CoreWeave (CRWV)

Last updated: 2026-10-25 · Author: Claude (claude-fable-5-1 subagent) · Status: reviewed · Reviewer: Codex (gpt-6-astra) cross-check, applied 2026-10-17 · Updated 2026-10-21 from signals/2026-10-21.md · Updated 2026-10-25 from signals/2026-10-25.md

Summary

• CoreWeave buys Nvidia GPUs and rents compute under committed contracts to customers including Microsoft, OpenAI, Meta, Nvidia and Jane Street. Q2 2026 revenue was $2.575bn (+112% y/y); 2026 revenue guidance is $12.4–13.2bn. June backlog was $104.2bn, excluding >$25bn of commitments announced in early Q3; those announcements are not an October backlog balance. The three largest customers supplied 72% of Q2 revenue (claims 1, 11–19).
• Expansion depends heavily on external financing. June debt principal was $35.551bn and operating lease liabilities $16.319bn, or $51.870bn combined. Identified subsequent gross borrowing additions of $6.6bn imply $42.151bn debt before repayments and other movements. Capex guidance is $35–39bn; another $35.5bn of undiscounted uncommenced lease payments sits outside recognized lease liabilities. The September ATM is a financing option, not evidence of cash raised (claims 10, 20–25, 41, 47).
• CoreWeave is not GAAP-profitable at its current scale and financing cost. Q2 adjusted EBITDA was $1.510bn (59% margin), but D&A of $1.393bn and net interest of $640m explain most of the bridge to a $626m net loss. Growth can create costs before revenue contribution; it does not make profitability impossible (claims 3–8).
• Q2 adjusted EBITDA covered net interest 2.36×; this does not establish cash coverage of principal and interest. H1 CFO less cash PP&E purchases was −$10.454bn before principal repayments. Delivery spending, cash collection timing, prepayments, restricted cash and borrower-level obligations determine liquidity; RPO is a revenue-recognition schedule (claims 13, 24, 48, 52).
• October credit and equity markets sharply repriced risk. CNBC reports five-year CDS ~1,082bp on 19 October, nearly 1,081bp on 20 October and 1,065.84bp on 21 October; the 9.75% 2031 yield was 15.97% on 20 October. These are secondary snapshots, not synchronized closes. Shares closed at $61.39 on 16 October, down 25.2% for the week. IFR’s 23 October report describes sector financing delays, while the CEO says CoreWeave retains borrowing, equity and prepayment options. This establishes financing pressure, not that every funding channel has closed or a default has occurred (claims 31, 33–34, 41, 55–58).
• Equity is residual and exposed to dilution; unsecured creditors lack secured lenders' collateral priority, but recoveries depend on each borrower, collateral and guarantees. June DDTLs totalled $13.643bn; identified unsecured notes/converts total $20.817bn after September. Nvidia's January holdings would equal about 8.6% of June total shares if unchanged, and its $6.3bn capacity arrangement has delivery conditions and termination rights (claims 17–18, 21–22, 48–49).
• GPU renewal pricing remains a risk, particularly for DDTL 5.5's five-year debt against contracts averaging three years. Six-year depreciation does not prove that every GPU needs renewals to recover its cost. Management reports A100/H200 renewals and attractive initial-term returns; those disclosures challenge the original thesis but do not establish audited fleet-wide returns. OCPI settled H100 at $3.02, H200 at $6.20 and B200 at $9.57/GPU-hour on 24 October; dated Silicon Data H100 was $2.88 on its separate basis. These support current paying demand without establishing fleet-wide renewal returns. The timing of the first large renewal cohort remains unverified (claims 28, 30, 39, 43–45, 53, 57).

Key claims

Analysis

What the business is and where it sits in the stack

CoreWeave converts Nvidia GPUs into contracted rental revenue. All GPUs in its fleet are Nvidia's (the 10-Q describes this as a contractual obligation to customers), and three unnamed suppliers accounted for 60% of 2025 purchases. Gross technology equipment was $33.823bn at June 2026, plus $11.918bn construction in progress. CoreWeave leases buildings, recognizes $16.319bn of operating lease liabilities and increasingly co-invests in data-center joint ventures (up to $1.7bn committed in H1). Active power grew from 420MW in March 2025 to 1.5GW in June 2026; June contracted power of 3.7GW was superseded by a disclosed 4.2GW at 11 August. The company targets more than 1.85GW active at year-end and more than 5GW of AI factories by 2030 under its January Nvidia framework. Q2 10-Q; releases 12, 19; pricing update.

Long-lived commitments extend beyond recognized liabilities: existing operating leases represent $29.135bn undiscounted payments; another $35.5bn relates to leases not yet commenced, excluding specified uncertain arrangements including a site with a $14.7bn contractual maximum. These are separate undiscounted commitments, not amounts to add to discounted liabilities and call current debt. The planned 240MW India deployment announced on 7 October, first phase expected mid-2028 with an opportunity for another 240MW, adds execution and funding exposure without a disclosed capex or end-customer backlog amount (claims 46–47). Q2 10-Q, Note 8; India announcement.

Ninety-eight per cent of revenue comes from committed contracts, whose initial terms generally range from one to six years. Microsoft supplied 67% of 2025 revenue; Q2 2026's largest customers supplied 36%, 26% and 10%. The A–D labels need not identify the same company across periods, so a continuing Microsoft share trend is an inference. Cumulative announced OpenAI commitments total up to $22.4bn. Meta's September $14.2bn agreement was followed by a March 2026 order, announced in April, covering new capacity and exercise of an existing additional-capacity option; Reuters reports the ~$21bn as additional to September's agreement. Jane Street committed ~$6bn, Nvidia's initial order was $6.3bn and the original Microsoft MSA's value is undisclosed. These announced totals are not remaining RPO. The 10-K explicitly flags OpenAI's credit risk. 10-K; claims 14–19, 50; April Meta 8-K; Reuters.

Profitability: EBITDA is real, profit is not

Adjusted EBITDA margins ranged from 56–62% over the six quarters, with $1.510bn in Q2 2026. The full Q2 bridge in millions is 1,510 − 1,393 D&A − 640 net interest − 165 SBC − 62 tax − 1 acquisition costs + 125 other income = −626. D&A and net interest explain 95.2% of the $2,136m gap. Annualized D&A is $5.572bn, but comparing total D&A with quarter-end gross technology equipment does not validate the six-year useful-life estimate: D&A includes other assets and the fleet changes during the quarter. Q2 release, reconciliation; Q2 10-Q, Note 5 (A components, B arithmetic).

Adjusted operating income was $128m, a 4.97% margin versus $200m/$1,212m = 16.50% a year earlier (company-displayed prior margin 16%). It excludes SBC, acquired-intangible amortization and acquisition costs, so it retains most, not all, D&A. The accumulated deficit was $4.009bn. Q2 net interest absorbed 24.85% of revenue. Q3 guidance midpoints imply 25.53% and interest 3.91× adjusted operating income; endpoint ranges are 23.89–27.25% and 3.31–4.70× respectively. Growth creates deployment costs before full revenue contribution, but growth alone does not rule out future GAAP profitability. Q2 release; August outlook (A reported/guided values, B ratios).

H1 CFO was $3.663bn and benefited from prepayments and other working-capital movements. Deferred revenue increased from $8.185bn to $9.692bn, a $1.507bn balance-sheet rise; the cash-flow reconciliation contribution was separately $1.365bn. Management reports prepayments in about 70% of Q2-signed deals and expected contribution margins 5–10 percentage points above recent deals. These terms can reduce upfront funding needs, but deferred revenue also represents future delivery obligations; neither measure establishes future unencumbered cash (claims 44, 52). Q2 10-Q; September presentation, slide 7.

Financing: the loan stack

June debt principal was $35.551bn; the carrying amount after discounts and issuance costs was $35.068bn. Adding the September $4.2bn convert, $1.2bn DDTL 5.5 draw and $1.2bn revolver draw gives $42.151bn. This is an identified gross-additions illustration before repayments, other draws and FX movements, not a reported September or October balance. Q2 10-Q; closing 8-K.

1. GPU-backed delayed-draw term loans. June balances total $13.643bn, or $14.843bn including the disclosed August draw before other subsequent movements. SPVs pledge particular equipment and contracts; most June DDTLs also have parent guarantees. Blackstone, Magnetar and Coatue participated in the early facilities; Morgan Stanley, MUFG and JPMorgan lead later bank syndicates. DDTL 4.0 is non-recourse except for customary carve-outs, rated A3/A(low), with ~$7.5bn initially available and up to $8.5bn at stabilization, priced at SOFR+2.25% or ~5.9% fixed. The facility rating does not establish the unnamed customer's own rating or guarantee recovery. DDTL 5.0 is $3.1bn, Ba2/BB+, SOFR+4.5%, serving two non-IG customers. DDTL 5.5 is $2.6bn, Ba2/BB+, SOFR+5.5%, with approximately five-year debt against contracts averaging three years: lenders explicitly underwrite renewal risk. Pledged collateral excluding 4.0 was $18.2bn non-current and $2.6bn current. The old ~$25bn committed-DDTL total is not reconciled and cannot be treated as undrawn liquidity. Q2 10-Q, Note 10; releases 20, 25, 26.
2. Unsecured notes and converts. June straight notes of $10.029bn plus converts of $6.588bn and September's $4.2bn convert total $20.817bn, before subsequent movements. The September conversion price of $97.85 was 22.5% above issuance-day reference $79.88 and is 59.4% above $61.39; the stock is 37.3% below conversion. April capped calls cost $492m; September capped calls cost $566.2m. September proceeds after purchaser discounts were $4,137.0m; subtracting capped calls leaves $3,570.8m before remaining offering expenses, materially below principal (claim 42). Q2 10-Q; closing 8-K (A inputs, B arithmetic).
3. Vendor and bank lines. OEM/software financing was $5.102bn at 9–11% effective rates, with a separate $0.189bn Magnetar loan. The revolver is $2.5bn committed, with availability reduced by borrowings and letters of credit; $1.2bn was drawn in August and June letters of credit were $533m. These balances reconcile the $42.151bn gross-additions illustration. Q2 10-Q; 10-K.

The April noteholder presentation frames December debt/$3.09bn FY2025 adjusted EBITDA as 7.00× and a fully drawn $53.5bn stack/$16.1bn “LTM run-rate” EBITDA as 3.32×. The latter EBITDA is 2.67× annualized Q2 actual EBITDA of $6.04bn. Using the actual denominator gives 5.89× June debt, 6.98× gross-additions debt, or 8.86× the April fully drawn illustration. These are different dates and scenarios, not interchangeable leverage measures; the review checked the April-input arithmetic but could not independently read the numeric slide. April supplement; Q2 release; Q2 10-Q (B calculations on stated inputs).

DDTL 3.0's amendment aligned financing with delivery timing: $100m minimum liquidity applied to March/April payment dates, DSCR testing starts 31 October 2027, and contract-realisation testing was deferred to 28 February 2026. Unlimited equity cures apply before 28 October 2026; thereafter cures are permitted no more than three consecutive calendar months in any four consecutive calendar months (claim 51). The date alone is neither a default nor the end of all cures. Amendment 8-K.

The September 35m-share ATM, including collared forwards, provides an actual equity-financing mechanism. At $61.39 it would represent ~$2.15bn gross and shares equal to 6.35% of June's existing share count if fully sold at that price. It authorizes issuance rather than proving proceeds; the announcement anticipated at least a 30-day sales restriction after the convert purchase agreement. June disclosures separately identify $5.524bn cash/equivalents, $0.015bn securities and $10.014bn facility availability, plus $1.380bn restricted cash. That availability is dated and conditional, and restricted cash is not freely available parent liquidity. Management expected resources and borrowing availability to cover the year following the filing; that assertion needs a cash-flow stress test (claims 41, 48). ATM release; September 8-K; Q2 10-Q.

Can contracted revenue service the debt? The arithmetic

RPO of $103.7bn × 41% gives $42.517bn of revenue recognition over two years, averaging $21.2585bn annually versus a $10.300bn Q2 revenue run-rate, or 2.064×. There is no evidence of even annual recognition or matching cash collections. At an assumed 55–60% EBITDA margin, this scenario produces $11.692–12.755bn annual EBITDA; it is an estimate, not “contracted EBITDA.” RPO includes already-billed amounts, so some future recognized revenue will not generate new cash in that period. Q2 10-Q, Note 2; Q2 release (B revenue arithmetic; C assumed-margin scenario).

For illustration, 8% on $42.151bn is $3.372bn annual interest; annualized Q3 net-interest guidance is $3.44–3.76bn. The earlier blended-rate calculation, $592m ×4 / an estimated $30bn average debt, is 7.89%, not 7.8%, but that average debt is undocumented and cannot validate a rate. Contractual interest, net P&L interest and cash interest differ. Adding $6.184bn of June-scheduled 2027 principal to $3.372bn assumed interest gives $9.556bn service and 1.22–1.33× hypothetical EBITDA coverage. H2 2026 principal of $4.413bn is a half-year amount: annualizing it plus the same interest gives $12.198bn and only 0.96–1.05× illustrative coverage. These scenarios mix a two-year revenue average with incomplete post-June debt schedules. Neither proves liquidity or borrower-level debt-service coverage. Q2 EBITDA/net interest is 2.36×, not a contracted-cash-flow result or an SPV DSCR. Q2 10-Q, Note 10; August outlook (C financing/coverage scenarios).

Company capex/revenue guidance is 2.89× at midpoints, with an endpoint range of 2.65–3.15×, but that is not an external funding requirement. H1 CFO of $3.663bn less cash PP&E of $14.117bn leaves −$10.454bn before $5.219bn principal repayments, or −$15.673bn after principal. CFO already includes $806m operating cash interest; $176m capitalized interest is part of investment. Do not subtract $6.201bn total debt service again. Debt issuance proceeds of $16.747bn were before repayments, leaving $11.528bn net debt cash inflow; equity proceeds were $2.982bn. Net financing cash flow after capped calls and other uses was $13.985bn. Known Q3 gross borrowing additions are $6.6bn, not a complete quarterly cash bridge. Q2 10-Q; closing 8-K (A disclosed, B sums/netting).

The funding dependence is clear, but a further $30–40bn funding need does not follow directly from $35–39bn capex guidance. It requires cash generation, prepayments, committed draws, repayments and other uses to be modeled. High secondary yields raise the prospective cost of new unsecured funding without resetting existing fixed coupons or proving issuance impossible. September's successful convert and ATM authorization predate the October selloff; neither proves post-selloff access, nor does the evidence establish that every channel closed. The cross-check found no verified post-selloff failed financing, missed payment or new corporate downgrade in its bounded search. A 21 October re-fetch of the SEC submissions inventory found the latest issuer filing dated 14 October (Form 4), with the latest 8-K dated 22 September; no 17–21 October 8-K or financing disclosure was located there. This is a bounded filing search, not proof about private negotiations or a new funding event. SEC inventory. IFR’s 23 October report adds evidence of sector issuance delays and reluctant lenders (claim 58), while CNBC reports the CEO describing several financing options and Huang frames the constraint as scarce working capital rather than absent demand (claim 56). These are reporting and attributed views, not proof of a closed CoreWeave financing or a declared default. Bernstein’s quoted $85bn estimate uses a $104bn backlog perimeter; it does not supersede the dated backlog and separate early-Q3 commitments in claim 12. IFR; CNBC; Huang interview.

Per MW, net in-service PP&E is $46.736bn − $11.918bn = $34.818bn, or $23.212m/MW on 1,500MW quarter-end power. Annualized Q2 revenue and actual EBITDA are $6.867m and $4.027m/MW-year. Net in-service PP&E/annualized EBITDA is 5.76 years, about 5.8; using gross in-service PP&E of $40.704bn gives 6.74 years. Neither ratio measures original-investment payback: mixed cohorts, prior depreciation, non-GPU assets and ramping-quarter earnings prevent that inference. A stylized fully debt-funded 9% charge on $34.818bn costs $3.134bn/year and leaves $2.906bn from $6.04bn EBITDA, a roughly 12-year ratio; actual debt amortizes and is not all priced at 9%. It does not prove that initial contracts fail to recover capital. Q2 10-Q, balance sheet/Note 5; Q2 release (B accounting ratios; C financing scenario).

Selected Q3 short contracts priced at ~$40m/MW annualized revenue, stronger expected contribution margins and disclosed renewals materially qualify that static fleet calculation. They cannot be extrapolated to the entire fleet. Cohort costs, stabilized margins, debt amortization and residual values are needed to establish returns (claims 43–45, 53). Pricing update; September presentation; corrected call.

Management's unaudited operating claims about initial-term returns, renewals and ~$40m/MW contracts should be read in light of the CEO/CFO conclusion that disclosure controls were ineffective because of continuing material weaknesses (claim 54). Q2 10-Q, Item 4.

Renewal risk, 2027–2029

The original 2027–2029 window is a risk horizon, not a verified first large renewal cohort: DDTL maturities and RPO recognition dates do not identify contract expiries. Six-year depreciation versus a hypothetical three-to-five-year contract leaves one-to-three accounting years, but initial contracts generally range from one to six years, and an accounting tail need not represent unrecovered investment. DDTL 5.5's five-year debt against contracts averaging three years is direct evidence of renewal exposure. 10-K; DDTL 5.5 release.

Claim 39 preserves primary OCPI settlements of H100 $3.12, H200 $6.16 and B200 $8.83/GPU-hour on 20 October and appends 21–24 October observations. The latest 24 October settlements are H100 $3.02, H200 $6.20 and B200 $9.57; H100 is up 8.6% over seven days and 3.4% over thirty, B200 up 12.9% and 19.5%. Claim 57 adds the separate Silicon Data H100 dated card/chart, $2.88 on 24 October, while retaining its conflicting undated $2.53 FAQ. These retrospectively retrieved index readings do not establish CoreWeave realized contract prices, utilization, collateral recoveries or complete-month benchmark coverage. The older tertiary $1.70→$2.35 H100 contract-index comparison remains unverified and is not spliced into OCPI. Ornn H100; Ornn B200; Ornn panel. Management's ~25% July price increase, A100 renewal into 2029 and three-year H200 renewal above original pricing are primary company disclosures. Management also says initial contracts repay asset debt and generate attractive returns, with near-term renewal exposure limited. Those are material counterevidence to universal renewal dependence, not independently audited cohort returns. September presentation, slides 7/10; corrected call, pp. 8/11.

A 50% rental-price decline can be used as an explicit stress assumption, not an observed outcome or a deduction from each GPU generation's performance. Its effect depends on cohort costs, outstanding debt and replacement demand. Nvidia's $6.3bn capacity arrangement may support utilization but requires delivery/availability and permits termination for breach and specified insolvency events. It neither guarantees renewal prices nor CoreWeave debt; enforceability in a particular insolvency has not been assessed. Nvidia order 8-K; claim 49.

Who holds the debt and who loses first

Equity is residual and faces valuation loss or dilution. Multiplying June's rounded 551m total common shares by $61.39 gives an illustrative $33.826bn market value, not a verified October capitalization. Nvidia's disclosed 47,213,353 January shares would equal 8.57% of that June share count if unchanged; its filed 11.5% was a January Class A percentage. Nvidia is supplier, investor and conditional capacity purchaser, but “main beneficiary” is a comparative judgment this note cannot establish. Q2 10-Q; Nvidia 13G/A; claims 17–18, 31.

Identified unsecured notes/converts total $20.817bn after September. Secured creditors have priority in their collateral, but there is no guaranteed sequential loss waterfall across separate SPVs and parent guarantees. Most June DDTLs retain parent recourse; DDTL 4.0's limited recourse has carve-outs and its rating is not a recovery guarantee. Vendor financing of $5.102bn includes equipment security; short maturity alone does not create legal seniority. Customers' $9.692bn deferred revenue represents service obligations whose recovery requires contract/entity analysis rather than an assumed universal unsecured ranking. The 78-cent tertiary quote applies to a particular bond snapshot, not all unsecured paper. Q2 10-Q, Notes 2/10; closing 8-K; claim 34.

What the securities discount

Using June's rounded share count and the 16 October price gives $33.826bn illustrative market cap, or 2.64× midpoint 2026 revenue guidance. Combining this with $42.151bn gross-additions debt and June $5.524bn cash/$0.015bn securities gives $70.438bn EV excluding leases, or 11.66× annualized Q2 EBITDA. These mixed-date illustrations omit subsequent cash flows and share changes; restricted cash is not deducted as freely available cash. They cannot by themselves establish whether equity is fairly valued. Q2 10-Q; August outlook; claims 22, 31 (B arithmetic; C valuation illustration).

CDS above 1,000bp and bond yields above 15% indicate severe repricing. With flat hazard and assumed 40% recovery, five-year risk-neutral default probability is 1 − exp[−spread ×5/0.60]: 56.5% at 1,000bp and 63.0% at 1,194.5bp. These are simplified pricing estimates, not real-world default forecasts; a bond yield alone cannot identify a probability without benchmark rates, recovery and liquidity assumptions. Positive equity value can coexist with substantial default risk because equity retains upside in survival states. Dilution, liquidity premia and divergent beliefs are possible explanations, not an exhaustive pair of reconciliations (claims 33–34; C model assumptions).

Conflicts and open questions

• Financing pressure versus access (25 October). IFR’s 23 October sector freeze report qualifies the note’s earlier lack of verified post-selloff financing obstacles, but the CEO’s range-of-options memo and Huang’s working-capital argument support a less absolute interpretation than universal closure. IFR’s Bernstein $85bn/$104bn funding/backlog estimate is an analyst scenario with an older backlog perimeter, not a filed replacement for claim 12. No new closed CoreWeave financing or declared default is established by these sources. IFR; CNBC; Huang interview.

• Silicon Data dated display versus FAQ (25 October). The dated 24 October H100 card/chart says $2.88; the same page’s undated FAQ says $2.53. The dated display supersedes the prior source-access limitation, while its Saturday chart point does not establish publication freshness or complete-month coverage. Neither benchmark is CoreWeave’s realized rent. Silicon Data; claim 57.

• New rental evidence versus the prior source limitation (21 October). The note previously had only D-grade tertiary rental-index figures. OCPI now supplies primary 20 October settlements and dated history (claim 39). This improves the evidence for current rental demand while leaving CoreWeave utilization, reserved-contract renewals and resale recoveries unresolved. At the 21 October check, Silicon Data’s page said daily readings were temporarily unavailable; the prior 17 October $2.82 observation remains historical. The 25 October re-fetch recovers dated 21–24 October card/chart readings (claim 57), with a conflicting undated FAQ. This supersedes the prior access limitation without proving uninterrupted publication coverage. Ornn H100; Ornn B200; Silicon Data.

• Debt perimeter (21 October). CNBC’s 20 October ~$51.6bn headline explicitly includes about $35bn in notes, bonds and loans. It is broader than filed June debt principal of $35.551bn (claim 20); the existing June debt-plus-operating-lease calculation is separately $51.870bn (claim 25). These are not comparable successive principal balances, and neither replaces the mixed-date gross-additions illustration. CNBC 20 Oct; Q2 10-Q.

• Cash coverage: original note versus cross-check. The original judged contracts to cover debt service comfortably and said CoreWeave could service debt if customers paid and no more equipment was bought. The review agrees that demand is contracted but finds no maturity-matched cash proof: RPO recognition and assumed EBITDA margins omit prepayment amortization, delivery spending, restricted cash and borrower-level service. The revised Analysis retains the scenarios and 2.36× EBITDA/net-interest ratio without calling them cash coverage.

• Profitability: original note versus cross-check. The original said CoreWeave “cannot be” GAAP-profitable while growing. The review finds that growth alone does not prohibit profit; utilization, pricing, depreciation and financing costs jointly matter. The evidence establishes current losses and a large financing burden, not an accounting impossibility.

• Initial-term returns and renewals. The original inferred a 5.6-year payback, universal unrecovered one-to-three-year GPU tails and first large roll-offs in 2027–2029. The review corrects the mixed-cohort ratio to 5.8 years and rejects those payback/timing deductions. Management says initial contracts repay asset debt and produce attractive returns, while renewal exposure is limited; it reports A100/H200 renewal examples. The review cannot independently prove management's returns either. Cohort cash flows are needed to resolve this disagreement. September presentation, slide 10; corrected transcript, pp. 8/11.

• Funding access. The original called the unsecured market closed at acceptable cost and inferred a next $30–40bn raise. The review supports substantial external-financing dependence but finds neither a complete funding-needs calculation nor evidence that all markets closed. The convert and 35m-share ATM are counterevidence to inevitable exclusion, but predate the selloff; authorization, the anticipated 30-day restriction and actual issuance must be distinguished. Closing 8-K; ATM release; September 8-K.

• Recovery and Nvidia support. The original described a sequential equity/unsecured/SPV loss waterfall, treated short vendor maturities as seniority and presented Nvidia as backstop and main beneficiary. The review finds collateral priority directionally relevant but stresses parent recourse, separate borrowers and guarantees. Nvidia's conditional purchases and termination rights weaken unconditional rescue assumptions; the note lacks evidence to rank comparative beneficiaries. Insolvency enforceability remains unassessed. Q2 10-Q, Note 10; Nvidia order 8-K.

• What triggered the 14 October fall. The original/coordinator framed it as CPI-driven; contemporary reporting also framed it as credit-market stress. Claim 32's CPI surprise coincided with the share decline but does not establish causality. The 14 October CNBC page was previously unavailable, so those historical quotes remain snippet-based; the 19–20 October AMP bodies were fetched directly on 21 October. The review supports repricing while rejecting certainty about the trigger.

• CDS level. Reported snapshots range from above 1,000bp through 1,140bp to ICE's reported 1,194.5bp intraday peak. The review corroborates broad stress (B) but has no primary ICE series: exact quote/model estimate are C and June's ~452bp remains unverified. The review fetched Bloomberg's lead; the application fetch was unavailable, with TNW reporting the same peak. A simplified risk-neutral probability is not a real-world forecast. Bloomberg; TNW.

• Bond price and dates. The original lead said 76 cents/~16%, and the note used TradingNews's ~78/>16% for 2032s; the review grades that tertiary claim D rather than the note's C. BondTerminal shows 81.504/14.45% YTW dated 15 October (also D); CNBC's 14 October yield snippets are another snapshot. MT Newswires already reports the 2031s' 15.78% on 15 October, so it is not uniquely a 16 October observation. Times/methodologies may differ; no TRACE tape was obtained. None of these marks establishes the price of all ~$21bn unsecured paper. MT Newswires; BondTerminal; claim 34.

• Daily moves. One aggregator (roic.ai) reported an 8.3% fall on 15 October; Yahoo closes and CNBC show +5.9%. The Yahoo close series remains in use; the review checked the supplied-price arithmetic, not the exchange's original daily tape.

• Capex and prepayments. Q2 cash PP&E purchases of $6.422bn differ from company-defined $9.4bn capex because financed/accrued additions are excluded from the cash line. The CSV uses cash purchases. Likewise, deferred revenue's $1.507bn balance growth is not its $1.365bn cash-flow contribution. Expected improved new-contract margins and prepayments in ~70% of Q2 deals qualify a static fleet-wide funding calculation without proving free cash flow. Q2 10-Q; corrected call, p. 9; September presentation, slide 7.

• Meta $21bn. The original considered replacement versus addition unresolved. The review points to the April 8-K's new capacity plus prior-option exercise and Reuters's explicit report that ~$21bn is additional to September's $14.2bn. This supports an incremental announced commitment, while neither cumulative figure equals current remaining backlog. April 8-K; Reuters.

• Microsoft trend. Customer A fell from 72% to 36%, but labels may identify different customers across periods. The original inference of falling Microsoft dependence remains unverified for 2026; the observed largest-customer concentration fell.

• Run-rate versus actual. The company's April framing is 3.3× on fully drawn $53.5bn debt/$16.1bn run-rate EBITDA. The original's ~7× actual-EBITDA comparison uses subsequent gross-additions debt; June debt on annualized Q2 EBITDA is 5.89×, and fully drawn debt on that same denominator is 8.86×. The review validates arithmetic on supplied April inputs but could not independently read the image-only numeric slide. Delivery assumptions, dates and denominators must remain explicit. April supplement.

• Equity and credit pricing. The original said the two could reconcile only through dilution or credit illiquidity. The review considers that exhaustive framing unsupported: survival-state upside can preserve positive equity value alongside substantial default risk. Mixed-date valuation multiples do not settle which market is right.

• Insider selling. Form 144 volume (100 notices in three months) is primary; dollar totals circulate through aggregators. The cited 10b5-1 plans were adopted in November 2025 and March 2026, so sales were pre-planned but not therefore uninformative.

• Verification limits and grades. The FT McBee interview was not fetched; June debt plus operating lease liabilities is independently reproducible at $51.870bn, but no “>$35bn raised this year” quote is adopted. Nvidia's January holdings/Class A percentage now have a primary 13G/A; an October stake does not. Corporate ratings and older tertiary rental indices remain D; the newly fetched OCPI readings in claim 39 are A for the publisher’s observations. Primary guidance is A that management issued it, not proof of achievement.

• Core Scientific. CoreWeave remains Core Scientific's largest hosting tenant after the failed merger; megawatt energisation in 2026 was not verified.

Signals to watch

Sources

Original sources accessed 2026-10-17 unless otherwise identified. Update sources CF22, CU23, IFR23, HU22, OH, OB, OHH, SDH, CA, MS28, GO28, AM29 and FED27 re-fetched 2026-10-25; CNBC/IFR full bodies and Silicon Data embedded chart were accessible by direct request. SEC requests used generic User-Agent research-bot [email protected]. Dated 21 October CDS is separate from undated 1,145bp/~15% publication snapshots; neither undated quote becomes a CSV observation. Update sources CN19, CN20, FT20, OH, OB, OCPI, GS20, SDH, SECINV and amendment CA fetched 2026-10-21; CNBC AMP bodies were accessible by direct fetch despite browser-fetch restrictions. New figures were fetched from the linked sources when applying the cross-check. Bloomberg’s direct page was unavailable on the application fetch; exact market marks retain the review’s verification limits.

1. CoreWeave, Form 10-Q for the quarter ended 30 June 2026, filed 2026-08-12. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000366/crwv-20260630.htm

2. CoreWeave, "CoreWeave Reports Strong Second Quarter 2026 Results", 8-K Exhibit 99.1, 2026-08-11. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000362/coreweave2q26earningspress.htm

3. CoreWeave, Form 10-Q for the quarter ended 31 March 2026, filed 2026-05-08. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000222/crwv-20260331.htm

4. CoreWeave, Q1 2026 results release, 8-K Exhibit 99.1, 2026-05-07. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000220/coreweave1q26earningspress.htm

5. CoreWeave, Form 10-K for the year ended 31 December 2025, filed 2026-03-02. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000104/crwv-20251231.htm

6. CoreWeave, Q4 2025 results release, 8-K Exhibit 99.1, 2026-02-26. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000094/coreweave4q25earningspress.htm

7. CoreWeave, Form 10-Q for the quarter ended 30 September 2025, filed 2025-11-13. https://www.sec.gov/Archives/edgar/data/1769628/000176962825000062/crwv-20250930.htm

8. CoreWeave, Q3 2025 results release, 2025-11-10. https://www.sec.gov/Archives/edgar/data/1769628/000176962825000059/coreweave3q25earningspress.htm

9. CoreWeave, Form 10-Q for the quarter ended 30 June 2025, filed 2025-08-13. https://www.sec.gov/Archives/edgar/data/1769628/000176962825000041/crwv-20250630.htm

10. CoreWeave, Q2 2025 results release, 2025-08-12. https://www.sec.gov/Archives/edgar/data/1769628/000176962825000039/coreweave2q25earningspress.htm

11. CoreWeave, Form 10-Q for the quarter ended 31 March 2025, filed 2025-05-15. https://www.sec.gov/Archives/edgar/data/1769628/000176962825000014/crwv-20250331.htm

12. CoreWeave, Q1 2025 results release, 2025-05-14. https://www.sec.gov/Archives/edgar/data/1769628/000176962825000010/coreweave1q25earningspress.htm

13. CoreWeave, 8-K Item 1.01, NVIDIA order form ($6.3bn, residual capacity through 13 Apr 2032), 2025-09-15. https://www.sec.gov/Archives/edgar/data/1769628/000176962825000047/crwv-20250909.htm

14. CoreWeave, 8-K Item 1.01, OpenAI order form ($6.5bn through 31 May 2031), 2025-09-25. https://www.sec.gov/Archives/edgar/data/1769628/000119312525216497/d17274d8k.htm

15. CoreWeave, 8-K Item 1.01, Meta order form ($14.2bn through 14 Dec 2031), 2025-09-30. https://www.sec.gov/Archives/edgar/data/1769628/000176962825000050/crwv-20250925.htm

16. CoreWeave, 8-K Item 1.02, termination of Core Scientific merger agreement, 2025-10-31. https://www.sec.gov/Archives/edgar/data/1769628/000095010325014006/dp236631_8k.htm

17. CoreWeave, 8-K Item 8.01, termination of Series C put right, 2025-10-06. https://www.sec.gov/Archives/edgar/data/1769628/000176962825000052/crwv-20250925.htm

18. CoreWeave, 8-K Item 1.01, First Amendment to DDTL 3.0 credit agreement, 2026-01-02. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000003/crwv-20251231.htm

19. CoreWeave, 8-K, NVIDIA $2bn private placement and collaboration framework, 2026-01-26. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000044/crwv-20260123.htm

20. CoreWeave, "CoreWeave Closes Landmark $8.5 Billion Financing Facility" (DDTL 4.0), 8-K Exhibit 99.1, 2026-03-31. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000129/ex991.htm

21. CoreWeave, "CoreWeave and Meta Announce $21 Billion Expanded AI Infrastructure Agreement", 8-K Exhibit 99.1, 2026-04-09. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000154/ex991.htm

22. CoreWeave, supplemental investor information for senior notes offering (capitalization and leverage), 8-K Exhibit 99.3, 2026-04-09. https://www.sec.gov/Archives/edgar/data/1769628/000119312526148912/d136714dex993.htm

23. CoreWeave, 8-K, closing of 9.75% senior notes due 2031 and 1.75% convertible notes due 2032, 2026-04-14. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000164/crwv-20260409.htm

24. CoreWeave/Jane Street, "Jane Street Signs $6 Billion AI Cloud Agreement with CoreWeave", 8-K Exhibit 99.1, 2026-04-15. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000167/ex991.htm

25. CoreWeave, "CoreWeave Closes $3.1 Billion Loan Facility" (DDTL 5.0), 8-K Exhibit 99.1, 2026-05-18. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000236/ex9912.htm

26. CoreWeave, "CoreWeave Closes $2.6 Billion Loan Facility" (DDTL 5.5), 8-K Exhibit 99.1, 2026-08-10. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000357/ex991pr.htm

27. CoreWeave, 8-K, closing of $4.2bn 2.875% convertible senior notes due 2033 and capped calls, 2026-09-22; pricing release 2026-09-18. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000432/crwv-20260917.htm ; https://www.sec.gov/Archives/edgar/data/1769628/000176962826000432/ex991pricing.htm

28. CoreWeave, "CoreWeave Closes $2.6 Billion Secured Debt Financing Facility" (DDTL 3.0), 8-K Exhibit 99.1, 2025-07-31. https://www.sec.gov/Archives/edgar/data/1769628/000176962825000033/ddtl30pressrelease-ex991x6.htm

29. SEC EDGAR, CoreWeave Form 144 filings list. https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001769628&type=144&dateb=&owner=include&count=100

30. SEC EDGAR, Form 4, Michael Intrator, transactions through 2026-08-25. https://www.sec.gov/Archives/edgar/data/0001769628/000176962826000412/form4.xml

31. Yahoo Finance chart API, CRWV daily closes, 2025-03-28 to 2026-10-16. https://query1.finance.yahoo.com/v8/finance/chart/CRWV?range=2y&interval=1d

32. CNBC, "CoreWeave shares tumble as bond yields near 15% on concern about AI borrowing binge", 2026-10-14 (search snippet; page blocked to fetch). https://www.cnbc.com/2026/10/14/coreweave-stock-bond-yields-ai-debt.html

33. CNBC, "CoreWeave credit default swaps hit distress levels as investors sour on AI debt", 2026-10-15 (snippet). https://www.cnbc.com/2026/10/15/coreweave-credit-default-swaps-hit-distress-levels-as-investors-sour-on-ai-debt.html

34. CNBC Investing Club, "Why did CoreWeave stock get crushed this week?", 2026-10-16 (snippet). https://www.cnbc.com/investingclub/2026/10/16/coreweave-credit-default-swaps.html

35. CNBC Daily Open, 2026-10-16 (snippet). https://www.cnbc.com/2026/10/16/cnbc-daily-open-ai-inflation-coreweave-.html

36. The Next Web, "Nvidia's Huang says CoreWeave needs no backstop, as its credit swaps hit a record" (citing Bloomberg and ICE Data Services), 2026-10-15. https://thenextweb.com/news/huang-coreweave-backstop-stockholm

37. TheStreet via Yahoo Finance, "CoreWeave stock plunges 12%, but Wall Street analyst says buy the dip" (citing Bloomberg CDS data), 2026-10-16. https://finance.yahoo.com/markets/stocks/articles/coreweave-stock-plunges-12-wall-223927844.html

38. MT Newswires via Yahoo Finance, "CoreWeave Sharp Sell-Off About Cost of Funding, Not Demand, Analysts Say", 2026-10-16. https://finance.yahoo.com/markets/stocks/articles/coreweave-sharp-sell-off-cost-204945626.html

39. CNN Business via Yahoo Finance, "Nasdaq sinks more than 2% as investors raise questions about AI debt", 2026-10-16. https://finance.yahoo.com/markets/stocks/articles/nasdaq-sinks-more-2-investors-191031142.html

40. TradingNews, "CoreWeave Sinks to $60.96 as CDS Spreads Top 1,000bp", 2026-10-15 (tertiary). https://www.tradingnews.com/news/coreweave-sinks-to-60-usd-as-cds-spread-top-1000bp-and-35b-usd-debt

41. investinglive, "US September CPI 3.7% vs 3.6% expected", 2026-10-14. https://investinglive.com/news/us-september-cpi-vs-3-6-expected-0/

42. MarketBeat, "CoreWeave Q2 earnings call highlights", 2026-08-11 (auto-generated call summary). https://www.marketbeat.com/instant-alerts/coreweave-q2-earnings-call-highlights-2026-08-11/

43. Uvik Software, "GPU Price Index", as of 2026-10-07 (cites SemiAnalysis contract index). https://uvik.net/blog/gpu-price-index/

44. CoreWeave IR, "CoreWeave Expands Agreement with OpenAI by up to $6.5B", 2025-09-25 (via search snippet). https://investors.coreweave.com/news/news-details/2025/CoreWeave-Expands-Agreement-with-OpenAI-by-up-to-6-5B/default.aspx

45. Blackstone, "CoreWeave Secures $7.5 Billion Debt Financing Facility led by Blackstone and Magnetar", May 2024 (via search snippet). https://www.blackstone.com/news/page/3/

46. Motley Fool, "Nvidia just increased its stake by 95% in this AI company" (13F), 2026-05-26. https://fool.com/investing/2026/05/26/nvidia-just-increased-its-stake-by-95-in-this-ai-c

47. PitchBook via Yahoo Finance, "CoreWeave's not-so-secret weapons: price hikes and an Nvidia trump card", 2026-10-16. https://finance.yahoo.com/markets/stocks/articles/coreweaves-not-secret-weapons-price-233000657.html

48. Data file: data/coreweave-quarterly.csv (this project), compiled from sources 1–12.

49. CoreWeave, Q2 2026 outlook presentation, 2026-08-11, p. 3 (primary guidance). https://s205.q4cdn.com/133937190/files/doc_financials/2026/q2/CoreWeave-Q2-26-Earnings-Outlook-Presentation.pdf

50. CoreWeave, Q1 2026 outlook presentation, 2026-05-07, p. 3 (prior guidance). https://s205.q4cdn.com/133937190/files/doc_financials/2026/q1/CoreWeave-1Q26-Outlook-Presentation.pdf

51. CoreWeave, corrected Q2 2026 earnings-call transcript, 2026-08-11, pp. 8–11. https://s205.q4cdn.com/133937190/files/doc_financials/2026/q2/CRWV-US-CORRECTED-TRANSCRIPT-CoreWeave-Q2-2026-Earnings-Call-11August2026.pdf

52. CoreWeave, September convertible-offering presentation, 2026-09-17, especially slides 7, 10–11, 28, 30 and 37. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000429/crwv2026convertpresentat.htm

53. Nvidia, Schedule 13G/A, event date 2026-01-23 (47,213,353 shares; 11.5% of Class A). https://www.sec.gov/Archives/edgar/data/1045810/000104581026000008/xslSCHEDULE_13G_X01/primary_doc.xml

54. CoreWeave, “CoreWeave Announces At-the-Market Offering Program,” 2026-09-17. https://investors.coreweave.com/news/news-details/2026/CoreWeave-Announces-At-the-Market-Offering-Program/default.aspx

55. CoreWeave, 8-K, 2026-09-17 (ATM and anticipated initial sales restriction). https://www.sec.gov/Archives/edgar/data/1769628/000176962826000429/crwv-20260917.htm

56. CoreWeave, “CoreWeave Continues to Contract New Compute Capacity at Higher Prices,” 2026-09-17. https://investors.coreweave.com/news/news-details/2026/CoreWeave-Continues-to-Contract-New-Compute-Capacity-at-Higher-Prices/default.aspx

57. CoreWeave, “CoreWeave Enters India, Expanding AI Cloud Platform With AdaniConneX,” 2026-10-07. https://investors.coreweave.com/news/news-details/2026/CoreWeave-Enters-India-Expanding-AI-Cloud-Platform-With-AdaniConneX/default.aspx

58. CoreWeave, April 8-K, filed 2026-04-09, Item 8.01 (Meta order dated 31 March). https://www.sec.gov/Archives/edgar/data/1769628/000176962826000154/crwv-20260331.htm

59. Reuters via Investing.com, “Meta, CoreWeave deepen AI cloud partnership with fresh $21 billion deal,” 2026-04-09. https://www.investing.com/news/stock-market-news/coreweave-signs-21-billion-ai-cloud-deal-with-meta-4605144

60. Bloomberg News, “CoreWeave Risk Gauge Hits Record, Pricing 60% Odds of Default,” 2026-10-15 (lead fetched by cross-check; direct application fetch unavailable). https://news.bloomberglaw.com/daily-tax-report-international/coreweave-risk-gauge-hits-record-pricing-60-odds-of-default

61. Axios, “Warning signs are flashing on the AI trade. Here's where to spot them,” 2026-10-16. https://www.axios.com/2026/10/16/ai-coreweave-bonds-debt

62. Axios, “CoreWeave roils the markets, raises warning flag for the AI trade,” 2026-10-15. https://www.axios.com/2026/10/15/ai-coreweave-bonds-stocks

63. MT Newswires via Yahoo Finance, “Update: CoreWeave Execution Outweighs Financing Worries, BofA Says,” 2026-10-15, 5:14 p.m. EDT (retrieved search result; 15.78% quote). https://ca.finance.yahoo.com/news/coreweave-execution-outweighs-financing-worries-211401274.html

64. BondTerminal, CoreWeave issuer page (15 October-dated aggregator snapshots; D-grade). https://bondterminal.com/issuers/coreweave-inc