Signals: 2026-10-21 (17-21 October)
Verification cutoff: 2026-10-21, 15:00 UTC (11:00 US Eastern), before the US close. Sources searched and fetched, or access attempted, on 21 October. USD unless specified. Grok is a source of leads, never evidence.
What changed since 17 October
- SpaceX's Q3 disclosure puts SI revenue at $7.560bn and operating profit at $431m, following Q2's $1.257bn operating loss; its capital spending remains much larger (17).
- SpaceX's September related-party financing balance is $12.897bn, below June's $13.329bn; the new exhibit does not name the creditor (18).
- Musk says SpaceX and Tesla have independent merger committees; there are no agreed terms (20).
- Reuters reports accelerated OpenAI fundraising talks; ~$30bn at >$1.4tn remains a proposal, not money received (21).
- OpenAI models enter Google Cloud private preview, with purchases eligible against existing GCP commitments (22).
- Anthropic proposes a 5% share allocation to a public-benefit vehicle; fresh draft-prospectus reporting adds historical losses, not an IPO price or cash-burn measure (23, 25–26).
- Oracle CDS widened to reported 315/322bp on 19 October; CoreWeave protection remained above 1,000bp and its 2031 bond yielded 15.97% on 20 October (1–5).
- OCPI's 20 October settlements are H100 $3.12/B200 $8.83 per GPU-hour; a fresh Silicon Data H100 print could not be established (30–33).
- S&P's 19 October opacity warning is new commentary, with no rating action; Jupiter's revised hearing order and Oracle's emissions proposal predate the cutoff (11, 34–35).
- Meta Q3 has not been released at this cutoff. Its after-close report and next week's results remain scheduled events, not evidence of changed capex (29 and calendar).
Reading this verification
The 61 numbered bundles below cover both concatenated scans, consolidating overlaps. G denotes the broad scan's item number; C denotes the company pass's material-item number. The company pass's calendar and outside-window context are reconciled separately. Counts refer to bundles, not independent facts or sources.
Confirmed means the material rendering was traced; partly means an important component, period, definition or figure needs correction or remains unchecked; unconfirmed means the substantive numerical/absence claim was not established (including a contradicted claim); lead only means an unverified attributed opinion. A failed fetch does not prove a claim false. Private-company leaks remain reported, unverified, C, even when the publication of the report is confirmed. A primary company's forecast or index display earns A for what was published, not for the future outcome or market representativeness.
Novelty is measured against the 17 October information set, with the 17 October log and 18 October verification as comparators: Yes = verified new observation/disclosure after 17 October; No = older/recirculated; Mixed = fresh reporting alongside older underlying figures; Unestablished = no verified new observation. Old facts newly found in this verification are labelled explicitly. These categories prevent a fresh article date from turning an old contract or permit order into a new event.
Credit prices, financing access and ratings
| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 17 Oct? |
|---|---|---|---|---|---|---|
| 1. CoreWeave, 20 Oct (G1) | CNBC: 9.75% 2031 YTM 15.97%; five-year CDS nearly 1,081bp, +13% week. Lloyd describes overreaction, while owning the bonds. The ~$51.6bn debt headline includes a broader perimeter than its ~$35bn notes/bonds/loans; do not replace the filed debt-principal figure with it. | CNBC, full AMP body | secondary | B attribution; C quoted marks | confirmed | Yes — new snapshots, continuing stress |
| 2. CoreWeave/Oracle, 19 Oct (G2) | CNBC: CoreWeave 9.625% 2032 YTM 14.9%, CDS ~1,082bp; Oracle CDS 315bp, +311% y/y. Article published 19 Oct, updated 20 Oct, explicitly identifies Monday observations. | CNBC, full AMP body | secondary | C market snapshots | confirmed | Yes |
| 3. Goldman view (G3) | 19 Oct Buy upgrade and project-finance argument verified. Exact $139→$106 target revision and 100–200bp financing-cost estimate were not independently reproduced. This is analyst judgment, not a newly arranged loan. | Axios, 20 Oct; Barron's, attempted | secondary analyst attribution | B attribution; D unchecked precision | partly | Yes — new analyst view |
| 4. Prior-week recap (G4) | The scan's ~25% 9–16 Oct fall and 1,194.5bp intraday CDS high concern the previous week, already in the 17 Oct log. Its high is dated 15 Oct in the earlier Bloomberg/ICE reporting, not newly established on 16 Oct. $104bn backlog is stale against the already-reported >$129bn; TIKR does reproduce 79.5¢ and the memo quotation, but its underlying trade/document was not independently retrieved. | Reuters, 16 Oct, syndicated; TIKR recap, full direct fetch; 17 Oct comparison | secondary / unverified recap | B prior reporting; D extras | partly | No |
| 5. CDS trading and broader protection (G5) | FT, 20 Oct: Oracle 322bp Monday 19 Oct, versus 144bp start-year; CoreWeave ~1,090bp at publication, fixing time unspecified. SpaceX 226bp early week; Meta 108bp/Nvidia 101bp this month, without precise dates. Seven-name weekly CDS notional >$9bn September vs <$2bn July; AI issuance $466bn YTD vs $101bn, attributed to Morgan Stanley. Trading turnover is not net protection purchased. Distinct Oracle 315/322 quotes are not synchronized closes. | FT via Financial Post, full body fetched | secondary | B reporting; C individual quotes | confirmed | Mixed — new reporting/weekly quotes, September aggregates |
| 6. Baker credit commentary (G6) | 19 Oct CEPR does say CoreWeave yield passed 16% and Oracle yield exceeded 8.5% last week. It does not identify an Oracle bond/date sufficiently to create a yield observation; a high yield itself does not change an issuer's rating. | CEPR original; linked Barron's, partial | primary commentary; secondary underlying marks | A attribution; D precise unidentified yield | partly | No — retrospective price claims |
| 7. Trader recap (G7) | The 452→1,195bp June-to-15-Oct comparison is retrospective. The alleged 19 Oct close ~1,090bp cannot be established as a close: FT reports approximately that level without a fixing timestamp. 23–25% equity declines refer to the prior week. | FT report; 17 Oct credit comparison | secondary; unverified social attribution | C FT snapshot; D close attribution | partly | Mixed — current approximate quote, old drawdowns |
| 8. HSBC/Goldman gap and bond price (G8) | $39/$106 targets, below 80¢ on an unspecified 2031 bond and the exact >20% since April comparison were not traced to a fetched Reuters body or bond tape. Verified yields in 1–2 are not a substitute for an observed transaction price. | Reuters original not located; CNBC contextual yield reporting | unverified secondary attribution | D | unconfirmed | Unestablished |
| 9. New CoreWeave filing/financing (company pass) | Inspected issuer SEC inventory: latest filing 14 Oct Form 4, latest 8-K 22 Sep; no 17–21 Oct 8-K or financing disclosure located. This establishes the search result, not absence of private negotiations. | SEC submissions inventory | primary filing inventory | A inventory only | confirmed | No — no new disclosed transaction located |
| 10. Cure terms | The 31 Dec 2025 amendment, filed 2 Jan, permits unlimited equity cures before 28 Oct 2026, then no more than three consecutive calendar months in any four. First DSCR test 31 Oct 2027. 28 Oct is not an automatic default or debt maturity. | CoreWeave 8-K | primary | A | confirmed | No — old contract, upcoming transition |
| 11. S&P commentary, 19 Oct (C7) | Reuters reports Habib's concern that interconnections obscure AI liabilities; no quantified hidden-debt total. Six firms (Amazon, Microsoft, Alphabet, Oracle, Meta, SpaceX) expected to spend >$7tn over 2025–30, combined FCF negative at least through 2027. No rating changes warranted now. Oracle's July BBB− downgrade is historical. The capex/FCF thesis was already discussed in the 18 Oct log. | Reuters, full Boursorama syndication | secondary reporting of agency analysis | B attribution; C projections | confirmed | Mixed — new remarks; existing forecast thesis |
Financing links, SpaceX and lab disclosures
| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 17 Oct? |
|---|---|---|---|---|---|---|
| 12. IMF financing loop (G9) | October GFSR ch.1: 14 disclosed deals since 2023, ~$300bn equity, $650bn additional debt, $1.4tn future revenue commitments; ~$320bn debt belongs to investees. Wrong-way risk concerns correlated investment and revenue losses. The ~10% issuance comparison needs its stated market denominator; do not turn the sample into a complete debt census. Already verified 18 Oct B3. | IMF, pp.3–4 | primary institutional analysis | A | confirmed | No — recirculated report |
| 13. Nvidia accounting recap (G10) | July-quarter filing: revenue $96.22bn, receivables $63.059bn, DSO 60 days, nonmarketable equity $47.898bn, conditional guarantee caps $108.5bn. Debt carrying amount $33.366bn / face $33.5bn, not $38.9bn. Nonmarketable equity is not all lab cash investment; >$500bn platforms are August MOUs/ambition, not funded commitments. Huang's exact quotation was not traced to primary speech. | 10-Q; CFO commentary | primary | A disclosures; D scan's debt/quote rendering | partly | No — July/August disclosures |
| 14. Firmus (G11) | 20 Oct The Information reports Nvidia investment talks after Firmus's withdrawn IPO. No completed investment or exact amount independently established. The failed listing itself predates this window. | The Information, restricted original; Reuters Breakingviews, 20 Oct | secondary, private-company report | C reported, unverified | partly | Mixed — talks report new, IPO failure old |
| 15. IREN financing/pricing (G12) | 27 Aug issuer release already states 45–55% prepayments, estimated ~2-year paybacks, 2026 largely sold out; Microsoft financing $3.6bn/6% weighted interest excluding fees, covers 96% with prepayments; Mackenzie $2.4bn/9% fixed, 90% before prepayments and >100% with them. Exact A-rating and “more than double last year” were not established. | IREN FY26 results | primary management disclosures/estimates | A disclosed terms/claims; D extras | partly | No — August facts reposted |
| 16. North Star (G13/C9) | Same 16 Oct FT/Reuters proposal: Blackstone $9bn, >10% yield, approximately $60bn package. Existing note distinguishes $18bn junior/$42bn senior proposal and coupon versus fees/discounts. No closing located. A 17 Oct post is not a new loan. | Reuters, 16 Oct relay; FT relay; 17 Oct item10 | secondary | C reported, unverified | partly | No |
| 17. SpaceX Q3 / acquired revenue (C1) | 20 Oct: group revenue $14.575bn, operating income $2.791bn, net income $4.404bn. SI revenue $7.560bn, operating profit $431m, versus Q2 loss $1.257bn; adjusted EBITDA $4.953bn. SI capex $16.444bn of $19.593bn total; nine-month capex $48.069bn, operating cash $13.097bn. Cursor contributes $608m since 14 Aug closing; acquisition SBC $1.217bn. Nameplate compute 1.7GW. Filed diluted EPS $0.35, not the scan's unsupported adjusted $0.45. | SEC earnings exhibit | primary unaudited results | A; B arithmetic | partly | Yes — Q3 results; acquisition closing itself old |
| 18. SpaceX related-party/Valor balance | 30 Sep: related-party current financing $2.181bn plus noncurrent $10.716bn = $12.897bn; Q3 related-party interest $351m. June 10-Q identifies $13.329bn Valor equipment financing. Comparable aggregate declined $432m; September exhibit does not name Valor, so attribution of the whole new balance to Valor remains an inference pending Q3 notes. Historical $20.206bn lifetime undiscounted lease payments are a different measure, not remaining September debt. | September exhibit; June 10-Q, Notes9/17; prospectus | primary; arithmetic/inference | A balances; B sum/comparison/attribution | confirmed | Yes — September aggregate |
| 19. SpaceX–Anthropic contract | Prospectus: $1.25bn/month through May 2029, reduced ramp fees; after initial three months, either party can terminate on 90 days' notice; approximately 325,000 GPUs. Already in 17 Oct item16. No separately disclosed Q3 Anthropic revenue: $3.75bn cannot be inferred as booked sales just by multiplying the fee. | SEC prospectus | primary | A contract; D extrapolated sales | partly | No — existing contract |
| 20. SpaceX/Tesla committees and ambition (C2/C1 call) | 20 Oct call coverage: both boards have independent-director special committees, Musk excluded; no terms agreed and no assured merger. The reported $500bn–$1tn end-2027 ARR ambition is not realized revenue; a primary transcript establishing Grok's SI-only scope was not retrieved. | Ars, Eric Berger; call projection relay | secondary call reporting | B committees; C forecast; D SI-only scope | partly | Yes — exploratory process and stated ambition |
| 21. OpenAI round (G17/C4) | 19 Oct Reuters: accelerated ~$30bn talks at >$1.4tn, SoftBank/MGX discussions, aiming ahead of expected mid-Nov Anthropic IPO. No close; OpenAI/SoftBank declined comment. $50bn annualized/$70bn year-end are a reported pitch/target, not two realized observations. The round concept/valuation target and March $122bn/$852bn are older; acceleration and counterparties are the update. | Reuters, full syndication | secondary | C reported, unverified | confirmed | Mixed |
| 22. OpenAI on Google Cloud (C3) | 21 Oct official posts: frontier models in private preview, general availability planned in coming weeks; Codex integration. Marketplace purchases can draw down existing GCP commitments. Microsoft remains primary cloud partner. Distribution access is not a new financing round or disclosed incremental revenue. | OpenAI; Google Cloud | primary company announcement | A | confirmed | Yes |
| 23. Anthropic draft S-1 (G14) | 20 Oct reporting: H1 revenue $16.3bn vs ~$1.1bn; net loss $80.7bn, including ~$72bn noncash financing remeasurement; compute ~$9bn, versus $7.3bn FY 2025. Relay reports ~$450bn cloud contracts (Google $111bn, AWS $110bn, Microsoft $31bn); provider amounts already appeared in September reporting. Up to $750bn includes deals still in talks; the scan's >$600bn signed total not established. Not a public SEC filing or $80.7bn cash burn. | WSJ, partial original; Reuters relay; Investing WSJ relay | secondary leaked private accounts | C reported, unverified | partly | Mixed — new disclosure vintage; historical figures/provider splits |
| 24. Anthropic Q3/investor day (G15) | $16.8bn Q3, prior $11.6bn Q2/$4.7bn Q1, adjusted profit and positive cash-flow claim were already recorded 17 Oct. $1.5tn investor discussion versus $2tn desired IPO mark now has a fetched Reuters attribution, but concerns last week's discussions. Neither is a priced IPO. | Reuters, 19 Oct; 17 Oct item14; Ctech recap | secondary | C reported, unverified | confirmed | No — recirculated underlying information |
| 25. Anthropic Public Benefit Fund (C5) | 18 Oct: subject to stockholder approval, a proposed 5% share commitment to a wholly owned LLC; board may add 2% per 50% value increase above undisclosed baseline. Combined with founder/employee pledges, approximately 20% devoted to public purposes. Prior $200m labor initiative; no lobbying/political use. ~$48bn is paper-value arithmetic, not cash donated. | Anthropic original | primary proposed equity/governance policy | A proposal; B arithmetic | confirmed | Yes |
| 26. Anthropic IPO timing (G16/C4) | Reuters expects mid-November, not an official listing date. 21 Oct Fortune quotes Renaissance Capital’s Kennedy saying a filing could come as soon as Friday: analyst speculation, not a company promise or a public S-1. $100bn proceeds/$2tn priced offering remain unestablished. Confidential submission is not a public prospectus. | Reuters; Fortune, full direct fetch | secondary expectations / unverified filing lead | C expected timing/speculation; D priced proceeds | partly | Mixed — new analyst speculation, old November expectation |
Hyperscaler spending and GPU prices
| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 17 Oct? |
|---|---|---|---|---|---|---|
| 27. Big-four capex (G20) | Existing July guidance: Microsoft ~$175bn CY 2026, lease reclassification without lower physical investment; Meta $130–145bn, including finance-lease principal; Amazon ~$220bn, Alphabet $195–205bn. $732.5bn midpoint sum combines company-specific definitions and is not audited AI-only capex. Nvidia's top-five $800bn/$1.3tn is an older estimate, not new hyperscaler guidance. | Microsoft July call; Meta Q2; AP, July 30 | primary guidance / secondary | A primary components; B reporting | partly | No |
| 28. Capex/OCF model (G21) | 18 Oct TokenPost publishes $800.5bn capex vs $707.1bn OCF, $93.4bn gap, and $10.3tn 2025–32 scenario. Companywide spending, including non-AI, using results/guidance/analyst estimates; leased datacenters excluded. Underlying company-by-company bridge not reproduced. This is a model, not a new actual cash-flow release. | TokenPost original | published estimates | C | partly | Yes — new estimate publication only |
| 29. Meta preview/results status (G22/company calendar) | Primary Q2 release confirms capex $31.08bn, FCF $784m, Q3 revenue guide $61–64bn. Preview's $63.3bn revenue/$41bn Q3 capex/$205bn 2027 are estimates, not actuals; exact attribution not reproduced. No Q3 results released at 15:00 UTC/11:00 ET 21 Oct; official notice schedules after close, call 4:30pm ET. | Meta Q2; official October 12 notice; preview, failed fetch | primary historical figures/schedule; unverified estimates | A history/schedule; D extras | partly | No — historical figures; results still pending |
| 30. OCPI daily settlements (G23) | 20 Oct: H100 SXM $3.12/GPU-hour, +4.7% day/−1.6% week/+17.7% month; B200 $8.83, +5.0%/+12.8%/+14.5%; H200 $6.16. Public tables verify 17–20 Oct observations archived in the CSV. These are index settlements, not every provider's rent or utilization. | Ornn H100; B200; cross-index panel | primary index publisher | A readings | confirmed | Yes |
| 31. B200/B300 “ATH” (G24) | B200 primary table: 18 Oct $8.54, 19 Oct $8.41, 20 Oct $8.83. Exact 12.9% last-week and B300/all-history record claims not established; displayed B200 history begins July. The fresh rise does not validate an arbitrary weekly window or causation for CoreWeave's equity fall. | Ornn B200 | primary observations; unverified social interpretation | A observed prices; D extras | partly | Yes — verified prices |
| 32. Listed-price aggregators (G25) | 21 Oct Price of Compute: index 139.4, +4.3% 7d/+15% 30d; H100 SXM $4.07, B200 $7.54, B300 $6.60. $9.45 B300 upper bound and getdeploying weekly medians not established. These are posted/list observations, distinct from OCPI settlements; not spliced into that series. | Price of Compute | primary aggregator display | A displayed readings; D unchecked comparisons | partly | Yes |
| 33. Silicon Data H100 check | Product page says daily readings temporarily unavailable. No fresh 17–21 Oct SDH100RT print independently established in this pass. The previously recorded 17 Oct $2.82 is a historical observation, not a 21 Oct fixing; no carry-forward row added. | Silicon Data; 17 Oct observation | primary availability statement | A availability; D new numerical print | unconfirmed | Unestablished |
Power, permits and China
| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 17 Oct? |
|---|---|---|---|---|---|---|
| 34. Oracle Jupiter proposal (G26) | Original is 15 Oct, not a new 19 Oct announcement: proposed ~35% reduction in permitted emissions, modeled GHG 8.8m→5.7m tons/year, preserving IT capacity; carbon-free matching target 2031. This supplements a pending permit application. Exact repost's 2GW RFP/nine-household comparison not independently established from the fetched release. | Oracle original | primary company proposal | A proposal/targets; D unchecked extras | partly | No |
| 35. Jupiter hearing (G27) | Primary 16 Oct order sets hearing 7 Dec, 9am, Sunland Park; technical testimony 17 Nov, rebuttal 30 Nov. Separate 16 Oct notices set dismissal motions 26 Oct, 8:30am MDT, discovery 29 Oct. SourceNM's 19 Oct article is fresh reporting of older orders. No permit approval or newly priced loan established; September force-majeure/89–91¢ stories already in both prior logs. | NMED hearing order; dismissal notice; discovery notice; SourceNM | primary orders / secondary | A dates; B reporting | confirmed | No — pre-cutoff orders newly surfaced |
| 36. Cumberland gas plant (G28) | County primary page explicitly records 20 Oct approval of CUP26-03. Dominion describes approximately 3,000MW, two combined-cycle units. This is local land-use approval of a proposed plant, not an operating generator, completed interconnection or guaranteed datacenter delivery. | County decision log; Dominion project | primary | A | confirmed | Yes — approval 20 Oct, reported 21 Oct |
| 37. DHS DeepSeek removal (G29) | Federal Register notice published20 Oct/filed 19 Oct excludes DeepSeek/affiliates and third-party products using DeepSeek implementations from civilian procurements/IT. Signed order's original date is 24 July. Scope excludes specified national-security systems and is not a private-user ban. Grok's implied new 60-day clock starting20 Oct was not established; original order and implementation deadlines must be distinguished. | Federal Register notice; signed-order text in daily issue, p66347 onward | primary government notice | A publication/scope; D reset-clock claim | partly | Mixed — newly public notice, older order |
| 38. Huawei–DeepSeek plans (G30) | The Information's 19 Oct report concerns chip co-design and prospective Ascend training. Exact engineer deployment, 950DT end 2026/960DT Q12027 allocation, and 160,000-chip site cannot all be checked in the restricted original. Training plans are not demonstrated training or deliveries; September site reporting is older. | The Information, restricted; MT Newswires relay, attempted | secondary private-company report | C core report; D unchecked details | partly | Mixed — new report, old site claim |
| 39. DeepInfer (G31) | Official repository supports Nvidia SM100 and Huawei Ascend950, Apache2.0, V4.1-Flash offline inference. H100/Hopper is not listed as supported. This verifies release/support scope, not equal measured performance: published benchmarks use Nvidia hardware. | DeepSeek repository | primary developer publication | A | confirmed | Yes —20 Oct release |
| 40. China export consultation (G32) | 16 Oct MOFCOM: 40 proposed technologies, including 23 AI; 14 prohibited/26 restricted across all sectors. Interim pre-export registration, consultation, and hosted API/web/SaaS exemptions are explicit. This was already checked17 Oct; not a newly enacted universal model-export ban. | MOFCOM Q&A | primary | A | confirmed | No |
Macro releases and market commentary
| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 17 Oct? |
|---|---|---|---|---|---|---|
| 41. Inflation/rates recap (G33) | 14 Oct BLS: September CPI +0.6% m/m/+3.7% y/y, versus August 3.4% y/y. This is old. The scan's exact Nasdaq100−5.5%, ~35% hike→83% hold historical odds, and 5.11–5.19% Treasury range were not reproduced from a dated primary series in this pass; do not treat them as new fixings. | BLS archived CPI; TradeStation, fetch restricted | primary statistics; unverified market recap | A CPI; D unchecked tape | partly | No — released 14 Oct; newer quotes unestablished |
| 42. Oil tape (G34) | 20 Oct Cinco Días reports Brent moving from ~$111 to ~$106 intraday, emphasizing volatile diplomacy headlines. It does not support the scan's >$111 closing price. Exact WTI ~$108, CoreWeave premarket−4%, and synchronized Treasury move not established from the blocked DowJones page. | Cinco Días,20 Oct; DowJones/Morningstar, attempted | secondary | B intraday reporting; D closing/synchronized extras | partly | Yes — intraday observation; close claim unsupported |
| 43. Elice IPO (G34) | 20 Oct: reputable technology reporting, citing Bloomberg/Reuters, gives 32% debut fall, 61,500won close vs90,500won offer, 201.1bn won/$151m raised. A new risk-appetite observation; it does not establish the same balance-sheet risk as CoreWeave. | The Next Web, with original-report links | secondary | B reported price/filing figures | confirmed | Yes |
| 44. Griffin (G35) | 20 Oct bubble warning is corroborated in 21 Oct Bloomberg syndication: he regards AI investment as almost certainly a bubble. The scan's full survivor/Amazon/billing quotations were not all retrieved from the original event. Verifies a view, not an impending crash. | Bloomberg via The Edge, full body fetched; original Griffin article, restricted | secondary attributed opinion | B core attribution; D unchecked quotations | partly | Yes — new remarks |
| 45. Fink/Sewing (G36) | 21 Oct Bloomberg TV reporting: Sewing rejects an imminent bursting bubble but accepts company setbacks; Fink argues faster datacenter building lowers compute-cost inflation. Both also flag fiscal deficits/capital competition. No measured ROI or future inflation effect demonstrated. | Bloomberg via The Edge | secondary attributed opinions | B attribution | confirmed | Yes |
| 46. Yardeni meter (G37) | 21 Oct primary briefing: Conditions 83/100, Triggers 34/100; high/low historical mix in 54 quarters since 1965, 9% subsequent two-year bear incidence, 18.3% median real gain. These are an analyst/model backtest, not calibrated probabilities for this project's forecasts. 17 Oct bond-discipline essay is older. | Yardeni briefing archive, direct fetch; 17 Oct essay | primary analyst model/opinion | A attribution; C modeled inference | confirmed | Mixed — meter new, essay old |
| 47. All-In E293 (G38) | Original video title/episode was fetched, but no original transcript verifying $250bn Anthropic/$35bn CoreWeave and named speakers was retrieved. Search recaps agree broadly, but remain tertiary leads. Do not enter these as financing requirements or a signed funding gap. | original episode, metadata only | primary metadata; unverified opinion attribution | D substantive claims | lead only | Unestablished |
| 48. Ben Inker (G39) | Original video metadata identifies Inker/GMO interview. Fetched podcast guide dates it 14 Oct and describes equity-supply/overbuild argument and 35–50% conditional decline; original substantive transcript unavailable. The −50% is a scenario, not an unconditional forecast. | original video, metadata; podcast guide | primary metadata / tertiary guide | D substantive attribution | lead only | No —14 Oct interview |
| 49. Elson (G18) | 20 Oct original essay locates valuation risk in labs, cites 15% CoreWeave yield and 62% implied default estimate, and compares lab sales multiples. The default figure lacks a verified model/recovery assumption here; a multiple comparison is opinion, not a cash-flow valuation or observed impairment. | Simply Put original, direct fetch | primary commentary / secondary numerical assertions | A view attribution; C valuation; D implied-default model | partly | Yes — new essay, older inputs |
| 50. Ctech round-up (G19) | 19 Oct original describes lab IPO, chip and neocloud-credit risks; reports Altman ruling out 2026 IPO and investor $1.5tn Anthropic discussions. Its BofA/MorganStanley arguments are attributed views, not matched current-vs-Cisco historical valuation measurements. Underlying earnings/credit/lab reports largely predate the window. | Ctech original | secondary round-up/opinion | B attribution; C private valuations | partly | No — recap of older inputs |
Prediction markets
| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 17 Oct? |
|---|---|---|---|---|---|---|
| 51. Polymarket bubble contract (G40) | 21 Oct fetched display: 10% end 2026 (Yes10.1¢), 22% June 2027; year-end volume $2,488,198, whole event $3,111,645. Prior opened18 Oct display was 7%/17%. Grok's 7→8→11% dated path, 21% June bucket and causal Griffin attribution were not reproduced from historical trades. Full resolution-rule text did not appear in fetched page; do not certify its numerical thresholds from Grok or its AI-generated context. No exact match to our forecasts. | actual Polymarket page; prior opened display | primary market display | A current display; D unverified path/rule rendering | partly | Yes — changed snapshot, historical path unchecked |
| 52. Kalshi Anthropic IPO (G41) | 21 Oct page: official-announcement contract 73% before 1 Dec 2026, 76% before 1 Jan 2027, 79% before 1 Feb 2027. Thus Grok's71% near-date quote is not the current display. The 19 Oct 75%/OpenAI3% and ranking across23 contracts were not independently established. Announcement probability is not listing probability or an observed IPO. | actual Kalshi contract; CNBC, attempted | primary market display | A current display; D historical/ranking extras | partly | Yes — current observation |
Company announcements and statistical releases
| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 17 Oct? |
|---|---|---|---|---|---|---|
| 53. Netflix Q3 (C6) | 20 Oct letter: revenue $12.874bn, operating income $4.459bn, net income $3.664bn, diluted EPS $0.87. Original shareholder letter does not substantiate “AI workflows on300 titles,” InterPositive acquisition, or exact ad-guidance cut in accessible text. These AI extras are not independently verified by the financial figures. | issuer shareholder letter | primary financial statements; unverified AI claims | A results; D extras | partly | Yes — Q3 actuals |
| 54. GE Aerospace, not Vernova (C10) | 20 Oct: orders $19.0bn, GAAP revenue $14.1bn, adjusted revenue $13.3bn, adjusted EPS $2.25, FCF $2.0bn; guidance adjusted EPS $8.00–8.10, FCF $9.1–9.4bn. These are aerospace results, not a fresh GE Vernova turbine/backlog report. | GE Aerospace release | primary | A | confirmed | Yes — limited direct AI relevance |
| 55. IBM llm-d, not earnings (C11) | 20 Oct IBM Research: contributors announced llm-d1.0, distributed Kubernetes-native inference serving. Production readiness/cost efficiency are developer claims, not measured industry returns; no IBM Q3 actuals are established by this product post. | IBM Research original | primary developer announcement | A publication | confirmed | Yes |
| 56. Blackstone energy fund (C8) | 19 Oct: final close $8.7bn, >50% larger than predecessor, hard cap; platforms Kindle/Fisterra/Viridon and cooling-related FlowControl holdings identified. Strategy has committed >$11bn new equity sinceJan 2025. Fund size is not all new AI-only capex deployed this week. | Blackstone release | primary | A | confirmed | Yes — closing announcement |
| 57. Housing (C12) | 20 Oct Census/HUD: September starts 1.221m SAAR, −5.2% m/m/−7.4% y/y; permits 1.400m, −0.2%; single-family starts 859,000, completions 1.192m. Starts' monthly/yearly decline confidence intervals include zero; the statistical significance caveat matters. | dated Census PDF | primary statistics | A | confirmed | Yes — September data released 20 Oct |
| 58. EIA petroleum (C13) | 21 Oct, week ended16 Oct: crude inventories −3.1m to412.5m barrels, gasoline −3.0m, distillate +1.7m, total commercial petroleum −8.5m; refinery runs 16.6m b/d, utilization 93.4%. Live summary fetched directly; these are quantities, not oil-price settlements. | EIA summary, direct fetch | primary statistics | A | confirmed | Yes — new release of prior-week quantities |
| 59. Weekly earnings (C14) | 21 Oct BLS: Q3 median $1,232 NSA, +1.5% y/y versus same-period CPI +3.5%; SA $1,239 vs$1,258 Q2. Women $1,127, men $1,352, ratio 83.4%. Quarter-average CPI comparator differs from September's3.7% point reading. | BLS archive, direct fetch | primary statistics | A | confirmed | Yes |
| 60. State employment (C15) | 20 Oct BLS: unemployment down in 5 states, stable45+DC; statistically significant payroll rises in 2, fall in 1: Wisconsin−13,700. California's table change −9,300 is not a statistically significant decrease; avoid presenting it as the single falling state. | BLS dated release | primary statistics | A | partly | Yes — new release, corrected significance |
| 61. “No named company announcement” claim | Blanket absence claim fails for Nvidia21 Oct: official DSX POD reference design targets early 2027 partner delivery; six months→as little as14days applies at a prepared site. OpenShell architecture also announced, naming several cloud ecosystem firms. Product/partner announcements are not verified orders/revenue. This disproves the blanket, without asserting every listed company had material financial news. | Nvidia DSX POD; OpenShell | primary corrective evidence | A announcements; D blanket absence | unconfirmed | Yes — verified corrective evidence |
Scheduled and not yet released
Schedules below are excluded from the 61-bundle verification counts. Status is at the stated cutoff, not a claim about the remainder of 21 October.
| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 17 Oct? |
|---|---|---|---|---|---|---|
| Meta Q3 | 21 Oct after US close, call 4:30pm ET; not yet released. Do not turn consensus or a capex preview into results. | issuer notice | primary schedule | A | scheduled | No — earlier notice |
| FOMC | 27–28 Oct; no October decision yet. Grok's shorthand28 Oct is decision day, not the full meeting. | Federal Reserve calendar | primary schedule | A | scheduled | No |
| Hyperscaler results,27–30 Oct watch window | Microsoft28 Oct; Alphabet28 Oct,4:30pm ET; Amazon29 Oct,5pm ET. No new results for them in this scan window. The27–30 Oct monitoring range is not evidence that each reports on27or30 Oct. | Microsoft; Alphabet issuer notice, syndication; Amazon | primary schedules / issuer syndication | A | scheduled | No |
| CoreWeave cure transition | 28 Oct ends unrestricted cure period; subsequent frequency limit is 3 consecutive months in4, not an automatic default. | filed amendment | primary contract | A | scheduled contractual transition | No |
| Jupiter | 26 Oct dismissal motions;29 Oct discovery;7 Dec permit hearing. These distinct proceedings do not constitute approval. | NMED order; dismissal; discovery | primary orders | A | scheduled | No —16 Oct notices |
| Other company-pass calendars | Tesla/IBM/Lam/TI were described as due after close21 Oct; their exact event times were not independently verified here and no actuals are used. GEV is a separate issuer from GE Aerospace; its claimed28 Oct result remains a watch item, not this week's release. CoreWeave9 Nov/Nebius10 Nov/IREN5 Nov/Cerebras12 Nov/Nvidia17–18 Nov dates remain scan leads unless separately checked. | Tesla IR; IBM IR; Lam IR; TI IR — calendar leads, not fetched actuals | unverified schedule leads | D exact unverified dates/times | watch list | Unestablished |
| Macro follow-up | 29 Oct GDP first release and personal income/PCE on NYFed calendar; no in-window BEA GDP/PCE actuals established. | NYFed October calendar | primary calendar | A | scheduled | No |
The company pass correctly puts ASML14 Oct/TSMC15 Oct before this window; their supply-demand conclusions were already checked in the18 Oct log. Tesla2 Oct deliveries, SoftBank1 Oct tranche, Cipher's purported mid-October convert and the IREN/CoreWeave/Adani/TeraWulf earlier announcements are not new17–21 Oct events. This pass does not re-certify every number in those context bullets. In particular, IREN's900MW WaggaWagga proposal reported 14 Oct is early development targeting 2029 grid connection, not900MW energized this week.
Implications for research notes and forecasts
These are proposed row updates, not edits to the notes/register. None of the evidence below by itself resolves a forecast; no probabilities were changed. A financing discussion, equity donation, permit hearing, risk quote or secondary-market discount is not a distressed credit event.
| Research note | Row or named watch row that should change | Evidence and forecast relevance |
|---|---|---|
| labs-other | Claims 2–3 Q3 SI results/composition/capacity; 7,52,54 related-party financing; 31 Cursor; 9,47 liquidity watch; add merger process | Replace pending Q3 watch with17–20; distinguish acquired growth, operating profit and financing cash needs. September related-party aggregate is direct; Valor attribution inferred. F4, frontier-lab layer: no distress or markdown resolved. |
| anthropic | 4,6–7,10 historical accounts/concentration; 18 IPO; 27–28 obligation perimeter; 40 public-benefit governance; IPO/compute watch rows | Add23–26; preserve gross/net accounting and adjusted/GAAP distinction. Reconcile$450bn/$518bn/$600bn/$750bn scope/vintage rather than overwrite. D2,F3b,F5: H1 is historical, targets/IPO talk do not resolve. |
| openai | 4 financing proposal; 3 run-rate basis; add Google distribution/commitment-drawdown watch row | 21–22 add accelerated fundraising and new channel; no funded-round or same-basis revenue decline established. D1,D2,F3,F5 remain open. |
| coreweave | 33–34 CDS/yields; 39 GPU rents; financing-access watch; 27,51 cure wording | Add dated1–2 quotes and current30 readings; retain cure mechanics and distinguish~$51.6bn broad debt headline from filed debt principal. No new8-K funding row. F1,F4,M1b, neocloud layer: stress is evidence, not a qualifying event. |
| oracle | 25 CDS; 40,61,63 Jupiter gate/timetable; 24,52 rating context | Add315/322bp observations, Dec 7/Oct 26/Oct 29 dates; mark35% proposal15 Oct. F2,F3,F4,M1b, Oracle/power layers: no downgrade or disclosed qualifying cancellation. |
| debt-markets | 27,30–31,39,53–55,64; named CDS/project-mark watch rows | Add18 balance, credit prints and11 commentary; retain individual fixing limitations, gross notional vs net demand, and no new Jupiter loan mark. F1,F2,F4, credit-led headline/layers remain open. |
| circular-financing | 34,37,42–43,51–55,59,61 and new distribution/opacity watch | SpaceX financing/results and S&P warning change monitoring; Nvidia/IMF recap is already old. Do not count marketplace commitment drawdown as new cash. F3,F3b,F4,F5: no qualifying event. |
| gpu-pricing-and-utilization | 1,3–5,32,47,55–59,64; daily-archive watch | Add17–20 Oct OCPI table, unavailable Silicon print, index-method distinctions and constrained ATH claim. P1/P1b cannot resolve from four daily readings; Silicon display access limits this pass’s coverage verification, not proof of ceased publication. Recover publisher prints/flags before scoring or invoking fallback. Correct the register’s “neither keeps a public history”: OCPI offers a rolling three-month public history; durable archiving remains necessary. |
| power-and-datacenters | 14,32,40,51–52, Jupiter permit and power-delivery watch; add Cumberland/DSX | Update procedural timetable, fund close and local approval. Correct claim14’s assertion that Dec 7 makes the Nov 23 deadline “moot”: the scheduling order does not establish the legal effect on that deadline. Prepared-site deployment speed does not remove power/permit gating. F4, power layer and C1/C2 context; no default or capex cut resolved. |
| nvidia | Product/deployment watch, financing 12,18,26,52 cross-check | Add61 as product plans; retain correct older filing debt/equity/guarantee perimeters. C3,M2,M1a require future released revenue/closing market values, not a design launch. |
| microsoft, alphabet | AI distribution/partnership watch and upcoming earnings/capex guidance rows | Add22, retain July guidance until new releases. D3,C1,C2: no newly released segment growth or qualifying capex cut. |
| amazon, meta | Upcoming results and capex/cash-flow watch rows | Preserve27/29 as prior guidance, add scheduled dates, await actuals. D3,C1,C2 remain open; analyst forecasts are not issuer guidance. |
| neoclouds-other | IREN prepayment/financing and pipeline watch; add Elice/Firmus risk-appetite context | 15 is old,43 new; no inference that all neocloud financing is closed. F4, neocloud/neolab layer questions unchanged by another firm's IPO performance. |
| china | 13–14,31–32,50–53; US market-access watch | Add newly public DHS notice and DeepInfer; distinguish prospective Huawei plans/old MOFCOM proposal. Equipment support is not independently demonstrated training economics. C3/C4,M1a only indirect context, no resolution. |
| public-markets | 21,27,35–38,49,51; prediction-market comparison/watch | Add Elice, remarks and51–52 live snapshots, retain old-week drawdowns. M1,M1a,M1b,M2,M3 require actual defined closing series; none certified resolved here. Polymarket/Kalshi do not match them. |
| rates-sovereign-geopolitics | Inflation/energy/Fed-decision and dated Treasury watch rows | 41–42,57–60 provide context; no unsupported yield close or historical Fed-odds series. M1/M3, credit forecasts indirectly exposed; no resolution. |
| demand-side | 1–3,44,59–60,64–65, distribution/final-demand watch | Add23 and17; avoid adding intercompany SI/cloud revenue to labs as independent final demand. Netflix AI adoption specifics remain unverified. D1,D2,D3: no qualifying new decline/$200bn pair/growth deceleration. |
| ai-unit-economics | 22,33,44,46,50–52, cash-recovery watch | Q3 SI profitability coexists with much greater capex; historical accounting remeasurement is not cash burn. P1/P1b,C1/C2,F3/F3b context, no threshold met. |
| bubble-mechanics | Credit-discipline/trigger watch and sentiment indicators | 11,43–46 show disagreement and financing stress, not a calibrated clock to a crash. Headline conditional tree remains unresolved; no new empirical claim requires changing the historical-precedents note. |
Register audit: relevant IDs are D1,D2,D3; C1,C2,C3,C4; P1,P1b; F1,F2,F3,F3b,F4,F5; M1,M1a,M1b,M2,M3. P2 has no new fixed-product DRAM contract assessment in these scans. No verified$10bn binding compute cancellation/one-year deferral, qualifying down round, distressed credit event or Oracle junk downgrade was found. Price forecasts/layer thresholds have not been evaluated from a complete closing-price ledger in this task, so this is no resolution established, not a claim that every possible price trigger was exhaustively disproved.
Data additions and verification counts
Credit CSV additions use only dated, fetched CNBC19/20 Oct and FT19 Oct Oracle snapshots. 315bp/322bp coexist; neither is silently promoted to an official close. CoreWeave~1,090bp and SpaceX/Meta/Nvidia marks lack exact fixing dates and are not added. No fresh Oracle bond yield or individual bond-price observation met the dating/source test.
GPU CSV archives the publisher's 17–20 Oct H100/H200/B200 tables and 21 Oct Price of Compute composite (index points, 10 August=100), accessed 21 Oct; no synthetic Silicon Data row or prematurely dated21 Oct OCPI settlement. OCPI prices are USD per GPU-hour, and retrospectively retrieved publisher histories are identified in the CSV header; their preservation is not an independent audit of every underlying trade.
Counts: 61 verification bundles: 26 confirmed, 30 partly confirmed, 3 unconfirmed, 2 lead-only. Novelty: 30 new, 18 recirculated, 10 mixed, 3 unestablished. “New” includes newly published estimates or attributed views, clearly labelled in their rows; it does not mean 30 independent economic discoveries. Schedules are excluded. The three unconfirmed bundles are the unsupported bond/target precision, the unavailable fresh Silicon Data print, and the contradicted blanket company-announcement claim.