guidance | 1,050–1,150; midpoint +30% QoQ | (1100-846)/846 |
| Interest / adj. operating income | 2.23x | 846/379 |
| Interest / revenue | 23.1% | 846/3661 |
| Total debt | 46,873 | 9,183+32,417 (recourse) + 1,389+3,884 (non-recourse) |
| Cash & equivalents | 4,122 | Release (plus restricted cash 1,405+877) |
| Net debt | 42,751 | 46,873-4,122 |
| Deferred revenue, total | 10,002 | 2,269 (current) + 7,733 (non-current) |
| 9M revenue | 8,314 | 2,078+2,575+3,661 |
| Implied Q4 revenue | 4,686–4,986 | 13,000/13,300-8,314 |
| 9M capex / Q3 capex | 23,558 / ~14,511 | Release; TradingKey |
| Implied Q4 capex | 13,442–15,442 | 37,000/39,000-23,558 |
| 9M debt issued | 27,048 | Release |
| Backlog | $132.5B (+138% YoY; Q2'26: $104.2B) | Release; secondary source (Q2) |
| Backlog / run-rate midpoint | 6.46x | 132.5/20.5 |
| Market cap | ~33,517 | yfinance (~551.5M shares × 60.77) |
| (Market cap + net debt) / run-rate midpoint | 3.72x | (33,517+42,751)/20,500 |
| yfinance EV (incl. leases) | 89,275 | yfinance |
| Mean / median analyst target | $125.46 / $125 (37 analysts) | yfinance; may be stale |
| Short interest | 74.21M shares; 23.8% of float | yfinance |
| Largest customer share of revenue, 2026 | Q1 45%; Q2 36% (2nd 26%, 3rd 10%) | Q2 10-Q via sec-api.io |
Quarterly series (inputs for Charts 2–3): revenue Q1'25 982, Q2'25 1,212, Q3'25 1,365, Q4'25 1,572, Q1'26 2,078, Q2'26 2,575, Q3'26 3,661; net interest 264, 267, 311, 388, 536, 640, 846 (EDGAR "Interest Income (Expense), Nonoperating, Net"). Net interest / revenue for Q3'25–Q3'26: 22.8%, 24.7%, 25.8%, 24.9%, 23.1%. Chart 1 change = close ÷ 10/12 close - 1; 10/12 closes: CRWV 80.26, NBIS 210.06, IREN 34.80, ORCL 137.39.
A2. Timeline
| Date (ET) | Event |
|---|---|
| 2025-11 | IREN signs a five-year, ~$9.7B GB300 contract with Microsoft (20% prepaid) |
| 2026-04 / 06 | CoreWeave issues 2032 unsecured notes at 9.75% / 9.625% coupons |
| 2026-08-12 | Q2 10-Q: revenue $2.575B; largest customer 36% of revenue |
| 2026-10-14 to 16 | CRWV falls for three sessions; CDS hits a record ~1,195bp; 2031 bond yields top 16% |
| 2026-11-05 | CRWV closes at $43.66, the low of the period (intraday 40.58) |
| 2026-11-06 | 2032 notes yield more than 16% |
| 2026-11-10 after close | Q3 results and earnings call |
| 2026-11-11 | CRWV +24.71% to $60.77; NBIS +15.44%, ORCL +8.82%, IREN +5.63% |
| 2026-11-12 pre-market | Nebius reports |
A3. Aya heat data (US, close of 2026-11-11)
| Rank | Ticker | Heat | Discussion/Trading/Options | Change (list basis) | Rel. volume | Bull/Bear |
|---|---|---|---|---|---|---|
| 1 | MU | 88 | 89/84/100 | +11.7% | 0.9x | 100% bull |
| 3 | CRWV | 80 | 93/96/0 | +23.4% | 2.4x | 86% bull / 14% bear |
| 4 | NBIS | 75 | 87/90/0 | +15.6% | 1.4x | 87% bull |
| 5 | SNDK | 73 | 87/85/0 | +13.5% | 1.0x | 90% bull |
| 15 | ORCL | 61 | 77/67/0 | +8.5% | 0.6x | 97% bull |
Note: the list's price changes come from a program snapshot and differ slightly from final closes (CRWV +24.71%); the text uses closing prices.
A4. Market commentary (X, by handle)
- @thekilldigest: citing Reuters, CoreWeave will avoid unsecured debt markets after borrowing costs jumped and says cash flow from existing contracts covers current and future debt payments; "Demand is one story; paying for it is another."
- @Teddy_oost (quoting @Kairos_Res): the CEO expects $9B of prepayments and deposits over the next 3 months; with rising EBITDA and supplier financing, that should let CoreWeave avoid leaning on ATM sales or "prohibitively expensive unsecured debt."
- @sdub2019: interest $846M vs $310M (+173%); "interest is still eating about 23 cents of every revenue dollar."
- @graboosky: interest grew a little faster than sales; which line are you watching next quarter?
- @TapeAndTicker: NBIS +15.4%, ORCL +8.8%; SNDK and MU up on the "memory bid."
- Bloomberg @business (via this platform's 11-12 daily): a healthier CoreWeave reassures the AI market, especially Anthropic as it prepares its IPO.
A5. Sources
- CoreWeave IR, "CoreWeave Reports Strong Third Quarter 2026 Results," 2026-11-10, https://investors.coreweave.com/news/news-details/2026/CoreWeave-Reports-Strong-Third-Quarter-2026-Results/default.aspx
- CoreWeave Q3 2026 earnings call transcript (AOL / Motley Fool), 2026-11-10/11, https://www.aol.com/articles/coreweave-crwv-q3-2026-earnings-031911000.html
- Investing.com call transcript, 2026-11, https://www.investing.com/news/transcripts/earnings-call-transcript-coreweave-tops-q3-2026-forecast-and-shares-rise-57-93CH-5017354
- SEC EDGAR (edgartools): CoreWeave quarterly financials Q1'25–Q2'26; 10-Q (2026-08-12, 0001769628-26-000366); 8-K (2026-11-10)
- sec-api.io, CoreWeave customer concentration (based on Q2'26 10-Q), https://sec-api.io/insights/coreweave-crwv-2026-revenue-analysis-top-customer-concentration-exposure
- TIKR, "CoreWeave Borrowed at 9% Earlier This Year. Its Debt Now Yields 15%…," 2026-11, https://www.tikr.com/blog/coreweave-borrowed-at-9-earlier-this-year-its-debt-now-yields-15-going-into-tuesdays-report
- Motley Fool, "Why Is CoreWeave Stock Falling Today," 2026-10-16, https://www.fool.com/investing/2026/10/16/why-is-coreweave-stock-falling-today/
- TradingKey, CoreWeave Q3 capex of $14.511B, 2026-11
- yfinance: CRWV/NBIS/IREN/ORCL daily prices and CRWV company info, through 2026-11-11
- Web search: Nebius 2026 targets (Finviz/Moomoo and other secondary sources), IREN–Microsoft contract (techinformed and others), 2025-11 / 2026-11
- grok-x x_search: X posts, 2026-11-05 to 11-12
- Aya heat data, close of 2026-11-11
For information and research discussion only. Not investment advice.
<!-- AYA_PODCAST -->🎙️ Podcast script (Chinese, ~20 min, 泓观阳略): read it here
AI-generated by Aya Research from public data and disclosures, for information and research purposes only. Not investment advice.