Goldman Sachs and Bernstein have put very different numbers on how much money CoreWeave still has to raise. A Goldman Sachs note dated 21 October, as reported by Cailianshe and relayed by BigGo Finance on 22 October 2026, estimates CoreWeave “would still need to raise roughly $19.5 billion to $23 billion in new debt to fulfill its $129 billion backlog as of August.” Bernstein analyst Madison Rezaei, according to TipRanks on 22 October 2026, estimates the company needs about $85 billion to fulfil existing customer contracts, leaving an estimated $76 billion to finance using outside capital.
The two figures measure different things. Goldman’s range nets out customer prepayments, available funds, committed but undrawn loans and operating cash flow. Without prepayments, and with all of CoreWeave’s locked-in power capacity brought online, Goldman’s own new-debt figure “would rise to about $84.5 billion.” A separate JPMorgan note, reported by Yahoo Finance, looks at the lenders and estimates that large US investment banks and Japan’s MUFG could each have exposure of about $1 billion to $3 billion.
Business Pill · NEOCLOUDS
A short Business Engineer explainer on neoclouds: companies that rent AI chips by the hour. It teaches the general idea only and says nothing about any company or estimate in this piece.
The key insight: As we read it, the three notes disagree less than the headline numbers suggest: Goldman’s own figure rises from roughly $19.5 billion to $23 billion to about $84.5 billion once prepayments are taken out and all 4.2 gigawatts are built, so much of the gap between the banks comes from what each one counts.
What Goldman Sachs Estimated
Goldman’s note is titled “CoreWeave: How Big Is the Funding Gap?”, according to the Cailianshe report. These are the figures BigGo Finance relays from it:
- The backlog case. Assuming customer prepayments equal 20% of contract value and a GPU useful life of 4.5 to 6 years, CoreWeave would need roughly $19.5 billion to $23 billion in new debt to fulfil its $129 billion backlog as of August.
- The full build-out. If CoreWeave brought all of its locked-in 4.2 gigawatts of power capacity into operation, new debt needs would expand to about $38 billion. Excluding prepayments, “that figure would rise to about $84.5 billion.”
- Prepayments. About 70% of CoreWeave’s contracts include prepayment arrangements, averaging 15% to 25% of total contract value, according to the note as reported.
- Debt load. Total short- and long-term debt is forecast to rise from about $50.5 billion to about $97.3 billion by 2028, while net debt to EBITDA falls from 6.2x in 2026 to 2.8x in 2028.
- The rating. Goldman upgraded CoreWeave to Buy on 19 October and kept a $106 price target.

What Bernstein Estimated
Bernstein’s Madison Rezaei lowered her price target from $74 to $51 and kept a Sell rating, TipRanks reported. According to Rezaei, the company needs about $85 billion to fulfil its existing customer contracts, and only $9 billion of that is covered by the financing facilities already in place. TipRanks added that CoreWeave’s unsecured bonds recently traded at yields of around 15%.
Who Holds the Debt: JPMorgan’s View
JPMorgan’s note, by analysts led by Kian Abouhossein and published on Tuesday, looks at the banks lending to CoreWeave, according to Yahoo Finance. It estimates large US investment banks and MUFG could each have exposure of about $1 billion to $3 billion, while most European and Asian banks are estimated to have less than $500 million. Yahoo Finance notes these are JPMorgan’s assessment, not figures reported by the banks.
- Loan structure. Much of the exposure is in delayed draw term loans, largely held through special-purpose vehicles, with funds released as projects meet development milestones.
- The amounts. CoreWeave has about $24 billion in committed delayed draw term loans: $15 billion drawn and $9 billion available but unused.
- The verdict. “From the perspective of financing banks, we view bank exposures to CoreWeave as well managed and very well secured,” the analysts wrote.
- The bonds. The yield on CoreWeave’s bonds maturing in 2031 approached 16% last week, and the cost of insuring against a default rose to about 1,000 basis points, Yahoo Finance reported.

The Table: Three Banks, Three Numbers
The table sets the three notes side by side. Each one answers a different question: how much new debt, how much outside money, and who already holds the loans.
| Bank | What it estimates | Key assumption or context | Rating |
|---|---|---|---|
| Goldman Sachs (note dated 21 October, via Cailianshe) | About $19.5 billion to $23 billion of new debt to fulfil the $129 billion backlog as of August; about $38 billion if all 4.2 gigawatts of locked-in power come online; about $84.5 billion excluding customer prepayments | Assumes prepayments of 20% of contract value and a GPU useful life of 4.5 to 6 years | Buy, $106 target |
| Bernstein (Madison Rezaei, via TipRanks) | About $85 billion needed to fulfil existing customer contracts; $9 billion covered by facilities in place; about $76 billion to finance with outside capital | Unsecured bonds recently traded at yields of around 15% | Sell, target cut to $51 from $74 |
| JPMorgan (Kian Abouhossein, note published Tuesday, via Yahoo Finance) | About $24 billion of committed delayed draw term loans: $15 billion drawn, $9 billion available but unused; about $50 billion of debt in total | Free cash flow negative until 2030, by JPMorgan’s forecast | Overweight, $125 target for the end of 2027 |
Goldman Sachs figures as reported by Cailianshe and relayed by BigGo Finance on 22 October 2026; Bernstein figures as reported by TipRanks on 22 October 2026; JPMorgan figures as reported by Yahoo Finance on 22 October 2026. The three estimates measure different things and are not directly comparable.
The Debt Backdrop in Our Coverage
- Our piece on CoreWeave’s debt yield covers CNBC’s 21 October report that CoreWeave’s unsecured debt was “sporting a yield of about 15%.”
- Our piece on neocloud short sellers covers ORTEX data, reported by Reuters, showing close to $1.7 billion of October paper profits for CoreWeave short sellers.
- Our piece on Nvidia’s data center backstops covers Nvidia’s guarantees and its $7 billion of direct investments or pledges in neocloud companies.
Benzinga ran the headline “CoreWeave Debt Stress Mounts: Bond Yield Hits Record as Stock Suffers $44 Billion Wipeout” on 22 October 2026. We could not read the article behind it, so its figures are not used here.
The Structural Read
As we read it, customer prepayments are the swing factor in Goldman’s model: the note as reported calls them “the most important pressure-relief variable,” and about 70% of CoreWeave’s contracts include them.
As we read it, the lenders and the bond market are telling different stories: JPMorgan describes bank exposures as “well managed and very well secured,” while the 2031 bonds approached a 16% yield last week, according to Yahoo Finance.
As we read it, the $9 billion figure appears on both sides: Bernstein says $9 billion of the need is covered by existing facilities, and JPMorgan puts the unused part of the delayed draw term loans at $9 billion. The reports do not say whether these are the same facilities.
JPMorgan analysts led by Kian Abouhossein, as reported by Yahoo Finance, 22 October 2026
“From the perspective of financing banks, we view bank exposures to CoreWeave as well managed and very well secured”
Three Implications
FOR COREWEAVE Goldman’s note as reported says even its most optimistic scenario leaves about $19.5 billion of new debt that must be raised in the capital markets.
FOR ITS LENDERS JPMorgan estimates large US investment banks and MUFG at about $1 billion to $3 billion of exposure each, and most European and Asian banks at under $500 million, according to Yahoo Finance.
FOR NVIDIA Our earlier piece reports Nvidia’s guarantees and its $7 billion of direct investments or pledges in neocloud companies, the group CoreWeave belongs to.
The Business Engineer Lens
This story maps onto the Business Engineer framework The AI Capex Map.
The framework’s thesis: “the capex race is no longer a corporate spending cycle. It is a sovereign-scale financing problem dressed in tech-company clothing”.
As we read it, the three banks’ notes on CoreWeave are estimates of that financing question for one neocloud, one of the groups the framework counts in its 2026 compute capex total. The framework was published in May 2026 and does not discuss these figures.
What Is Not Established
We have not read the Goldman Sachs, Bernstein or JPMorgan notes themselves. Goldman’s figures come from Cailianshe’s report as relayed by BigGo Finance, Bernstein’s from TipRanks and JPMorgan’s from Yahoo Finance. Each estimate rests on its own assumptions about prepayments, GPU life and capacity, so the three cannot be read as one range. Nothing here is investment advice.
The Bottom Line
Goldman Sachs estimates CoreWeave needs roughly $19.5 billion to $23 billion in new debt to fulfil its $129 billion backlog, while Bernstein’s Madison Rezaei puts the amount to be raised from outside at about $76 billion. Goldman’s own figure rises to about $84.5 billion for the full build-out without customer prepayments. JPMorgan estimates the largest lending banks each hold about $1 billion to $3 billion of exposure, which it describes as well secured.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. This piece is based on BigGo Finance’s 22 October 2026 relay of Cailianshe’s report on a Goldman Sachs note, TipRanks’ report on Bernstein, Yahoo Finance’s report on a JPMorgan note and CNBC’s 21 October report, each read in full; Benzinga’s headline (the article was not readable); and our own earlier pieces. We have not read the bank notes themselves. Nothing here is investment advice.
Sources: BigGo Finance relaying Cailianshe on Goldman Sachs’ 21 Oct note (22 Oct 2026) · TipRanks on Bernstein’s Madison Rezaei (22 Oct 2026) · Yahoo Finance on JPMorgan’s bank-exposure note (22 Oct 2026) · CNBC (21 Oct 2026)









