Sources: $CRWV & The $5 Trillion IOU
Every figure in the piece is keyed below to its source and as-of date. Figures derived by calculation are labeled as such, with inputs shown so the arithmetic can be reproduced. Prices are as of the October 23, 2026 close. Two sources used in this piece are not primary and are labeled as such throughout: Bank of America research notes are cited via press coverage because the notes themselves are not public, and Jensen Huang’s GTC Berlin remarks are cited via event coverage and independent transcripts.
Figure by figure
Figure 1. Share price since IPO. Month-end Nasdaq closing prices, March 2025 through October 2026, from exchange data via Yahoo Finance, pulled October 24, 2026. IPO priced at $40.00 (first trade March 28, 2025); all-time closing high $183.58 on June 20, 2025; October 23, 2026 close $54.235 exactly (a half-cent print), rendered $54.24 in the piece and cross-checked against a second provider showing $54.24. The decline from high is 1 minus 54.235/183.58, or 70.5%, reported as 70%. Placement reference points: NVIDIA purchased 22,935,780 Class A shares at $87.20 (aggregate $2.0B) per the Form 8-K dated January 26, 2026; Jane Street purchased 9,174,311 shares at $109.00 (aggregate ~$1.0B) per the Form 8-K dated April 15, 2026. $54.235 against $87.20 is a 37.8% discount, reported as 38%; against $109.00 it is 49.8% of the price, reported as half. CoreWeave Forms 8-K · Nasdaq price history
Hyperscaler capex (in text). Purchases of property and equipment, most recent reported quarter, from SEC EDGAR XBRL: Alphabet $44.9B and Microsoft $35.8B (quarters ended June 30, 2026), Oracle $28.5B (quarter ended August 31, 2026), CoreWeave $6.4B (quarter ended June 30, 2026). Fiscal calendars differ; the chart subtitle says so. Amazon is excluded because its capex line (purchases of property and equipment, including finance leases) is tagged differently across periods and a clean like-for-like quarter was not assembled; its absence understates total hyperscaler spend. As of the dates above. SEC EDGAR company facts
Figure 2. Oracle remaining performance obligations. $137.8B (quarter ended May 31, 2025), $455.3B, $523.3B, $552.6B, $638.0B, $664.0B (quarter ended August 31, 2026), from Oracle filings via EDGAR XBRL, concept RevenueRemainingPerformanceObligation. Oracle does not disclose the customer composition of the balance, and the piece makes no attribution of it. Oracle Forms 10-Q and 10-K
Figure 3. Neocloud revenue scale. Most recent reported quarterly revenue: CoreWeave $2,575M (Q2 2026 Form 10-Q), Nebius $582.3M (Q2 2026 results, Form 6-K furnished August 2026), Applied Digital $341.9M total revenue (quarter ended August 31, 2026, Form 10-Q filed October 7, 2026; comprises $262.8M of services and other revenue, of which approximately $157.2M is one-time tenant fit-out work, plus $79.1M of data center rental revenue, the landlord line the piece describes), Core Scientific $164M (Q2 2026 Form 10-Q), IREN $137M (fiscal Q4 ended June 30, 2026, derived from FY2026 Form 10-K full-year revenue of $707.0M less the nine-month total; IREN’s FY26 results release reports Q4 AI cloud revenue of $70.5M and bitcoin mining revenue of $66.7M, consistent with the derived total). The chart’s claim is “more than twice the rest combined”: 582+342+164+137 = $1,225M, and 2,575/1,225 = 2.10x. Two corrections were made in review: an earlier draft said “more than four times,” which was wrong, and an earlier chart used Applied Digital’s services line of $263M instead of total revenue. As of the periods above. CoreWeave 10-Q · Nebius 6-K · Applied Digital 10-Q · Core Scientific 10-Q · IREN 10-K
The Nvidia relationship table. Each row sources to a filing: supplier relationship and equipment base per the FY2025 Form 10-K business description and the Q2 2026 10-Q property note ($52.6B gross property and equipment at June 30, 2026, of which technology equipment $33.8B); customer MSA dated April 10, 2023 per the 10-K exhibit index; the September 9, 2025 order form with initial value $6.3B and the obligation to purchase residual unsold capacity through April 13, 2032 per the Form 8-K filed September 15, 2025 (the quoted phrase “is obligated to purchase the residual unsold capacity” is the filing’s own language; the same filing conditions the obligation on delivery and availability of service requirements and permits either party to terminate the MSA on a bankruptcy of the other not dismissed within 90 days, which the piece states); the $2.0B equity purchase per the Form 8-K filed January 26, 2026; the quoted phrase “leverage NVIDIA’s financial strength” from the joint press release furnished as Exhibit 99.1 to that 8-K; Applied Digital warrants (13,062,521 shares at $7.19 and 8,393,611 shares at $10.75, issued May 28 and August 28, 2025 in connection with data center leases) per Applied Digital’s Form 10-Q filed October 7, 2026 (21.46M warrant shares against 302.4M shares outstanding on the cover page, roughly 7%, unexercised); customer prepayments of $9.7B per the CoreWeave Q2 2026 10-Q deferred revenue disclosure. The piece’s “a company worth $5 trillion” for Nvidia: $210.96 October 23, 2026 close (exchange data via Yahoo Finance, pulled October 25, 2026) times 24.1B shares outstanding per the cover page of Nvidia’s Q2 FY2027 Form 10-Q (shares as of August 21, 2026), approximately $5.1 trillion.
Figure 4. Backlog by period. Revenue backlog of $9.9B (December 31, 2023), $15.1B (2024), $66.8B (2025), $104.2B (June 30, 2026) per the September 2026 convertible notes investor presentation (Form 8-K, September 17, 2026), which defines revenue backlog as remaining performance obligations plus other amounts the company estimates will be recognized under committed contracts. The Q2 2026 earnings release (August 11, 2026) states the $104.2B backlog excludes more than $25B of net new customer commitments added in early Q3 and the September 17 pricing release and deck report contracted power of approximately 4.2 GW as of the August call against 3.7 GW at June 30. RPO alone, the GAAP-defined subset, was $103.7B at June 30 per the 10-Q. The 246% growth is confirmed: the September deck shows June 2025 backlog of $30.1B, and 104.2/30.1 = 3.46x, or +246%. Q2 2026 earnings release · September 2026 investor presentation via Form 8-K
Figure 5. Customer concentration. Q2 2026: Customer A 36%, Customer B 26%, Customer C 10% of revenue; Q2 2025: Customer A 71%, no other customer at 10% or more (the chart stacks each quarter’s disclosed customers, so the 2025 column shows only Customer A). Customers A and B were each 32% of net accounts receivable at June 30, 2026. The 10-Q warns the letter labels may refer to different customers across periods. Microsoft is named in the FY2025 10-K as the top customer at approximately 67% of FY2025 revenue; the top-two concentration for FY2024 was approximately 77%. The piece’s “top three were 72%” is 36+26+10. As of June 30, 2026. CoreWeave Q2 2026 Form 10-Q · FY2025 Form 10-K
Named contract commitments (in text). OpenAI: master services agreement dated March 7, 2025 with commitments of up to approximately $11.9B through October 2030, and a second MSA (May 2025) with a September 2025 order form of up to approximately $6.5B through May 31, 2031, both per the FY2025 Form 10-K. Meta: order form of September 25, 2025 with an initial commitment of up to approximately $14.2B through December 14, 2031 and an option to materially expand through 2032, per the Form 8-K filed September 30, 2025; a further order form dated March 31, 2026 with an initial commitment of approximately $21B, per the Form 8-K filed April 9, 2026. The piece’s “roughly $35 billion” for Meta is 14.2 plus 21. Contract terms beyond these disclosures are redacted in the filed exhibits.
Contract texture (in text). 98% of revenue from committed, take-or-pay contracts (Q2 2026 10-Q and MD&A). RPO recognition timing: 41% expected within the initial 24 months ending June 30, 2028, 39% in months 25 through 48, remainder through month 78 (10-Q Note 2). Weighted-average prepayment of 15% to 25% of total contract value across active contracts as of year-ends 2023 through 2025 (10-Q risk factors). Deferred revenue $9.7B at June 30, 2026 against $8.2B at December 31, 2025. The $2.0B of operating cash flow on $189M of revenue is Q1 2024, from XBRL cash flow data, and reflects prepayment timing. Q3 short-duration contract pricing of roughly $40M per megawatt annualized and the phrase “significant demand for prior generations” are from the September 17, 2026 pricing release and investor presentation. The two-tier pricing in the piece is from Nebius’s Q2 2026 results (Form 6-K): four $1B+ multi-year deals at “a yield of $20-25 million per megawatt,” a seen “price opportunity in the $40-50 million per MW range” on new short-term capacity deals, and a stated expected payback on Q2 deals of 1 year 10 months.
The debt table. All balances, effective interest rates, and maturities from the Q2 2026 Form 10-Q, Note 10, as of June 30, 2026. Effective rates are as presented there, rounded to whole percents by the filing. Recourse debt net of discounts $31,405M; non-recourse $3,663M; total future principal $35,551M. The 2026 facilities’ terms (DDTL 4.0 $8.5B at SOFR+2.25%, non-recourse, March 2026; DDTL 5.0 $3.1B at SOFR+4.50%, May 2026; DDTL 5.5 $2.6B at SOFR+5.50%, August 2026, not yet drawn at June 30; 9.75% notes $2.75B, April; 9.625% notes $1.25B and EUR 8.50% notes EUR2.0B, June; 2.875% convertible $4.2B, September) are each from the Form 8-K announcing them. The covenant amendment tied to delivery timing was signed December 31, 2025 and filed January 2, 2026 (DDTL 3.0: minimum liquidity reduced to $100M for early-2026 payment dates, the debt service coverage test postponed to October 31, 2027, the contract realization test to February 28, 2026, and unlimited equity cures for those covenants permitted only before October 28, 2026, capped thereafter; all per the 8-K).
Figure 6. Debt and its cost. Total debt of $1.6B / $8.0B / $21.6B / $35.6B and weighted-average effective rates of 14.9% / 12.2% / 9.0% / 8.3% are the company’s own presentation (September 2026 convertible investor deck, which states its method: effective rates weighted by principal outstanding at period end, OEM financing estimated at 10%, FY23 rate computed on FY24 effective rates). The deck also shows an illustrative pro forma of $38.6B at 7.8% assuming a $3.5B convertible at 2.625%; the actual convertible printed upsized at $4.2B and 2.875%, and the deck figure does not reconcile to $35.6B plus either amount, so the pro forma point is excluded from the chart and the piece notes the actual print instead. “Up 22x” is 35.6/1.6, end-2023 to June 2026.
Figure 7. Maturity ladder. Future principal payments as of June 30, 2026: remainder of 2026 $4,413M, 2027 $6,184M, 2028 $4,416M, 2029 $2,421M, 2030 $3,221M, thereafter $14,896M, total $35,551M (10-Q Note 10). The piece’s “$10.6 billion maturity wall across 2027 and 2028” is 6,184 plus 4,416.
Figure 8. Funding gap. H1 2026 purchases of property and equipment $14,117M (7,695 + 6,422) against net cash provided by operating activities $3,663M (2,984 + 679), from XBRL quarterly values derived from the filings’ year-to-date statements. Reported as $14.1B and $3.7B.
Interest mechanics (in text). Q2 2026 contractual interest expense $592M; amortization of discounts and issuance costs $45M; capitalized interest $79M; interest expense on debt as reported $558M; the earnings release reports an “interest expense, net” line of $640M, which exceeds the debt note’s $558M because it includes interest items beyond the debt obligations (the release does not decompose the difference); the piece uses it as the most conservative run-rate basis. H1 2026 contractual interest $1,075M, capitalized $176M. The “42% of run-rate adjusted EBITDA” is (640x4)/(1,510x4) = 42.4%. All per the 10-Q Note 10 and the Q2 earnings release.
Figure 9. Revenue, D&A, interest. Quarterly values Q1 2024 through Q2 2026 from EDGAR XBRL company facts. Where only year-to-date values are tagged, quarters are derived by differencing consecutive year-to-date periods, and fourth quarters by subtracting nine-month from annual values. Reconciliation: the derived 2025 quarters sum to $5,131M against reported FY2025 revenue of $5.13B, and 2024 to $1,915M against $1.92B. “Up 13.6x” is 2,575/189.
Figure 10. Margins. Adjusted EBITDA margin 62% (Q2 2025) to 59% (Q2 2026); adjusted operating income margin 16% to 5%; adjusted EBITDA $753M to $1,510M; GAAP operating margin negative 2%; net loss $626M. All from the Q2 2026 earnings release, which defines the non-GAAP measures and provides reconciliations. Six-year straight-line depreciation for technology equipment and the full useful-life table per the FY2025 Form 10-K summary of significant accounting policies; the Q2 10-Q states no material changes. Technology equipment gross $9,146M (December 2024), $20,903M (December 2025), $33,823M (June 2026) per the property notes. Vera Rubin validation per the Q2 2026 earnings release highlights.
Figure 11 and the bonds table. Bond levels are secondhand, from press coverage of TRACE trade data during October 18 to 23, 2026, and moved daily within that window: the 9.75% senior notes due October 1, 2031 reported near 80.6 cents with a 15.58% yield to maturity in one account, a 15.97% yield on an 83 handle in another, and “about 15%” against the five-year Treasury at 4.92% in a third; the piece uses “near 80 cents” and “around 15.6%”; the chart notes the quotes come via press coverage of TRACE data. The mid-August yield of approximately 11.3% is from the same coverage. Hold-to-maturity arithmetic: ten semiannual coupons of 4.875 points equal 48.75 points, plus pull to par of 19.4 points from 80.6, totaling 68.15 points on an 80.6 outlay, or 84.6%, reported as roughly 84%; this ignores reinvestment of coupons and any default, and the separately quoted 15.6% YTM is the compounding-consistent figure. Equity comparisons use the scenario table ($191, $102, $32 against $54.24). The ~$19B of structurally or contractually senior claims sums the recourse DDTL facilities ($10.8B: 1,300+3,190+3,000+2,215+1,101), the non-recourse DDTL 4.0 ($2.8B), and the OEM financing arrangements ($5.1B: 4,220+882) from the Q2 2026 10-Q debt note, totaling $18.7B, reported as roughly $19B; the Magnetar loan ($189M) is excluded as unsecured. The 144A/QIB restriction is per each note offering’s Form 8-K. The distressed-exchange and recovery cautions are analysis, not events. Benzinga on the yield move · Morningstar on the slide · CNBC on a holder’s view
Valuation arithmetic
Market math. 458,871,690 Class A plus 92,664,912 Class B shares as of July 31, 2026 (10-Q cover page), total 551.54M. At $54.235: market cap $29.9B. Pro forma debt $39.8B ($35,551M principal plus the $4.2B September convertible). Pro forma cash of approximately $9.5B is an estimate: $5.5B of cash and equivalents at June 30 (XBRL) plus an assumed $4.0B of net convertible proceeds after fees and the cost of capped call transactions; the company has not disclosed the net figure, and marketable securities are excluded. Enterprise value of approximately $60.1B follows. EV to run-rate revenue of 5.8x uses Q2 annualized revenue of $10.30B; EV to run-rate adjusted EBITDA of 10.0x uses Q2 adjusted EBITDA annualized at $6.04B. “The equity is worth three quarters of the debt” is 29.9/39.8.
Scenarios. Author’s construction, stated as such in the piece. 2027 revenue of $16B / $21B / $26B against the RPO-implied average of roughly $21.3B per year from the first-24-month recognition window ($103.7B times 41% divided by 2). Adjusted EBITDA margins of 50% / 55% / 58% bracket the reported 59% with room for mix and pricing pressure. EV/EBITDA multiples of 6x / 7.5x / 9x are the author’s judgment, argued in the text. Equity value is scenario EV less pro forma net debt of $30.3B; per-share figures divide by 551.5M shares (a 10% dilution sensitivity in an earlier draft was cut in edit). The growth-debt sensitivity in the piece: $10B of additional net debt divided by 551.5M shares is $18.13 per share, reported as roughly $18; the H1 2026 funding gap of $10.5B (capex $14.1B less operating cash flow $3.7B, unrounded 14,117 less 3,663 = 10,454) is the stated pace reference.
GAAP cross-check. An income-statement cross-check table (scenario adjusted EBITDA less D&A at 40 to 45% of revenue less pro forma interest of $3.1B) appeared in an earlier draft and was cut in edit.
Price target mechanics. An earlier draft quantified floating-rate exposure (roughly $12.1B of contractual floating debt, about $30M of annual interest per 25 basis points); it was cut in edit along with the paragraph that used it.
Secondary sources, labeled
Bank of America. Four items, all via press coverage because the underlying notes are not public: the September 11, 2026 “Funding the Future” estimate of approximately $5.0 trillion of cumulative AI data center capex from 2026 through 2030 with roughly $1.2 trillion of external financing, primarily debt; the October 16 note arguing the buildout echoes prior infrastructure investment cycles in which the technology succeeded and investors still lost money; the October 23 note citing the withdrawn Firmus IPO as evidence of growing investor reluctance and describing flatlining U.S. business AI spending; and the October 23 projection of data center spending exceeding $1 trillion in 2026 and reaching approximately $2.1 trillion by 2028, with a build cost near $57 million per megawatt. Investing.com on the October 23 note · Coverage of the October 16 note · Coverage of the spending forecast · Coverage of the September report
Jensen Huang. GTC Berlin, October 2026, with provenance stated per quote. Verified against the full keynote transcript: a one-gigawatt AI factory “costs about $50 billion,” and “it’s $40 billion per year of revenue generation that comes out of that infrastructure.” The piece’s unquoted line “Demand is off the charts” is the author’s own characterization; it echoes Huang’s reported remarks at the event (via CNBC) but is not presented as a quotation. The piece no longer quotes “the inference inflection has arrived”; the trillion-dollar order figure from the March keynote is described as his stated vision, per event coverage. “King of AI” is the author’s nickname, not a quotation. The framing of financing as the binding constraint is likewise from event coverage and is paraphrased without quotation marks in the piece. GTC March 2026: approximately $1 trillion of order visibility, via event coverage. The January quote (“leverage NVIDIA’s financial strength”) is primary, from the joint press release filed as Exhibit 99.1 to the January 26, 2026 Form 8-K. CNBC GTC coverage · Berlin keynote transcript · GTC March coverage
Other secondary items. Nebius Q2 2026 figures are from its furnished Form 6-K results release (primary); the four $1B+ contracts and customer names per the same release and contemporaneous coverage. IREN FY2026 figures per its results release filed on Form 8-K; the FY2027 capex plan of $25B to $30B per earnings call coverage. The “worst day since June” for chip stocks and the phrase “neocloud distress” are both from CNBC’s October 23 GTC coverage, linked above under Jensen Huang. The securities class action (Masaitis v. CoreWeave, D.N.J., filed January 12, 2026) per the Q2 2026 10-Q legal note.
Not sourced, and limits
The identity of quarterly Customers A, B, and C is not disclosed by the company; the piece names only what filings name (Microsoft for FY2025, OpenAI and Meta as committed counterparties) and does not map letters to names. CoreWeave’s brand color (#0541E9) used in charts was sampled from coreweave.com stylesheets on October 24, 2026. Nebius, Applied Digital, Core Scientific, and IREN bars use neutral grays because verified brand colors were not established. The claim that OpenAI’s industry-wide compute commitments exceed its revenue many times over rests on widely reported aggregate figures for its contracts and financing rather than on a single filed document, since OpenAI is private; it is presented as context, not computation. The net proceeds of the September convertible and therefore pro forma cash are estimated, as stated above. The Fed funds path is deliberately not forecast. Two spec deviations are noted rather than hidden: comp margins are decomposed for CoreWeave only (Nebius, IREN, Applied Digital, and Core Scientific appear at the revenue and financing level, not with identically constructed margin stacks), the final figure count is 11 against a spec target of 12 to 16: the bond-market figure was added in review and the hyperscaler capex chart was cut in edit, with its numbers kept in the text.