Macro Weekly · Research

Market Overview
Facts & Positioning

Week of August 3, 2026 · Issue #2

Fact layer only — every exhibit carries Source + as-of. Interpretation and views are published separately (Chinese weekly) and appear here solely on the page marked OUR VIEW. Data cutoff: Jul 31 close, plus Aug 3 intraday releases (QRA, ISM) and the Aug 3 close.

Macro Weekly | Market Overview | Aug 3, 202601

This Week in One Page

Two Full Round Trips — in Oil and in Positioning — Ended at a Record Close

+0.43%
SPX over two weeks, via a −1.9% drawdown; Monday close 7,600.50, a new high
54.5%
Polymarket odds of a September +25bp hike — above "no change" (42.5%)
−$158bn
Bank reserves over two weeks — last issue's "August drain" arrived early
203bp
Oracle 5Y CDS, a record; S&P cut to BBB−
  • Brent ran 88 → 105+ → ~84: escalation to a 13-night bombing peak, a truce window, a broken truce, and canceled strikes (Aug 2).
  • The CTA trigger (7,427) fired on Jul 23 — and the modeled supply was absorbed by record retail inflows and reopened buybacks.
  • All four hyperscalers raised 2026 capex; Kimi K3's parity claim was refuted by third-party evals.
  • The equity-credit divergence flipped sides: last issue equities panicked while credit stayed calm; now equities sit at highs while AI credit deteriorates.
Source: FRED (Jul 31 / Aug 3); Polymarket API (Aug 4); Treasury (Aug 3); GS via Tickmill (Jul 31)02

Scorecard · Last Issue's Seven Triggers

Five Verdicts, Two Pending — the First Public Settlement

#TriggerOutcomeVerdict
1Any SPX close below 7,427Jul 23 closed 7,408.30; below trigger 4 of the next 5 sessions; trough 7,316.15 (Jul 29)HIT
2≥50% of Jul 17 drawdown recovered in 3 sessions68% recovered within two sessions (Jul 21)HIT — bid intact
3Any hyperscaler cuts 2026 capex guideAll four raised or lifted the floor; MSFT's optical cut is a depreciation reclassNOT HIT (inverse)
4Hike language or ≥2 dissents at FOMC9-3 hold; Hammack, Kashkari, Logan dissented for a +25bp hikeHIT
5HY OAS closes above 300bpPeaked 287bp (Jul 29); +13bp over two weeksNOT HIT — rolled
6July core CPI m/m ≥0.4%Prints Aug 12PENDING
7Lab revenue second derivativeAnthropic run-rate $47B (May) → ~$69B (end-June)HIT — accelerating
  • Framework revision 1: a fired trigger is a flow forecast, not a price forecast — mechanical supply must be netted against counter-flows (record retail buying + buybacks).
  • Framework revision 2: "maintaining capex guidance = bullish" is obsolete. The market now splits capex into revenue-validated (MSFT +7%, AMZN +15%) and not (GOOGL −7%, META −10%).
Source: FRED SP500 (Jul 31); Federal Reserve (Jul 29); Yipit via Jukan (end-June, relayed)03

Flows · Systematic

CTAs Flipped From Buyers to Sellers; Down-Tape Supply Grew to $172bn

Scenario1W Global ($bn)1W US ($bn)1M Global ($bn)1M US ($bn)
Flat tape−1.3−5.2+16.6−3.2
Up tape+2.3−1.9+37.3+10.7
Down tape−24.9−15.7−172.3−68.4
−172.3
+37.3
1M down-tape supply (was −127.7)1M up-tape demand (was +41.7)
  • Pivots moved up: short-term 7,453 · medium 7,204 · long 6,765. Friday closed 7,489.72, back above the short-term line; Monday 7,600.50.
  • Last issue's price-insensitive US down-tape bid (+$4.1bn) is gone — now −$15.7bn.
  • Positioning de-risked: systematics $196bn long US equities (48th pct, 3y); CTAs at the 44th pct. Asymmetry now ~4.6:1 (was 3:1).
Source: GS via Tickmill, Jul 28 & Jul 31 notes · Estimates; for illustration · 1W-flat US>global row as published04

Flows · Reaction Function

The Bad-News Test Passed: Maximum Dosage, Fully Absorbed

  • The dosage: oil at $105 intraday, the CTA pivot broken, a hawkish triple dissent — modeled 1-week down-tape supply ~$31.5bn (Jul 28 basis).
  • Realized: SPX net +0.43% over two weeks; the Jul 29 trough (−1.9%) was fully recovered in two sessions; a record close on Monday.
  • SOX jumped +8.19% on Jul 30 — the largest single day since April 2025; Micron +18.4% the same day.
  • The counter-flows: a record $12bn into semiconductor ETFs in the final July week (25% of all ETF inflows); buyback windows reopened post-earnings; systematic length already de-risked.
  • Residual strongly positive — the mirror image of last issue's "no bid" reading. Caveat: retail as the marginal absorber is itself a crowding signal (see harvest-review watch).
  • VIX: 20.66 on FOMC day → 15.99 by Friday. Gamma and Prime Book: no fresh reads this issue (data gap).
Source: FRED (Jul 31 / Aug 3); Fortune / Benzinga (Aug 2, relayed)05

Liquidity

The August Drain Arrived Two Weeks Early — and the First Wave Is Already Absorbed

GaugeLatestRead
SOFR − IORB0bpWas −3bp; converging, not stressed
ON RRP$2.1bnBuffer still effectively zero
Reserves (H.4.1)$2.98T−$158bn in two weeks — the TGA rebuild hit reserves directly
TGA$911bn wk · $998bn daily+$155bn rebuild; sitting on Treasury's "$1T late-July peak" guidance
  • QRA (Aug 3): Q3 borrowing $739bn (+$68bn vs May); Q4 debut estimate $628bn. Cash targets: $950bn end-Sep, $850bn end-Dec — the TGA itself turns from a drain into a mild release; remaining pressure is net issuance with a zero RRP buffer.
  • Coupon sizes unchanged through FY2026 (TBAC); the gap runs through bills. Refunding statement Aug 5.
  • Calendar: Aug 5 refunding + ISM services · Aug 7 July payrolls · Aug 12 CPI · Aug 13 PPI · Aug 11 CoreWeave earnings.
Source: FRED (Jul 29-Aug 3); Treasury sb0584 (Aug 3); Treasury Fiscal Data API (Jul 30)06

Rates & Fed

Three Hawkish Dissents — and a Market That Now Prices a September Hike as the Coin-Flip Favorite

9 – 3
FOMC hold at 3.50-3.75%; Hammack, Kashkari, Logan preferred +25bp
54.5%
Polymarket: September +25bp; no-change 42.5%; cut ~1%
4.28%
2Y yield after a 4.16 → 4.37 → 4.22 → 4.28 round trip
  • Statement cites Middle-East-driven uncertainty and supply shocks lifting energy prices; Chair Warsh keeps statements short and offers no forward guidance, by stated philosophy.
  • What pushed pricing hawkish: ECI +0.9% q/q (cons. 0.8; private wages accelerating 0.7 → 0.9), initial claims 187k — the lowest since 1969 — and ISM manufacturing 55.6 (Aug 3), the highest since May 2022, employment sub-index expanding for the first time in ~3 years.
  • What the price data itself says: June headline CPI −0.4% m/m (largest drop since Apr 2020), core CPI 2.6% y/y and ~flat m/m; core PCE +0.1% m/m (below consensus), 3.3% y/y. Activity and wages are hot; measured prices are not.
  • Q2 GDP advance +1.5% (cons. 2.1) — but private domestic final sales +3.9% and equipment investment +15.2%; the miss is an import surge (−1.01pp).
Source: Federal Reserve (Jul 29); Polymarket gamma-api (Aug 4); BLS/BEA (Jul 14-31); ISM (Aug 3); FRED (Jul 31)07

Geopolitics & Oil

Brent Ran 88 → 105 → 84: a Full Escalation Cycle That Stopped One Step Short of the Big Strike

  • Escalation peak Jul 20-24: 13 consecutive nights of US strikes, first B-1 use, Houthi attacks opening a Red Sea second front; Brent settled above $100 on Jul 23 (FRED spot 105.32).
  • De-escalation Jul 25-27: US halted nightly strikes — Patriot interceptor inventory cited — and a five-day battle pause began; Brent fell 9-11% on Jul 27.
  • Pause broken Jul 29 (IRGC missiles at Jordan; a drone hit an LNG facility at Damietta, Egypt — a third chokepoint at risk); Trump ordered large-scale strike prep Jul 31, then canceled on Aug 2 after Qatari mediation and Saudi pressure. Brent −4.4% to ~$84 on Aug 3.
  • Hormuz transits collapsed from ~88/day to a 7-day average of ~11; war-risk premia jumped from 1-3% of hull value to 7.5-10%; underwriters withdrew.
  • Supply cushion is thin: SPR at 307.7M bbl (lowest since 1983) after the IEA's largest-ever 400M bbl release; Vitol: effectively all global spare capacity sits inside Hormuz. OPEC+ approved its final +188k b/d quota step (Sep).
  • Iran's negotiating demand — Hormuz control written into any deal — remains the unresolved core. No formal ceasefire as of the cutoff.
Source: FRED DCOILBRENTEU (Jul 27); TradingEconomics (Aug 3); war timeline via Wikipedia/UKMTO/Hormuz monitors (relayed, as of Aug 3)08

AI Trade · The Lifeline

The Lifeline Accelerated Through the Panic: Anthropic ~$69bn Run-Rate, All Four Hyperscalers Raised Capex

Company2026 capex guideChange & validationStock
Alphabet$195-205bnRaised from 180-190; "significant increase" flagged for 2027; FCF negative for the first time−7.1%
Microsoft~$175bnOptical cut = depreciation reclass only; FY27 guided $255-260bn; Azure +43%, AI run-rate $37bn+7%
Meta$130-145bnFloor raised; FCF −91% to $0.8bn, buybacks zero, no 2027 number−10%
Amazon~$220bnRaised from ~200, increment attributed to memory prices; AWS +36.7%, fastest in 18 quarters+15.3%
  • Anthropic run-rate (Yipit relay): $14B (Feb) → $30B (Apr) → $47B (May) → ~$69B (end-June); daily ARR adds accelerating $0.4B → $0.55B. Confidential S-1 filed Jun 1; $965B valuation; Nasdaq pricing possibly Oct-Nov. SemiAnalysis: Q3 EBITDA ~$1B; gross margin −94% (2024) → ~60%.
  • OpenAI: data gap — freshest primary print remains ">$25B" (late Feb, The Information). No acceleration verdict either way.
  • The new split: the market rewards revenue-validated capex and punishes the rest — the same week, same guidance direction, opposite stock reactions.
Source: company reports via CNBC (Jul 22-30); Yipit via Jukan (relayed); SemiAnalysis via Dealroom · unaudited run-rates, for illustration09

AI Trade · The Ghost Story, Settled

Kimi K3's Parity Claim Was Refuted — and the Semi Carnage Was in Memory, Not Compute

  • Third-party verdict (Artificial Analysis Index): K3 scores 57, ranked #4 — behind Claude Opus 5 (61), Claude Fable 5 (60), GPT-5.6 Sol (59); level with the prior flagship generation. Largest gap: agentic tasks. One genuine #1: frontend-code blind arena.
  • Moonshot itself concedes K3 "still trails" the top proprietary flagships.
  • Jevons paradox, live: K3 demand hit GPU capacity limits and Moonshot paused new subscriptions — the cheap model amplified compute demand. TSMC and ASML raised guidance the same week the panic peaked.
  • July SOX: −21%, worst month since Oct 2008; $2.2T of value erased; then +8.3% over the final two sessions.
  • The dispersion tells the real story: Intel −41%, Astera −48%, Marvell −45%, KLA −44%, Micron −36% — vs Nvidia −5%, Broadcom −2%. The selling repriced the memory/equipment inflation chain, not AI compute. Nvidia reclaimed the #1 market cap ($4.86T) from Apple, whose guidance cited DRAM/NAND inflation.
  • TrendForce (Jul 30): DRAM and NAND diverge in 2027 — DRAM stays short (HBM crowd-out), NAND loosens in 2H27.
Source: Artificial Analysis (late Jul); Fortune/Benzinga (Jul 31-Aug 2); TrendForce (Jul 30)10

Credit

The Divergence Flipped: Equities at Records While AI Credit Deteriorates

284bp
HY OAS, +13bp in two weeks; widening attributed to Alphabet's capex raise
203bp
Oracle 5Y CDS record (144bp at NY start); S&P downgrade to BBB−, the last IG notch
<2×
Hyperscaler bond cover ratios, down from ~5× in February
  • Issuance forecasts re-based: Morgan Stanley now sees $570bn of AI-linked 2026 issuance (last issue's baseline: $400bn); AI datacenter deals ≈45% of net IG supply.
  • Demand is the constraint: Amazon's $25bn 8-tranche needed +18-21bp concessions and drew a 1.6× book; its 30-year curve widened 20bp on announcement; the company then pledged no further 2026 issuance.
  • Oracle: RPO $638bn (+363%) with reportedly ~$300bn from OpenAI alone; FY26 capex $55.7bn vs $21.2bn prior year.
  • CoreWeave: quarterly interest expense doubled to $536M; a $2.6bn loan repriced +125bp to ~SOFR+550 to clear. Earnings Aug 11.
  • The framework note: the shield's stated failure condition — "FCF turns negative, so the build must be debt-funded" — is now literally in effect at two of five hyperscalers.
Source: FRED (Jul 30); Seeking Alpha / Fortune / CNBC / Yahoo (late Jul, relayed); Man Group H2 outlook11

Falsifiable Checkpoints

What Would Change Our Mind: Nine Triggers Into the Next Two Weeks

#TriggerVerifiable conditionDue
1Hike-pricing collapseJul core CPI m/m ≤0.2% and 2Y −10bp+ on the dayAug 12
2Metric-distortion playbookJul core CPI m/m ≥0.4%Aug 12
3Labor-softening revivalPayrolls <50k or unemployment ≥4.4%Aug 7
4Credit confirmation (rolled)HY OAS >300bp, or ORCL 5Y CDS >250bprolling 4w
5CoreWeave verdict2026 capex guide-down or liquidity deterioration at earningsAug 11
6Liquidity escalation lineAny weekly reserves print below $2.9Trolling
7Forced harvest reviewNew SPX high plus another $10bn+ retail week into semi ETFsrolling
8Geopolitical re-shockBrent closes above 100, or Hormuz transits <10/day againrolling
9Lifeline datapoint (rolled)Anthropic public S-1, or a primary-source OpenAI revenue printrolling
Scored publicly in the next issue — misses become framework revisions12

Interpretation — Clearly Separated From the Fact Pages

OUR VIEW

Tactical · into Aug 12

Defense-to-neutral, but no chasing. Last issue's flip-to-constructive conditions both triggered — yet the upgrade lands into three verdict days (refunding, payrolls, CPI) with a hike priced >50% and up-tape CTA demand only a quarter of down-tape supply.

Shift constructive if: Jul core CPI ≤0.2% m/m with oil staying on its round-trip lows.

Tactical · 2-4 weeks

Rotate the AI thesis internally: own the compute core (the −5% layer), avoid chasing the memory/equipment inflation chain (the −36~48% layer) whose price surge is now taxing its own customers.

Void if: Q4 DRAM contract-price guidance turns negative, or Nvidia misses the buy-side whisper.

Tactical-to-strategic · into Aug 12

Last issue's long-2Y call is wounded, not dead: the wage half of its kill condition (ECI acceleration) triggered. Stand flat; re-enter only on a soft CPI, or run the metric-distortion decomposition first on a hot one.

Branch: core CPI ≤0.2% → re-long the 2Y against the 54.5% hike pricing; ≥0.4% → playbook 07 before following.

Strategic · Aug-Q4

The liquidity caution upgraded from forecast to confirmed: reserves −$158bn in two weeks. But the steepest leg — the TGA rebuild — is complete at the $1T peak; what remains is issuance-paced attrition. Watch line: reserves below $2.9T promotes liquidity from backdrop to main pricing variable.

Void if: reserves are back above $3.0T at quarter-end.

No price targets, by discipline. Known limits this issue: the credit call leans on one name (Oracle); Polymarket's hike pricing is unverified against CME FedWatch; no fresh gamma or Prime Book reads.

Views are the author's; facts on preceding pages stand independently13

Appendix

Data Snapshot & Disclaimer

FRED latest: SOFR 3.65 (7/30) · IORB 3.65 (8/3) · ON RRP $2.1bn (8/3) · Reserves $2.985T (wk 7/29) · TGA $911bn (wk 7/29; daily $998bn 7/30) · VIX 15.99 (7/31) · HY OAS 284bp · IG OAS 80bp (7/30) · 2Y 4.28% · 10Y 4.75% (7/31) · 5Y/10Y breakevens 2.26%/2.28% (7/31) · Brent spot 91.82 (7/27; futures ~$84 Aug 3) · Unemployment 4.2% (Jun) · Core CPI 2.6% y/y, ~0.0% m/m (Jun) · Core PCE 3.3% y/y, +0.1% m/m (Jun).

Markets: SPX 7,489.72 (Jul 31), +0.43% over two weeks, trough 7,316.15 (Jul 29), Monday close 7,600.50 (Aug 3) · Nasdaq 25,373.85 (Jul 31), −0.57% over two weeks · SOX −21% in July (worst since Oct 2008), +8.19% on Jul 30 · SMH 540.29 (Jul 31).

Relayed exhibits: GS CTA matrix and pivots via Tickmill (Jul 28, Jul 31); Polymarket September odds via gamma-api (Aug 4); QRA borrowing estimates via Treasury sb0584 (Aug 3); ISM (Aug 3); war timeline via public monitors. Relayed desk data are estimates, for illustration, and not independently verifiable.

Data tiers: T1 official · T2 own calculations/API pulls · T3 relayed (source named) · T4 first-hand sensors. Methodology distilled from @shanghaojin E1-E10 (Mar-Jun 2026); production: macro_weekly pipeline, issue #2 (agent-assisted run). Views pending owner review.

Disclaimer: Personal research. Not investment advice. No solicitation. Figures as of the dates shown and subject to revision.

Macro Weekly | Market Overview | Aug 3, 202614