Rates, sovereign capital and geopolitics: outside triggers for the AI cycle

Last updated: 2026-10-25 · Author: Codex / GPT-6 · Status: reviewed · Reviewers: Codex (gpt-6-astra) cross-check applied 2026-10-17; Claude (Opus 5.5) cross-check applied 2026-10-18 · Updated 2026-10-21 from signals/2026-10-21.md · Updated 2026-10-25 from signals/2026-10-25.md

Summary

• On September 16, 2026 the Fed raised its target 25 bp to 3.75–4.00% by a 12–0 vote. Its October 7 minutes link surging AI investment and oil to inflation and say most participants expected another hike by year-end, alongside supportive financial conditions. September CPI was 3.7% y/y, core 2.6%; the ten-year reached 5.31% on October 5, its highest close since May 2002, with the real ten-year at 2.95%. S2, S4, S50; claims 1, 42, 52

• During October 9–16, 2026, M1 fell 12.7% versus Nasdaq-100's 5.5%, while the ten-year fell 15 bp to 5.09%. Real yields fell 13 bp through October 15 and physical Brent fell 2.7% through October 14. High rate and oil levels remain funding headwinds; their changes do not explain the week's sell-off as a rising-discount-rate or oil shock. Equity inputs retain C; the financing interpretation is an inference. Claims 52–54, 57

• Physical Brent averaged $126.98/bbl over ten October observations through October 14, 2026, versus EIA's $105 Q4 forecast completed October 1, before the October 2 spike to $135.51. That Q4 average would require roughly $101 for the rest of the quarter; September shut-ins had already declined to 4.8m b/d. S14, S12; claims 15, 44, 56

• Through October 18, 2026, the disclosed sovereign lab-check floor here is HUMAIN's historical $3bn xAI investment, subsequently converted into SpaceX equity. MGX reportedly joined OpenAI's March 2026 $122bn round (C, reported, unverified); QIA disclosed Anthropic Series H participation (A). Both individual allocations are unknown, so the floor understates sovereign-linked lab funding by an unknown amount. S24, S30, S64; claims 24, 28, 55

• SoftBank supplied $60bn to OpenAI in 2025–October 1, 2026 (cumulative $64.6bn, approximately 13%). The final $10bn was funded with September notes carrying a 9.41% weighted USD coupon, about 4.2 percentage points over the September 24 ten-year; its OpenAI-linked loan carries valuation-linked collateral and prepayment terms. This is a leverage-financed purchase of lab equity. S34, S37, S38; claims 30, 32, 33, 48

• TSMC’s October 15, 2026 release reports Q3 revenue of $46.72bn and a 58.7% operating margin. That is a profitable upstream counterweight; purchasers’ incremental AI returns remain unquantified. S53; claim 50

• China's October 16, 2026 rule registers outbound transfer/licensing of Chinese AI technology and foreign takeovers of its owners across 23 AI categories. Its AI provisions do not specify chips or hardware; a small direct channel to US capex is an inference. Taiwan remains a physical supply concentration: TSMC expects Arizona fab two volume production in Q3 2027. S44, S52; claims 38, 49

• New October 20–21 releases add mixed demand evidence: September housing starts were 1.221m SAAR, down 5.2% m/m with an interval including zero; Q3 median weekly earnings rose 1.5% y/y, below the same-quarter CPI increase of 3.5%. State payrolls were statistically unchanged in 47 states and DC. These qualify the earlier August consumption counterweight without establishing a broad contraction. FOMC meets October 27–28; GDP and PCE are scheduled for October 29. Claims 58, 60, 62–63

• In the week to October 23, rates again fell while AI equities sold off. The ten-year par yield went 5.15% (Oct 19) → 5.12% → 5.13% → 4.96% (Oct 22, −17bp) → 5.00% (Oct 23), and the two-year fell 16bp on October 22 to 4.65%. On the same day the M1b financing basket fell 6.9%, after Meta's results. Initial claims were 194,000 in the week to October 17. No dated Brent settlement for October 19–23 was verified. The October FOMC outcome remains a forecast: venue cards on October 25 priced a hold at 93–96%, and ING expects a hold followed by a December hike to 4.00–4.25%. Neither is a decision. Claims 7–8, 12, 64–66

Key claims

Original sources accessed 2026-10-17; additions and identified re-checks accessed 2026-10-18. A confirms the disclosed reading or announcement, not fulfillment of a forecast or investment ambition. B identifies arithmetic; directly disclosed components retain A where specified. C marks limited futures verification, vendor equity prices and reported, unverified funding or AI-credit estimates; D marks the fetched tertiary AI-credit commentary. B arithmetic does not upgrade C inputs. The October 21 update re-fetched the primary calendars/statistical releases and the dated Treasury table; items predating October 17 remain historical context. The October 25 update fetched the Treasury par-yield CSV, the DOL claims release and the ING preview directly; the venue-card and EIA Table 1 additions rest on the verified October 25 signals log and are marked as such. Monthly sequences below run chronologically; currency amounts are nominal.

Analysis

Rates, inflation and financing

The Fed held through summer and raised in September. CPI and PPI measure different prices; neither is the Fed's PCE inflation objective. BEA's September 30 release reports August PCE inflation of 3.4% headline and 3.0% core y/y, with monthly increases of 0.3% and 0.2%; real PCE rose 0.6% m/m. The release incorporates annual revisions beginning January 2021. The minutes’ staff estimates preceded that release: August headline/core PCE was estimated at 3.8%/3.4% on the old basis, or 3.6%/3.2% under the new methodology. BEA’s September 30 print supersedes both; 3.8% must not propagate as the actual August reading. September core CPI of 2.6% is below August core PCE of 3.0%, the reverse of the usual ordering; the reference months and index construction differ. The CPI reading alone does not establish benign underlying inflation. Persistent inflation coexists with consumption growth, rather than demonstrated aggregate contraction from the energy shock. S1, S2, S4, S5, S50, S51

The minutes released October 7 describe supportive financial conditions and broadly available credit, while identifying AI investment as an inflation and financing-demand driver. Most participants expected another hike by year-end; several judged policy not restrictive or only mildly restrictive. The manager also reported that “yield spreads on hyperscaler debt used to finance AI infrastructure remained wide”. Market commentary cited competition for capital from “heavy private debt issuance to finance the development of artificial intelligence (AI) infrastructure”; the staff reported that “changes in real rates contributed to most of the net increase in longer-maturity Treasury yields”. Some participants warned the buildout “could cause aggregate demand to outpace aggregate supply over the medium term”. These are separately attributed assessments (A), not measured causal attribution. Together they describe late-cycle feedback: AI demand can raise the inflation and financing hurdles the buildout then faces. Supportive aggregate conditions can coexist with wide AI borrowing spreads. S50

The ten-year’s October 5 close of 5.31% was the highest since 5.32% on May 14, 2002, above the 2007 peak of 5.26% (comparison B); the real ten-year was 2.95% that day. The real yield is the relevant observed starting point for discounting real cash flows, alongside risk premiums and expected growth; it is not a measured AI-equity discount rate. The minutes’ real-rate attribution concerns their intermeeting period, not the later October sell-off. S55, S56; claim 52

CME's browser-rendered October distribution favors a hold. Futures-implied probabilities change with prices; a hike is not the October base case in that snapshot. This does not answer the separate year-end policy question in the September minutes. S6, S50

Duration links higher discount rates to lower present values, especially where distributable cash lies far ahead. This is an inference, not a measured AI-index duration. SOFR reprices floating debt; Treasury yields and spreads affect new fixed debt. CoreWeave has fixed as well as floating debt. A 25 bp increase adds $2.5m annual interest per $1bn of unhedged floating principal at full repricing (claim 41, B). Overnight SOFR is not the contractual Term SOFR fixing for DDTL 5.5. S9, S19

DDTL 5.0 also requires ≥95% swap coverage after specified deadlines; future requirements do not establish current hedge coverage. DDTL 4.0 limits recourse except for carve-outs, whereas DDTL 5.0 is parent-guaranteed. These distinctions change loss allocation; aggregate October net exposure remains unverified. S19

Profitable incumbents withstand funding shocks better than projects requiring repeated raises. Microsoft's and Nvidia's net income does not measure cash remaining after capex or isolate incremental AI returns. Overextension combines construction delays, fixed customer pricing, floating debt and equipment obsolescence. Equity loses first; creditor, landlord and vendor losses depend on collateral, guarantees and recourse. S19, S43, S47

The September directive rolls Treasury maturities over, reinvests agency principal into bills and permits reserve-management purchases. The New York Fed distinguishes those purchases from stimulus: reserve availability can improve independently of the policy-rate stance. Several September participants judged that stance not restrictive or only mildly restrictive. S3, S48, S50

Historical precedents illustrate mechanisms rather than timing rules. Japan's asset prices initially rose after BOJ tightening. The technology boom coexisted with Fed restraint: productive technology does not protect leveraged buyers. The local note's “quiet” current tightening indicator conflicts with September's hike. S20, S21, S2

Market reaction, 9–16 October

The M1 basket fell 12.7% over October 9–16, versus 5.5% for Nasdaq-100 and 11.4% for SOX. CoreWeave fell 25.2%, Vertiv 15.0%, Micron 14.6%, AMD 13.0%, Oracle 9.3% and Nvidia 6.6%. M1 uses NVDA, AVGO, TSM, ORCL, CRWV, AMD, MU and VRT, rebalanced equally each session: compound the daily mean of constituent gross returns, rather than averaging their weekly returns. Yahoo histories were fetched again and the arithmetic reproduces; vendor equity and SOX inputs retain C. Official Nasdaq EOD levels support NDX return arithmetic at B. S62; claim 53

Treasuries rallied: the ten-year fell 5.24%→5.09% (15 bp) through October 16 and the thirty-year 5.60%→5.43% (17 bp). Through October 15, real ten-year yields fell 2.91%→2.78% (13 bp) and the two-year 4.80%→4.77% (3 bp). Physical Brent fell $130.03→$126.55 (2.7%) through October 14. ICE whole-market HY OAS widened 3.11%→3.23% (12 bp) through October 15. The review’s Goldman comparison is AI-related HY +25 bp to 360 bp on October 13–14 versus +2 bp non-AI; the same figures appear in fetched VLT Capital commentary (tertiary, D). The reviewed secondary estimate retains C, but the Goldman note and the unspecified TipRanks wrap were not obtained. Different windows and issuer mixes prevent equating this pair with ICE’s weekly index change. S55–S63; claims 53–54

October 14 is the plausible macro catalyst within that week: headline CPI was 3.7% versus 3.6% consensus and core rose 0.3% versus 0.2%. M1 fell 6.2% (CRWV −15.5%, ORCL −8.8%) against NDX −2.0%, while the two-year rose 3 bp and HY OAS widened 9 bp. Yet the ten-year fell 2 bp and the real ten-year 4 bp. Reuters/LSEG’s roughly 33% October hike probability on October 14 fell to a separately timed CME 15.5% snapshot on October 17; this is not a like-for-like weekly endpoint comparison. S4, S6, S55, S56, S58, S60, S62, S65; claim 57

Falling yields during the sell-off are consistent with a flight to quality. The week’s repricing cannot be attributed to rising rates or oil prices on these observations: their high levels set the refinancing backdrop, but their changes ran the other way. The concentration in AI shares and reported AI credit widening favor an AI- and financing-specific interpretation (inference), without identifying causality or excluding the CPI surprise. A daily regression of M1 returns minus S&P equal-weight returns on changes in real ten-year yields, two-year yields and Brent front-month, with CPI/PPI-day dummies, would test that reading. Non-AI duration controls (unprofitable growth, biotech/XBI, REITs and utilities), matched-rating AI/non-AI credit pairs, CoreWeave CDS versus CDX HY, intraday release/headline timing, and breakeven versus real-yield decomposition remain unperformed checks.

October 22 repeat. The day after Meta's results, the ten-year par yield fell 17bp to 4.96% and the two-year 16bp to 4.65%, while the M1b financing basket fell 6.88% and NDX 2.90%. Yields partly retraced on October 23 (ten-year 5.00%). Falling rates during an AI-equity selloff repeat the October 9–16 pattern. This again argues against a rising-discount-rate explanation for the day, without identifying the cause: the same session carried Meta's negative quarterly free cash flow and higher capex floor, a record Meta CDS print and the equipment selloff. No dated oil settlement or HY OAS observation for the day was verified here. S61; claim 64

Oil, fiscal supply and the dollar

EIA attributes the oil shock to infrastructure and tanker attacks, notably Saudi Arabia's East–West pipeline bypassing Hormuz. Freight, insurance and inventory depletion amplify the shortage, connecting energy, inflation, rates and sovereign budgets. S14

EIA completed its $105/bbl Q4 forecast on October 1, before physical Brent reached $135.51 on October 2. The ten available October spot observations through October 14 average $126.977, rounded to $126.98. With an illustrative 64 observation days in Q4, the remaining 54 would have to average (64×$105 − 10×$126.977)/54 = $100.93, roughly $101 (B). This makes the forecast look low relative to the observed October path, not impossible; it is a conditional calculation, not a revised EIA forecast. S14, S59; claims 15, 56

The requested oil range is a futures quote from FT's indexed table; full access failed and exchange settlement was not retrieved. EIA/FRED physical Brent is higher. The quoted $105.92–109.95 futures closes are above the roughly $101 remaining-quarter sensitivity, but they concern different delivery horizons and dates. Those differences may explain the physical/futures gap; the precise basis remains unverified. S11, S12

EIA estimates Middle East shut-ins fell from 5.8m b/d in August to 4.8m in September and reports partial East–West pipeline resumption on September 22. Recovery is already partly observed, not just forecast; it does not establish normalization. S14

Power exposure depends on local generation, PPAs and backup fuel. September gasoline explained over half the monthly CPI rise. Commercial electric power PPI fell 0.6% m/m, but consumer electricity CPI rose 0.2% m/m and 4.4% y/y, qualifying a benign power-price reading. These measure different customers and price concepts; none is a data-center PPA index; uniform oil-to-power cost transmission is unsupported. Weaker spending and dearer funding/construction remain conditional channels, while August real PCE growth provides a demand counterweight. S4, S5, S51

Deficits imply competition for savings. Net borrowing excludes refinancing included in gross issuance. Treasury's advisory committee recommended unchanged coupon sizes despite possible later increases. S15, S16, S17

Long yields combine expected short rates and a model-estimated term premium. Fiscal supply, inflation uncertainty and growth can all lift yields. No contemporaneous ACM premium was extracted; fiscal stress cannot explain the entire move by assumption. S18

The dollar reading is a reference, not a full-year trend claim. Appreciation raises foreign buyers' local-currency GPU and debt costs; depreciation reverses that burden. Hedges and revenue currencies determine exposure. S10

Sovereign and strategic capital: count the check, not the headline

PIF's HUMAIN spans models, cloud and infrastructure. Its AMD envelope has no disclosed sovereign allocation. Its identified xAI check subsequently converted into SpaceX equity, changing the exposure. S22, S23, S24

MGX's founders Mubadala and G42 connect sovereign finance with G42's Stargate UAE compute partnership. Avoid counting these related pools repeatedly. AIP includes outside equity and debt. Aligned has closed, but enterprise value includes financing and payments to sellers, rather than solely new capex or MGX capital. S25, S26, S27, S28

The December 9, 2025 $20bn Qai–Brookfield JV combines partners’ capital without disclosing QIA’s check. QIA identifies Anthropic and Crusoe participation without check sizes. MENA Startup Digest’s April 1 report also places MGX in OpenAI’s $122bn March 2026 round, with no individual check disclosed (C, reported, unverified); participation is excluded from dollar addition rather than treated as zero. S64 GIC and Mubadala Capital also appear among participants. Whole rounds cannot be attributed to states, nor must Mubadala-managed capital all be sovereign-owned. S29, S30, S31

The reproducible funding reconciliation is:

Claims 23–39 support this partial sum: $3bn + $8.9bn = $11.9bn; ¥100bn + ¥150bn = ¥250bn; $11.9bn + $60bn = $71.9bn. Both cross-checks found no arithmetic error. Claude questions the analytic usefulness of these mixed subtotals; they remain here solely to make the original reconciliation auditable and are removed from the Summary. They are bookkeeping subtotals across beneficiaries, instruments, dates and execution status, not a measure of current holdings or global AI investment. Missing allocations, beneficial ownership and a matched global denominator prevent estimating the sovereign share. Japan remains in yen.

Stargate's ambition is not a sovereign appropriation. SoftBank funded its final OpenAI tranche with bonds and repaid/canceled its bridge. Its September notes carry $941.25m plus €75.625m in gross annual coupons (claim 48, B), before hedging, fees or tax. The weighted USD coupon is $941.25m/$10bn = 9.4125%, or 9.41%, with maturities in April 2030, 2032 and 2034. Relative to the September 24 ten-year Treasury at 5.18%, that is roughly 4.2 percentage points; the notes’ 3.5–7.5-year terms make this an approximate benchmark comparison, not a tenor-matched credit spread. Buying lab equity with debt at that cost makes SoftBank a concrete leverage-financed-purchase exposure, even after bridge repayment. S55 S32, S33, S34, S38

The Arm facility is non-recourse to the parent; the OpenAI-linked loan has cash-account collateral and a parent guarantee. Valuation declines can trigger collateral or prepayment requirements, but numerical thresholds remain unverified. The memorandum also discloses a fully drawn $11.9bn senior term facility with a two-year maturity; the OpenAI-linked loan matures in August 2028. Repayment of the named bridge therefore does not demonstrate equivalent consolidated deleveraging (inference). S37

Bloomberg's October 6 report of UAE funds including MGX discussing up to $10bn collectively in OpenAI is a C-grade, reported, unverified watch lead. It adds a possible inflow alongside the pullback scenario, but no funded check to any subtotal. S54

Intel equity reallocates prior grants/program funding: avoid counting both. Japan's Rapidus equity is executed; EU InvestAI is a mobilization ambition. Strategic autonomy could sustain uneconomic capacity and transfer losses to public investors; this is a conditional mechanism, not an observed outcome established by these announcements. S39, S40, S41

Geopolitical tails and scenario transmission

Taiwan disruption threatens physical supply. On October 15, TSMC scheduled Arizona fab two volume production for Q3 2027 and described building 13 leading-edge/advanced-packaging fabs in Taiwan and five to six overseas. Diversification is proceeding, but the disclosed plans do not establish immediate substitutability for Taiwan supply. Its Q3 revenue of $46.72bn and 58.7% operating margin are a profitable upstream counterweight; they do not prove customers' incremental AI returns. S46, S52, S53

BIS permits case-by-case chip licensing, but Nvidia reports Chinese restrictions and tariffs it cannot pass through. AMD distinguishes uncodified revenue-share expectations from inspection tariffs. Neither is equity or fresh funding. S42, S43, S45

China’s MOFCOM/MOST Announcement No. 48 of October 16 proposes catalogue changes while imposing immediate temporary pre-export registration for Chinese technology. Its 23 AI categories concern models, agents, compute clusters and information processing; registration includes unfinished outbound transfers, offshore licensing and changes of technology ownership/control to foreign entities. The AI provisions do not specify chips or AI hardware (the separate photovoltaic provisions do mention certain equipment). The direct channel is Chinese labs’ offshore licensing and foreign acquisitions of Chinese technology owners; a small direct effect on US AI capex is an inference, not a measured impact. Public-domain technology is exempt. Registration is already effective; the proposed catalogue is not a final blanket export prohibition. Final thresholds and practical offshore licensing effects remain uncertain; existing open models are not universally banned. S44

Russia–Ukraine affects refinery, sanctions and shipping risks. The latest US Russia–Iran sanctions order includes extensive waivers: headline penalties do not establish lost oil supply (primary, A). Middle East escalation threatens oil, LNG routes and Gulf investment. These transmission hypotheses have no measured Russian contribution here. Nvidia reports no significant impact on Israel operations so far. S49, S14, S43

The scenarios below are conditional mechanisms, not probability forecasts. Equipment includes semiconductor tools; facilities includes power and data centers.

Conflicts and open questions

• October 25 update — forecasts versus decisions: venue cards (93–96% hold) and ING's preview agree on an October hold, but ING forecasts a December hike to 4.00–4.25%, consistent with the October 7 minutes' "most participants" expecting another hike by year-end. All are expectations; the September 3.75–4.00% range remains the latest verified policy until the Oct 28 statement. The CME 15.5% hike figure is an older, different-venue snapshot and is not a point on the same series. Claims 1, 7, 42, 66

• Oil precision: a scan's Friday Brent $116.72 / WTI >$100 could not be traced to a dated settlement; the latest verified physical spot here remains $126.55 (Oct 14) and the latest futures close $109.95 (Oct 16, C). The gap between these benchmarks persists and is not resolved by intraday reports. Claims 12–13, 59

• Labor-market dating: continuing claims of 1.728m (+25,000) refer to the week ended Oct 10, not the Oct 17 reference week of the 194,000 initial-claims figure; low initial claims coexist with rising insured unemployment. Claim 65

• Fed interpretation — Codex versus Claude: Codex stresses supportive financial conditions, broadly available credit and fixed-rate/hedge distinctions, rejecting a general financing-bust inference. Claude weighs the same October 7 minutes more adversely: wide hyperscaler AI spreads, heavy private debt issuance, real-yield increases and the demand-versus-supply warning point to late-cycle feedback. Both retain A for the recorded assessments; their cycle judgments differ and neither is a causal estimate. Claims 42, 52–54

• Oil and power — Codex versus Claude: Codex emphasizes the decline in September shut-ins to 4.8m b/d, partial pipeline recovery and commercial-power PPI −0.6% m/m. Claude adds electricity CPI +4.4% y/y and an October spot mean of $126.98, and notes that EIA completed the $105 Q4 forecast before the October 2 spike. Both recovery and inflation pressure are observed; their weight differs. Claims 15, 44–45, 56

• Sovereign perimeter and subtotals — Codex versus Claude: Codex regards the $11.9bn plus ¥250bn and $71.9bn plus ¥250bn reconciliations as useful selected bookkeeping with explicit limits. Claude finds these mixed lab/chipmaker/foundry totals analytically unhelpful and the lab perimeter incomplete without reported MGX participation. The arithmetic remains correct (B); Summary now separates the $3bn disclosed-check floor from unquantified participation. Claims 39, 55

• China transmission — Codex versus Claude: Codex’s recommended Summary grouped October 16 registration with Taiwan as a deployment constraint. Claude limits the AI rule to outbound Chinese technology/licensing and foreign takeovers; the fetched announcement supports that narrower direct channel. Taiwan remains a separate physical supply risk, with Arizona fab two due Q3 2027. Claims 38, 49

• Rates versus AI repricing — shared blind spot identified by Claude: The earlier Codex review did not examine the October 9–16 market changes. Claude disputes the theory page’s inference from high yield/oil levels: the ten-year fell 15 bp while M1 lost 12.7%. The new market-reaction analysis records that evidence and treats the financing interpretation as descriptive; it does not claim a completed causal decomposition. The theory page is outside this edit’s scope. Claims 53–54, 57

• Source provenance / dates: Claude’s AI-credit comparison is attributed to an unspecified TipRanks wrap; this pass obtained the same +25 bp/360 bp versus +2 bp figures from VLT Capital, not the Goldman note or TipRanks item. The reviewed secondary estimate stays C; the fetched analyst commentary is tertiary (D) under the methodology and must not be presented as a verified Goldman dataset. Reuters attributes the October 14 roughly 33% hike odds to LSEG, whereas the October 17 15.5% snapshot is CME; they are not a uniform weekly series. The MENA MGX report was updated April 1 and concerns the March round. S63–S65; claims 54–55, 57

• PCE vintage: The minutes’ 3.8%/3.4% old-basis and 3.6%/3.2% new-basis staff estimates were prepared before BEA released 3.4%/3.0% on September 30. The later Claude synthesis review’s 3.8% headline citation is a superseded estimate, not an alternative actual reading; both the estimate and release are preserved here. S50, S51; claim 43

• Financing interpretation — original note versus Codex cross-check: the original note characterized the external financing backdrop as tightened and policy as restrictive, emphasizing higher rates and distant profits. The Codex cross-check gives greater weight to the September minutes: supportive financial conditions, broadly available credit and AI-driven inflation/financing demand; several participants saw policy as not restrictive or only mildly restrictive. Higher rate hurdles remain, but neither view establishes general credit contraction or a cycle-ending shock. October hold odds do not settle the year-end hike question. S2, S6, S50

• Borrower vulnerability — original note versus Codex cross-check: the original summary generalized floating-rate pressure to neocloud and construction borrowing. The Codex reviewer stresses fixed debt, swap deadlines and recourse; net October exposure is unknown. Incumbent net income supports resilience but is not post-capex cash, and TSMC's profitable Q3 adds upstream counterevidence. S19, S43, S47, S53

• Energy interpretation — original note versus Codex cross-check: the original note emphasized disruption and projected recovery. The Codex reviewer gives more weight to already declining shut-ins, partial pipeline resumption, falling commercial-electric-power PPI and positive August real PCE. These qualify uniform cost or demand-damage claims without establishing normalization. Benchmark/date differences still prevent treating the futures quote as physical spot. S4, S5, S11, S12, S14, S51

• Patient capital versus market dependence: the original note emphasized bridge repayment reducing rollover risk; the Codex cross-check agrees for that bridge but rejects inferring equivalent consolidated deleveraging given other borrowing and collateral provisions. Sovereign mandates may stabilize investment, while uneconomic-capacity support remains a hypothesis rather than an A-grade observed outcome. S34, S37, S38, S39, S40, S41

• Sovereign attribution — original note versus Codex cross-check: the original summary said announcements “substantially overstate” attributable cash. The Codex reviewer finds missing allocations and incompatible scopes, not a measurable overstatement. The historical funding arithmetic passes; neither current holding values nor a global sovereign share follows from it. S24, S29; claim 39

• Sovereign inflow watch: Bloomberg reported October 6, 2026 that UAE funds including MGX discussed up to $10bn collectively in OpenAI (secondary, C, reported, unverified). This is a counterpoint to the pullback scenario, excluded from all sums pending a primary closing disclosure. S54

• Funding uncertainty — original note versus Codex cross-check: the original summary ranked unconditional funding as the “biggest” unknown; the Codex reviewer finds that ranking unestablished. Undisclosed MGX/QIA/GIC checks, external LPs, overlapping facilities and staged commitments remain unresolved, preventing a global sum or percentage. The stated floor is deliberately incomplete.

• Review scope: the earlier Codex audit targeted selected original claim rows; it did not certify CME's timestamped distribution, the futures settlement, all other rows or detailed Russia–Iran sanctions waivers. Its White House fetch failed; that access limitation does not contradict the original note's direct-fetch claim. Codex cross-check. The Claude cross-check re-verified twelve selected rows and added the market, vintage and scope critiques above; neither review certifies every row

• Unverified: exact term-premium reading, exchange-verified Brent settlement, aggregate sovereign share, margin-call thresholds, current Russian energy flows and final Chinese export-control thresholds. These require datasets, agreements or final rules not obtained here.

• October update versus earlier demand/energy interpretation: August real PCE +0.6% remains the latest spending observation used here, but the Oct 20–21 housing/earnings releases add weaker signals. Housing-start declines are not statistically significant, state payrolls are mostly unchanged, and wage/CPI comparison uses quarterly averages; none establishes a broad contraction. The Oct 20 Brent ~$111→~$106 intraday move also conflicts with describing >$111 as a verified close. No new historical Fed-odds series or undated Treasury range is inferred. Claims 43, 58–63

Signals to watch

CSV: month-start labels identify reference months; Fed values are month-end carry-forwards, GS10/Brent monthly averages. Daily snapshots are labeled; PPI uses the latest vintage; missing CPI annual history is not interpolated.

Sources

Original sources accessed 2026-10-17; the re-checks and additions below were accessed 2026-10-18 as marked. Issuer-distributed press releases are primary evidence of the issuer's statement. Search-index-only access is specifically flagged above.

1. Federal Reserve, “FOMC statements”. Published/as of 2026-01-28 / 03-18 / 04-29 / 06-17 / 07-29. https://www.federalreserve.gov/newsevents/pressreleases/monetary20260128a.htm. Accessed 2026-10-17.

2. Federal Reserve, “September FOMC statement”. Published/as of 2026-09-16. https://www.federalreserve.gov/newsevents/pressreleases/monetary20260916a.htm. Accessed 2026-10-17; re-fetched 2026-10-18.

3. Federal Reserve, “September implementation note”. Published/as of 2026-09-16. https://www.federalreserve.gov/newsevents/pressreleases/monetary20260916a1.htm. Accessed 2026-10-17.

4. BLS, “CPI: September”. Published/as of 2026-10-14. https://www.bls.gov/news.release/archives/cpi_10142026.htm. Accessed 2026-10-17; re-fetched 2026-10-18.

5. BLS, “PPI: September, including revised history”. Published/as of 2026-10-15. https://www.bls.gov/news.release/archives/ppi_10152026.htm. Accessed 2026-10-17.

6. CME Group, “FedWatch, live QuikStrike widget; observed 02:37:45 CT”. Published/as of 2026-10-17. https://www.cmegroup.com/markets/interest-rates/cme-fedwatch-tool.html. Accessed 2026-10-17.

7. Federal Reserve / FRED, “DGS10: daily Treasury yield”. Published/as of updated 2026-10-16. https://fred.stlouisfed.org/series/DGS10. Accessed 2026-10-17.

8. Federal Reserve / FRED, “GS10: monthly Treasury yield; CSV fetched directly”. Published/as of 2026-09 observation. https://fred.stlouisfed.org/graph/fredgraph.csv?id=GS10&cosd=2026-01-01&coed=2026-10-17. Accessed 2026-10-17.

9. New York Fed / FRED, “SOFR”. Published/as of updated 2026-10-16. https://fred.stlouisfed.org/series/SOFR. Accessed 2026-10-17.

10. Federal Reserve / FRED, “Nominal broad dollar index”. Published/as of updated 2026-10-13. https://fred.stlouisfed.org/series/DTWEXBGS. Accessed 2026-10-17.

11. Financial Times / Markit Digital, “Brent futures historical prices; indexed table fetched, full page returned forbidden”. Published/as of 2026-10-16 observation. https://markets.ft.markitdigital.com/data/commodities/tearsheet/historical?c=Brent+Crude+Oil. Accessed 2026-10-17.

12. EIA / FRED, “Brent Europe physical spot”. Published/as of updated 2026-10-15. https://fred.stlouisfed.org/series/DCOILBRENTEU. Accessed 2026-10-17.

13. EIA / FRED, “MCOILBRENTEU: monthly Brent spot; CSV fetched directly”. Published/as of 2026-09 observation. https://fred.stlouisfed.org/graph/fredgraph.csv?id=MCOILBRENTEU&cosd=2026-01-01&coed=2026-10-17. Accessed 2026-10-17.

14. EIA, “Short-Term Energy Outlook: global oil markets”. Published/as of 2026-10-06. https://www.eia.gov/outlooks/steo/report/global_oil.php. Accessed 2026-10-17; re-fetched 2026-10-18.

15. CBO, “Monthly Budget Review: September”. Published/as of 2026-10-08. https://www.cbo.gov/publication/61985. Accessed 2026-10-17.

16. US Treasury, “Marketable borrowing estimates”. Published/as of 2026-08-03. https://home.treasury.gov/news/press-releases/sb0584. Accessed 2026-10-17.

17. US Treasury, “Treasury Borrowing Advisory Committee minutes”. Published/as of 2026-08-05. https://home.treasury.gov/news/press-releases/sb0592. Accessed 2026-10-17.

18. New York Fed, “Treasury Term Premia: ACM methodology”. Published/as of undated, current page. https://www.newyorkfed.org/research/data_indicators/term-premia-tabs?mod=livecoverage_web. Accessed 2026-10-17.

19. CoreWeave / SEC, “Quarterly report, debt and interest-rate risks”. Published/as of 2026-08 filing; period ended 2026-06-30. https://www.sec.gov/Archives/edgar/data/1769628/000176962826000366/crwv-20260630.htm. Accessed 2026-10-17.

20. Bank of Japan IMES, “Okina, Shirakawa and Shiratsuka: Asset Price Bubble and Monetary Policy, Table 7”. Published/as of 2001-02. https://www.imes.boj.or.jp/research/papers/english/me19-s1-14.pdf. Accessed 2026-10-17.

21. Federal Reserve, “FOMC tightening announcement”. Published/as of 2000-05-16. https://www.federalreserve.gov/boarddocs/press/general/2000/20000516/default.htm. Accessed 2026-10-17.

22. PIF, “Crown Prince launches HUMAIN”. Published/as of 2025-05-12. https://www.pif.gov.sa/en/news-and-insights/press-releases/2025/hrh-crown-prince-launches-humain-as-global-ai-powerhouse/. Accessed 2026-10-17.

23. AMD, “AMD and HUMAIN strategic collaboration”. Published/as of 2025-05-13. https://newsroom.amd.com/news/amd-and-humain-form-strategic-10b-collaboration/. Accessed 2026-10-17.

24. HUMAIN, issuer release via PRNewswire, “Investment in xAI before SpaceX merger”. Published/as of 2026-02-18. https://www.prnewswire.com/news-releases/humain-backs-xai-with-3-billion-series-e-investment-ahead-of-historic-spacex-merger-302691510.html. Accessed 2026-10-17.

25. Mubadala, “Abu Dhabi launches MGX with Mubadala and G42”. Published/as of 2024-03-11. https://www.mubadala.com/en/news/abu-dhabi-launches-comprehensive-global-investment-strategy-on-artificial-intelligence?os=0. Accessed 2026-10-17.

26. MGX, “Aligned acquisition announcement and AIP targets”. Published/as of 2025-10-15. https://www.mgx.ae/news-insights/ai-infrastructure-partnership-aip-mgx-and-blackrocks-global-infrastructure-partners. Accessed 2026-10-17.

27. MGX, “Aligned acquisition completion”. Published/as of 2026-07-21. https://www.mgx.ae/news-insights/aip-mgx-and-blackrocks-gip-close-acquisition-aligned-data-centers. Accessed 2026-10-17.

28. G42, issuer release via PRNewswire, “Global Tech Alliance Launches Stargate UAE”. Published/as of 2025-05-22. https://www.prnewswire.com/news-releases/global-tech-alliance-launches-stargate-uae-302463328.html. Accessed 2026-10-17.

29. QIA, “Brookfield–Qai AI infrastructure partnership”. Published/as of 2025-12-09. https://www.qia.qa/en/Newsroom/Pages/Brookfield-and-Qai-Form-$20-Billion-Strategic-Investment-Partnership-for-AI-Infrastructure.aspx. Accessed 2026-10-17.

30. QIA, “Anthropic Series H participation”. Published/as of 2026-05-28. https://www.qia.qa/en/Newsroom/Pages/QIA-Increases-Investment-in-Anthropic-Through-$65-Billion-Series-H-Financing.aspx. Accessed 2026-10-17.

31. QIA, “Crusoe Series F participation”. Published/as of 2026-09-17. https://qia.qa/en/Newsroom/Pages/QIA-Participates-in-Crusoe’s-$3.9-Billion-Series-F-Funding-Round-to-Support-AI-Infrastructure.aspx. Accessed 2026-10-17.

32. SoftBank, “Stargate announcement”. Published/as of 2025-01-22. https://group.softbank/en/news/press/20250122. Accessed 2026-10-17.

33. SoftBank, “Completion of additional OpenAI investment”. Published/as of 2025-12-31. https://group.softbank/en/news/press/20251231. Accessed 2026-10-17.

34. SoftBank, “OpenAI final tranche executed; bridge canceled”. Published/as of 2026-10-01. https://group.softbank/en/news/press/20261001. Accessed 2026-10-17; re-fetched 2026-10-18.

35. SoftBank, “Bridge facility agreement”. Published/as of 2026-03-27. https://group.softbank/en/news/press/20260327. Accessed 2026-10-17.

36. SoftBank, “Early repayment announcement”. Published/as of 2026-09-09. https://group.softbank/en/news/press/20260909. Accessed 2026-10-17.

37. SoftBank / SGX, “Offering memorandum; Recent Developments and Capitalization; PDF downloaded and text extracted”. Published/as of 2026-09-23. https://links.sgx.com/FileOpen/SoftBank Group Corp. - Offering Memorandum dated 23 September 2026.ashx?App=Prospectus&FileID=71217. Accessed 2026-10-17.

38. SoftBank, “Foreign currency senior notes pricing”. Published/as of 2026-09-24. https://group.softbank/en/news/press/20260924. Accessed 2026-10-17; re-fetched 2026-10-18.

39. Intel, “US government equity agreement”. Published/as of 2025-08-22. https://www.intel.com/content/www/us/en/newsroom/news/corporate/intel-and-trump-administration-reach-historic-agreement.html. Accessed 2026-10-17.

40. Rapidus, “Completed government funding round”. Published/as of 2026-06-05. https://www.rapidus.inc/en/news_topics/information/20260605-2/. Accessed 2026-10-17.

41. European Commission, “InvestAI launch”. Published/as of 2025-02-11. https://digital-strategy.ec.europa.eu/en/news/eu-launches-investai-initiative-mobilise-eu200-billion-investment-artificial-intelligence. Accessed 2026-10-17.

42. AMD, “Quarterly report: China licenses and proposed revenue share”. Published/as of 2026-08-05. https://ir.amd.com/financial-information/sec-filings/content/0000002488-26-000123/amd-20260627.htm. Accessed 2026-10-17.

43. Nvidia, “Quarterly report: China tariffs, financials and Middle East risks”. Published/as of 2026-08-27; period ended 2026-07-26. https://investor.nvidia.com/files/doc_financials/2027/NVDA-2027-Q2-10Q-Final-including-exhibits.pdf. Accessed 2026-10-17.

44. MOFCOM / MOST, “Draft technology export catalogue and immediate temporary registration”. Published/as of 2026-10-16. https://www.mofcom.gov.cn/zcfb/dwmygl/art/2026/art_61446ab4beee4f63973d4d0ab3097648.html. Accessed 2026-10-17; re-fetched 2026-10-18.

45. BIS, “Revised China semiconductor license review policy”. Published/as of 2026-01-13. https://media.bis.gov/press-release/department-commerce-revises-license-review-policy-semiconductors-exported-china. Accessed 2026-10-17.

46. TSMC, “Annual report: continuing Taiwan investment and overseas expansion”. Published/as of 2025 reporting year; publication date not displayed. https://investor.tsmc.com/static/annualReports/2025/english/index.html. Accessed 2026-10-17.

47. Microsoft, “Full-year financial results”. Published/as of 2026-07-29. https://www.microsoft.com/en-us/investor/earnings/fy-2026-q4/press-release-webcast. Accessed 2026-10-17.

48. New York Fed, “Perli: reserve-management purchases are not monetary easing”. Published/as of 2026-03-26. https://www.newyorkfed.org/newsevents/speeches/2026/per260326. Accessed 2026-10-17.

49. White House, “Implementing the Lindsey O. Graham Sanctioning Russia and Iran Act of 2026”. Published 2026-10-16; primary HTML fetched directly after web fetch failed. https://www.whitehouse.gov/presidential-actions/2026/10/implementing-the-lindsey-o-graham-sanctioning-russia-and-iran-act-of-2026/. Accessed 2026-10-17.

50. Federal Reserve, “September FOMC minutes; direct HTML fetch after web-tool access failure”. Published/as of 2026-10-07; meeting 2026-09-15–16. https://www.federalreserve.gov/monetarypolicy/fomcminutes20260916.htm. Accessed 2026-10-17; re-fetched 2026-10-18.

51. BEA, “Personal Income and Outlays, August 2026; annual update”. Published/as of 2026-09-30. https://www.bea.gov/news/2026/personal-income-and-outlays-august-2026. Accessed 2026-10-17; re-fetched 2026-10-18.

52. TSMC, “Q3 2026 earnings-call transcript, capacity plans (p. 5)”. Published/as of 2026-10-15. https://investor.tsmc.com/english/encrypt/files/encrypt_file/reports/2026-10/eba4fac72873bf546b669bd2b5d973df5c93215d/TSMC 3Q26 Transcript.pdf. Accessed 2026-10-17.

53. TSMC, “Q3 2026 earnings release”. Published/as of 2026-10-15. https://investor.tsmc.com/english/encrypt/files/encrypt_file/reports/2026-10/526b65af7305df2dde6b9ece6812b1fdf1e49f9b/3Q26 EarningsRelease.pdf. Accessed 2026-10-17.

54. Bloomberg, republished by Moneycontrol, “OpenAI looks to UAE for up to $10 billion as it lines up $30 billion round; BlackRock in talks; reported, unverified”. Published/as of 2026-10-06. https://www.moneycontrol.com/news/business/openai-looks-to-uae-for-up-to-10-billion-as-it-lines-up-30-billion-round-blackrock-in-talks-14045202.html. Accessed 2026-10-17.

55. Federal Reserve / FRED, “DGS10 daily history, 2000–October 15, 2026”; downloaded CSV scanned for last higher close and 2007 maximum. https://fred.stlouisfed.org/graph/fredgraph.csv?id=DGS10&cosd=2000-01-01&coed=2026-10-17. Accessed 2026-10-18.

56. Federal Reserve / FRED, “DFII10: ten-year inflation-indexed Treasury yield”; observations through October 15, 2026. https://fred.stlouisfed.org/graph/fredgraph.csv?id=DFII10&cosd=2026-01-01&coed=2026-10-17. Accessed 2026-10-18.

57. Federal Reserve / FRED, “DGS30: thirty-year Treasury yield”; observations through October 15, 2026. https://fred.stlouisfed.org/graph/fredgraph.csv?id=DGS30&cosd=2026-01-01&coed=2026-10-17. Accessed 2026-10-18.

58. Federal Reserve / FRED, “DGS2: two-year Treasury yield”; observations through October 15, 2026. https://fred.stlouisfed.org/graph/fredgraph.csv?id=DGS2&cosd=2026-01-01&coed=2026-10-17. Accessed 2026-10-18.

59. EIA / FRED, “DCOILBRENTEU daily physical Brent”; observations through October 14, 2026; October mean calculated from ten non-missing values. https://fred.stlouisfed.org/graph/fredgraph.csv?id=DCOILBRENTEU&cosd=2026-01-01&coed=2026-10-17. Accessed 2026-10-18.

60. ICE Data Indices / FRED, “ICE BofA US High Yield Index Option-Adjusted Spread”; observations through October 15, 2026. https://fred.stlouisfed.org/graph/fredgraph.csv?id=BAMLH0A0HYM2&cosd=2026-01-01&coed=2026-10-17. Accessed 2026-10-18.

61. US Treasury, “Daily Treasury Par Yield Curve Rates, 2026”; October 16 ten-year 5.09% and thirty-year 5.43%. https://home.treasury.gov/resource-center/data-chart-center/interest-rates/TextView?field_tdr_date_value=2026&type=daily_treasury_yield_curve. Accessed 2026-10-18; dated Oct 9–16 ten-year observations re-fetched 2026-10-21.

62. Yahoo Finance, “Daily equity and SOX histories”; observations through October 16, 2026, C price inputs, with reproducible B basket arithmetic. https://query1.finance.yahoo.com/v8/finance/chart/NVDA?range=2y&interval=1d. Also re-fetched https://query1.finance.yahoo.com/v8/finance/chart/AVGO?range=2y&interval=1d, https://query1.finance.yahoo.com/v8/finance/chart/TSM?range=2y&interval=1d, https://query1.finance.yahoo.com/v8/finance/chart/ORCL?range=2y&interval=1d, https://query1.finance.yahoo.com/v8/finance/chart/CRWV?range=2y&interval=1d, https://query1.finance.yahoo.com/v8/finance/chart/AMD?range=2y&interval=1d, https://query1.finance.yahoo.com/v8/finance/chart/MU?range=2y&interval=1d, https://query1.finance.yahoo.com/v8/finance/chart/VRT?range=2y&interval=1d, https://query1.finance.yahoo.com/v8/finance/chart/^SOX?range=2y&interval=1d. Nasdaq official NDX EOD levels (A; returns B): https://indexes.nasdaq.com/Index/History/NDX, fetched by POST to https://indexes.nasdaq.com/Index/HistoryData with id=NDX, startDate=2026-10-09, endDate=2026-10-16, timeOfDay=EOD. Accessed 2026-10-18.

63. VLT Capital, “Le crédit, signal d’avertissement sur le financement du cycle de l’IA”; published October 16, 2026; tertiary analyst commentary (D) repeats the review’s C-grade secondary estimate of +25bp/360bp versus +2bp attributed to Goldman, primary note unavailable; direct HTML fetch. https://vltcapital.substack.com/p/le-credit-signal-davertissement-sur. Accessed 2026-10-18.

64. MENA Startup Digest, “UAE’s MGX Joins OpenAI in Historic $122 Billion Funding Round”; updated April 1, 2026, reporting the March round; reported, unverified participation, C; direct HTML fetch. https://menastartupdigest.com/?p=34969. Accessed 2026-10-18.

65. Reuters, republished by AOL, “Asia shares edge lower as chip selloff deepens”; October 14, 2026 market report, CPI consensus and LSEG policy-odds snapshot; direct HTML fetch. https://www.aol.com/articles/asia-shares-edge-chip-selloff-015255000.html. Accessed 2026-10-18.

66. Federal Reserve, FOMC meeting calendar; Oct 27–28, 2026. https://www.federalreserve.gov/monetarypolicy/fomccalendars.htm. Re-fetched 2026-10-21.

67. New York Fed, October 2026 economic-indicators calendar; GDP first release and personal income/PCE Oct 29, 08:30. https://www.newyorkfed.org/research/calendars/i-oct26.html. Re-fetched 2026-10-21.

68. Cinco Días, Oct 20 market report; intraday Brent observations, not settlements. https://cincodias.elpais.com/mercados-financieros/2026-10-20/la-bolsa-y-el-ibex-35.html. Re-fetched 2026-10-21.

69. Census/HUD, September new residential construction; released Oct 20; first PDF page and uncertainty intervals checked. https://www.census.gov/construction/nrc/pdf/newresconst_202609.pdf. Re-fetched 2026-10-21.

70. EIA, weekly petroleum summary; week ended Oct 16; live text fetched directly. https://ir.eia.gov/wpsr/summary.txt. Re-fetched 2026-10-21.

71. BLS, Q3 usual weekly earnings; Oct 21 archive fetched directly. https://www.bls.gov/news.release/archives/wkyeng_10212026.htm. Re-fetched 2026-10-21.

72. BLS, September state employment and unemployment; Oct 20 release, statistical-significance table. https://www.bls.gov/news.release/archives/laus_10202026.htm. Re-fetched 2026-10-21.

73. Polymarket, Fed decision in October 2026 (2026 market); cards read Oct 25 per signals 2026-10-25 item 66. https://polymarket.com/event/fed-decision-in-october-20260617190323537. Not re-fetched.

74. Kalshi, Fed meeting October 2026; Chance headings read Oct 25 per signals 2026-10-25 item 66. https://kalshi.com/markets/kxfeddecision/fed-meeting/kxfeddecision-26oct. Re-fetch refused (connection) 2026-10-25.

75. EIA, Weekly Petroleum Status Report Table 1, dated archive for Oct 21 release. https://www.eia.gov/petroleum/supply/weekly/archive/2026/2026_10_21/pdf/table1.pdf. Per signals 2026-10-25 item 54; not re-fetched.

76. US Department of Labor, Unemployment Insurance Weekly Claims, release of Oct 22, 2026 (PDF text extracted). https://www.dol.gov/sites/dolgov/files/OPA/newsreleases/ui-claims/20261613.pdf. Fetched 2026-10-25.

77. ING Think, Knightley, Garvey and Turner, "Federal Reserve preview: a pause before a December hike". https://think.ing.com/articles/federal-reserve-preview-a-pause-before-a-december-hike/. Fetched 2026-10-25 (relative date "Yesterday, 14:41").

78. Bloomberg, Oct 23 essay on AI inflation and the Fed decision; restricted, not retrieved. https://www.bloomberg.com/news/articles/2026-10-23/federal-reserve-interest-rate-decision-faces-challenges-from-ai-inflation-war. Treasury S61 was re-fetched on 2026-10-25 as the 2026 CSV download from the same page for Oct 16–23 rows.

Additional FOMC statements supporting S1: March, April, June, July. Published on the dates embedded in each URL; accessed 2026-10-17. FOMC calendar, undated current page, accessed 2026-10-17.