# Signals: 2026-10-25 (21-25 October)

*Verification date/cutoff: 25 October 2026; no information dated after 25 October is used. Comparison set: the [21 October log](/signals/2026-10-21), whose cutoff was 15:00 UTC, before that day's earnings. USD unless specified. Sources searched and fetched on 25 October; some restricted pages are identified below. EDGAR requests used the generic User-Agent `research-bot contact@example.com`. Grok supplies leads, never citations.*

## What changed since 21 October

- Meta released Q3 revenue of **$64.601bn**, capex including finance-lease principal of **$34.906bn**, and **−$2.120bn FCF**; strong advertising growth coexists with much heavier investment.
- Meta raised its 2026 capex floor **$130bn→$135bn**, keeping the ceiling at $145bn: midpoint **$137.5bn→$140bn, +1.82%**. This triggers **neither C1 nor C2**.
- Meta disclosed nonvoting Class C plans and September commitments: **$810.25bn** firm contracts plus uncommenced leases, or **$824.97bn** including up to $14.72bn contingent cloud commitments. These are not all debt or annual AI capex.
- Tesla lowered its open-ended 2026 capex threshold from **over $25bn to over $20bn**; its newly filed SpaceX committee confirms exploration already known on 21 October, not agreed merger terms.
- Lam reported **$8.219bn** quarterly revenue and **35% China exposure**, and forecasts preliminary **2027 WFE above $200bn**; the latter is management's outlook, not orders.
- Credit stress persisted and broadened: Meta CDS reached **114.79bp Markit / ~119.3bp ICE intraday** on 22 October. Bank access nevertheless remained open selectively: **Nebius $3.5bn** and **TeraWulf ~$150m added committed capacity**.
- OCPI settlements on 24 October were **H100 $3.02 / B200 $9.57 per GPU-hour**; Silicon Data's dated H100 card/chart showed **$2.88**, with a stale conflicting FAQ. No verified spot-price collapse.
- OpenAI's **up-to-$30bn at $1.25tn pre-money** is revised funding talk, not a closed down round; Anthropic's November IPO remains reported plans and its leaked accounts largely repeat the baseline.
- China's listed-model rout deepened, but MiniMax rebounded **16.74% on 23 October**; its below-IPO observation belongs to 22 October. Suffolk approved a temporary datacenter ban; Loudoun's proposed pause remains deferred.
- Microsoft/Alphabet **28 October**, Amazon **29 October**, FOMC **27–28 October**, Jupiter motions **26 October**, and CoreWeave's cure-rule transition **28 October** remain future events.

## How to read the verification

Numbered rows consolidate the broad scan (**G1–G36**) and company pass (**C1–C18**); repeated leads are handled once. **Confirms** means the cited source supports the material claim; **partly** identifies a corrected date, scope, definition, unsupported precision or incomplete fetch; **does not confirm** means the asserted fact was not established, not that access failure proves it false. Grades follow [methodology](/report/methodology): A directly stated primary; B reputable reporting or simple primary-source arithmetic; C leak, estimate or individual reported market snapshot; D tertiary, contradicted or unreproduced detail. A management forecast receives A for its publication, not its eventual accuracy. Private-company leaks are **reported, unverified, C** even when the existence of the report is confirmed.

**New** is measured against the 21 October **15:00 UTC information set**, not the start of that calendar day. **Yes** = later verified observation/disclosure; **No** = older/recirculated; **Mixed** = new disclosure/reporting containing older underlying events; **Unestablished** = no independently verified new observation. Old facts newly retrieved are labelled. Quotes are snapshots, not synchronized closing marks. Research-note changes below are instructions for later upkeep; this entry does not edit those notes or the forecast register.

## Meta: results, cash recovery, governance and financing

| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 21 Oct? |
|---|---|---|---|---|---|---|
| 1. Q3 operations (G1–2, C1) | **21 Oct release**, quarter to 30 Sep: revenue **$64.601bn, +26% y/y**; ads **$63.061bn**; operating income **$12.988bn**, margin **20%**; net income **$9.860bn**, diluted EPS **$3.83**. | [Release][meta-release]; [SEC earnings exhibit][meta-exhibit] | primary | A | confirms | Yes — after baseline cutoff |
| 2. Legal charge and earnings bridge | **$9.89bn Q3 expense** for August agreement with **52 attorneys general**, potential total payments up to **$18.59bn**. Ex-charge operating income **$22.88bn**, EPS **$7.40**. The scan's ~$19.05bn ex-charge net income is approximate arithmetic, not a separately disclosed GAAP result. | [Release][meta-release]; [10-Q][meta-q] | primary / derived | A disclosed; B arithmetic | partly | Mixed — new recognition, older settlement |
| 3. Cash flow and capex | Q3 OCF **$32.786bn** less PPE purchases **$33.942bn** and finance-lease principal **$0.964bn** gives company-defined capex **$34.906bn** and FCF **−$2.120bn**, versus **+$10.625bn** Q3 2025. Nine-month capex **$85.824bn**, FCF **$11.049bn**. Reuters describes the first cash burn in over a decade; the release alone does not establish a full since-2012 history. | [Release][meta-release]; [Reuters 22 Oct][meta-reaction] | primary; secondary historical characterization | A amounts; B history/arithmetic | confirms financial deterioration; historical superlative attributed | Yes |
| 4. Guidance and exact midpoint test | FY2026 capex including lease principal **$135–145bn**, previously **$130–145bn**. Floor +$5bn, ceiling unchanged, midpoint **$137.5→140bn (+1.818%)**. Expenses **$175–179bn** including legal charge; Q4 revenue **$73–76bn**. This is a raise, not a capex cut. | [Release][meta-release] | primary / arithmetic | A guidance; B midpoint | confirms | Yes |
| 5. 2027 and returns | Release says aggressive infrastructure investment continues; AI helps core advertising, while MSL returns take longer. **No quantified 2027 capex guidance**. CFO says quantitative budgeting information waits for the Q4 call; Reality Labs losses expected **~$15bn in 2027**, versus **$18–19bn in 2026**. Neither analysts' $196bn/$225bn estimates nor this qualitative commentary is C1 guidance. | [Release][meta-release]; [21 Oct full call transcript][meta-call] | primary release; secondary-hosted management transcript | A release; B call attribution | confirms qualitative statements; does not confirm numerical 2027 capex | Yes |
| 6. Management's AI claims | Call: **>10m people set up Muse agents** in US/Canada in six weeks; **~25% active on a given day**. This is not 10m DAU. Nearly **10m advertisers** used a generative creative tool in Q3; video-generation advertiser adoption **+25% q/q**. Expected **>10× cost-efficiency improvement next year** is a target, not realized savings. | [Full call][meta-call] | secondary-hosted management transcript | B attribution | partly — correct usage denominator and forecast/actual distinction | Yes |
| 7. Class C mechanics | Board approval **19 Oct**, controlling-shareholder consent **20 Oct**, charter effective **19 Jan 2027**. One nonvoting C share per A/B share; record/payment dates still require subsequent action. Initial founder allocation **20% freely saleable / 80% stapled**. Voting control preserved; subsequent issuance can dilute economic ownership. The stock dividend itself raises **no cash**; CFO announced no specific financing plan. | [8-K][meta-8k]; [PRE 14C][meta-proxy]; [call][meta-call] | primary; transcript | A documents; B call | partly — January effectiveness is not an already priced equity raise | Yes disclosure; underlying approvals 19–20 Oct |
| 8. Capital competition | PRE 14C describes increasing competition for capital and **long-term forecasts discussed in June** projecting hundreds of billions of financing needs over several years, potentially starting fall 2026. Not a newly completed raise or commitment. This disclosure, rather than an unsupported paraphrase assigned to the 10-Q, is the primary basis for the crowded-capital warning. | [PRE 14C][meta-proxy] | primary planning disclosure | A disclosure | partly | Mixed — new public disclosure of earlier planning |
| 9. Future commitments, 22 Oct 10-Q (G3) | At **30 Sep**: uncommenced leases **$355.73bn** + noncancelable contracts **$454.52bn** = **$810.25bn**; separately **up to $14.72bn contingent cloud**, yielding **$824.97bn** if included. Noncancelable commitments due **$45.85bn remainder-2026 / $121.41bn 2027**. Different horizons, contingencies and scopes; not an $825bn funded debt balance or AI-only spending this year. | [Filed 22 Oct 10-Q][meta-q] | primary / arithmetic | A components; B totals | partly — ~$825bn headline needs contingent qualification | Yes — new September balances |
| 10. Ventures and liquidity | September maximum exposure to **Louisiana and Texas combined** unconsolidated ventures **$68.404bn**, including **$41.493bn conditional residual-value guarantees**; no RVG liability booked because payment not probable. Cash/securities **$82.449bn**; notes **$84bn face / $83.669bn carrying / $73.31bn estimated fair value at 30 Sep**. Combined-venture exposure cannot be compared directly with June Hyperion-only $46.03bn. Fair value is not an October traded bond price. | [10-Q][meta-q] | primary / arithmetic | A disclosed; B carrying sum | confirms with perimeter qualification | Mixed — new balances, existing venture structures |

## Tesla, equipment and enterprise demand

| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 21 Oct? |
|---|---|---|---|---|---|---|
| 11. Tesla Q3 (G6, C2) | **21 Oct** update: revenue **$28.046bn**, GAAP net income **$787m**, capex **~$4.73bn**, FCF **~$1.53bn**. Automotive/energy/software economics are distinct; all Tesla revenue is not AI revenue. | [Tesla Q3 update][tesla-update] | primary | A | confirms | Yes — released after cutoff |
| 12. Tesla capex reduction | September 10-Q says **in excess of $20bn** in 2026; June filing said **in excess of $25bn**. This lowers an open-ended threshold, not a precise midpoint or a hard $20bn ceiling. September filing accepted **21 Oct 20:12:41**, SEC filing date **22 Oct**. Tesla is outside C2's four eligible issuers. | [Q3 10-Q][tesla-q]; [Q2 10-Q][tesla-q2] | primary | A | partly — cut verified, exact $25→20bn midpoint claim rejected | Yes |
| 13. Merger committee and related parties | Filed disinterested committee explores deeper SpaceX relationship, potentially a combination; no assurance or agreed transaction. Q3 **$221m** / nine-month **$626m** Megapack revenue from SpaceX; **$142m Q3 investment loss** on its investment. These are related-party links, not merger consideration. | [Tesla 10-Q, Item 5 and notes][tesla-q]; [21 Oct baseline item 20](/signals/2026-10-21) | primary; baseline comparison | A | confirms exploratory status | Mixed — new filed corroboration/quarterly amounts; committees already known |
| 14. SpaceX Q3 recap (C3) | **20 Oct** release and merger-call statements precede baseline: total revenue **~$14.6bn**, SI revenue **$7.560bn**, operating SI profit **$431m** already recorded. Neither repeating the release nor Musk's favorable view of a combination establishes a merger, completed consideration or a new equity-funded compute purchase. | [Issuer results PDF][spacex]; [21 Oct items 17–20](/signals/2026-10-21) | primary, previously verified | A results; B novelty comparison | confirms as old information | No |
| 15. Lam September quarter (C4) | **21 Oct**, quarter ended **27 Sep**: revenue **$8.219418bn**, **+22% q/q/+54% y/y**; GAAP GM **52.3%**, OM **40.0%**, EPS **$2.19**; systems **$5.732907bn**. December revenue guide **$8.60bn ±$400m**. | [Lam release][lam-release] | primary | A | confirms | Yes |
| 16. Lam China share | China **35%** September versus **26% June**; management expects significantly lower share in December as domestic-China demand moderates. A changing geographical mix is not a quantified loss of all China business. | [Release][lam-release]; [official call PDF][lam-call] | primary | A actuals/outlook as stated | confirms | Yes |
| 17. Lam WFE outlook | Official call: **2026 mid-$150bn**, **preliminary 2027 >$200bn**, percentage growth led by DRAM, then leading-edge foundry/logic and NAND; cleanroom/component constraints remain. Calendar-2026 systems revenue expected to grow **well over 40%**. Forecasts are not accepted orders or qualifying hyperscaler capex guidance. | [Official call PDF][lam-call] | primary management outlook | A attribution | confirms | Yes |
| 18. IBM (C5) | **21 Oct**: revenue **$17.103bn, +5%/+6% cc**; software **$7.804bn, +8%/+9% cc**; Q3 FCF **$3.0bn**, nine-month **$7.8bn**. FY revenue growth still **4–5% cc**; FCF growth now **more than $1bn y/y**. AI deployment commentary does not supply an AI-only revenue series. | [IBM release][ibm] | primary | A | confirms core figures; no isolated AI-sales figure | Yes |
| 19. Texas Instruments (C6) | **21 Oct** revenue **$6.055bn, +28% y/y/+11% q/q**, operating profit **$2.874bn**, EPS **$2.62**; growth includes industrial, auto and datacenter. Q4 revenue **$5.80–6.30bn**. TI's **$8.422bn TTM FCF includes CHIPS incentives**, unlike plain CFO less capex. No AI-only sales disclosed. | [TI official release][ti] | primary | A | confirms with FCF-definition correction | Yes |
| 20. SAP (C7) | **21 Oct**: revenue **€10.270bn, +13%/+12% cc**, cloud **€6.668bn, +26%/+25% cc**, current cloud backlog **€24.144bn**, non-IFRS operating profit **€3.030bn**. FY cloud toward upper end **€25.8–26.2bn cc**, FCF **~€10bn**. Cloud/ERP totals are not incremental AI monetization. | [SAP-issued statement][sap] | primary company release | A | confirms | Yes |
| 21. Western Digital (C13) | **22 Oct**, fiscal Q1 to **2 Oct**: revenue **$4.381bn, +55% y/y**, GAAP GM **57.8%**, non-GAAP EPS **$4.64**, OCF **$1.60bn**, FCF **$1.38bn**. Q2 revenue guide **$5.1bn ±$100m**. Scan's **88% cloud / 248EB** not established in the fetched release. HDD results do not measure HBM or DDR5 prices. | [WD release][wd] | primary | A verified figures; D unchecked extras | partly | Yes |
| 22. ServiceNow Arc (C14) | **21 Oct** general availability, governed by ServiceNow's control platform and Nvidia OpenShell. AI Passport integration is prospective. No dollar revenue, uptake count or measured ROI supplied; Q3 financial results remain future. | [ServiceNow announcement][arc] | primary product announcement | A announcement | confirms | Yes — specific launch; generic architecture older |
| 23. Vertiv/Polar (C11) | **21 Oct** modeled configuration **30→3 days**, deployment **90→45**, combined **120→48**; **up to 30%** more cumulative first-year tokens through earlier start, equal installed capacity. Initial Polar Herøya phase **41MW**, scalable to **100MW**. Vendor modeling, not measured fleet productivity or commissioned output. | [Vertiv release][vertiv] | primary vendor model | A publication; C realized benefit untested | confirms modeled claim | Yes — specific application; generic DSX already known |
| 24. Google–Voltus (G23) | **22 Oct** expanded agreement from **100MW to 1GW** of PJM flexible capacity. Demand response via distributed customers, not 1GW of new generating plant. Canary confirms customer commitments through **2040**, but says Google has **not shared a timeline**. Scan's **additional 1,070MW / 3–4-year ramp** not supported by these fetched sources. Do not assume operational capacity today. | [Google announcement][voltus]; [Canary report][canary] | primary; secondary | A announcement; B reporting; D unchecked increment/ramp | partly — capacity type and timeline corrected | Yes; 6 Oct nuclear deal is older |

## Financing access, credit and circular links

| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 21 Oct? |
|---|---|---|---|---|---|---|
| 25. Nebius Citi facility (C10) | **Signed 17 Oct/disclosed 22 Oct**: **$3.5bn** senior-secured 364-day bridge/term facility; maturity **16 Oct 2027**. SOFR margin **2.75%**, steps **0.25pp each 90 days**, cap **3.50%**, subject to adjustments. Minimum-liquidity covenant/security over substantially all borrower assets with exceptions; not evidence all $3.5bn drawn. Issuer's **>$40bn IG-customer commitments** are older contract context. | [6-K][nebius-k]; [issuer release][nebius] | primary | A | confirms, distinguishing capacity from cash | Mixed — new disclosure; agreement 17 Oct, contracts older |
| 26. TeraWulf revolver (C12) | **23 Oct** amendment: committed capacity **$250m→~$400m**, plus **~$150m uncommitted accordion** subject to lenders/conditions; **$150m LC sublimit removed**. Additional committed capacity **~$150m**, not $300m funded cash or $550m committed. | [23 Oct 8-K][wulf] | primary | A | confirms | Yes |
| 27. Blackstone Q3 (C8) | **22 Oct**: AUM **$1,368.3bn**, inflows **$63.3bn**, DE **$1.716bn, −9% y/y**, FRE **$1.572bn, +6%**. Schwarzman identifies AI/energy infrastructure as investment themes. These are firmwide sums, not AI-only capital deployed. | [Issuer release/presentation][bx] | primary | A | confirms | Yes |
| 28. Blackstone AI/credit commentary | Call describes AI debt/equity exposure **~7% of AUM**, mostly physical datacenters; initial leases **15–20 years**, **96% IG counterparties**. Reuters quotes Gray on higher spreads offering opportunities on new, not-yet-leased assets. Management's portfolio characterization is not an audit of immunity to loss. Repeated **~6GW** leasing ambition existed in May. | [Full call transcript][bx-call]; [Reuters 22 Oct][bx-reuters] | secondary-hosted transcript/reporting | B attribution | confirms view, not future credit outcome | Mixed — fresh commentary, old ambition |
| 29. Meta CDS and spillover (G2/12, C15) | **22 Oct Reuters/Markit:** five-year **114.79bp, +7.6bp**. **Bloomberg/ICE:** intraday **~119.3bp, +8.9bp**, record since trading began. Distinct vendors/times, not competing closes. Bloomberg says Oracle/Microsoft/Broadcom protection also rose but gives no numerical marks for them. | [Reuters full syndication][meta-cds]; [Bloomberg full Yahoo body][meta-ice] | secondary market reporting | B reporting; C individual marks | confirms | Yes |
| 30. CoreWeave CDS/yield (G10–12, C15) | **CNBC 22 Oct, FactSet: 1,065.84bp on 21 Oct**, fixing time unspecified, a dated snapshot eligible for CSV. Separate CNBC article originally **21 Oct 17:12 EDT**, updated **23 Oct 14:57 EDT**, says CDS last **1,145bp**, unsecured yield **~15%**, without fixing date/security. Those latter quotes cannot become 22/23 Oct closes or CSV observations. Scan's **~1,080–1,100bp** largely repeats 19–20 Oct baseline. | [CNBC dated FactSet observations][cds-factset]; [CNBC updated article][crwv-cnbc], both full bodies fetched directly | secondary reported snapshots | B reporting; C marks; D close attribution | partly — dated pair recovered, undated updated marks kept separate | Yes — newly disclosed 21 Oct snapshot; older ranges recirculated |
| 31. Oracle CDS and bond prices | **CNBC/FactSet: 308.25bp on 21 Oct**, versus **141bp 2 Jan**. **22 Oct Reuters** says near a record without a numerical fixing. **315/322bp** remain distinct **19 Oct** observations already archived. No dated 21–25 Oct individual Oracle cash-bond price/yield was traced to primary/reputable secondary evidence; undated **~295bp / >8.5%** recaps are not fresh closes. | [CNBC 22 Oct][cds-factset]; [Reuters][meta-cds]; [baseline](/signals/2026-10-21) | secondary; baseline comparison | B reporting; C dated mark; D fresh bond inference | partly — CDS confirmed; cash-bond gap remains | Yes CDS; no new cash-bond mark established |
| 32. CoreWeave cash bonds, retrospectively reported | Morningstar/PitchBook **22 Oct** identifies **20 Oct** 9.75% 2031 at **79.25 per $100 / ~15.97% yield**, and 9.625% 2032 at **77.5 / 15.70%**. These are prior-cutoff marks; 15.97% was already known. Thursday bonds traded above those lows, without exact updated bond marks. Search supplied substantive body; direct Morningstar fetch returned 403. | [Morningstar/PitchBook original][crwv-bonds] | secondary, search-index body; direct fetch restricted | B attribution; C quotes | partly — old marks, not this week's closing tape | No — older observations newly traced |
| 33. IFR financing squeeze (G10) | Full direct fetch: **23 Oct 08:28 UTC publication**, issue dated **24–30 Oct**. **Five datacenter IPOs pulled/delayed**; banks underwrote **almost $330bn** US public AI bonds/loans YTD, **almost $100bn** datacenter/neocloud. CoreWeave bonds **below 80 / yields above 16% in recent days**, without security/date for a new CSV mark. Bernstein's **$85bn funding / $104bn backlog** is an estimate on its chosen/stale perimeter, not a new filed backlog replacing already-reported **>$129bn**. | [IFR full article fetched directly][ifr] | secondary reporting; analyst estimate | B reporting; C estimate | confirms financing pressure; partly on timing/backlog | Mixed — new reporting; some IPO pauses/price stress older |
| 34. CDS turnover and issuance (G11) | FT **20 Oct**: weekly seven-name CDS turnover **>$9bn September vs <$2bn July**; AI-linked issuance **$466bn vs $101bn** comparable YTD. Already baseline item 5; turnover is not net protection bought. Reposting on 21 Oct supplies no fresh stress event. | [FT syndicated original][ft-cds]; [baseline](/signals/2026-10-21) | secondary | B aggregate attribution | confirms as old report | No |
| 35. Supplier guarantees / circular stack (G13–14) | The cited Reuters feature is **1 Oct**; Nvidia up-to-$100bn OpenAI investment, up-to-$105bn campus backstop, $500bn financing ambition, Broadcom convertibles/lease support and AMD warrants are older structures. The social circularity argument adds no verified new signed deal or cash transfer in this window; contingent support and ambitions cannot be summed as funded final demand. | [Reuters 1 Oct feature][vendor-old]; [existing circular-financing note](/research/macro/circular-financing) | secondary prior reporting; unverified social interpretation | C reported terms; D fresh-deal inference | partly — recirculation, not new financing | No |
| 36. Huang's capital constraint (G9) | **22 Oct CNBC interview**, Berlin: management argues neocloud demand is exceptionally strong but working capital is scarce, expects constraint to be solved, and flags European power allocation. No new quantified customer order or bank commitment. Earlier filing exposures in the scan remain earlier-period figures, not a new October balance sheet. | [CNBC original, full direct fetch][huang] | secondary attributed supplier opinion | B attribution; C prognosis | confirms core view; does not establish funded demand | Yes remarks; older financial context |
| 37. Rating-action claim corrected | **Moody's upgraded TSMC Aa3→Aa2 on 21 Oct**, outlook positive→stable, including guaranteed subsidiaries' debt. Agency projections are not realized growth. This contradicts the company pass's blanket absence of rating actions on named firms. No fresh Oracle/CoreWeave/Meta issuer action was established in the bounded agency search. **21 Oct KBRA ABS East recap** is commentary, not a rating action. | [Taipei Times 22 Oct][tsmc-rating]; [Moody's research feed, indexed headline/JS body][moodys]; [KBRA recap][kbra] | reputable secondary action; primary commentary/inventory | B action; A commentary publication | partly — blanket absence contradicted; issuer-specific absence not asserted | Yes TSMC action/research; not Oracle junk status |

## Labs: financing proposals, IPO plans and distribution

| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 21 Oct? |
|---|---|---|---|---|---|---|
| 38. OpenAI round (G15) | **22 Oct CNBC:** nearing up-to-**$30bn** round at **$1.25tn pre-money**, UAE investors/Coatue leading; signatures targeted end of following week, size unsettled, simultaneous UAE compute talks. Illustrative full-round post-money **$1.28tn**, not a closed price. Lower than earlier **>$1.4tn talk**, still above March's reported **$852bn post-money**. No completed down round or money received established. | [CNBC original, full direct fetch][openai-round] | secondary private-company sourcing, reported, unverified | C | confirms report; does not confirm closing | Yes revised reported terms; round concept old |
| 39. Anthropic accounts (G16) | **21 Oct Brew** repeats leaked **H1 revenue $16.3bn**, **Q3 $16.8bn**, **H1 net loss $80.7bn**, including **~$72bn financing-accounting charge**; H1 infrastructure **~$9bn vs $2.3bn**. Core numbers already baseline. **≥$518bn commitments / possibly ~$750bn including potential deals** differ in scope/vintage; not all signed debt, and accounting loss is not cash burn. | [Brew full body][anthropic-accounts]; [21 Oct items 23–26](/signals/2026-10-21) | secondary leaked private accounts, reported, unverified | C | partly — scope and cash-burn interpretation corrected | No material new financial observation; older comparator newly located |
| 40. Anthropic IPO / window (G17) | **22 Oct Reuters:** mid-November ambition, prospectus as early as late October, marketing expected **week of 9 Nov**, subject to change. **$1.5tn** is investor valuation discussion, not priced offering; **under $1.5tn / $100bn proceeds / firm 26 Oct filing** not established. Same report adds Switch/Nscale delays (Nscale partly SEC comments); SB Energy/AirTrunk/Firmus pauses older. No public S-1 was verified in this pass; failed search is not proof none exists. | [Reuters full syndication][anthropic-ipo] | secondary private-company reporting, reported, unverified | C | partly | Mixed — fresh reaffirmation/details, much timetable/valuation already known |
| 41. OpenAI on Google Cloud (C9) | **21 Oct** private preview/Marketplace distribution, eligible against GCP commitments; general availability still future. Already baseline item 22. Distribution adds no disclosed new compute contract value or independent final-demand revenue to sum across providers. | [OpenAI statement][openai-gcp]; [Google Cloud statement][google-openai]; [baseline](/signals/2026-10-21) | primary company announcement | A announcement | confirms as already known | No |

## GPU rental prices: retain separate instruments

| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 21 Oct? |
|---|---|---|---|---|---|---|
| 42. OCPI H100 daily | **21 / 22 / 23 / 24 Oct: $3.09 / $3.15 / $3.17 / $3.02 per GPU-hour**, daily 16:00 America/New_York settlement. **24 Oct** publisher changes **+8.6% 7d / +3.4% 30d**. Retrieved retrospectively, not independently captured on those dates. No 25 Oct settled print displayed. | [Ornn H100 dated table][h100] | primary benchmark publication | A prints; sample representativeness separate | confirms | Yes — four additional dated settlements |
| 43. OCPI B200 and other generations | B200 **21 / 22 / 23 / 24 Oct: $9.40 / $9.32 / $9.58 / $9.57 per GPU-hour**; **24 Oct +12.9% 7d / +19.5% 30d**. Additional **24 Oct H200 $6.20 / A100 $1.08** verified from their dated publisher pages. Transaction-based benchmarks; cannot substitute Price of Compute's $7.50 B200 posted-price median. | [Ornn B200][b200]; [H200][h200]; [A100][a100] | primary benchmark publications | A prints | confirms | Yes — six additional dated settlements |
| 44. Silicon Data H100 | Dated neocloud SDH100RT card/chart: **21 / 22 / 23 / 24 Oct $2.86 / $2.88 / $2.90 / $2.88**. Chart values recovered from publisher page's embedded data, including Saturday24. The same page's undated FAQ still says **$2.53**: preserve conflict; use the explicitly dated card/chart, not FAQ, and do not infer missing days or independent same-day publication/freshness flags. | [Silicon Data original H100 page and embedded chart][silicon] | primary publisher display, retrospectively retrieved | A dated display; C freshness/coverage suitability | partly — dated values verified, stale FAQ/calendar conflict remains | Yes observations; no proof of complete P1 archive |
| 45. Price of Compute (G18) | **25 Oct** composite **135.9** (10 Aug=100), **+1.4% 7d/+9.5% 30d**. Posted medians H100 SXM **$3.99** (9 providers), B200 **$7.50**, B300 **$8.02**, H200 SXM **$5.95**. H100 row range **$1.73–12.29**, while page introduction says **$1.99–11.68**: internal inconsistency. List/ask composite, not transacted OCPI or utilization. | [Publisher dated display][poc] | primary posted-price aggregator | A display; C market-wide inference | partly — main values confirmed; range conflict explicit | Yes new display versus 139.4 on 21 Oct |

These ten OCPI settlements, four dated Silicon chart values and one posted-price composite are retained as **15 additional unique date/series observations** in the GPU CSV. Overlapping 24 October H100/B200 rows and a second name for the same Silicon series are deduplicated. Silicon's 24 October point is explicitly publisher-dated weekend data, not an inferred business-day fixing. Prior daily values are preserved; no carry-forward or reconstructed missing 25 October settlement is added. The recovered rolling Silicon chart does not establish scheduled publication coverage or revision flags for a complete month.

## China: prices, financing and model restrictions

| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 21 Oct? |
|---|---|---|---|---|---|---|
| 46. Listed-model crash (G19) | **23 Oct early Bloomberg** describes MiniMax/Zhipu combined **$150bn** peak-to-date value loss, Zhipu **−81% from June peak**, MiniMax below IPO price **on Thursday22**. MiniMax's **Friday23 close HK$187.60, +HK$26.90/+16.74%**, versus Thursday160.70, means the below-IPO statement must not be carried forward as Friday's close. Approximate **87%** MiniMax drawdown is not a newly synchronized two-name closing calculation. | [Bloomberg/Yahoo full body][china-models]; [FT/LSEG dated history][minimax-history] | secondary market reporting | B reporting; C approximate peak measure | partly — observation date and rebound corrected | Yes prices; cumulative fall began much earlier |
| 47. CSI 300 (G20) | Fetched **22 Oct Bloomberg syndication** says CSI300 **closed −1.8%**, bringing losses from June high to **20%**; Hang Seng Tech intraday **as much −5.3%**. Does not establish scan's exact **ChiNext−36%**, individual-name drops or first-bear-since-2020 superlative. | [Bloomberg/The Edge full body][china-market] | secondary | B reported close; D unchecked extras | partly | Yes — dated market observations |
| 48. DeepSeek round (G21) | **22 Oct Reuters:** approximately **RMB100bn/$14.98bn**, **RMB500bn pre-money**, **>15 investors**; completion expected coming days, mutable terms. CATL/Tencent plus Meituan/Geely/GigaDevice/Midea among names. A separately restricted Information report claims nearly $16bn; precise reconciliation/closing not established. RMB80–100bn/500bn talk already existed **6 Oct**. | [Reuters full syndication][deepseek-round]; [Information, restricted][deepseek-info] | secondary private financing, reported, unverified | C Reuters report; D unreconciled precision | partly — near-close report, not certified cash received | Mixed — investor/timing reporting new; range/valuation older |
| 49. Export registration (G22) | **23 Oct Reuters**, citing CCTV-affiliated **22 Oct** post: DeepSeek/MiniMax cleared overseas/open-source registration. Company/regulator original approvals not fetched. **16 Oct draft, 23 AI technologies** is old; registration is not universal permission for all future model exports or a newly enacted blanket ban. | [Reuters full syndication][china-export] | secondary attributed state-media report | B attribution; C underlying completion without regulator document | partly | Yes registration reporting; No underlying draft |
| 50. DHS, DeepInfer and Huawei context | DHS notice published **20 Oct**, original order **24 July**; developer's **20 Oct DeepInfer** release and Huawei chip-co-design report **19 Oct** already baseline items37–40. No new ban on private US users, new 60-day clock, proven Huawei training run or fresh model-efficiency result was verified from these recaps. | [Federal Register notice][dhs]; [official DeepInfer repository][deepinfer]; [baseline](/signals/2026-10-21) | primary older publications; secondary old plans | A publication/scope; D new-event inferences | confirms old context; does not confirm new outcome | No |

## Power, permits and macro

| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 21 Oct? |
|---|---|---|---|---|---|---|
| 51. Project Jupiter (G24) | ABQ Journal **17 Oct** tour reports **~4,400 workers** and first phase originally hoped for end-2026 apparently pushed back by design/regulatory issues. Older story, not a fresh completion/default. **16 Oct NMED orders** explicitly set motions **26 Oct**, merits hearing **7 Dec**; technical testimony17 Nov/rebuttal30 Nov. Scan's **23 Nov permit deadline** and **1 Feb 2027 pipeline in-service** not established by the fetched tour body/orders; do not let them override the later primary hearing calendar. | [ABQ Journal full direct fetch][jupiter-tour]; [NMED scheduling order][jupiter-order]; [motions notice][jupiter-motions] | primary procedural documents; secondary older site report | A dates; B reporting; D unchecked deadline/pipeline | partly | No — older documents/report; upcoming gates |
| 52. Virginia: separate decisions | **Suffolk council21 Oct**, WHRO22 Oct: unanimous temporary ban pending local rules, not a statewide prohibition. **Loudoun20 Oct notice:** proposed pause on legislative datacenter/substation decisions deferred to **17 Nov**; September vote was to consider it. **Virginia Beach20 Oct rally** sought statewide pause; advocacy is not enacted law. Cumberland20 Oct gas-plant land-use approval already baseline item36. | [WHRO original][suffolk]; [Loudoun primary notice][loudoun]; [WTKR original][va-rally]; [baseline](/signals/2026-10-21) | primary county notice; reputable local reporting | A Loudoun notice; B reporting | confirms distinctions | Mixed — Suffolk new; other events20 Oct or older/newly retrieved |
| 53. Inflation/rate context (G25) | BLS **14 Oct** September CPI **+0.6% m/m/+3.7% y/y**, core **+0.3%/+2.6%**. Fed September range **3.75–4.00%** remains policy until another decision; no October decision yet. These are old data, not this week's inflation release. | [Archived BLS CPI][cpi]; [Fed September statement][fed-sept]; [calendar][fed-calendar] | primary | A | confirms as old context | No |
| 54. EIA weekly (C17) | **21 Oct,10:30 EDT (14:30 UTC) release/week 16 Oct:** crude stocks **412.5m barrels, −3.1m**, gasoline **−3.0m**, distillates **+1.7m**; SPR **282.7m, −0.3m**; production **13.269mbpd, −0.410**; four-week products supplied **21.122mbpd, +3.2% y/y**. Table does not verify hurricane attribution. | [Dated EIA archive Table1][eia-table]; [summary, direct fetch][eia-summary]; [contemporaneous10:30 report][eia-time] | primary; secondary release timestamp | A amounts; B timestamp; D causal attribution | partly | No — released before 15:00 UTC cutoff; newly retrieved |
| 55. Jobless claims (C18) | **22 Oct:** initial **194,000** week 17 Oct, **−6,000** from revised200,000; four-week mean **197,750**. Continuing **1.728m,+25,000** belongs to **week 10 Oct**, not17 Oct. | [DOL dated archive PDF][claims] | primary | A | confirms, period distinction retained | Yes |
| 56. Housing (C16) | **20 Oct** September SAAR starts **1.221m, −5.2% m/m/−7.4% y/y**, permits **1.400m**, completions **1.192m**. Starts confidence intervals cross zero; point estimate is not statistical proof of decline. Before window and cutoff. | [Census/HUD dated PDF][housing] | primary | A published estimates | confirms as older release | No |
| 57. Treasury rates / oil (G4/25) | Official 10-year par yield **21Oct5.13%→22Oct4.96%→23Oct5.00%**; Thursday fall **17bp**, not an exact16bp close. Indicative par benchmarks, not corporate yields. Scan's Friday Brent **$116.72 / WTI>$100** not verified from an exchange/EIA dated closing series in this pass. Intraday oil commentary must not substitute for closing settlement. | [Treasury dated par table][treasury] | primary rates; unverified oil recap | A rates; B subtraction; D unverified oil close | partly | Yes rates; oil precision unestablished |
| 58. FOMC preview (G26) | Fetched **ING preview, circulating23–24 Oct:** expects October hold, **25bp December hike to4.00–4.25%**, with inflation/energy uncertainty. This is an economist forecast. Restricted Bloomberg23 Oct AI-inflation essay not fully retrieved; exact **Barclays >90% OCF** estimate not independently reproduced as a new observation, with similar argument circulating in September. | [ING original][ing]; [Bloomberg essay, attempted][fed-essay] | primary analyst opinion; restricted secondary | A ING attribution; C forecast; D unverified Barclays precision | partly | Yes fresh preview; No decision; Barclays novelty unestablished |

## Investor views and recirculated quantitative arguments

| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 21 Oct? |
|---|---|---|---|---|---|---|
| 59. US market reaction (G2/4/5) | Reuters final **22 Oct**: Meta **−7.0%**, Tesla **−3.1%**, SOX **−6.7%** (worst since 23 Jun), Nasdaq Composite **−2.60%**, S&P500 **−1.21%**. SOX **>26% below 22 Jun peak/+51.5%YTD**. Reuters company report's **~$114bn** Meta loss was conditional at15:06EDT; FT's **$124bn** not independently established as a close. Separate Reuters CDS story gives **>$688bn** prospective SOX value loss, a timed estimate, not an official closing market-cap computation. | [Reuters final wrap][us-close]; [Reuters company report][meta-reaction]; [Reuters timed CDS story][meta-cds] | secondary | B price reporting; C value estimate; D unchecked FT precision | partly — final closes distinguished from intraday estimates | Yes |
| 60. Five-company capex/FCF projection (G8) | **23 Oct Seoul Economic Daily quoting 22 Oct Reuters** prints **$945.55bn capex / −$2.87bn FCF**, versus **+$223.11bn**, described as projected “in January next year.” Underlying Reuters chart, time aggregation and comparable accounting basis unrecovered. Do not label this one month's spending, verified CY2027 guidance or an audited combined cash-flow result. | [Full Seoul Economic Daily body][five-capex] | secondary relay of estimate | C estimate; D period definition | partly | Mixed — new report; underlying estimate vintage unestablished |
| 61. Griffin and Bessent (G27–28) | Griffin **20 Oct** bubble/transformation/uneven business-value argument already baseline44; GPU-price-freefall/portfolio extras not reproduced. Bessent's no-systemic-risk/Pets.com remarks are **reported private-event opinion**, FT21 Oct; relays disagree on20/21 Oct event date, and original transcript not located. His **99% wealth** rhetoric is not verified ownership arithmetic. | [TechSpot21 Oct][griffin]; [FT original attempted][bessent-ft]; [Asia Business Daily relay][bessent-relay]; [baseline](/signals/2026-10-21) | secondary attributed opinion; private remarks relay | B Griffin attribution; C Bessent report; D extras/exact date | partly | No Griffin; Mixed Bessent new reporting/older event |
| 62. Ives, Munster and Wagner (G7/29–30) | Ives's **~$850bn spend/>$2tn backlog/$1→$5–6 spillover**, Munster's **>$225bn 2027 Meta capex vs~$196bn Street**, and Wagner's stronger coverage argument are estimates/views. No filed source verifies an economy-wide multiplier; **15–25% earnings-upside** precision not independently traced. Munster's **10m DAU** rendering fails Meta's setup/active-share distinction in6. Street earnings previews are not next week's actuals or company capex guides. | [TipRanks original, restricted][ives]; [named reporting of Ives interview][ives-relay]; [Benzinga full body][munster]; [Meta call][meta-call] | secondary attributed opinion / estimates | B attribution; C forecasts; D unchecked precision/DAU | partly | Yes views; repeated demand framework |
| 63. Tokic / Delevska (G31–32) | Tokic's Strong Sell and delayed frontier-payback thesis verified as his view; **“capex not increased” is wrong for floor/midpoint**, true only of ceiling. Delevska24 Oct argues tighter finance weeds out weaker projects. Her **Nasdaq100−5.5% / long chip losing streak** context is older; eight-session chip streak ended **19 Oct**, not a newly verified week-to23 Oct return. | [Tokic original summary][tokic]; [Delevska original][delevska]; [Reuters19 Oct][streak] | primary author opinions; secondary old tape | A opinion attribution; C thesis; D misleading timing/guidance | partly | Mixed — new opinions, recirculated market context |
| 64. Koduri / buyback / podcast recap (G33) | Official **29 Sep** podcast: estimated **$50–60bn/GW**, **>400GW / ~$24tn** through 2030; **China target <$10bn/GW**, not observed completed Chinese economics. Nvidia **28 Sep** authorized **additional$150bn**, **$235bn remaining** throughFY2028, not cash already spent. Compound **$360bn FCF** / Collum bell remarks already in30 Sep recap; originals not fetched. | [Official podcast page][koduri]; [Nvidia authorization][buyback]; [30 Sep recap, leads only][matterfact] | primary estimate/authorization; tertiary recap | A attribution/authorization; C estimate; D untraced podcast claims | partly | No — September material recirculated |

## Prediction-market snapshots

*Read from the named market pages on 25 October; continuously changing displays, not historical closing probabilities. “Chance” headings and executable asks can differ. Market context/AI summaries are not rules or evidence about company finances. No trade placed.*

| Item | Figure as verified | Source (linked) | Type | Grade | Status | New since 21 Oct? |
|---|---|---|---|---|---|---|
| 65. Polymarket bubble (G34) | Current outcome cards: **9% end-2026** (Yes ask9.5¢), **23%30Jun2027**; volumes **$2,496,737 / $37,684**, event **$3,122,417**. FAQ instead says10% forDecember: prefer named card and flag inconsistency. Baseline cards10%/22%; scan10–11%/~20% is an earlier snapshot, not today's fixing. Full composite trigger rules not recovered in this fetch; retain prior rule audit, not a fresh certification of every threshold. | [Polymarket original][poly-bubble]; [baseline rule/price audit](/signals/2026-10-21) | primary market display | A displayed quotes; D exact historical-change/rule reconstruction | partly | Yes fetched display; not a matched register probability |
| 66. October Fed (G36) | Correct **2026** Polymarket event cards **96% no change /5%25bp hike**, asks96¢/5¢; page's AI summary still94.5%. Kalshi named Chance headings **93% hold /5% hike**, asks94¢/6¢. Scan94.5%/92.5% is not the fetched pair. Rounded/independently priced outcomes need not sum to100. Generic Polymarket October slug resolves to **2025** and is excluded. | [Correct2026 Polymarket market][poly-fed]; [Kalshi October2026][kalshi-fed] | primary market displays | A display | partly — refreshed, contract/date and card/summary corrected | Yes snapshots; meeting still future |
| 67. IPO markets (G36) | Polymarket searched source displays **95% Anthropic first**, **$384,627** volume; open page did not reliably reproduce the live card, so no certified25 Oct fixing. Rules resolve50–50 if neither lists by **31Dec2027**, so price is not a pure first-IPO probability. Kalshi's separate **official-announcement** ladder shows **68% before 1 Dec /79% before 1Jan2027 /87% before 1 Feb**, asks68¢/79¢/89¢; volume **$4.122m**. Not completed IPOs or offerings. | [Polymarket original/indexed rules][poly-ipo]; [Kalshi announcement ladder][kalshi-ipo] | primary indexed rules/display | A Kalshi display; C Polymarket search snapshot | partly | Yes fetched ladder; head-to-head fixing unestablished |
| 68. Best model (G35) | Polymarket named outcome cards now **Anthropic50% /Google39% /OpenAI6%**, event volume **$2.975m**, not scan74%. Lines article is dated **21 Oct**, not22 Oct, reports Kalshi43%/41%; direct Kalshi contract not recovered, so no current cross-platform comparison. **$2.7m single bet** not traced to trade/account data. Capability-market prices are not commercial revenue or solvency measures. | [Polymarket original][poly-model]; [Lines dated article][lines] | primary current display; tertiary historical comparison | A Polymarket display; D unverified Kalshi/bet precision | partly | Yes current display; historical comparison not independently verified |

## Scheduled and not yet released

All remain future at the 25 October cutoff. A scheduled event, predicted budget or reported roadshow date is not a realized result.

| Date | Event and present status | Source (linked) | Type / grade |
|---|---|---|---|
| **26 Oct,8:30am MDT** | Jupiter pending motions to dismiss; **not yet heard**, no outcome established. | [NMED16 Oct notice][jupiter-motions] | primary A |
| **27–28 Oct** | FOMC meeting; **no October decision**. September target3.75–4.00% remains latest verified policy. | [Fed calendar][fed-calendar]; [September statement][fed-sept] | primary A |
| **28 Oct** | Microsoft FY2027 Q1; **not released**. Azure growth/capex consensus in scan is preview, not actual. | [Microsoft event][msft-date] | primary A |
| **28 Oct,4:30pm ET call** | Alphabet Q3; **not released**. Earlier capex guidance remains earlier guidance, not a new Q3 revision. | [Company-issued6 Oct announcement, syndicated][goog-date] | primary announcement A |
| **28 Oct** | CoreWeave unlimited equity cures permitted **prior to** this date; thereafter restriction to **no more than3 consecutive months in4**. This is **not a debt maturity, automatic default, or first DSCR test** (first DSCR test31Oct2027). | [2Jan8-K describing31Dec2025 amendment][cure] | primary A |
| **28 Oct,10:55am Central** | Oracle Investor Day, **announced23 Oct**, event not yet held; corrects scan's overbroad absence of fresh issuer announcements. | [Oracle primary notice][oracle-day] | primary A |
| **28 Oct** | ServiceNow/Vertiv Q3 results still future; product releases above are not financial results. | [ServiceNow IR][now-ir]; [Vertiv IR][vertiv-ir] | issuer schedules A where displayed |
| **29 Oct,after close** | Amazon Q3 **not released**, call5pm ET; Intel Q3 **not released**, call2pm PDT. No Intel Q3 economics are adopted. | [Amazon notice][amzn-date]; [Intel6 Oct notice][intel-date] | primary A |
| **29 Oct,8:30am ET** | BEA Q3 advance GDP and September personal income/outlays/PCE **not released**. | [BEA release schedule][bea] | primary A |
| **29 Oct /17 Nov /30 Nov /7 Dec** | Jupiter discovery motions /technical testimony /rebuttal /permit merits hearing respectively; the scan's23 Nov deadline is not an established replacement schedule. | [NMED scheduling][jupiter-order]; [discovery notice][jupiter-discovery] | primary A |
| **Reported9 Nov roadshow /mid-Nov IPO** | Anthropic remains unconfirmed timetable, not a completed listing; planned late-October public filing not treated as published. | [Reuters22 Oct][anthropic-ipo] | secondary private-company report C |

## Implications for research notes and forecasts

These are **specific proposed updates, not edits performed here**. Where a claim number is supplied it refers to that note's current Key claims table; named signal rows are also identified. Preserve older observations and append history rather than overwriting their dates.

| Research note | Row / required change |
|---|---|
| [research/stack/meta.md](/research/stack/meta) | Claims **10,12–17,19–20,28,37,41–43,48–49**: add Q3/9M cash figures, new guide/midpoint, legal charge, actual Muse denominator and no numerical2027 capex. **14,23–24,29**: September liquidity and combined venture exposure. Add **Class C / capital-market planning** row, distinguishing dividend from raised cash and voting from economic dilution. Remove Q3-pending signal. |
| [research/stack/semiconductor-equipment.md](/research/stack/semiconductor-equipment) | Current **46–48** already contain Lam September actuals, China mix and December guidance: reconcile, do not duplicate them beside June-only26. **49** can now replace secondary-only call attribution with the recovered official transcript; retain forecasts/supply constraints as management statements, not realized orders. |
| [research/stack/neoclouds-other.md](/research/stack/neoclouds-other) | **7** append Nebius signed/disclosed dates, secured364-day facility, steps/covenant and capacity-versus-draw distinction. Add **TeraWulf revolver** row beside 22–23 separating400m committed from 150m accordion. |
| [research/stack/coreweave.md](/research/stack/coreweave) | **CDS / unsecured-debt yield / financing-access signal rows**: append21Oct1065.84bp, keep undated1145bp and unidentified~15% as publication snapshots only; add IFR/CNBC financing-access commentary with no declared default. Keep28 Oct cure transition distinct from 31Oct2027 DSCR test. |
| [research/stack/oracle.md](/research/stack/oracle) | **Credit spreads / rating / Jupiter procedural signal rows**: append21Oct308.25bp, no fresh traded cash-bond price established; add28 Oct Investor Day. Preserve rating status until agency evidence changes it; no Jupiter permit decision or qualifying lab cancellation. |
| [research/stack/openai.md](/research/stack/openai) | **4** latest proposal up-to30bn/1.25tn pre-money, preserve earlier>1.4tn talk and last priced round as separate vintages; no cash-received/down-round assertion. Distribution row stays21 Oct, already known. |
| [research/stack/anthropic.md](/research/stack/anthropic) | **18** append22 Oct IPO reaffirmation/9 Nov expected marketing; **47–48** preserve old accounts and loss-versus-cash / signed-versus-potential commitment distinctions; no fresh revenue observation from recaps. |
| [research/stack/nvidia.md](/research/stack/nvidia) | **39** and **Hyperscaler2026 capex / Meta FCF** signal rows: new Meta guide/negative quarter. Add **Huang capital / Vertiv-Polar / Arc** partner-watch entries as views/announcements. **43** buyback remains28 Sep authorization, not fresh spending. |
| [research/stack/tsmc-and-foundry.md](/research/stack/tsmc-and-foundry) | **Credit/financial-strength row** add Moody21OctAa2stable upgrade (B; agency body restricted). **31 / Intel competitive-profitability signal** retain Q2 actuals and29 Oct schedule; no Q3 result yet. |
| [research/stack/memory-hbm.md](/research/stack/memory-hbm) | **Demand/outlook and NAND-capacity signal rows** add Lam management WFE outlook. WD is HDD storage, not a new HBM/DDR5 contract-price reading; no P2 observation. |
| [research/macro/gpu-pricing-and-utilization.md](/research/macro/gpu-pricing-and-utilization) | **1,3–5,56,58–59,64,66–67 / P1 benchmark, newer-capacity and archive rows**: add21–24 Oct OCPI/SDH100RT, dated-versus-FAQ conflict and weekend flags;25OctPoC135.9 separate index. Do not assert complete-month publication coverage, utilization or collapse from these snapshots. |
| [research/macro/debt-markets.md](/research/macro/debt-markets) | **30–31,54–55,68** append explicitly dated CDS and retrospective bond observations; **47,59–60** replace latest-balance summaries with Meta September scopes and fair values. Add **selective bank access / Blackstone / TSMC rating differentiation** rows; **70** old turnover stays September, **71** Jupiter schedule stays prospective. |
| [research/macro/circular-financing.md](/research/macro/circular-financing) | **26 / root-Meta obligor table**: combined ventures, conditional RVG, avoid double counting rent/guarantees as principal. **64–65** append Tesla filed committee/related-party sales as corroboration, not merger. **42 / funding-platform** add Blackstone views and new bank facilities; UAE compute/funding remains simultaneous talks. |
| [research/macro/power-and-datacenters.md](/research/macro/power-and-datacenters) | **5–6** Google/Voltus demand response; **14,51–52 / Jupiter and Virginia watch rows**: future merits/motions and separate Suffolk/Loudoun statuses. **36** Meta Louisiana/Texas perimeter; **54,58,60** new modeled Vertiv/Polar and Blackstone portfolio-quality statements. Existing Cumberland approval not recounted as new. |
| [research/macro/demand-side.md](/research/macro/demand-side) | **Enterprise adoption/monetization rows;50,60,64–66 context**: add Meta ad evidence and IBM/TI/SAP/Arc disclosures without assigning whole-company/cloud revenue to final AI spending. Anthropic recap does not change quarter bridge; product GA not realized ROI. |
| [research/macro/china.md](/research/macro/china) | **31–32** add registration reporting; **49** DeepSeek near-close/investor update preserving old range; append **dated listed-model/CSI price row** with MiniMax Friday rebound. Keep59H1 economics separate; DHS/DeepInfer existing rows remain20 Oct context. |
| [research/macro/public-markets.md](/research/macro/public-markets) | **37,49,51,55 / credit-dispersion and prediction-market rows**: explicitly dated new CDS,22 Oct close reaction, China prices and refreshed contract-specific market cards. Older peak drawdowns are not origin-based basket resolutions; $114bn intraday estimate not certified closing loss. |
| [research/macro/rates-sovereign-geopolitics.md](/research/macro/rates-sovereign-geopolitics) | **1,7–8,12,42 / next-FOMC and rates rows**: hold remains forecast; update Treasury21–23 Oct and primary market displays, not unverified oil settlement. Add EIA21 Oct/DOL22 Oct with actual observation periods. CPI/PPI retain14–15 Oct release dates. |
| [research/theory/ai-unit-economics.md](/research/theory/ai-unit-economics) | **Cash recovery / capex / deployment-timing signal rows**: Meta's negative quarter alongside continued ad growth; Tesla lower open-ended threshold as outside-cohort context; Vertiv gains are modeled timing, not measured economy-wide ROI. |

No new evidence requires a fresh historical-precedent claim, a new AMD/Broadcom warrant outcome, or a Microsoft/Alphabet/Amazon earnings result before their releases.

| Forecast ID / register rule | What these facts bear on; resolution assessment |
|---|---|
| **C1** | **Meta's change does not count.** C1 uses the first **calendar 2027** capex midpoints delivered **with Q4 2026 results by15Mar2027**, across specified companies on consistent bases. Q3 2026 capex floor, qualitative2027 language, Lam WFE and analyst2027 estimates are not those observations. No qualifying initial2027 total yet. |
| **C2** | **Meta's change does not count.** Requires **≥10% cut in current-year midpoint**, excluding reclassification. Meta midpoint **137.5→140bn,+1.818%**, a raise. Tesla's lower open-ended threshold is outside eligible MSFT/Alphabet/Amazon/Meta cohort and has no stated midpoint. No Yes established. |
| **C3, C4** | Lam/WD/supplier strength and TSMC upgrade are context. No new Nvidia Data Center y/y revenue decline (**C3**) or three declining TSMC monthly NT$ releases (**C4**) established. |
| **D1, D2** | Funding/IPO talk and repeated Anthropic quarters add no new comparable dated run-rate. Neither a decline on OpenAI's own basis (**D1**) nor qualifying45-day paired **$200bn** total (**D2**) established. |
| **D3** | IBM/SAP/TI and Google product distribution are not Azure/GoogleCloud/AWS growth reports. Wait28–29 Oct; no qualifying10pp sequential deceleration established. |
| **P1, P1b** | New H100/B200 prints are above respective thresholds, countering an immediate glut claim. **P1** needs3 complete covered months of Silicon readings or calibrated fallback; **P1b** needs3 complete covered B200 months. Four days and a retrospective weekend chart point resolve neither; no missing coverage filled. |
| **P2** | Lam memory outlook/WD storage results supply no new final-month TrendForce DDR58GBSO-DIMM contract-price session; no resolution. |
| **F1, F4** | New CDS stress, selective bank facilities and permitting constraints affect financing-risk assessment, but market marks, ordinary financing and cure-rule changes are not distressed credit events. No CoreWeave distress/closed sub-$40 acquisition (**F1**) or qualifying$1bn+ project/GPU/SPV distress (**F4**) verified. |
| **F2** | Oracle CDS is pricing, not a below-IG agency rating. TSMC's upgrade is another issuer. No Oracle junk trigger established. |
| **F3, F3b** | Financing negotiations, rigid commitments, supplier delays and permit hearings do not disclose **$10bn+ binding compute cancellation/reduction or ≥1year deferral** by OpenAI/Anthropic or counterparties. No qualifying amendment verified. |
| **F5** | OpenAI revised talk versus earlier talk is not a **closed priced primary down round** versus its previous priced round; Anthropic IPO unpriced. Neither resolves F5. |
| **M1, M1a, M1b, M2, M3** | Equity selloff bears on risk, especially financing-sensitive names; no daily-rebalanced basket calculation from 16 Oct closes was performed, no Nvidia closing market-cap/share-count test or NDX30%-from-prior 2026-high close established. China/SOX/Meta moves are not substitutes for those exact contracts. No resolution claimed. |
| **Layer and headline rows** | Qualitative update to hyperscaler, equipment, neocloud, Oracle, lab and power risk; no demonstrated further40%-from 16 Oct constituent trigger. No distressed event established, so no credit-led headline outcome. New strong supplier/rental/bank-access evidence should accompany financing stress in any later probability review. |

**Register action:** no probability revision or resolution is made here. This log separates new evidence from resolution rules; the register is outside the authorized write scope.

## Verification and data counts

The **68 numbered bundles** cover both scans and additional checks: **33 confirm the core claim and 35 partly confirm it**, with unsupported subclaims explicitly marked D. Novelty: **43 Yes /12 Mixed /13 No**. These are bundle counts, not independent facts; a confirmed attributed opinion is not a confirmed prediction. No unsupported bundle is silently omitted.

**Credit CSV: 8 added observations** — four CDS snapshots, three Treasury par yields and one retrospectively reported 20Oct2032 bond yield. Undated1145bp/~15% and untraced fresh Oracle cash-bond prices are excluded. **GPU CSV: 15 added unique observations** — ten OCPI, four Silicon H100 chart values, one posted-price composite; overlapping24 Oct records consolidated. CSV columns, unique keys and Markdown tables/references checked. Only this signal note and the two requested CSVs were edited; no git commands run.

[meta-release]: https://investor.atmeta.com/investor-news/press-release-details/2026/Meta-Reports-Third-Quarter-2026-Results/
[meta-exhibit]: https://www.sec.gov/Archives/edgar/data/1326801/000162828026066700/meta-09302026xexhibit991.htm
[meta-q]: https://www.sec.gov/Archives/edgar/data/1326801/000162828026066773/meta-20260930.htm
[meta-call]: https://www.fool.com/earnings/call-transcripts/2026/10/21/meta-meta-q3-2026-earnings-call-transcript/
[meta-8k]: https://www.sec.gov/Archives/edgar/data/1326801/000162828026066700/meta-20261019.htm
[meta-proxy]: https://www.sec.gov/Archives/edgar/data/1326801/000162828026066710/meta-20261021.htm
[meta-reaction]: https://www.marketscreener.com/news/meta-falls-as-investors-balance-muse-growth-with-rising-ai-costs-ce785cdeda88f522
[tesla-update]: https://ir.tesla.com/_flysystem/s3/sec/000162828026066738/tsla-20261021-gen.pdf
[tesla-q]: https://www.sec.gov/Archives/edgar/data/1318605/000162828026066775/tsla-20260930.htm
[tesla-q2]: https://www.sec.gov/Archives/edgar/data/1318605/000162828026049270/tsla-20260630.htm
[spacex]: https://ir.spacex.com/files/content_files/SpaceX-Reports-Third-Quarter-2026-Results.pdf
[lam-release]: https://investor.lamresearch.com/2026-10-21-Lam-Research-Corporation-Reports-Financial-Results-for-the-Quarter-Ended-September-27,-2026
[lam-call]: https://investor.lamresearch.com/image/LRCX-USQ_Transcript_2026-10-21.pdf
[ibm]: https://newsroom.ibm.com/2026-10-21-IBM-RELEASES-THIRD-QUARTER-RESULTS
[ti]: https://ti.gcs-web.com/news-releases/news-release-details/ti-reports-third-quarter-2026-financial-results-and-shareholder
[sap]: https://www.prnewswire.com/news-releases/sap-quarterly-statement-q3-2026-302915956.html
[wd]: https://www.westerndigital.com/company/newsroom/press-releases/2026/2026-10-22-wd-reports-fiscal-first-quarter-2027-financial-results
[arc]: https://newsroom.servicenow.com/press-releases/details/2026/ServiceNow-delivers-industry-first-personal-enterprise-agent-and-policy-enforcement-with-NVIDIA-extending-AI-governance-across-software-and-silicon/
[vertiv]: https://www.vertiv.com/en-us/about/news-and-events/corporate-news/2026/vertiv-targets-up-to-30-more-cumulative-ai-token-output-in-first-12-months-using-nvidia-dsx-polar-deploys-approach-at-norway-site/
[voltus]: https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/virtual-power-plant-1-gigawatt/
[canary]: https://www.canarymedia.com/articles/virtual-power-plants/google-plans-big-virtual-power-plant-in-energy-strapped-pjm-grid
[nebius-k]: https://www.sec.gov/Archives/edgar/data/1513845/000110465926117978/tm2628078d1_6k.htm
[nebius]: https://nebius.com/newsroom/nebius-secures-3-5-billion-financing-facility-with-citi
[wulf]: https://investors.terawulf.com/sec-filings/all-sec-filings/content/0001628280-26-067100/wulf-20261023.htm
[bx]: https://www.blackstone.com/blackstone3q26earningspressrelease/
[bx-call]: https://stockanalysis.com/stocks/bx/transcripts/714082-q3-2026/
[bx-reuters]: https://uk.marketscreener.com/news/blackstone-s-profit-falls-as-choppy-market-tempers-investment-exits-ce785cdedb8bf323
[meta-cds]: https://www.marketscreener.com/news/meta-s-credit-risk-indicator-jumps-to-record-high-after-results-ce785cdedd89fe23
[meta-ice]: https://finance.yahoo.com/technology/ai/articles/meta-credit-risk-gauge-hits-125603234.html
[cds-factset]: https://www.cnbc.com/2026/10/22/ai-debt-and-french-bonds-stress-test.html
[crwv-cnbc]: https://www.cnbc.com/2026/10/21/coreweaves-market-confidence-erodes-as-bond-yields-spike-stock-sinks.html
[crwv-bonds]: https://www.morningstar.com/markets/coreweave-debt-tumbles-anew-korean-neocloud-ipo-tanks
[ifr]: https://www.ifre.com/bonds/2502504/coreweave-rout-freezes-data-centre-financing
[ft-cds]: https://financialpost.com/financial-times/ai-debt-worries-spark-surge-in-default-insurance-trading
[vendor-old]: https://www.reuters.com/legal/transactional/nvidias-bet-that-its-chips-can-finance-ai-boom-gets-wall-street-reality-check-2026-10-01/
[huang]: https://www.cnbc.com/2026/10/22/nvidia-jensen-huang-neocloud-demand-capital.html
[tsmc-rating]: https://www.taipeitimes.com/News/biz/archives/2026/10/22/2003866384
[moodys]: https://ratings.moodys.com/research
[kbra]: https://indices.kbra.com/publications/mFWWHdyC/kbra-releases-abs-east-2026-conference-recap
[openai-round]: https://www.cnbc.com/2026/10/22/openai-funding-round-uae-investors.html
[anthropic-accounts]: https://www.brewmarkets.com/stories/anthropics-monster-bill
[anthropic-ipo]: https://www.streetinsider.com/Reuters/Analysis-Anthropic%2BIPO%2Bstands%2Balone%2Bas%2BAI%2Bmarket%2Bjitters%2Bsideline%2Bothers/27220169.html
[openai-gcp]: https://openai.com/index/openai-models-google-cloud/
[google-openai]: https://cloud.google.com/blog/products/ai-machine-learning/google-cloud-brings-openai-models-to-customers/
[h100]: https://data.ornn.com/markets/h100-sxm
[b200]: https://data.ornn.com/markets/b200
[h200]: https://data.ornn.com/markets/h200
[a100]: https://data.ornn.com/markets/a100-sxm4
[silicon]: https://www.silicondata.com/products/silicon-index/h100
[poc]: https://priceofcompute.com/
[china-models]: https://hk.finance.yahoo.com/news/%E4%B8%AD%E5%9C%8Bai%E6%A8%A1%E5%9E%8B%E9%96%8B%E7%99%BC%E5%95%86%E8%82%A1%E5%83%B9%E4%BC%BC%E6%9C%AA%E8%A6%8B%E5%BA%95-minimax%E5%92%8C%E6%99%BA%E8%AD%9C%E5%B8%82%E5%80%BC%E5%85%B1%E8%B7%8C1500%E5%84%84%E7%BE%8E%E5%85%83-015155881.html
[minimax-history]: https://markets.ft.markitdigital.com/data/equities/tearsheet/historical?s=100%3AHKG
[china-market]: https://theedgemalaysia.com/node/822807
[deepseek-round]: https://www.streetinsider.com/Reuters/Deepseek%2Braises%2B%2415%2Bbillion%2Bin%2Bchallenge%2Bto%2Bleading%2BUS%2Blabs%2C%2BBloomberg%2BNews%2Breports%C2%A0/27217623.html
[deepseek-info]: https://www.theinformation.com/briefings/deepseek-raises-nearly-16-billion-second-funding-round
[china-export]: https://www.marketscreener.com/news/china-s-deepseek-minimax-win-approval-for-ai-global-use-state-media-affiliated-entity-says--ce785cded089ff21
[dhs]: https://www.govinfo.gov/content/pkg/FR-2026-10-20/html/2026-21447.htm
[deepinfer]: https://github.com/deepseek-ai/DeepInfer
[jupiter-tour]: https://www.abqjournal.com/news/first-look-at-project-jupiter-site/3143545
[jupiter-order]: https://www.env.nm.gov/opf/wp-content/uploads/sites/13/2026/07/2026-10-16-AQB-26-57-2nd-Amnd-Sch-Order-filed.pdf
[jupiter-motions]: https://www.env.nm.gov/opf/wp-content/uploads/sites/13/2026/07/2026-10-16-AQB-26-57-NOH-on-MTDs-filed.pdf
[jupiter-discovery]: https://www.env.nm.gov/opf/wp-content/uploads/sites/13/2026/07/2026-10-16-AQB-26-57-NOH-on-Mtns-for-Disco-filed.pdf
[suffolk]: https://www.whro.org/local-government/2026-10-22/suffolk-approves-temporary-data-center-ban
[loudoun]: https://www.loudoun.gov/m/newsflash/home/detail/10938
[va-rally]: https://www.wtkr.com/news/in-the-community/virginia-beach/hampton-roads-protesters-call-for-statewide-pause-on-new-data-centers
[cpi]: https://www.bls.gov/news.release/archives/cpi_10142026.htm
[fed-sept]: https://www.federalreserve.gov/newsevents/pressreleases/monetary20260916a.htm
[fed-calendar]: https://www.federalreserve.gov/monetarypolicy/fomccalendars.htm
[eia-table]: https://www.eia.gov/petroleum/supply/weekly/archive/2026/2026_10_21/pdf/table1.pdf
[eia-summary]: https://ir.eia.gov/wpsr/summary.txt
[eia-time]: https://www.streetinsider.com/Commodities/Crude%2BInventory%2BFell%2B3.06%2BMillion%2BBarrels%2BLast%2BWeek%2B-%2BEIA/27214731.html
[claims]: https://www.dol.gov/sites/dolgov/files/OPA/newsreleases/ui-claims/20261613.pdf
[housing]: https://www.census.gov/construction/nrc/pdf/newresconst_202609.pdf
[treasury]: https://home.treasury.gov/resource-center/data-chart-center/interest-rates/TextView?field_tdr_date_value=2026&type=daily_treasury_yield_curve
[ing]: https://think.ing.com/articles/federal-reserve-preview-a-pause-before-a-december-hike/
[fed-essay]: https://www.bloomberg.com/news/articles/2026-10-23/federal-reserve-interest-rate-decision-faces-challenges-from-ai-inflation-war
[us-close]: https://www.marketscreener.com/news/s-p-500-nasdaq-end-sharply-lower-as-revived-ai-doubts-ignite-tech-selloff-ce785cdedc80ff25
[five-capex]: https://en.sedaily.com/international/2026/10/23/big-tech-cant-stop-ai-spending-but-the-cash-is-running-out
[griffin]: https://www.techspot.com/news/114280-citadel-ken-griffin-investors-almost-certainly-ai-bubble.html
[bessent-ft]: https://www.ft.com/content/5b7c2fdf-2300-4208-860e-6389ceb18886
[bessent-relay]: https://www.asiae.co.kr/en/article/2026102205323821185
[ives]: https://www.tipranks.com/news/article/hyperscaler-stocks-microsoft-amazon-alphabet-report-next-week-heres-what-dan-ives-expects
[ives-relay]: https://stocktwits.com/news-articles/markets/equity/dan-ives-calls-ai-sell-off-saas-apocalypse-2-0-oracle-in-toughest-spot/cZxyK1wRBxF
[munster]: https://www.benzinga.com/markets/tech/26/10/62523526/metas-145-billion-ai-bet-doesnt-scare-this-bull-why-he-thinks-bears-are-wrong
[tokic]: https://seekingalpha.com/article/4956401-meta-confirms-the-ai-bubble-burst-thesis-why-ignore-it
[delevska]: https://www.investing.com/analysis/ai-spending-vs-nasdaq-selloff-why-wall-street-may-be-misreading-the-boom-200690185
[streak]: https://www.reuters.com/business/wall-st-futures-climb-ai-stocks-rebound-oil-yields-fall-2026-10-19/
[koduri]: https://www.techsurgepodcast.com/podcasts/gaming-chip-pioneer-raja-koduri-on-chinas-ai-cost-advantage-and-moving-beyond-the-gpu
[buyback]: https://nvidianews.nvidia.com/news/nvidia-announces-a-150-billion-share-repurchase-authorization-increase/
[matterfact]: https://www.matterfact.com/newsletter/2026-09-30-ai-capex-token-prices-halve-open-models-micron
[poly-bubble]: https://polymarket.com/event/ai-bubble-burst-by
[poly-fed]: https://polymarket.com/event/fed-decision-in-october-20260617190323537
[kalshi-fed]: https://kalshi.com/markets/kxfeddecision/fed-meeting/kxfeddecision-26oct
[poly-ipo]: https://polymarket.com/event/will-anthropic-or-openai-ipo-first
[kalshi-ipo]: https://kalshi.com/markets/kxipoanthropic/anthropic-ipo/kxipoanthropic-date
[poly-model]: https://polymarket.com/event/which-company-has-best-ai-model-end-of-2026
[lines]: https://www.lines.com/best-ai-model-2026-polymarket-and-kalshi-split
[msft-date]: https://www.microsoft.com/en-us/investor/events/fy-2027/earnings-fy-2027-q1
[goog-date]: https://www.marketscreener.com/news/alphabet-announces-date-of-third-quarter-2026-financial-results-conference-call-ce785dd9dc89f62d
[cure]: https://www.sec.gov/Archives/edgar/data/1769628/000176962826000003/crwv-20251231.htm
[oracle-day]: https://investor.oracle.com/investor-news/news-details/2026/ORACLE-INVESTOR-DAY-MEETING-TO-BE-HELD-OCTOBER-28-2026/default.aspx
[now-ir]: https://newsroom.servicenow.com/press-releases/details/2026/ServiceNow-to-Announce-Third-Quarter-2026-Financial-Results-on-October-28/default.aspx
[vertiv-ir]: https://investors.vertiv.com/news/news-details/2026/Vertiv-Announces-Date-of-Third-Quarter-2026-Earnings-Release-and-Conference-Call/
[amzn-date]: https://www.aboutamazon.com/news/company-news/amazon-earnings-q3-2026-report
[intel-date]: https://www.intel.com/content/www/us/en/newsroom/news/corporate/intel-reports-third-quarter-2026-financial-results.html
[bea]: https://www.bea.gov/news/schedule