Froth City Research · Short sale note · CRWV · August 24, 2026
CoreWeave: A Ranked Short
Five reasons, in the order we would underwrite them. The principal schedule comes first because it is already larger than the cash the business produces. Community opposition and insider selling do not make this list.
Executive summary
We are short CoreWeave. The ranking below is our judgment. The dollars under it are not. At June 30, 2026, future principal payments were $35,551 million, including $4,413 million for the rest of 2026 and $6,184 million in 2027. In the first half, cash from operations was $3,663 million. Cash purchases of property and equipment were $14,117 million. Principal repayments were $5,219 million, and cash interest paid, net of amounts capitalized, was $806 million. The operations do not fund the build or the debt service.1
Interest expense, net was $640 million in the June quarter, against a $49 million operating loss. A 100 basis point move in rates would change quarterly interest expense by about $30 million on the floating-rate balance then outstanding. Three customers were 36%, 26%, and 10% of that quarter’s revenue. Technology equipment is depreciated over six years, a life the company lengthened from five years in 2023 and has not shortened since. And 355 MW of electrical power under certain leases was still undelivered, with delivery expected in phases from 2026 to 2028.12
We looked hard at the other bear arguments in circulation. Local opposition in Kenilworth and Lancaster is real, and construction was still moving. The IPO put right, and the lock-up attached to it, terminated in September 2025. NVIDIA is the sole GPU supplier and a $2.0 billion shareholder. The June 2026 filing does not call NVIDIA a related party. Those points are in the last section because they do not, on the record we can cite, outrank the cash gap.
| # | Reason | Likelihood | Impact | Timing |
|---|---|---|---|---|
| 1 | Principal the operations do not fund | High | Severe | Rest of 2026 and 2027 |
| 2 | Interest already larger than operating income, with a floating slice | High | Severe | Now, and at each rate move |
| 3 | Concentrated revenue, and a backlog stated net of delays | Medium-high | Severe | Visible now; named contracts run to 2031–2032 |
| 4 | Six-year depreciation while the platform moves to Rubin | Medium | High | Rubin in the second half of 2026, then the next cycle |
| 5 | Undelivered power and capex still larger than operating cash | Medium-high | High | 2026–2028 |
- Scheduled principal
- $35.6B
- Due by year-end 2027
- $10.6B
- Q2 interest, net
- $640M
- Q2 operating loss
- $49M
- Unsatisfied RPO
- $103.7B
Contents
- 1 Principal the operations do not fund
- 2 Interest is the operating result
- 3 Concentration, and a backlog net of delays
- 4 A six-year life on a shorter product cycle
- 5 Power not delivered, capex not finished
- 6 What the price assumes
- 7 Risks to the short
- 8 The same pattern at other AI data-center names
- 9 Other considerations
01 Principal the operations do not fund
This is the short. At June 30, 2026, future principal payments were $4,413 million for the rest of 2026, $6,184 million in 2027, $4,416 million in 2028, $2,421 million in 2029, $3,221 million in 2030, and $14,896 million thereafter. The total is $35,551 million. That is a schedule of principal. It is not a facility commitment, and it is not the carrying value after discounts.1
Carrying value is lower because of unamortized discount and issuance costs. Recourse debt, net, was $31,405 million, of which $6,235 million was current. Non-recourse debt, net, was $3,663 million, of which $1,278 million was current. Debt net of discount was $35,068 million. Non-recourse debt is limited to specified subsidiaries except for bad-act guarantees. It still has to be serviced if those subsidiaries are to keep the assets. Management reported $13.6 billion outstanding under delayed-draw term loans and $16.6 billion of aggregate note principal.1
| Instrument | Maturity | Effective rate | June 30, 2026 | Dec 31, 2025 |
|---|---|---|---|---|
| DDTL 1.0 | Mar 2028 | 15% | 1,300 | 1,553 |
| DDTL 2.0 | Aug 2030 | 11% | 3,190 | 5,037 |
| DDTL 2.1 | Mar 2031 | 9% | 3,000 | 2,741 |
| DDTL 3.0 | Aug 2030 | 9% | 2,215 | 340 |
| DDTL 5.0 | Nov 2031 | 9% | 1,101 | None |
| 2030 Senior Notes | Jun 2030 | 10% | 2,000 | 2,000 |
| 2031 9.00% Senior Notes | Feb 2031 | 10% | 1,750 | 1,750 |
| 2031 9.75% Senior Notes | Oct 2031 | 10% | 2,750 | None |
| 2032 9.625% Senior Notes | Jul 2032 | 10% | 1,250 | None |
| 2032 EUR Senior Notes | Jul 2032 | 9% | 2,279 | None |
| 2031 Convertible Senior Notes | Dec 2031 | 2% | 2,588 | 2,588 |
| 2032 Convertible Senior Notes | Oct 2032 | 2% | 4,000 | None |
| OEM and software financing | Dec 2026–Jul 2030 | 11% | 4,220 | 3,518 |
| Magnetar loan | Jan 2029 | 12% | 189 | 273 |
| Unamortized discount and issuance costs | (427) | (242) | ||
| Recourse debt, net | 31,405 | 20,726 | ||
| DDTL 4.0, non-recourse | Mar 2032 | 7% | 2,837 | None |
| OEM financing, non-recourse | Aug 2026–Aug 2028 | 9% | 882 | 647 |
| Non-recourse debt, net | 3,663 | 647 |
Two 2026 facilities are easy to misread as cash already in hand. DDTL 4.0 is an $8,500 million commitment, available in draws through June 30, 2027, priced at SOFR plus 2.25% or at Treasury plus 2.00%. At June 30 the carrying amount was $2,837 million, and the filing describes $1.4 billion of floating-rate loans and $1.5 billion of fixed-rate loans outstanding. DDTL 5.0 is a $3,100 million commitment at SOFR plus 4.50%, with $1,101 million outstanding at June 30. After the quarter, a subsidiary signed DDTL 5.5, a $2.6 billion facility at SOFR plus 5.50% maturing in September 2031, and the company had drawn $1.2 billion. The stack got larger after the balance-sheet date.1
Set the schedule next to the cash flow statement, not next to the backlog. First-half operating cash flow was $3,663 million. Purchases of property and equipment were $14,117 million. Debt issuance, net, brought in $16,747 million, and repayments took out $5,219 million. Cash interest paid, net of capitalized amounts, was $806 million, and another $176 million of interest was capitalized. Unrestricted cash was $5,524 million. Operating lease liabilities were $16,319 million. Stockholders’ equity was $5,024 million. The accumulated deficit was $4,009 million.1
Likelihood is high because none of this is a forecast. The principal dates are contractual, and the half already closed with operating cash below debt service and far below capex. Impact is severe because the equity, at $5.0 billion of book value, is the residual under that schedule. The timing is the red bars in Figure 1.
02 Interest is the operating result
Interest expense, net was $640 million in the June quarter, up from $267 million a year earlier. For the six months it was $1,176 million, against $531 million. Operating income in the quarter was a loss of $49 million. Revenue was $2,575 million. Interest was 24.9% of revenue in a quarter that did not produce an operating profit.1
| Q2 2026 | Q2 2025 | H1 2026 | H1 2025 | |
|---|---|---|---|---|
| Revenue | 2,575 | 1,212 | 4,653 | 2,194 |
| Operating income (loss) | (49) | 19 | (193) | (8) |
| Interest expense, net | (640) | (267) | (1,176) | (531) |
| Net loss | (626) | (290) | (1,366) | (605) |
The debt footnote is a different cut of the same bill. Contractual interest was $592 million in the quarter and $1,075 million in the half. Capitalized interest was $79 million and $176 million. After capitalization and after amortization of discounts, interest on debt obligations in the footnote was $558 million for the quarter. The income-statement line is larger. Both moved the same way, and management attributed the increase primarily to more borrowing.1
The rate piece is quantified. For every 100 basis point increase or decrease, quarterly interest expense could move by about $30 million, and six-month interest expense by about $61 million, based on the floating-rate balance at June 30, 2026. The delayed-draw loans are the floating paper: DDTL 4.0 at SOFR plus 2.25% on the floating tranche, DDTL 5.0 at SOFR plus 4.50%, and the later DDTL 5.5 at SOFR plus 5.50%. The senior notes are the fixed, expensive slice, with coupons from 9.00% to 9.75% and effective rates of 9% or 10% in the table above. The two convertible series, at a 2% effective rate, are $6,588 million of the recourse lines. They are the cheap slice, not the book.1
We rank this second, not first, only because the principal calendar is the larger cash call. The interest result has already happened. Likelihood is high. Impact is severe. A lower rate helps, and the company’s own sensitivity says how much: about $30 million a quarter per 100 basis points, which does not retire a $640 million quarterly bill.
03 Concentration, and a backlog net of delays
The customer base is not “unprofitable companies, full stop.” Microsoft and Meta are profitable public companies. Saying otherwise would be false. The short is that revenue is concentrated in a few AI budgets, that the company expects that concentration to continue, and that the backlog investors cite is already defined to exclude part of what can go wrong.
In the June quarter, three customers were 36%, 26%, and 10% of revenue. For the six months the top two were 40% and 23%. Customers A and B were each 32% of accounts receivable at June 30. A year earlier, one customer was 71% of the quarter. The 2025 Form 10-K names that year’s dominant customer: about 67% of revenue came from Microsoft, and no other customer was 10% or more. The June 2026 letters are not a second source for that name. The filing says the letters can change identity between periods. Committed contracts, including capacity delivered before a commitment start date, were 98% of revenue.12
The forward commitments the company does name are large, and they are “up to” amounts. OpenAI OpCo, under a May 2025 master agreement and a September 2025 order form, committed to pay up to about $6.5 billion through May 31, 2031. In March 2026 Meta signed an order form described as an initial commitment of up to about $21.0 billion, covering new capacity through December 20, 2032 and an existing option through April 10, 2032. In April 2026 Jane Street, identified as a private company, committed about $6.0 billion. The risk factor lists Microsoft and OpenAI among significant customers and says early-stage and private customers can carry a higher risk of non-payment. We do not describe Microsoft, Meta, or Jane Street as loss-making.1
OpenAI is the name that fits a credit worry, and it does not file these figures with the SEC. Audited 2025 statements obtained by Ed Zitron and reviewed by the Financial Times, as reported by Quartz and Ars Technica in June 2026, show revenue of $13.07 billion and an operating loss of $20.92 billion. The headline net loss was about $38.5 billion and included a large charge from the conversion out of a nonprofit structure. We use the operating loss, which those accounts agree on. We have not inspected the statements. A $6.5 billion commitment is not most of CoreWeave’s revenue today. It matters because the company says it expects OpenAI to be a significant customer, and because that buyer’s reported operations did not fund themselves in 2025.4
The backlog has to be read with its own footnote. Unsatisfied remaining performance obligations were $103.7 billion at June 30, 2026. CoreWeave expected to recognize 41% over the 24 months ending June 30, 2028, 39% between months 25 and 48, and the rest between months 49 and 78. RPO is the transaction price allocated to work not yet delivered, net of estimated variable consideration. That variable consideration includes estimated future service credits, amounts that may not be recognized because of delivery delays, and committed capacity the company has the right to resell. Deferred revenue, which is cash already collected or billed ahead of delivery, was $9.7 billion, up from $8.2 billion at year-end. The backlog is a real number. It is not cash, it is not the capex required to deliver it, and the company has already reserved inside it for delay.1
We rank this third. The concentration is a fact, so the likelihood that one customer’s budget matters is high. The likelihood of a break in that budget is the part we cannot read out of a filing, which is why the line in the table says medium-high rather than high. If it happens, the impact is severe, because 98% of revenue is committed capacity and the debt does not flex with it.
04 A six-year life on a shorter product cycle
Effective January 1, 2023, CoreWeave changed the useful life of computing equipment in its data centers from five years to six, citing hardware performance, software optimization, and data-center design. The change reduced 2023 expenses by $20 million. The 2025 Form 10-K still lists technology equipment at six years, straight-line. The June 30, 2026 Form 10-Q says there have been no material changes to those significant accounting policies. This is not a hidden assumption. It is a disclosed one, and it has been lengthened, not shortened.12
The dollars under the assumption got much larger. Technology equipment, gross, was $33,823 million at June 30, 2026, up from $20,903 million at December 31, 2025. Construction in progress was $11,918 million. Gross property and equipment was $52,622 million. Accumulated depreciation and amortization was $5,886 million, so net property and equipment was $46,736 million. Depreciation and amortization of property and equipment was $1.4 billion in the June quarter and $2.5 billion in the half, against $553 million and $996 million a year earlier. Technology and infrastructure expense rose mainly because depreciation of servers, switches, and networking equipment went to about $1.3 billion in the quarter.1
The company describes the other clock in its own risk factors. It cycles out older components for the latest technology, the useful-life estimate can be wrong, and an inability to redeploy equipment past its contracted life would hit results. It also says it expects to deploy the NVIDIA Rubin platform in the second half of 2026, and that customers may be slow to adopt a new generation. Current customer contracts require NVIDIA GPUs, and all of the GPUs in the infrastructure today are NVIDIA GPUs. A shift away from that supply would take time and money the company says it may not have on the customer’s schedule.1
We do not publish a revised life, a residual value, or an impairment charge. The filing does not give us those. Likelihood of a near-term write-down is medium: management reviewed the estimate, lengthened it in 2023, and left it in place through this 10-Q. Impact is high if the estimate is long, because the gross technology book is $33.8 billion and quarterly depreciation is already $1.4 billion. The timing is the Rubin deployment the company has put in the second half of 2026, and whatever cycle follows it.
05 Power not delivered, capex not finished
Some leases price off construction cost, and the payments during construction are variable until the contingencies are resolved. At June 30, 2026, 355 MW of electrical power at those sites remained undelivered. The company expects that power in phases between 2026 and 2028. Separately, an equipment lease with an unconsolidated lessor covers mechanical, electrical, and plumbing infrastructure, and the company’s committed payments under that lease were about $1.8 billion, inside the lease commitments that have not yet commenced. In the first half, operating-lease right-of-use assets acquired through lease liabilities were $8,359 million. Lease liabilities ended at $16.3 billion.1
The build is not a completed asset with a mortgage on it. Construction in progress was $11.9 billion. Cash capex was $14.1 billion in six months. The company has also committed up to $1.7 billion to two joint ventures that each hold a data-center development project, with maximum exposure limited to that commitment and a carrying value of $479 million at June 30. A separate New Jersey campus joint venture, in which CoreWeave’s stake had risen to 35%, expects to fund construction with more debt and equity. None of that is a cancelled project. It is a project list that still has to be energized while principal is being repaid.1
Likelihood is medium-high because the undelivered power is already a disclosed fact, not a rumor about a town council. Impact is high if a contracted campus slips, because RPO itself deducts amounts that may not become revenue when delivery is late, while contractual interest does not. We rank local politics below this. Politics can slow a site. The filing already says the power is not there.
06 What the price assumes
We do not publish a price target. A target would require a multiple we would be choosing. What follows is arithmetic on figures we can cite. It is not a forecast.
| Item | Figure | How it is derived |
|---|---|---|
| Class A close, Aug 21, 2026 | $87.85 | Yahoo Finance historical prices3 |
| Class A shares outstanding | 457 million | June 30, 2026 balance sheet1 |
| Basic shares | 551 million | June 30 ending shares in the equity statement1 |
| Stockholders’ equity | $5,024 million | June 30, 20261 |
| Book value per basic share | $9.12 | 5,024 ÷ 551, rounded |
| Class A close ÷ book value | 9.6× | 87.85 ÷ 9.12 |
| Class A stake at the close | $40,147 million | 457 × 87.85 |
| Illustrative equity, all basic shares | $48,405 million | 551 × 87.85, if Class B is worth the Class A price |
| Debt, net of discount | $35,068 million | 31,405 recourse + 3,663 non-recourse |
| Unrestricted cash | $5,524 million | Cash and cash equivalents1 |
| Illustrative enterprise value | $77,949 million | 48,405 + 35,068 − 5,524 |
The $48.4 billion equity figure is not a market-cap print. The share count is three weeks older than this close. The cover of the 10-Q says that on July 31, 2026 the company had 458,871,690 Class A shares and 92,664,912 Class B shares outstanding. Issuance after that date is not in the table. Class B may not be worth the Class A price. Restricted cash, $873 million current and $507 million non-current, is excluded from the cash deduction. The convertible notes can add shares. We do not have a conversion price we are willing to use here, so we do not guess a diluted count.1
Read it from the obligations up. An illustrative enterprise value around $78 billion sits on a quarter in which operations did not cover interest, and on a half in which operating cash did not cover debt service. Book equity of $5.0 billion is the accounting residual. At $87.85, Class A traded at about 9.6 times June 30 book value per basic share. We are not saying the stock should trade at book. We are saying book is thin next to $35.6 billion of scheduled principal and $16.3 billion of lease liabilities. The price assumes the $103.7 billion of RPO is delivered, collected, and used to amortize this stack.
07 Risks to the short
The backlog can carry a great deal of debt if the sites are delivered. RPO of $103.7 billion is not a rumor, and $9.7 billion of it has already shown up as deferred revenue. Management says the delayed-draw loans are collateralized by contributed contracts and pledged cash flows, generally from investment-grade counterparties, and that they amortize as those cash flows arrive.1
Microsoft and Meta can pay. A thesis that needs them to default is a different, weaker thesis, and it is not this one. Their commitments can fund a large share of the build if they take the capacity. Jane Street’s $6 billion commitment is a third contracted buyer we are not calling distressed. The six-year GPU life may be right. Management lengthened it and has kept it. If the equipment earns for six years and is redeployed after the first contract, reason 4 does not become an impairment.
The company has sold stock when it needed to. The March 2025 IPO of 37 million Class A shares at $40 produced $1.4 billion of net proceeds before $31 million of offering costs, and the underwriters later bought about 2 million more shares for $68 million net. In January 2026 NVIDIA bought about 23 million shares at $87.20, for $2.0 billion gross. In April 2026 the company sold about 9 million shares at $109, for $1.0 billion gross. First-half private-placement proceeds, net of issuance costs, were $2,982 million. A lower stock price makes the next sale more painful for existing holders. It does not make a raise impossible. Rates can fall. The disclosed sensitivity cuts both ways.1
Credit was still available after June 30: DDTL 5.5 was signed and $1.2 billion was drawn. Kenilworth and Lancaster have drawn protests and, in Lancaster, a lawsuit, and construction has continued. We may be early, or wrong, about a slowdown in AI spending. The OpenAI loss figures are leaked. If the audited statements differ, that paragraph has to be reread. The June-quarter customer letters are not a named list.
The short is wrong if CoreWeave delivers the capacity, collects the contracted cash, and refinances the 2026 and 2027 principal without a break in the price of GPU hours or in the willingness of credit and equity markets to extend. That is a possible outcome. The capitalization does not have much room if it slips.
08 The same pattern at other AI data-center names
CoreWeave is the most levered version of a structure that shows up across the listed AI data-center and neocloud group. The points below are one or two facts from each company’s own SEC filings or from an exhibit filed with them. Quotes pages on Finviz are linked so a reader can see the tape. They are not a source for the figures.
Nebius (NBIS) closed $5.75 billion of convertible notes on August 24, 2026: $3.45 billion of 0.50% notes due 2030 and $2.3 billion of 4.50% notes due 2034. The company said the proceeds fund data centers, GPUs, and the rest of the build. In the same transaction it exchanged $400 million of its 2.00% notes due 2029 and $400 million of its 3.00% notes due 2031 for about 15.8 million Class A shares. That is new debt for the build, and dilution to retire older converts.8
Oracle (ORCL) is not a neocloud, and its cloud build has the same shape at a larger scale. At May 31, 2026, notes payable and other borrowings were $7,199 million current and $122,342 million non-current. Cash capital expenditures in fiscal 2026 were $55,663 million. Additional lease commitments of $260 billion, substantially all related to data centers, were not on the balance sheet. They are generally expected to commence between the first quarter of fiscal 2027 and fiscal 2029, on terms of fifteen to nineteen years. Remaining performance obligations were $638 billion, of which Oracle expects to recognize about 12% as revenue over the next twelve months, 34% from month 13 through month 36, and 34% from month 37 through month 60.7
Core Scientific (CORZ) says its colocation revenue is concentrated with a single customer. The first high-density contract, announced in 2024 with CoreWeave, was later expanded to about 590 MW of leased customer power across five sites. About 395 MW had commenced billing by June 30, 2026. On May 6, 2026 a financing subsidiary issued $3.30 billion of 7.75% senior secured notes due 2031. Together with $460 million of 3.00% convertible notes due 2029 and $625 million of zero-coupon convertible notes due 2031, gross notes payable were $4,385 million. Amortization of the secured notes does not start until rent has commenced and certain revenue credits have abated. CoreWeave’s landlord is using CoreWeave’s rent to support its own bond.9
IREN (IREN) is delivering a five-year, $9.7 billion cloud-services contract with Microsoft. Horizon 1, the first of four 50 MW deployments at Childress, Texas, was delivered to and accepted by Microsoft in August 2026. The same company’s quarterly results exhibit describes a separate five-year, $3.4 billion AI cloud contract with NVIDIA. Two customers, both also central to CoreWeave’s own supply and demand, anchor the book.10
Applied Digital (APLD), in its Form 10-K for the year ended May 31, 2026, reported about 1,410 MW of contracted critical IT load and about $36.2 billion of contracted lease revenue over 15-year base terms. Polaris Forge 1 is 400 MW and about $11.0 billion of that revenue, and the anchor tenant is CoreWeave. The filing says each lease is take-or-pay and non-cancellable. Part of CoreWeave’s lease liability is this company’s revenue, and the rest of Applied Digital’s campuses are still scheduled for delivery into 2028.11
Boost Run (BRUN) is the small public version. For the quarter ended June 30, 2026, lease revenue from GPU rentals was $30.4 million and blockchain award revenue was $0.7 million. Cash was $120.2 million. Finance lease liabilities were $82.2 million current and $155.9 million non-current. Operating lease liabilities were $20.2 million current and $99.4 million non-current. The balance-sheet debt lines were zero. The risk in the filing is a lease-funded GPU fleet against a revenue base of about $31 million a quarter, not a CoreWeave-sized bond complex. The 10-Q does not give a customer-concentration percentage, so we do not invent one.12
09 Other considerations
These are real, and they are not the short. We put them here because a fact-check did not support the weight they sometimes carry.
Local opposition. None of the project facts in this paragraph come from the 10-Q. In Kenilworth, New Jersey, NJ.com reported an approximately $1.8 billion AI data center, with construction begun and operations expected in early 2027, and a petition that had passed 4,000 signatures when that story ran. Union News Daily reported protests before a May 19, 2026 planning-board meeting. Cranford, next door, has moved to take data centers out of the uses allowed in its commercial zones. In Lancaster, Pennsylvania, LancasterOnline reported CoreWeave projects at 216 Greenfield Road and 1375 Harrisburg Pike. A resident’s zoning appeal was blocked in court. Work on the first Greenfield Road building was underway, with an opening expected in summer 2027, and the developers told the court they had spent $1.3 billion on it. A later ordinance was written so that complete applications already on file would keep the old rules. Opposition changed the politics. On the facts we have, it did not stop the pour. We do not treat a petition as a balance-sheet event unless a filing says a site has stopped. That is why this sits below the 355 MW the company has already said is undelivered.56
NVIDIA. NVIDIA is the supplier and a large new shareholder. It is not, in the June 2026 10-Q, a related party. Magnetar was a related party through March 2025, when it gave up its board seat, fell below 10% of the vote, and lost significant influence. The January 2026 private placement was about 23 million Class A shares at $87.20, for $2.0 billion gross. Customer contracts require NVIDIA GPUs, and the installed base is NVIDIA. That is a supply concentration. The $2.0 billion check is also evidence that the strategic supplier was willing to fund the equity. Both belong in the note. Neither replaces the principal schedule.1
Dilution. The share sales in the risks section are how the first half was financed, alongside $16.7 billion of debt proceeds. Dilution is a cost of staying solvent. It is also the reason a default is not the base case. We do not rank it above the obligations those shares have not retired.
Lock-ups and insider selling. The redeemable Class A shares issued at the IPO carried a put at $38.95, about $1.2 billion in aggregate, subject to a lock-up. The put terminated automatically in September 2025, when the Class A stock’s 20-day volume-weighted average price reached at least $68.16 over a 30-trading-day window. Those shares were then ordinary Class A common stock. The June 2026 10-Q does not describe a lock-up that is still running. We are not using secondary write-ups of Form 4 sales. The counts in those write-ups were small next to roughly 458 million Class A shares, and we did not recompute them from the forms. A lock-up overhang is not a current reason for this short.1
An “AI bubble,” as a slogan. A general claim that AI spending must collapse is not in the filings. What the filings show is a customer book concentrated in that spending, a backlog already net of estimated delays, and obligations that do not shrink if the spending does. If the budgets hold and the power arrives, reasons 1, 2, and 5 can be refinanced. If they do not, the equity is the residual. That is the scenario. It is not a fifth independent fact, which is why it is not its own numbered reason.
10 Notes
- CoreWeave, Inc., Form 10-Q for the quarter ended June 30, 2026. Unless a sentence is footnoted elsewhere, the CoreWeave figures in this note are from that filing. sec.gov.
- CoreWeave, Inc., Form 10-K for the year ended December 31, 2025. Microsoft was about 67% of 2025 revenue. Technology equipment is depreciated straight-line over six years. The life of computing equipment was changed from five years to six effective January 1, 2023. sec.gov.
- Class A unadjusted closing price of $87.85 on August 21, 2026, the last trading day before this note, from the historical price table at Yahoo Finance.
- OpenAI figures are not from a CoreWeave filing and not from an OpenAI SEC report we inspected. Quartz, June 16, 2026, and Ars Technica, on statements obtained by Ed Zitron and reviewed by the Financial Times: 2025 revenue $13.07 billion, operating loss $20.92 billion.
- Kenilworth: NJ.com on cost, schedule, and the petition; Union News Daily on protests; Cranford Patch on the neighboring zoning change.
- Lancaster: LancasterOnline on the blocked appeal and the developers’ reported spend; companion article on the later ordinance.
- Oracle Corporation, Form 10-K for the year ended May 31, 2026, filed June 22, 2026. The August 31, 2026 quarter was not yet on file. sec.gov.
- Nebius Group N.V., Exhibit 99.1 to a filing on the August 24, 2026 convertible notes. sec.gov.
- Core Scientific, Inc., Form 10-Q for the quarter ended June 30, 2026. sec.gov.
- IREN Limited, Exhibit 99.1, August 13, 2026, on Horizon 1 and the $9.7 billion Microsoft contract, sec.gov. The $3.4 billion NVIDIA contract is in the company’s fiscal third-quarter 2026 results exhibit, sec.gov.
- Applied Digital Corporation, Form 10-K for the year ended May 31, 2026. Applied Digital investor relations, filing.
- Boost Run Inc., Form 10-Q for the quarter ended June 30, 2026. Dollar figures in this paragraph are the thousands reported in that filing, expressed in millions. sec.gov.
This note is opinion, not investment advice. Froth City Research and its affiliates may hold positions in securities discussed. Do your own diligence.