CoreWeave: a cash flow stress test
CoreWeave buys GPUs, the chips that run AI, and rents them out. It pays for them long before its customers pay it back, so it borrows a lot. We followed the money through a simple model of its cash.
Why now. In mid-October, the prices of CoreWeave's bonds fell sharply and people started asking: can CoreWeave pay its debts? We followed the money to find out.
Our conclusion: we built a cautious model of CoreWeave's money. Our stress tests suggest it could pay its debts as long as:
- New borrowing costs about 10% a year in our model. Our stress test suggests the numbers could still hold at up to around 14%, if everything else stays as we assume.
- New deals keep coming in at about the prices we assume, in line with its July price rise.
- It finds customers for almost all its power, at least about 93%.
- Its big customers stay.
- It can keep raising new money: about $35 billion of loans and bonds and $6 billion from new shares by the end of 2027.
- Its new capacity starts earning roughly on time.
These limits apply one at a time. If two things go wrong together, there is less room.
How we built this
What we know (from its reports and the market): at the end of June it owed $36 billion in loans and bonds, not counting building leases. It has secured 4.2 gigawatts of power, 1.5 gigawatts of it running in June. In mid-October, one of its existing 2032 bonds traded at a yield of just over 16% (Reuters). That means investors buying it wanted a return of just over 16% a year.
What our model assumes: to fill that power with GPUs, it spends about $62 billion on GPUs and related costs by the end of 2027. Its business covers about $5 billion of that and about $16 billion is already arranged. We assume it raises the rest, about $35 billion of new loans and bonds and about $6 billion from new shares, plus about $5 billion more in 2028. We assume all the new bonds are repaid in one go in 2032; that date is our assumption, not CoreWeave's. New borrowing costs about 10% a year.
Borrowing costs: that 16% is the market's price for an existing bond, not a rate CoreWeave is paying. At the end of June it paid about 8% a year on average on its debt. Our model assumes about 10% on new borrowing. Our stress test suggests the numbers could still hold at up to around 14%, if everything else stays as we assume. Above that, its own cash would not fully cover its 2032 bonds in our model, and it would need to refinance, as companies like this usually do.
Our approach: we model only what CoreWeave has today, simply to check the health of the business. In real life it might add more power, customers and GPUs; we leave that out. So after 2029 we assume no new GPUs: its income shrinks as old contracts end, but it also spends nothing on new GPUs. On that basis, in our model it could still pay off all its loans and bonds by the end of 2033; its building leases carry on after that. We also leave out income tax.
Here is the whole story in one picture: the cash the business brings in each year, what it owes its lenders, and the cash left in the bank.
Exhibit 1
In our model, once its GPUs are running, the business brings in more than its loan payments, and savings pay the new bonds we assume fall due in 2032.
Our estimate, $ billions a year
New money pays for GPUs
Brings in more than it owes, and saves
From its savings
| Late 2026 | 2027 | 2028 | 2029 | 2030 | 2031 | 2032 | 2033 | |
|---|---|---|---|---|---|---|---|---|
| Cash coming in | 4 | 21 | 26 | 26 | 24 | 21 | 15 | 6 |
| Owed to lenders | 6 | 14 | 15 | 15 | 12 | 14 | 36 | 5 |
| Cash in the bank | 3 | 3 | 3 | 10 | 23 | 30 | 9 | 10 |
Notes: Cash coming in is what the business brings in after running costs (line F of Exhibit 9). Owed to lenders is interest (net of interest it earns on its cash) and loan repayments, plus bonds (loans repaid in one go) in the year they fall due. The 2032 figure includes about $24 billion of new bonds that exist only in our model. Buying GPUs is left out of the bars. In late 2026 and 2027 new money pays for most of it (Exhibit 10); in 2028 and 2029 the business pays for most of it itself, plus about $5 billion of new money in 2028, so the bank balance only starts to grow in 2029. The line is cash in the bank at the end of each year. We assume no new GPUs after 2029, so cash coming in falls as old contracts end. Figures are rounded to the nearest billion. Late 2026 is July to December.
Source: CoreWeave's reports; our estimate.
The rest of this report shows how we got there, step by step.
1. How the business works
Picture a car-rental firm. It buys its cars, mostly with a loan, and leases a car park to keep them in. Customers pay to use the cars. Their money pays the car-park lease first and the car loan after that.
CoreWeave works the same way with servers full of GPUs. Its car park is a data centre. It measures its size by the electricity it can use: one megawatt is enough electricity for several hundred homes, and a gigawatt is 1,000 megawatts.
The money goes out in steps 1 and 2. It only starts coming back in step 4.
Exhibit 2
It has secured 4.2 gigawatts of power. 1.5 gigawatts were running in June.
Each square is 100 megawatts
CoreWeave calls this its contracted power: electricity and building space it has secured. 3.7 gigawatts at June 30 (Q2 results), 4.2 gigawatts by Aug 11 (call).
As far as we can tell from its reports, customers have not yet signed up for all of this power. We assume they will, based on what CoreWeave said on its August 11 call: “our near-term capacity remains effectively sold out.” That covers the near term, so assuming all its power finds customers is one of our main assumptions.
2. What we assumed
These are the main assumptions in our model. Where each one comes from is in our assumptions in detail at the end.
Exhibit 3
Our main assumptions
| What | We assume |
|---|---|
| Cost of GPUs and equipment, per megawatt | $30 million |
| What customers pay per megawatt a year, older deals | $10.3 million (April to June revenue ÷ about 1.0 gigawatt billing at the start of the quarter) |
| What customers pay per megawatt a year, new deals | $12.9 million (after the July price rise) |
| Running costs per megawatt a year (building rent, power, site and head office; not new GPUs) | $3.9 million |
| Contract length | 5 years |
| Paid up front by new customers | 15% of the contract |
| When a contract ends | The GPUs are rented again at half price. From 2031, only for one year, then they earn nothing. This is cautious (see the side note after Exhibit 13) |
| How new GPUs are paid for | The customer's up-front payment, then the same mix it has used in the past: about 85% loans and 15% from new shares |
| Interest on new borrowing | 10% a year |
| Customers for its power | All 4.2 gigawatts, as each building opens |
| Growth beyond that power | None. No new GPUs after 2029 |
| When the new bonds are paid back | All in 2032 (our assumption, not CoreWeave's) |
| Income tax | Left out |
3. One megawatt, start to finish
Start small. If one megawatt pays for itself, the rest is adding up. On a new deal, the GPUs and equipment for one megawatt cost about $30 million on day one (our estimate from its reports). The customer pays about $12.9 million a year to use them. That is also our estimate. We take about $10.3 million, what customers paid per megawatt in April to June. Then we add the price rise of about 25% across its products in July, which its CFO described on the August 11 call.
New customers usually pay part of the contract up front. We assume 15%, the low end of the 15% to 25% CoreWeave says customers paid in the past (10-Q).
CoreWeave borrows in two ways. A mortgage-style loan is backed by customer contracts and is paid down as those customers pay, so it is gone when the contracts end. A bond needs only interest each year, and the whole amount is paid back in one go a few years later. Here is how the $30 million gets paid on day one:
Exhibit 4
Day one: the $30 million is paid four ways
One megawatt on a new deal, $ millions
| Paid up front by the customer | 9.7 |
| New shares | 3.1 |
| Mortgage-style loan | 6.6 |
| Bond | 10.6 |
| GPUs and equipment for one megawatt | 30.0 |
The split is our estimate: shares at the mix it used in the first half of 2026, and loans split like its debt at June 30. We count all debt that is not mortgage-style as bonds.
Source: CoreWeave's quarterly report and its first-half 2026 cash flow; our estimate.
From then on, the megawatt has to pay back its loans out of what the customer pays. Think of the up-front payment as a deposit that gets used up: $1.9 million of each year's $12.9 million was already paid on day one. Exhibit 5 follows it year by year.
Exhibit 5
In our model, a new megawatt has about $4.3 million left each year after its loan payments. After paying back its bond in year 6, it is about $12.2 million ahead.
One megawatt on a new deal, $ millions
| Year 1 | Year 2 | Year 3 | Year 4 | Year 5 | Year 6 | |
|---|---|---|---|---|---|---|
| Customer pays | 12.9 | 12.9 | 12.9 | 12.9 | 12.9 | 6.5 |
| Running costs | −3.9 | −3.9 | −3.9 | −3.9 | −3.9 | −3.9 |
| Less: part already paid up front on day one | −1.9 | −1.9 | −1.9 | −1.9 | −1.9 | |
| Cash from the megawatt | 7.1 | 7.1 | 7.1 | 7.1 | 7.1 | 2.6 |
| Mortgage-style loan payment | −1.7 | −1.7 | −1.7 | −1.7 | −1.7 | |
| Bond interest | −1.1 | −1.1 | −1.1 | −1.1 | −1.1 | −1.1 |
| Bond paid back | −10.6 | |||||
| Left over | 4.3 | 4.3 | 4.3 | 4.3 | 4.3 | −9.1 |
| Added up | 4.3 | 8.5 | 12.8 | 17.1 | 21.3 | 12.2 |
Year 6: the contract is over and the same GPUs are rented again at half price. We assume the bond is paid back that year. Figures are rounded, so rows may not add exactly.
So in our model, one megawatt on a new deal pays back what it borrowed by the end of year 6 and still has about $12.2 million left.
4. Older deals and new deals
Most of the GPUs running today were rented out before July, at about $10.3 million per megawatt a year (our estimate from its April to June results). They were paid for with borrowed money; its debt cost it about 8% a year on average at the end of June (September investor presentation).
Exhibit 6
Both deals build up cash for five years. In our model, only the new deal builds up enough to pay back its bond at the end.
Cash left after loan payments, added up year by year, $ millions per megawatt
The "Added up" line of Exhibit 5, for each deal. Both examples assume the customer paid 15% up front, the low end of what customers paid in the past. The drop in year 6 is the bond being paid back.
Source: CoreWeave's quarterly report and its Aug 11, 2026 call (price rise); our estimate.
In our model, an older megawatt covers its bills each year, because it pays only interest on its bond. But in our model it does not quite save enough to pay the bond back, ending about $0.6 million short.
That small gap comes from the cautious rule in our one-megawatt example that the GPUs earn nothing after year 6. If an older megawatt keeps renting for just one more year at half price, it pays back its bond in full and ends about $0.7 million ahead. One recent contract suggests this is possible: CoreWeave signed a contract for A100 chips, introduced in 2020, that runs into 2029 (Aug 11 call). Otherwise, CoreWeave could cover the gap by refinancing when old bonds fall due.
Its last report showed a loss: in April to June it lost about $24 for every $100 customers paid (Q2 results), much of it interest on its debt. Our model follows cash, including interest, rather than accounting profit.
5. The whole company
Add up every megawatt, old and new, as each one switches on, and you get the whole company. We count only the 4.2 gigawatts it has already secured. CoreWeave plans to grow beyond that; we leave growth out.
Exhibit 7
In our model, revenue reaches about $45 billion a year in 2029
Revenue, $ billions
Late 2026 is July to December only. In our model, deals already signed bring in about $108 billion of the $167 billion in this chart, about 65%.
Source: CoreWeave's quarterly report (signed deals and when they pay) and its calls; our assumptions for the rest.
Revenue is simply the megawatts earning, times the average price per megawatt:
Exhibit 8
From megawatts to revenue
| Late 2026 | 2027 | 2028 | 2029 | 2030 | |
|---|---|---|---|---|---|
| Gigawatts earning (yearly average) | 1.58 | 2.66 | 3.72 | 4.11 | 4.20 |
| Average price per megawatt, $M a year | 10.30 | 10.73 | 11.09 | 10.99 | 10.43 |
| Revenue, $B | 8.1 | 28.5 | 41.3 | 45.2 | 43.8 |
Late 2026 is half a year, so its revenue is half of a full year at that price. Late 2026 revenue comes from the company's own forecast.
Why the average price falls after 2028. This comes from our cautious assumption, not a forecast. We model only the business CoreWeave has today. The average rises until 2028 as new deals at $12.9 million switch on. After that, the first 5-year contracts end, and we assume those GPUs are rented again at only half price. With no new GPUs added, nothing lifts the average back up.
In real life, CoreWeave says prices for older chips are holding up: its CFO says they remain “higher or at levels that we've seen about a year ago” (Aug 11 call). If that continues, the average price would not fall this much, which would be upside to our numbers.
Here is the whole company, year by year. Each line has a letter, so you can follow how one leads to the next.
Exhibit 9
In our model, after its building leases, running costs and loan payments, the business has money left every year from 2027
$ billions
| Late 2026 | 2027 | 2028 | 2029 | 2030 | ||
|---|---|---|---|---|---|---|
| A | Revenue | 8.1 | 28.5 | 41.3 | 45.2 | 43.8 |
| B | Building leases and running costs | −3.0 | −9.9 | −13.7 | −15.2 | −15.3 |
| C | Up-front payments received | 1.9 | 9.5 | 5.2 | 1.3 | |
| D | Less: revenue customers already paid for up front | −1.3 | −3.2 | −4.5 | −5.1 | −4.9 |
| E | Money customers owe but haven't paid yet (about 90 days of bills) | −2.1 | −4.3 | −1.9 | −0.2 | 0.6 |
| F | Cash from the business (A to E) | 3.6 | 20.6 | 26.4 | 26.0 | 24.2 |
| G | Interest | −1.6 | −4.9 | −5.8 | −5.3 | −4.2 |
| H | Paying back loans | −4.4 | −8.7 | −9.1 | −9.4 | −7.4 |
| I | Loan payments (G + H) | −6.0 | −13.6 | −14.9 | −14.7 | −11.6 |
| J | Left after all loan payments (F + I) | −2.4 | 7.0 | 11.5 | 11.3 | 12.6 |
Lines C, D and E are about timing: some customers pay early and some pay late, and over the years it evens out. Line G is interest on its loans minus interest it earns on its cash. Late 2026 is negative because most new GPUs are only just switching on. Running costs per megawatt fall a little as it grows, because head office costs (about $0.7 billion a year) are spread over more megawatts. Figures are rounded so that each column adds up.
So in our model, before buying GPUs, the business pays its own way from 2027. The new GPUs it is still installing are paid for mostly with new money. Exhibit 10 shows where that money comes from.
Exhibit 10
Where the $62 billion for new GPUs and related costs comes from
Late 2026 and 2027, $ billions
| The business itself (line J, late 2026 and 2027) | 5 |
| Already arranged: spare cash, money raised since June, loans already agreed | 16 |
| New loans, if it keeps its past funding mix | 35 |
| New shares, if it keeps its past funding mix | 6 |
| Spent on new GPUs and related costs | 62 |
Related costs include about $1.2 billion it committed to a joint venture in 2026. Spare cash means cash above the about $3 billion it keeps in the bank as a cushion. In our model it raises about $5 billion more in 2028, and saves after that.
We assume CoreWeave keeps funding itself the way it has in the past: after customers' up-front payments, about 85% from loans and about 15% from new shares, the mix it used in the first half of 2026 (10-Q). CoreWeave's own report warns it may not always get money on good terms.
6. Bond repayments
This section shows when its bonds fall due and whether, in our model, its own cash covers them. Based on our modelling assumptions, these are its bond obligations. In our model, its loans and bonds (not counting building leases) rise from $36 billion in June to about $72 billion at the end of 2027. They then fall to about $50 billion by the end of 2030 as loans are paid back. Most of what is left is bonds, which are paid back in one go in 2031, 2032 and 2033. About $24 billion of the 2032 bill is new bonds that we assume it borrows in 2026 to 2028; when they fall due is our assumption, not CoreWeave's.
We assume no new GPUs after 2029. That lowers income as old contracts end, but it also means no spending on new GPUs. Before 2031, re-rented GPUs earn half price; from 2031, each one earns half price only for the year after its contract ends, then nothing.
Exhibit 11
Bond payments in our model, 2031 to 2033
$ billions
| Year | Cash at start of year | Cash the business adds, after its other loan payments | Bonds due | Cash left at year end |
|---|---|---|---|---|
| 2031 | 22.9 | 13.7 | −7.1 | 29.5 |
| 2032 | 29.5 | 10.9 | −31.7 | 8.7 |
| 2033 | 8.7 | 5.4 | −4.5 | 9.7 |
Bonds due in 2032: about $7.5 billion already issued plus about $24.2 billion of new bonds we assume. Bonds already issued: its quarterly report and, for the 2033 convertible notes, its Sept 22 filing. Building leases carry on after 2033; we don't model those years. Figures are rounded, so rows may not add exactly.
If our assumptions hold, our model suggests CoreWeave should be able to manage these bond payments from its own cash, with about $9 billion left in 2032, its tightest year. If the new bonds were spread over more years, 2032 would be easier.
Our model leaves out a few things. Income tax would reduce the cash left. Some of its borrowing sits in separate units with their own lenders, so not every dollar may be free to pay the parent company's bonds. And in real life, companies like this usually refinance old bonds with new borrowing, which depends on lenders staying willing.
In mid-October the market was more worried than our model. One of its 2032 bonds traded at a yield of just over 16%. Reuters also reported that the price of insurance against a CoreWeave default implied a risk of default within five years of nearly 63%. That is market pricing, not our view.
7. Risks and upsides
This section shows what happens to the cash if one thing goes better or worse than we assume. Exhibit 12 changes one assumption at a time and shows the lowest cash left in 2031 to 2033 in our model. These are stress tests, not forecasts. Our estimate is about $9 billion; a minus means that, in that test, its own cash would not fully cover its bonds.
Exhibit 12
In our tests, prices and customers move the result most. Interest rates move it less.
Lowest cash left in 2031 to 2033 if one thing changes, $ billions
Each line changes one thing and keeps the rest as we assume. If two things go wrong at once, there is less room.
In our tests, customers move the result most: its biggest customer paid about 36% of its revenue in April to June, and its three biggest about 72% (36%, 26% and 10%, 10-Q). The customer test is hypothetical; it assumes CoreWeave collects nothing for a broken contract.
Exhibit 13 shows how far each assumption can move on its own in our model before the cushion is used up. The cushion is the cash left in its tightest year, about $9 billion in our model.
Exhibit 13
Stress test: how far each assumption can move before the cushion is used up
| Assumption | We assume | Cushion used up if | Room to spare |
|---|---|---|---|
| Average interest on new borrowing | 10% a year | Above around 14% | About 4 points |
| Price on new deals, per megawatt a year (summer deals keep their value) | $12.9 million | Below about $12.0 million | About 7% |
| Share of its power with customers (and of old GPUs re-rented) | All of it | Below about 93% | About 7 points |
| Cost of GPUs and equipment, per megawatt | $30 million | Above about $32.5 million | About 8% |
| New capacity from 2027 starting to earn late (rent and building costs on schedule) | On time | About 5 months late | About 5 months |
Each line changes one thing and keeps the rest as we assume. “Cushion used up” means that, in our model, its own cash would no longer fully cover its bonds in its tightest year, so it would need to refinance. It does not mean the business stops.
This is one of the biggest debates about CoreWeave. Its accounts spread the cost of a GPU over 6 years. Some investors worry that GPUs go out of date faster than that.
CoreWeave's executives see it differently. On its Aug 11 call, its CEO said pricing for older chips is “at or above where it was years ago”. Its CFO said prices for the older generation remain “higher or at levels that we've seen about a year ago”. It also signed a contract for A100 chips, introduced in 2020, that runs into 2029. If that holds, 5 to 6 years may prove too short rather than too long. The CFO also noted that only “a very limited part” of its fleet is up for renewal (same call).
It is too early to call this a trend, as newer and faster GPUs keep arriving. That is why our model stays cautious: from 2031, after each contract ends, the GPUs earn half price for one year and then nothing. If old GPUs earned 75% of their price instead of half, it could have about $16 billion left in 2032 instead of about $9 billion.
Conclusion
We built a cautious model of the business CoreWeave has today. If our assumptions hold, our stress tests suggest it could pay its debts, including the new bonds our model assumes it borrows and repays in 2032.
That depends on the six things listed at the top, each tested on its own. For borrowing costs, we assume about 10% a year; our stress test suggests the numbers could still hold at up to around 14%, if everything else stays as we assume.
Our assumptions in detail
We used CoreWeave's reports and calls to make our best assumptions. Some numbers come straight from its filings; others are our own estimates. The table says which.
Exhibit 14
Where each assumption comes from
| Assumption | Value | Type | Where it comes from |
|---|---|---|---|
| One megawatt | |||
| Cost of GPUs and equipment | $30 million per megawatt | Our estimate from its reports | Worked out from its 10-Q to June 2026 and 2025 annual report: about $25.5 million to $33 million per megawatt over the last year, depending on how you count, and about $30.4 million in 2025. |
| Price, older deals | $10.3 million per megawatt a year | Our estimate from its reports | April to June 2026 revenue ($2.58 billion, about $10.3 billion a year) divided by about 1.0 gigawatt that was running and billing at the start of the quarter. New power bills a month or two after it arrives, so we use start-of-quarter power (results, Aug 11; May 7 call). Divided by average power during the quarter it would be about $8.2 million, but that counts megawatts not yet billing. Cross-check: its own end-2026 forecast of $18.5 to $19.5 billion a year over more than 1.85 gigawatts active is also about $10 million per megawatt (Aug 11 call). Our results are sensitive to this number. |
| Price, new deals | $12.9 million per megawatt a year | Our assumption | The older price plus the about 25% price rise across its products in July (Aug 11 call; September presentation, slide 7). The same slide shows about $40 million for short 3 to 6 month contracts; we don't use that, as most of its business is 5-year contracts. |
| Running costs | $3.9 million per megawatt a year | Our estimate from its reports | Building leases, electricity, the site and head office at April to June 2026 levels (10-Q lease note and income statement). |
| Contract length | 5 years | Our assumption, based on its reports | Its CFO said the average contract length rose “from roughly 4 years to roughly 5 years” (Feb 26 call). We use 5 years for new deals; their exact terms are not disclosed. |
| Paid up front | 15% of the contract | Our assumption | The low end of the 15% to 25% customers paid in the past (10-Q risk factors). About 70% of deals signed in April to June included a payment up front (presentation, slide 7). |
| Life of the GPUs | 6 years | From its reports | 2025 annual report, property and equipment note. |
| After the contract ends | Rented again at half price | Our assumption | Before 2031, re-rented GPUs earn half price. From 2031 we count one year at half price after each contract ends, then nothing. |
| Borrowing | |||
| How new GPUs are paid for | 85% borrowed, 15% new shares | Our estimate from its reports | After the customer's up-front payment. The mix it used in the first half of 2026 (10-Q cash flow statement); we assume it stays similar. |
| What kind of borrowing | 38% mortgage-style loans, 62% bonds and other company-level debt | Our estimate from its reports | Our split of its debt at June 30 (10-Q debt note): its loans backed by customer contracts count as mortgage-style. These loans are typically paid down as customers pay, so the debt is “fully repaid by the time the customer contract expires” (presentation, slide 11). We assume they are paid off evenly over each contract. Some of its bonds are convertible notes, which CoreWeave can settle in cash or shares; our model assumes cash. About $3.7 billion of its debt is non-recourse: its lenders can claim only the assets of the unit that borrowed it (10-Q debt note). |
| Interest on new borrowing | 10% a year | Our assumption | Above the about 8% it paid on its debt at the end of June (presentation, slide 27). We also test 12%. |
| When new bonds are due | All in 2032 | Our assumption | Like its newest bonds (10-Q debt note). Spreading them over more years would make 2032 easier. |
| The whole company | |||
| Customers for all its power | All 4.2 gigawatts | Our assumption | Our reading of its reports is that customers have not signed for all of it yet. We assume they will. CoreWeave says “our near-term capacity remains effectively sold out” (Aug 11 call); that covers the near term, so this stretches it to all its power. |
| When building leases start | When each building is handed over | From its reports, plus our estimate | Signed leases start between 2026 and 2029 (10-Q lease note). For power secured after June we use our estimate of today's lease cost, about $1.6 million per megawatt a year (its April to June lease cost spread over its power). |
| Rest of 2026 | The company's own forecasts | From its reports (company forecast) | Its full-year 2026 revenue forecast of $12.4 to $13.2 billion, its July to September forecast and its yearly revenue pace at the end of 2026 (Aug 11 call). |
| After 2029 | No new GPUs | Our assumption | We follow every contract to its end, through 2033. Building leases, interest and loan payments carry on as scheduled. |
Where the numbers come from
CoreWeave's quarterly report to June 30, 2026, its 2025 annual report, its results and calls (February 26, May 7 and August 11, 2026), its September investor presentation and its September 22 filing. We don't know the exact terms of the deals signed since July, or how long the chips will keep earning.
Market data: bond prices and yields from Reuters (Tradeweb data), October 15, 2026.
- CoreWeave quarterly report (10-Q) to June 30, 2026, SEC
- CoreWeave Q2 2026 results call, August 11, 2026 (transcript)
- Reuters, “AI credit fears push tech default insurance to record highs”, October 15, 2026 (via BusinessDay)
- CoreWeave 2025 annual report (10-K); Q2 2026 results release; calls of February 26, May 7 and August 11, 2026; September 2026 investor presentation (filed with the SEC on Form 8-K); September 22, 2026 filing on its convertible notes
Important information. This report is for information and education only. It is not investment advice, a recommendation, or an offer or solicitation to buy or sell any security, bond, loan or derivative. It is not a credit rating or a forecast of default. The Buildout is not a registered financial adviser, investment adviser or broker-dealer, and CoreWeave has not reviewed this report.
Our estimates and forecasts for July 2026 onward are forward-looking and come from our own model, based on our reading of CoreWeave's public filings, results and calls, and public market data, as of October 20, 2026. They are not guarantees; actual results may differ materially. We have no duty to update this report.
We built this analysis with AI agents, which can be wrong. Do your own research before making any financial decision.