Closing Briefing — Tuesday, August 18, 2026

1. The rout everyone blamed was over by lunchtime, and the selling continued regardless. This is the single most important fact of the session and almost no wire copy carries it. The 30-year printed a fresh 19-year high near 5.32% in the pre-open, the 20-year touched 5.316% and the 10-year 4.736% (CNBC, 17 hours before the close). By the official 3:00 p.m. Treasury par close the 30-year was 5.28%, down 3 bp; the 10-year 4.71%, down 1 bp; the 2-year unchanged at 4.19%. Bloomberg's 4:59 p.m. board put the U.S. 10-year at 4.70%, −2 bp — while Germany's rose 4 bp, France's 5, Italy's 6 and South Korea's 7. The United States was the only major market whose long end rallied, and it rallied on the day its equity market fell hardest. The mechanism is worth naming: a duration-driven equity selloff should pressure long yields up, and a long-end rally into an equity decline is a flight-to-quality signature, not a term-premium signature. Forward catalyst: Wednesday's July FOMC minutes at 14:00, and Thursday's claims.

1. The AI trade repriced on a footnote, not on a forecast. The Wall Street Journal's analysis of nine large technology companies put about $3 trillion of off-balance-sheet commitments, mostly AI-related, against roughly $600bn of reported capex over the last reported year — obligations growing faster than capex and about triple the companies' outstanding leases plus long-term borrowings. Landing alongside Anthropic's $65bn annualised run rate and Reuters' report of 2028 guidance at $190–200bn, both short of privately circulated figures, it hit the suppliers hardest because the suppliers are the ones whose revenue those commitments become. The iShares Semiconductor ETF fell about 5.4% while the iShares Expanded Tech-Software ETF rose about 0.7%. Forward catalyst: Nvidia reports 26 August after the close; Marvell 27 August.

1. Precious metals would not act like havens. Gold settled −1.88% at $4,389.50 and silver −4.24% at $63.420, with gold closing within $5 of its low after opening $84 higher on the session's prior settle. Copper fell 2.43%. That happened on a day when equities fell, the long bond rallied and the dollar was flat — the exact configuration in which a haven bid should appear. It did not, and the reason is positioning: silver is still +69.5% year on year and gold's year-to-date gain has compressed to +0.49% on the spot series after a 30.9% twelve-month run. When the metal trades as a momentum asset, it de-risks with the momentum book rather than hedging it. Forward catalyst: the 26 August PCE deflator; the real-rate path after Wednesday's minutes.

1. Healthcare set all-time highs inside a 1.33% Nasdaq decline. The $44bn Health Care Select Sector SPDR, the $19bn Vanguard Health Care Index Fund, the iShares Biotechnology ETF and the Nasdaq Biotechnology index all reached records (CNBC). Three-month total returns: the healthcare SPDR 15.6%, Vanguard 16.6%, iShares Biotech 21.1%, Nasdaq Biotech 18.5%, against 4.8% for the S&P 500. Finviz put the sector +1.40% on the day, the best of eleven groups, and Johnson & Johnson +3.33% to $271.11 posted its first record close since 7 July, now roughly 31% higher year to date, more than double the index. Forward catalyst: the sector's defensive bid is a rate call as much as a fundamental one — watch whether it survives a hawkish minutes text.

1. Home Depot delivered the quarter and the stock did nothing, which is the read on the whole consumer block. Earnings and revenue both beat, same-store sales growth was the strongest in almost four years, and $730m of tariff refunds landed in the quarter — and management left full-year guidance unchanged for a second consecutive quarter, a forecast first given at an investor conference nine months ago. CFO Richard McPhail called conditions a "frozen housing market." The stock traded to $344.54 and closed $337.49, −0.12% — a 2.05% fade from high to close. Since the December guide the company has beaten EPS by 21 cents and revenue by nearly $1bn across the first half, and the guide has not moved. Forward catalyst: Lowe's, Target, TJX and Analog Devices Wednesday before the open; Walmart, Deere and Ross Stores Thursday (§5).

1. Funding is tightening under the surface while the coupon curve rallies. Secured overnight financing printed 3.66% for 17 August against an interest-on-reserves rate of 3.65% — SOFR now trades above the administered rate rather than through it, a four-basis-point move from 3.62% four sessions earlier, with the 75th percentile at 3.71% and the 99th at 3.74%. Tri-party and broad general collateral both cleared at 3.64%, up 4 bp. And the overnight reverse repo facility took in $0.155bn on 18 August, a fresh record low, from $0.255bn Monday. Reserve balances stand at $2,944.1bn, $198.6bn below the 15 July peak. Forward catalyst: the Fed's own Reserve Demand Elasticity release, Thursday 10:00 (§7), now the most informative scheduled item on the funding question.

1. Two Chinese-listed reads pulled in opposite directions on the same tape. Baidu fell 12.73% to $90.87 after second-quarter revenue of RMB31.3bn fell 4% year on year, online marketing services dropped 19% to RMB13.1bn and net income collapsed to RMB2.319bn from RMB7.322bn — while AI cloud infrastructure revenue rose 50% on GPU demand. Alibaba rose 2.78% to $128.18 into its own 20 August print. The split says the market will pay for AI infrastructure revenue and will not pay for the advertising business funding it — the same buyer-versus-seller logic driving the U.S. semiconductor tape, expressed inside a single P&L. Forward catalyst: Alibaba reports 20 August before the open.

1. The Dow's 0.22% decline was a composition accident and should not be read as resilience. Coca-Cola +2.13%, Chevron +1.50%, Apple +1.45%, Amgen +1.42%, JPMorgan +0.66%, McDonald's +0.55% and Microsoft +0.31% carried it; Caterpillar −4.58% and Nvidia −2.34% were the drags. The index has three semiconductor-adjacent members and thirty defensive-leaning ones, so a session that hits AI hardware and bids healthcare and staples mechanically flatters it. The Russell 2000 fell 1.24% on the same day, which is the better breadth read. Forward catalyst: any broadening of the AI-funding repricing into industrials with data-centre exposure — Caterpillar, GE Vernova and FTAI Aviation all fell 4–7% on Tuesday.

Three green, eight red, and a dispersion of 3.93 points from Healthcare +1.40% to Technology −2.53% — more than double Friday's 1.67 and the widest since 12 August. The rotation is unambiguous and it is defensive-plus-energy against everything with an AI multiple: healthcare, consumer defensive and energy up; technology, industrials and basic materials down.

Technology −2.53% is not a technology story, it is a hardware story, and the internal split is the whole point. Down: SanDisk −9.01% to $1,625.78, Seagate −9.16% to $903.68, SK Hynix −9.19% to $155.63, Teradyne −8.77% to $404.29, Marvell −7.81% to $216.03, Western Digital −7.43% to $496.16, Micron −6.94% to $941.51, Intel −6.58% to $96.68, AMD −4.21% to $484.68, Taiwan Semiconductor −4.05% to $413.50, Broadcom −3.21% to $380.00, Nvidia −2.34% to $219.74, with Coherent −12.75% to $306.43, Credo −13.03% to $245.98 and MACOM −10.61% to $292.77 at the extreme. Up, in the same sector, on the same day: Intuit +4.41%, GoDaddy +5.26%, Tyler Technologies +4.99%, Adobe +3.58%, ServiceNow +1.52%, with Monday.com about +7% and HubSpot about +6% (24/7 Wall St., intraday). The iShares Semiconductor ETF fell about 5.4% and the iShares Expanded Tech-Software ETF rose about 0.7% — a six-point single-day spread inside one GICS sector. Software is short AI capex; hardware is long it; the WSJ footnote analysis is a capex story; the market traded it correctly.

Industrials −1.81% is the contagion channel to watch. Caterpillar −4.58% to $841.23 was the Dow's worst member, with GE Vernova down about 6% and FTAI Aviation about 7% on 24/7 Wall St.'s behind-the-meter energy screen. These are not chip stocks. They are the names that sell turbines, engines and gearsets into the same data-centre commitments the WSJ was counting, and they repriced on the same footnote. If the AI-funding question is going to broaden beyond semiconductors, this is the group it broadens into first.

Energy +1.09% is the fourth consecutive advance in crude expressed in equities, and the composition inverted back to refiners: Marathon Petroleum +2.24% to $366.21, Phillips 66 +1.76% to $243.49, Valero +0.82% to $350.05, Chevron +1.50% to $205.75, Targa Resources +7.13% to about $297.77. Refiners over producers on a day the barrel rose only 0.81% is a crack-spread bid, not a crude bid. The exception is loud: Coterra Energy fell 8.62% to $32.56 on 73.3m shares, a volume figure roughly thirty times the next-largest energy name's, and public reporting ties it to an SEC deregistration and the Devon transaction rather than to the commodity — it is excluded from the sector read for that reason.

Reconciliation, stated. The last published sector table in this report is 14 August, so the YTD bridge spans two sessions rather than one. Compounding each group's 14 August YTD by its 18 August one-day move and solving for the implied 17 August move gives: Technology +0.02%, Energy +0.86%, Healthcare +0.06%, Financial −0.71%, Consumer Cyclical −0.91%, Communication Services −1.42%, Industrials +0.44%, Utilities −0.24%, Real Estate −0.69%, Basic Materials +0.64%, Consumer Defensive −1.49%. Every implied Monday move sits inside ±1.5%, consistent with a −0.52% index session, and the largest — Consumer Defensive −1.49% — reconciles to Constellation Brands' 6.2% decline on the Berkshire exit. The series is internally consistent; no group is drifting. Source: Finviz Groups, Performance table view (?g=sector&v=140&o=name). Finviz buckets are not GICS: Alphabet, Meta and Netflix sit in Communication Services, Amazon and Home Depot in Consumer Cyclical, and SanDisk, Seagate and Western Digital in Technology rather than in a hardware sub-index.

Up

Johnson & Johnson (JNJ) +3.33% to $271.11 — a record close, its first since 7 July's $267.24, and roughly +31% year to date, more than double the S&P 500's advance (CNBC). Targa Resources (TRGP) +7.13% to about $297.77 led the S&P 500 gainers. Insulet (PODD) +5.96% to $148.03, GoDaddy (GDDY) +5.26% to $97.23, Tyler Technologies (TYL) +4.99% to $336.49, Ulta Beauty (ULTA) +4.75% to $516.74, Intuit (INTU) +4.41% to $350.41, Monster Beverage (MNST) +4.09% to $47.38, Adobe (ADBE) +3.58% to $263.14.

UGI (UGI) +12% after the Wall Street Journal reported, citing people familiar, that KKR made a $9bn bid for the Pennsylvania utility, which serves roughly 700,000 customers and was up 5% on the year going in. Trading was briefly halted for volatility. KKR declined to comment. This is the session's only clean M&A print and it landed in the one sector — utilities — that closed lower anyway, at −0.52%.

Also higher: Alibaba (BABA) +2.78% to $128.18 into Thursday's print, Nike (NKE) +2.48% to $40.06, Netflix (NFLX) +2.30% to $77.77, Coca-Cola (KO) +2.13% to $88.84, Chevron (CVX) +1.50% to $205.75, Apple (AAPL) +1.45% to $310.03, Amgen (AMGN) +1.42% to $425.32, ServiceNow (NOW) +1.52% to $119.49, Walmart (WMT) +0.77% to $115.22 into Thursday, and Microsoft (MSFT) +0.31% to $481.82 — Monday's single largest index drag, and Tuesday's non-participant in the selling.

Down

Coherent (COHR) −12.75% to $306.43, Credo Technology (CRDO) −13.03% to $245.98 (non-S&P 500), MACOM Technology (MTSI) −10.61% to $292.77 (non-S&P 500), Seagate (STX) −9.16% to $903.68, SanDisk (SNDK) −9.01% to $1,625.78, Ciena (CIEN) −8.90% to $405.54, Teradyne (TER) −8.77% to $404.29, Marvell (MRVL) −7.81% to $216.03, Western Digital (WDC) −7.43% to $496.16, Micron (MU) −6.94% to $941.51, Intel (INTC) −6.58% to $96.68 on 100.0m shares — back below $100 — AMD −4.21% to $484.68, Taiwan Semiconductor (TSM) −4.05% to $413.50, Broadcom (AVGO) −3.21% to $380.00, Nvidia (NVDA) −2.34% to $219.74 on 82.5m shares. SK Hynix −9.19% to $155.63 (non-S&P 500). Coherent, Teradyne and Ciena were the S&P 500's three biggest decliners (24/7 Wall St.).

The valuation context matters because there was no company-specific news in any of them. Coherent entered Tuesday +276% over one year, +90% year to date, on a 79× trailing multiple; Teradyne +306% over one year, +129% year to date, 57× trailing; Ciena +390% over one year, +90% year to date, 143× trailing. SanDisk entered the session +653% year to date and +44% on the week, Micron +255% year to date and +18% on the week, Western Digital +211% year to date and +22% on the week. On a rotation day the most expensive names bleed first, and these were the most expensive names.

Baidu (BIDU) −12.73% to $90.87 on a second-quarter miss: revenue RMB31.3bn, −4% year on year; online marketing services −19% to RMB13.1bn; net income RMB2.319bn against RMB7.322bn a year earlier; AI cloud infrastructure revenue +50% on GPU demand (non-S&P 500). Meta Platforms (META) −4.45% to $543.67, a second session lower as the Oakland federal trial opened in which 29 state attorneys general allege Facebook and Instagram were designed to addict children. Caterpillar (CAT) −4.58% to $841.23. Corning (GLW) −7.68% to $159.90, Jabil (JBL) −8.45% to $338.62, Oracle (ORCL) −2.63% to $142.79, Strategy (MSTR) −5.25% to $92.55.

Non-S&P 500 extremes: Klarna −22.81% to $15.06, Applied Optoelectronics (AAOI) −15.16% to $131.41 — reversing Monday's +15.53% almost exactly — AXT −14.23% to $82.31, and Fabrinet down more than 9% after a fiscal Q4 beat and a fiscal Q1 2027 revenue guide of $1.375–1.425bn that the market read as a margin-seasonality warning; the read-through dragged Marvell and Amphenol.

Carvana (CVNA) fell more than 3% to $67.81 at 12:32 p.m. ET after SEC filings delivered Monday showed director Danforth Quayle disposed of 14,525 shares on Friday and director Ira Platt 30,000 shares on Thursday.

Analyst actions

D.A. Davidson's Wyatt Swanson upgraded Duolingo (DUOL) to Buy from Neutral, price target to $160 from $130 — 23% upside to Monday's close. "Underlying product work, marketing changes, and the continued efforts at refining the core monetization engine are underappreciated by investors," Swanson wrote. Citi's Quant team upgraded Bath & Body Works (BBWI) to Buy from Neutral, target unchanged at $25, citing Placer store-traffic data improving to about +1.1% in Q2 from +0.2% in Q1 and flagging that "2Q is unlikely to showcase any major turnaround" — the call is on the second-half plan, not the quarter. Daiwa upgraded Tapestry (TPR) to Outperform from Neutral, target $140. JPMorgan upgraded América Móvil (AMX) to Overweight, target $32 from $30. KeyBanc upgraded Acadia Realty (AKR) to Overweight, target $25. Downgrades: Raymond James cut Abercrombie & Fitch (ANF) to Market Perform, no target published; Baird cut Ameresco (AMRC) to Neutral, target $32 from $36; JPMorgan cut Millicom (TIGO) to Neutral, target up to $105 from $100. Initiations: Barclays started Centrus Energy (LEU) at Equal Weight, $207; Compass Point started Gorilla Technology (GRRR) at Buy, $44.

Times are ET where a clock time was verified. Sourcing change, disclosed: the Earnings Whispers day pages gate their calendar behind a cookie-and-usage-agreement consent banner that this unattended session did not accept, so the rosters below are captured from the Nasdaq earnings calendar API for each day and screened, name by name, against a rendered S&P 500 constituent list. Nasdaq publishes a before-open / after-close bucket rather than a clock time, so clock times shown in brackets are carried from the 14 August Earnings Whispers pull with that provenance stated. Confirm every time against company IR before trading a date.

Mon 8/17 — completed. No S&P 500 reporter on either bucket.

Tue 8/18 — completed. BMO: Home Depot (HD) [6:00] — beat on earnings and revenue, strongest same-store sales growth in almost four years, $730m of tariff refunds, full-year guidance left unchanged for a second straight quarter; closed −0.12% at $337.49 after trading to $344.54. AMC: Keysight Technologies (KEYS) [4:05] — shares fell about 7% into the print; Jack Henry & Associates (JKHY) [4:15].

Wed 8/19. BMO: Lowe's (LOW) [6:00], Estée Lauder (EL) [6:00], Target (TGT) [6:30], Analog Devices (ADI) [7:00], TJX Companies (TJX) [7:30]. AMC: Nordson (NDSN) [4:30].

Thu 8/20. BMO: Deere & Company (DE) [6:20], Walmart (WMT) [7:00]. AMC: Ross Stores (ROST) [4:00].

Fri 8/21. No S&P 500 reporter on either bucket.

Mon 8/24. No S&P 500 reporter on either bucket.

Tue 8/25. AMC: Intuit (INTU).

Wed 8/26. BMO: J.M. Smucker (SJM). AMC: Nvidia (NVDA), Salesforce (CRM), CrowdStrike (CRWD), Synopsys (SNPS), Agilent Technologies (A), Veeva Systems (VEEV), HP Inc. (HPQ), Williams-Sonoma (WSM).

Thu 8/27. BMO: Dollar General (DG), Dollar Tree (DLTR), Best Buy (BBY), Hormel Foods (HRL). AMC: Marvell Technology (MRVL), Autodesk (ADSK), Workday (WDAY), Ulta Beauty (ULTA).

Fri 8/28. No S&P 500 reporter on either bucket.

Changes vs. the prior calendar (8/17 report):

• Two names resolved, both additions to 8/18 AMC that this report had conservatively excluded: Toll Brothers (TOL) and Mercury Systems (MRCY) both appear on the Nasdaq 8/18 after-close list, ending the "absent from one capture" ambiguity carried for three editions. Neither survives the constituent screen — the rendered S&P 500 list does not carry either — so both are recorded here as confirmed reporters that are not index members, and are excluded from the roster above.
• The 24–28 August week is published for the first time. The prior edition stated it could not be captured; the full week is now screened and listed. Nvidia on 26 August after the close is the single largest scheduled event in this report's forward window.
• No removals and no additions to 8/19–8/21. Every name in the prior edition's roster for those three days reappears on an independent source, including Estée Lauder (EL), and nothing new joins them.
• Membership caveat. The constituent screen used this session does not carry Heico (HEI), Zoom (ZM), Dick's Sporting Goods (DKS), PVH, Coty (COTY), Bath & Body Works (BBWI), Kohl's (KSS), Abercrombie & Fitch (ANF), Urban Outfitters (URBN), Nutanix (NTNX), Okta (OKTA), Five Below (FIVE), Burlington (BURL), Gap (GAP), Affirm (AFRM), Elastic (ESTC), SentinelOne (S), HealthEquity (HQY), BILL Holdings (BILL) or Advance Auto Parts (AAP), each of which reports in the window covered. All are conservatively excluded and listed in Data Notes. The screen does now carry Coherent (COHR), SanDisk (SNDK) and Reddit (RDDT), resolving three caveats carried for eight editions.
• Index membership change, not an earnings item: Reddit (RDDT) joined the S&P 500 on 18 August, replacing AvalonBay Communities.
• Timing bucket unpublished: none. Every name above carries at least a before-open or after-close bucket from the reviewed calendar.
• What the forward calendar hands the desk. The current week's remaining risk is five consumer names and one analog semiconductor across 36 hours — Lowe's, Target, TJX, Estée Lauder and Analog Devices Wednesday morning, Walmart and Ross Stores Thursday, with Deere the only industrial. The reaction function is now measurable rather than assumed: Home Depot beat on both lines with the best comparable sales in four years and closed lower. Next week is the mirror image — nine S&P 500 names report on Wednesday 26 August alone, eight of them after the close, and Nvidia is one of them.

Name the shape: a long-end-led bull flattener, and the diagnostic is a failed breakout rather than a policy repricing. The 30-year fell 3 bp, the 20-year 2, the 10-year 1 and the 2-year not at all, which flattens 2s30s by 3 bp with the front end pinned. That combination has one clean interpretation and it is not about the Fed: the front end did not move because the September question did not change, so the entire day's action was in the compensation demanded for holding duration — and that compensation fell on a session that began with the 30-year at a 19-year high near 5.32%. A market that makes a multi-decade high in the morning and closes the long end 3 bp lower has rejected the level. It is the single most under-reported fact of the session.

The international cross-check is what makes it a domestic signal. Bloomberg's 4:59 p.m. board had the U.S. 10-year at 4.70%, −2 bp, against Germany +4, France +5, Italy +6, Spain +5, Greece +6, Japan +2, Australia +5 and South Korea +7. Only Canada joined the U.S. at −2. Every one of those long ends has its own fiscal story — Germany's 30-year at a post-2011 high, France's at a post-2008 high — and none of them rallied. The United States rallied because its equity market was down 1.33% on the Nasdaq and money went somewhere. That is a flight-to-quality bid inside a global bear market in duration, and it means the two moves should be modelled separately rather than as one "global bond rout."

The bill curve is the counterweight and it belongs with the funding data. The 3-month eased another basis point to 3.86% and is now −3 bp on the week, the 6-month −5 bp and the 1-year −4 bp — the bills led the whole curve lower over five sessions. That happened while overnight secured financing rose 4 bp to 3.66% and the reverse repo facility took a record-low $0.155bn (§9b). Bills richening while overnight repo cheapens is a straightforward collateral-scarcity signature: the money that used to sit at the facility is buying the front of the curve, and there is nothing left at the facility to buffer a settlement date. That is a financing signal, not a rates signal, and it is the thing to carry into September quarter-end.

Current week (Aug 17–21) — released

Current week (Aug 17–21) — remaining

Next week (Aug 24–28)

Current target range: 3.50–3.75% (IORB 3.65%; EFFR 3.63%, NY Fed, 17 August). The market prices hikes, not cuts, through the whole of 2026: the probability of a target range below 3.50–3.75% is 0.0% at every 2026 meeting on both vendors.

Provenance of every column, stated before any number is used. CME FedWatch published a complete numeric four-column table this session — the first time in several editions — so the NOW / 1 DAY / 1 WEEK / 1 MONTH headline below is read directly off CME's Current view with its own legend dates (17 Aug 2026, 11 Aug 2026, 17 Jul 2026), not chart-read and not carried from a prior edition. The card is stamped "Data as of 18 Aug 2026 05:03:36 CT"; contract ZQU6, expiring 30 September, mid price 96.3275, prior volume 27,222, prior open interest 243,136. A live CME read taken after the 5:00 p.m. ET hour is indicative rather than a settlement snapshot, and it is corrected in the next edition if it fails to match. The Investing.com matrix is stamped "Updated: Aug 18, 2026 05:45 PM EDT" and supplies its own current / previous-day / previous-week triplets for all eleven priced meetings. No cell in this section is estimated, chart-read or interpolated.

The headline — CME FedWatch, 16 September 2026 meeting

The cross-check — Investing.com Fed Rate Monitor, 16 September 2026

September contract price: CME mid 96.3275; Investing.com 96.330.

The gap, reconciled and quantified — and the sign disagreement, which matters more than the gap. On levels the two vendors are 0.3 points apart (CME hold 65.0%, Investing 64.7%), among the tightest readings this report has recorded. Work the arithmetic. The 16 September meeting sits on day 16 of a 30-day contract month, so a certain 25 bp hike lifts September's average effective rate by 25 × (14/30) = 11.7 bp, making one basis point of ZQU6 worth roughly 8.6 points of headline probability. CME quotes the contract 0.25 bp cheaper (96.3275 versus 96.330), which on its own argues CME should print about 2.1 points more hike probability. It prints 0.3 points less. The offset is the assumed base rate: CME anchors on the prevailing EFFR of 3.63%, half a basis point above the 3.625% range midpoint, and half a basis point of base absorbs roughly 4.3 points of probability in the opposite direction. Net the two conventions and CME should sit about two points below Investing; it sits 0.3 below. The unexplained residual is under two points of headline probability, about 0.29 bp on ZQU6 — inside a single half-tick. The vendors are not disagreeing about the Fed.

They are disagreeing about the direction of the day, and that is the finding. CME's own 17 August column has the hike at 36.1%, so CME reads Tuesday as a 1.1-point move toward a hold. Investing.com's previous-day snapshot has it at 34.0%, so Investing reads Tuesday as a 1.3-point move toward a hike. Both cannot be right, and the cause is the baseline rather than the current read: the two vendors take their prior-day snapshot at different clock times on a day when the long end round-tripped 4 bp. This report treats CME's series as primary because it is a single exchange source with a self-consistent legend, and it flags the Investing.com sign as the alternative. On CME's numbers, Tuesday nudged the hike out.

Multi-day momentum, which is unambiguous either way. The September hold has run 42.2% a month ago → 51.6% a week ago → 63.9% Monday → 65.0% Tuesday on CME's own four columns — a 22.8-point march in four weeks with no reversal, and the +50 bp September tail has gone from 6.6% on 17 July to 0.0%. The week-over-week arc is the same story at every 2026 meeting on the Investing.com matrix: the December cumulative-hike probability has fallen from 78.5% a week ago to 67.5%, and the December hold has risen from 21.6% to 32.6%. What Tuesday added was not direction but a change in the reason: for a fortnight the hike came out of the strip because demand data missed, and on Tuesday it came out on a day when the barrel rose and the long end rallied anyway. Bloomberg reported the same thing from the options market — traders "looking to hedge the risk that the Federal Reserve pivots to cutting rates in 2027," wagers the paper described as "at odds with recent moves in the Treasuries market."

(a) Current-year meeting distributions — 18 August, current / [prev day] / [prev week]

Modal ranges in bold. Contract prices: ZQU6 96.330, ZQV6 96.275, ZQZ6 96.160.

(b) Next-year meeting path — 18 August

The terminal rate is drawn by the 28 July 2027 contract at 95.985, implying 4.015% — five-tenths of a basis point higher than the prior edition's 4.010% at 95.990, so the strip's peak crept up while its 2026 hike probability fell. The modal range is 3.75–4.00% at every 2027 meeting and the modal probability decays monotonically from 42.3% in January to 34.5% in December. The easing tail is the live change: the first non-zero probability of a range below 3.50–3.75% is 0.6% at the 15 September 2027 meeting, rising to 1.0% in October and 2.4% by December 2027 — and every one of those cells is lower than the previous day's ([1.4], [1.8], [3.0]). On the day Bloomberg reported bond traders buying options against a 2027 pivot, the futures strip took cut probability out of late 2027. The options market and the futures market are pointing in opposite directions, which is usually what a hedge looks like rather than a view.

(c) Year-end probability ladders — 18 August

Year-end 2026 (9 December 2026 meeting), relative to the current 3.50–3.75%:

Year-end 2027 (8 December 2027 meeting), relative to the current 3.50–3.75%:

Rounding, stated transparently. Investing.com's 18 August cards do not all sum to exactly 100.0%. The September, October, January, March, April, June, July, September-2027 and October-2027 cards each sum to 100.0; December 2026 sums to 100.1; December 2027 sums to 99.9. CME's September headline sums to 100.0 in all four of its columns. Cumulative figures are computed by summing the vendor's published cells without re-normalising, so a cumulative can carry the same ±0.1 rounding as its parent card — the December 2026 cumulative-hike figure of 67.5% is 100.1 − 32.6 and inherits that tenth.

(a) IG and HY credit spreads

As-of date, stated rather than implied. FRED's ICE BofA series publish with a one-business-day lag, and all three carried 17 August as their last observation when read this session. Tuesday's oil-flat, equity-down, long-end-rallying session is therefore not in these numbers; it publishes on 19 August. Same-day direction was cross-checked against Bloomberg's credit coverage and WSJ's bond page, neither of which reported a spread move on 18 August.

IG has now widened on four consecutive prints and that is the story in this block. BAMLC0A0CM has gone 78 → 79 → 80 → 81 across 11, 13, 14 and 17 August after nine consecutive prints inside a 78–79 band. 81 bp is the widest since 21 July. Four basis points is nothing in isolation; four consecutive one-basis-point widenings after a nine-print flatline is a change in behaviour, and it is happening while the equity market reprices the funding structure of the largest IG issuers in the index. HY moved with it — +3 bp to 270 bp, the largest single-day HY move in this reporting window — and the CCC cohort widened 6 bp, taking the CCC-minus-HY differential to 748 bp, up 144 bp on the year while HY itself is 11 bp tighter. The dispersion inside credit is entirely in the tail, exactly as it has been all month, but the investment-grade end has stopped being inert.

CDX retrieval note — the six-step ladder, worked and named. (1) Bloomberg in Chrome: /markets/rates-bonds rendered fully and was verified programmatically to contain no CDX string anywhere in the page text; the Bloomberg Fixed Income Indices panel publishes Global Aggregate, U.S. Aggregate, Asian-Pacific, Pan-Euro and EM USD Aggregate but no credit-default-swap index. (2) WSJ Market Data bonds page: rendered; the article layer returned the day's bond story and no instrument table carrying a CDX level. (3) Cbonds / ICE / S&P Global: Cbonds' CDX.NA.IG 5Y and CDX.NA.HY 5Y pages remain entitlement-gated; the ICE Markit CDX.NA.IG product page and the S&P Dow Jones CDX North America index-family page publish methodology and RED-code announcements, not levels. (4) FT Markets Data and Reuters credit wraps: no instrument page and no wire wrap quoting a level appeared in the reviewed material. (5) TradingView / Barchart / CME CDS index products: documentation only. (6) Cash-market proxy: available but not published this session, because the two available proxies point in opposite directions — the FRED IG cash credit spread widened 1 bp while the equity market sold the same issuers' suppliers, and a proxy that needs an argument is not a proxy. The most recent third-party level found anywhere remains an undated 81 bp reference, and under this report's standing rule an undated digest number is not a CDX level and is withheld. Quoting conventions restated so no reader mis-signs the field: CDX IG 5y is quoted in basis points of spread; CDX HY 5y in price points, where a rising price means tightening credit spreads.

(b) Money-market and funding plumbing

NY Fed reference rates publish at roughly 8:00 a.m. ET for the prior business day; the overnight series below carry a 17 August effective date, and the SOFR averages and index carry 18 August.

Read this with §6, because it is the same story told from the other end. Four sessions ago SOFR printed 3.62%, three basis points through IORB, on a 11 bp first-to-99th percentile band, with tri-party and broad general collateral both at 3.60%. As of 17 August SOFR is 3.66%, one basis point above IORB, the percentile band has widened to 14 bp, the 75th percentile sits at 3.71% — six basis points above the administered rate — and both general-collateral benchmarks have risen 4 bp to 3.64% on $1.2tn a day. Meanwhile the reverse repo facility took $0.155bn, a fresh record low and effectively zero against a $2.4tn peak, and reserves are $198.6bn below July's high. This is not stress; EFFR is unchanged at 3.63% on a 5 bp band and nothing has broken. It is the absence of a cushion becoming visible in the price of secured overnight cash, mid-month, in the quiet part of the cycle, with no month-end or quarter-end distortion present in the 17 August fixings. September quarter-end will be the first test with nothing at the facility to absorb it, and the Fed publishes its own Reserve Demand Elasticity estimate on Thursday at 10:00.

(c) Rates volatility and swap spreads

The MOVE index published 75.63 for 17 August, up 8.70% from 69.58 on 14 August — a 6.05-point single-day jump and the largest in this reporting window, taking rates vol back above the 72–78 band it had spent early August unwinding. No 18 August value has been published; the Investing.com series is delayed and its last row is 17 August. Label the vintage before using the ratio. On a matched 17 August basis, MOVE 75.63 against a VIX of 15.19 gives 4.98×. Mixing dates — 17 August MOVE against Tuesday's VIX close of 15.84 — gives 4.77×, and that number should not be traded on.

The direction is the point and it is legible without a same-day print. Rates vol jumped 8.7% on Monday, the session that produced the 19-year high; equity vol rose 4.28% on Tuesday, the session in which the long end rallied and equities fell anyway. The two are no longer moving on the same driver. Rates vol is pricing a fiscal-supply and term-premium problem that is increasingly European; equity vol is pricing an AI-funding problem that is American. A MOVE/VIX ratio near 5× against a five-year average closer to 4× says the rates market still carries more risk premium than the equity market — but the composition of that premium changed on Tuesday, and a ratio cannot see composition. No verified 2y, 10y or 30y swap-spread level was obtainable this session; the vendors that publish them are behind entitlement walls and no wire quoted a level in the reviewed material. That is a gap and is recorded as one rather than filled with a proxy.

(d) Issuance, leveraged loans and private credit

Primary activity was live and it was structured, not vanilla. Boeing and Lockheed Martin's space joint venture priced $1.5bn of private bonds, and Ripple's brokerage arm tapped the private-placement bond market — two issuers with no natural public-market constituency going straight to private credit on the same session (Bloomberg). The Dulles airport board is seeking $14bn of bonds following a federal overhaul, a municipal supply event large enough to matter to the long end of the tax-exempt curve. On the distress side, the SEC sued former executives of Tricolor after the firm's collapse — the subprime-auto failure this report has tracked since July — and Mavik is circling a KKR real-estate lender's troubled book. Bloomberg Markets' own cover feature this week is titled "AI Looms Over Software Companies — and the Investors Who Piled Into Them," on buyout funds and lenders drawn to recurring revenue now facing model risk: that is the private-credit expression of the same trade that repriced in the public equity market on Tuesday. No verified IG or HY primary volume figure or forward-calendar total was obtainable for the session, and no updated Morningstar LSTA leveraged loan index level or bank CDS level was published in the reviewed material; those fields are gaps and are recorded as gaps.

Quote basis: TradingEconomics currency board, Aug/18 dated rows, pulled in the evening ET after the U.S. close; percentage changes are the vendor's own daily column on that dated row. USD/TWD is an Investing.com real-time mark at 18:17 ET and is labelled as such. Bloomberg's 4 p.m. New York marks are quoted alongside where they differ.

The take — read the crosses that did the opposite of what the tape says. Three of them did. The Swiss franc weakened 0.19% on a session with a 1.33% Nasdaq decline, a 4.98% semiconductor drawdown and a bid in the U.S. long end — a textbook haven setup that the textbook haven declined. Pair that with gold −1.88% (§11) and the conclusion is not that risk aversion was absent but that the hedges themselves are the crowded positions, and crowded hedges de-risk with the book instead of against it. The Korean won strengthened 0.32% on the day Seoul's equity market fell 1.55% and its small-cap index 3.5% — the mirror image of the pattern this report flagged three weeks ago, when the won weakened on the Kospi's best day ever. Both observations point the same way: Korean equity flows are domestically leveraged, so the won moves inversely to local risk appetite rather than with foreign inflow. If that holds, a Kospi drawdown is a won buy signal, not a sell. And the Canadian dollar weakened 0.16% into a fourth consecutive advance in crude, which strips the commodity story out of the loonie and leaves the rate story — and Canada's 10-year fell 2 bp alongside America's, the only other market that rallied.

The non-event is the dollar itself. DXY at 99.659 moved two hundredths of a percent on a day the U.S. equity market fell, the U.S. long end rallied, European long ends sold off hard and the U.S. curve flattened 3 bp at 2s30s. Every one of those inputs argued for a directional dollar move and none of them delivered one. The prior edition flagged speculative long USD positioning at +$48bn, the most crowded since 2015; a crowded position that will not move on a day full of catalysts is a position waiting for a different catalyst. Wednesday's minutes are it.

Basis caveats, stated before the analysis. Energy and metals settles are Investing.com per-contract historical boards, which publish the official close for the front contract. Weekly, monthly and YTD columns are TradingEconomics on its Aug/18 dated row, whose header order was verified programmatically as Price | Chg | %Chg | Weekly | Monthly | YTD | YoY | Date before any YTD was quoted. The contract-month basis on crude is a real gap and not a data conflict: Investing.com's front-month CL board has rolled to October, settling $84.42, while CNBC quotes September WTI at $85.02, +0.62% — a $0.60 calendar spread that also explains Monday's $83.74-versus-$84.50 divergence. Brent is October on both. TradingEconomics' crude level of $85.458 is withheld because it sits 1.23% above the Investing.com October settle, outside this report's 1% tolerance; only its ratio columns are used. Gold carries the reverse basis: TradingEconomics' spot XAU at $4,341.17 sits 1.10% below the Comex front-month settle of $4,389.50, which is contango on a 5%-plus funding rate rather than a vendor error — the futures number is used for the settle and the spot series only for YTD.

The take — positioning, not fundamentals, and the crack is where the money is. Nothing about the physical oil market changed on Tuesday: the memorandum is still unrenewed, Trump has still ruled out an extension, and Bloomberg's headline was simply "Oil Advances for Fourth Day With US and Iran Locked in Stalemate." What changed is everything around it. Crude rose 0.81% and the entire precious and industrial metals complex fell — gold 1.88%, silver 4.24%, copper 2.43% — which is not a commodity-index move and cannot be explained by the dollar, which was flat at 99.659. It is a positioning unwind inside the metals complex specifically, and the year-to-date column shows why: silver is −11.04% year to date after +69.5% year on year, gold has given back essentially all of 2026's gain to +0.49%, and both are down while crude is up nearly 50%. The metals rallied on a monetary thesis and are now being sold by books that also own semiconductors.

The crack is the trade the tape is actually paying for. Heating oil is +111.86% year to date and +5.69% on the week; gasoline +93.34% and +5.46%; crude itself +48.83%. Products are running at more than double the barrel's year-to-date gain, and the equity market confirmed it on Tuesday — Marathon Petroleum +2.24%, Phillips 66 +1.76%, Valero +0.82% against a 0.81% crude settle (§3). A Hormuz disruption constrains crude supply; it constrains refined product supply harder, because the region ships product as well as barrels and the distillate market has no spare capacity to absorb a diversion. The convexity is in the distillate crack and it is not priced as though it is. The invalidation is equally clean: a confirmed reopening collapses the crack faster than it collapses the barrel, because crack spreads are a scarcity premium and barrels are an inventory.

Desk-style ideas for institutional investors. Each carries an explicit catalyst and an explicit invalidation. These are not personalized investment advice; verify independently and size to your own mandate before acting.

1. The rates trade — mark it, keep it, and take the tail hedge off the table

Mark on the prior edition's rates idea. The book is long ZQZ6 against short ZQZ7, DV01-matched one-for-one at half of half size, entered at 96.165 / 96.060 for a spread of 10.5 bp. Monday's mark was unpublishable. Tuesday's is clean: ZQZ6 96.160, ZQZ7 96.015, a spread of 14.5 bp, for a gain of 4.0 bp, or $166.68 per contract pair at $41.67 of DV01 per basis point — up from 14.0 bp at the 14 August mark. The position has worked for exactly the reason it was put on: the front leg's driver, 2026 hike probability, has fallen (December cumulative hike 78.5% a week ago to 67.5%), while the back leg's driver, the 2027 terminal, has crept up to 4.015% from 4.010%.

Modal path, base case and the tails — this is where the §8 data becomes a position. Modal path: hold on 16 September (CME 65.0%, ease 0.0%); hold on 28 October at 51.3% modal with the cumulative hike at 48.7%; one 25 bp hike delivered by 9 December, 3.75–4.00% modal at 45.0% with the cumulative at 67.5%; terminal 4.015% drawn by the 28 July 2027 contract at 95.985; year-end 2027 modal 3.75–4.00% at 34.5%, with the cumulative easing probability at 2.4%. Base case: the hike keeps sliding to the right while the terminal creeps up, because the inflation impulse is real but arrives through energy and tariffs rather than through wages, and the committee's three dissenters keep the front end from pricing any cut. That is precisely the shape this spread is long. Tail one, dovish: claims through roughly 240,000 on Thursday would put a 2026 cut into a strip showing 0.0% at every meeting; the front leg gains but the back leg gains more and the spread compresses. Tail two, hawkish: a minutes text on Wednesday that reads like an endpoint rather than a pause puts September back toward a coin flip; the front leg loses and the back leg barely moves. Both tails compress, which is why the stop is the discipline.

Expression: long ZQZ6 against short ZQZ7, DV01-matched one-for-one at $41.67 per basis point per contract. Catalyst: FOMC minutes 8/19 14:00; initial claims 8/20 8:30; PCE deflator 26 August; Jackson Hole 27–29 August. Invalidation, unchanged: the spread back through 11.0 bp; or a core PCE deflator at or below +0.2% m/m on 26 August; or any 2026 meeting showing a non-zero cut probability. Sizing: take profit on a quarter of the position into Wednesday's minutes and hold the rest. A trade that is 4 bp in the money going into its single largest binary event should not be carried at full size through it.

2. New — long the distillate crack against short the barrel

Expression: long a basket of Valero, Phillips 66 and Marathon Petroleum against short front-month WTI, sized so the futures leg matches the basket's crude beta. This replaces the prior edition's version of the same idea, which was carried unmarked; the marked version is now materially better. Mark on the prior construction: the refiner basket returned +1.61% equal-weighted (MPC +2.24%, PSX +1.76%, VLO +0.82%) against a short leg that lost 0.81% on the October WTI settle — a net +0.80% on the session, and the first clean confirmation since the construction broke on 14 August. Thesis: products are running at more than double crude's year-to-date gain — heating oil +111.86%, gasoline +93.34%, crude +48.83% — and the equity expression has resumed tracking it. Catalyst: the EIA weekly report; any Hormuz development; the September distillate draw. Invalidation: a confirmed Hormuz reopening; or a distillate build above 3m barrels; or the basket underperforming the crude leg on two consecutive up-crude sessions. Sizing: small, raised from the prior edition's small-and-unmarked to small-and-confirmed — one marked session earns the right to keep the position, not to double it.

3. Long the memory suppliers against the hyperscaler buyers — stopped out, closed, at a loss

Expression, as opened Monday: long an equal-weight pair of Micron and SK Hynix against short Microsoft, dollar-neutral, small. Mark: Micron −6.94%, SK Hynix −9.19%, Microsoft +0.31%. The pair lost 8.38% in a single session. The invalidation was written as "the pair losing more than 5% in five sessions." It triggered on day one. The position is closed at the stop, and it is the worst single-session idea this report has published.

What went wrong, stated plainly, because the lesson generalises. The thesis was that if hyperscaler capex is committed regardless of payoff timing, you want the seller's revenue and not the buyer's cash flow. Monday's tape supported it perfectly. What the thesis did not contain was the possibility that the commitments themselves would be repriced — and that is exactly what the WSJ's $3 trillion off-balance-sheet analysis did. If the obligations are larger and less visible than believed, the supplier's forward revenue is less certain, not more, and the buyer's balance sheet is the one with the option to slow down. The long leg was short the buyer's optionality and did not know it. Microsoft rose 0.31% while the suppliers fell 7–9%: the market decided the buyer holds the cancellation right. That is a structural asymmetry in every long-supplier / short-hyperscaler pair, and it should be assumed in the construction rather than discovered in the P&L.

4. Own downside in the consumer block, not in the index — hold, thesis confirmed, one day left

Expression: a one- to three-week put spread on a consumer-discretionary basket or the sector ETF, struck beneath the 13 August level and financed by selling the wing. Mark: Consumer Cyclical −0.45% on the day and −2.64% on the week, the worst weekly performance of any sector; the structure gained. Thesis confirmed by the print, not the tape: Home Depot beat both lines, posted its best comparable sales in almost four years, booked $730m of tariff refunds — and closed −0.12% after fading 2.05% from its high. The invalidation was "Home Depot rising more than 3% on its print"; it did not come close. The trade's remaining life is 36 hours: Lowe's, Target and TJX Wednesday before the open, Walmart and Ross Stores Thursday. Catalyst: those five prints. Invalidation, unchanged: a control-group retail print above +0.5% m/m, or any two of the five reporters closing more than 3% higher on their print. Sizing: small, defined-risk, premium-at-risk only — do not roll past 8/20; the thesis expires with the block.

5. Long gold against the Swiss franc — cut to a token, thesis broken

Expression: long spot gold funded in CHF, notional-matched, previously a quarter size. Mark: the long leg lost 1.88% on the Comex settle; the funding leg gained 0.19% as the franc weakened; net roughly −1.7% on the session. Thesis, broken on its own narrowed terms: this position had already been cut to the inflation-hedge reading only, and Monday was its cleanest confirmation — gold rose on an oil shock while real rates rose. Tuesday was its cleanest refutation: the barrel rose again, the long end rallied, and gold fell 1.88% to close $5 off its low. An inflation hedge that falls on a day when the inflation input rises and the discount rate falls is not an inflation hedge; it is a momentum position. Invalidation status: spot gold below $4,300 has not triggered — the TradingEconomics spot series printed $4,341.17, 0.95% above it — but it is now within one ordinary session. Sizing: cut from a quarter to a token, effective at the close. Do not wait for the stop to print; the reason for the position stopped being true on Tuesday.

6. Protection on the CCC cohort funded in IG — hold at a half, and the funding leg is now the problem

Expression: long CCC-exposed credit protection (or short a levered-loan / CCC-heavy vehicle) against long IG cash. Mark: the CCC leg widened to 1,018 bp on 17 August, with the CCC-minus-HY differential at 748 bp, +3 bp on the day and +4 on the week — the protection leg is working. The funding leg is not: IG has widened four consecutive prints to 81 bp, its widest since 21 July. Thesis, on notice: this structure is built on IG doing nothing, which made the funding free. IG is no longer doing nothing, and the reason is the same $3 trillion of off-balance-sheet commitments that took the equity market apart. Catalyst: the 19 August FRED update, which carries the first post-selloff credit spread; the post-Labor Day IG supply calendar. Invalidation, tightened again: the differential back through 735 bp, or IG widening beyond 85 bp — four basis points away — which would make the funding leg the dominant P&L driver and break the structure's logic. Sizing: a half, unchanged.

7. On-balance-sheet AI funding against off-balance-sheet AI funding — hold at small; the thesis just got its headline

Expression: long AMD against short Blackstone, dollar-neutral, small. Mark: AMD −4.21% to $484.68; Blackstone +0.01% to $140.59. The pair lost 4.22% on the session. And yet the thesis was vindicated, which is the uncomfortable part. The whole idea was that the market would eventually distinguish between AI capacity funded on balance sheet and AI capacity funded off it. The WSJ published exactly that distinction on Tuesday — $3 trillion of off-balance-sheet commitments against $600bn of reported capex — and the market's first reaction was to sell the on-balance-sheet name and leave the off-balance-sheet financier unchanged. That is the wrong sign for one session and the right sign for the thesis: the repricing has started at the visible end, which is where it always starts. Catalyst: Nvidia 26 August after the close; Marvell 27 August; the post-Labor Day IG calendar; any rating action referencing a financing vehicle rather than earnings. Invalidation, unchanged: AMD's IG credit spread widening more than 25 bp from new issue; or the pair losing more than 5% in five sessions — it is 4.22% into that window after one day, so this is live. Sizing: small, unchanged, and do not add before Nvidia.

8. New — long healthcare against the semiconductor complex, beta-adjusted

Expression: long the Health Care Select Sector SPDR against short a semiconductor ETF, beta-adjusted rather than dollar-neutral, at small size. Thesis: Tuesday produced a 3.93-point sector dispersion, the widest since 12 August, and both ends of it are rate trades pointing in the same direction. Healthcare set all-time highs across four separate vehicles — the sector SPDR, the Vanguard fund, the iShares biotech ETF and the Nasdaq biotech index — with three-month total returns of 15.6% to 21.1% against 4.8% for the S&P 500, while semiconductors fell 4.98% on the SOX and about 5.4% on SOXX. The mechanism is that healthcare cash flows are near-dated and contractual while semiconductor cash flows are long-dated and, as the WSJ demonstrated, partly contingent on obligations that were not on anyone's balance sheet. A market that has just discovered $3 trillion of hidden forward commitments should pay more for certainty and less for duration, and that is the trade. The honest counter-argument: healthcare has run 15–21% in three months and is itself now a crowded defensive, and this pair is short the single best-performing equity complex of the last two years into its biggest catalyst. Catalyst: Nvidia, 26 August after the close — the cleanest possible test in either direction; the July FOMC minutes; any hyperscaler capex revision. Invalidation: Nvidia beating and guiding above consensus with the SOX up more than 4% in the session after; or the pair losing more than 4% in five sessions; or the healthcare complex making a new high on a day the SOX also rises, which would mean the trade is long two legs of the same beta. Sizing: small — one session of evidence, and the short leg is entering the event.

Volatility note. VIX 15.84, +4.28% from 15.19, a second consecutive bid but a strikingly small one: the Nasdaq fell 1.33%, the SOX 4.98%, and one-month implied vol still closed under 16. At 15.84 the index prices a daily move of about 1.00% against a realised S&P session of 0.69% — the short-vol carry is still positive but it has been positive for a fortnight and the cushion keeps thinning. The specific objection has changed since Monday. Monday's was correlation: a single headline moving the barrel, the 30-year and the index together. Tuesday's is dispersion: a 4.76-point spread between the SOX and the Dow, a 3.93-point spread across sectors, and an index that fell less than one percent because its winners and losers were nearly the same size. Index vol is the wrong instrument for this tape. If you want to own volatility here, own it in single names into the 26 August print, or own it in rates, where the MOVE has already jumped 8.7% (§9c) and the ratio to VIX is near 5×. The index straddle is priced for a market that moves together, and this one is not.

The crowded consensuses to stress-test, with the numbers.

1. "The AI capex cycle is fully disclosed." Tuesday's WSJ analysis put roughly $3 trillion of off-balance-sheet commitments across nine companies against $600bn of reported capex and about a third of that again in leases and long-term borrowings. If that number is directionally right, every model of hyperscaler free cash flow, every supplier revenue forecast built on announced capex, and every IG credit metric on those issuers is understated on the obligation side. The market's first-day reaction was to sell the suppliers 5–9% and leave the issuers' credit spreads 1 bp wider. One of those two reactions is wrong.

1. "The long end only goes one way." The 30-year printed a 19-year high near 5.32% and closed at 5.28%, −3 bp, on the day German and French long ends made multi-decade highs and the S&P fell 0.69%. The consensus fiscal-supply trade is correct about Europe and is now, on one session's evidence, being questioned in America. Fundstrat's Mark Newton is looking for 5.60–5.70% on the U.S. long bond; Ed Yardeni, who coined "bond vigilantes," says he is not "pushing the panic button" yet. A crowded short in duration that cannot hold a breakout is the setup for a squeeze, and the squeeze would hit the same equity books that are long the AI complex.

1. "Havens work." Gold −1.88%, silver −4.24%, the Swiss franc −0.19% against the dollar, all on a risk-off session. Three separate hedges failed on the same day, and they failed because they are held by the same books that hold the risk. Test your hedge portfolio for correlation to the position it is supposed to hedge, not to the risk it is supposed to hedge against.

1. "The consumer block will re-rate the retailers." Home Depot delivered the best quarter it has had in years and closed lower. Four more report in the next 36 hours into a sector that is −2.64% on the week and −3.71% year to date. The setup is not "will they beat" — it is that beats have stopped paying.

1. "There is no funding problem." ON RRP at $0.155bn, SOFR 1 bp above IORB with a 75th percentile 6 bp above it, general collateral +4 bp, reserves $198.6bn below the July peak — and EFFR unchanged with a 5 bp band. Nothing is broken. Nothing is left, either, and September quarter-end is the first settlement date with no facility cushion.

The two-sided geopolitical tape. The U.S.–Iran memorandum remains unrenewed with Trump ruling out an extension, and he added a threat to "bomb the s--- out of" Oman if it "gets in the way." Interactive Brokers' José Torres also flags "a fresh burst of violence in Lebanon." Crude has now risen four sessions running on this and is +48.83% year to date. The two-sided part is the fade risk: a stalemate that persists without escalation lets the risk premium decay, and the distillate crack decays faster than the barrel (§11). Separately, five grain ships were struck near Russian Black Sea ports, and Mexico is weighing tougher trade rules on China as U.S. talks grind on.

Structural watch items. The $14bn Dulles bond request and the $1.5bn Boeing-Lockheed private placement are two ends of a supply picture that competes directly with Treasuries for the same duration buyer, which CNBC's own explainer named as one of three drivers of the yield move alongside the deficit and sticky inflation. The SEC's suit against former Tricolor executives keeps the subprime-auto failure alive as a governance story. Bloomberg Markets' cover feature on AI risk to software companies and their private-credit lenders is the same trade as Tuesday's equity repricing, one layer down the capital structure and marked far less frequently. And Reddit joined the S&P 500 on 18 August, adding an index constituent whose float and volatility profile differ materially from the REIT it replaced.

Sources. CNBC (market live blogs for 17 and 18 August, closing bell, sector and healthcare desks); Bloomberg.com US edition (/markets, /markets/rates-bonds, rates story); WSJ.com (/market-data, /market-data/bonds); Investing.com (major indices, NDX, PHLX Semiconductor, VIX, trending and Dow-component boards, per-contract commodity historical boards, MOVE historical, USD/TWD, Fed Rate Monitor); Finviz Groups (Performance table view); U.S. Department of the Treasury (Daily Treasury Par Yield Curve Text View, August 2026); CME Group FedWatch; Federal Reserve Bank of New York (reference-rates API, reverse-repo operations API, August 2026 economic indicators calendar); FRED (BAMLC0A0CM, BAMLH0A0HYM2, BAMLH0A3HYC, WRESBAL); TradingEconomics (currencies and commodities boards); Nasdaq earnings calendar API; Slickcharts S&P 500 constituent list; stockanalysis.com (single-name closes); U.S. Census Bureau and BLS (via wire summaries); National Association of Realtors; 24/7 Wall St.; Reuters and Associated Press via wire summaries; Fast Company; Invezz; RTTNews.

Full Data Notes & Conflicts, the Overnight / Asia & Europe read-through and the complete categorised Source Links are in the companion file US_CrossAsset_Daily_2026-08-18_DataNotes.txt.