“Every banker knows that if he has to prove that he is worthy of credit, however good may be his arguments, in fact his credit is gone.”

Walter Bagehot, Lombard Street: A Description of the Money Market, 1873, Chapter II. Attribution: verified.

I. Pre-Market Indications

Individual equities had no meaningful pre-market print at the 06:23 ET capture, so none is shown. What follows is limited to the instruments that actually trade overnight. Every change column is measured against the Friday, August 21 settled close, which is the only authoritative basis available before the open.

UNSETTLED · PROVISIONAL. NRT column captured 06:23 ET, single-source (Yahoo v8 chart application programming interface). Not two-source confirmed, not settled, and not eligible to touch any locked mark. Foreign-exchange crosses trade continuously and have no settled close, so no Day change is shown for them.

The tape's own tell is in the second-to-last block of that table. Gold is up 1.78 percent before the open, to its highest level since the spring, on a morning when crude is down and equity futures are soft. That combination is not a growth scare and it is not an inflation scare. It is a credibility bid, and it is the same bid that ran through the whole of last week.

II. The Round Trip in the Long End

The most important number of the week is one that barely moved. The thirty-year Treasury yield finished the week at 5.276 percent against 5.265 percent the Friday before — a net change of 1.1 basis points. A reader who saw only the endpoints would conclude nothing happened. The path says otherwise.

Constant-maturity yield closes, two-source confirmed. Wednesday’s rally was 9.1 basis points in the thirty-year; the two sessions that followed took back 8.2 of them.

On Wednesday the Treasury announced that its maximum long-end liquidity-support buyback operations would be raised from $2 billion to at least $4 billion per operation, effective from September 9 through November 4. The long end rallied nine basis points on the news and had given all of it back within two sessions. That is the single cleanest piece of evidence produced last week, and it deserves to be stated plainly: the marginal buyer announced that it would double its size, and the market took the offer, sold into it, and finished the week at a higher yield than it started.

The context is not subtle. Federal debt outstanding passed $40 trillion for the first time, reported on August 18 and 19. Wednesday's minutes from the July 28 and 29 meeting — at which the target range was held at 3.50 to 3.75 percent on a nine-to-three vote, with all three dissents preferring an increase — recorded that many participants assessed that policy tightening would likely be necessary if inflation did not decline. And this is not solely an American problem, though the severity differs by market and the differences are worth respecting. Japanese ten-year yields reached 2.93 to 2.95 percent midweek, the highest since September 1996 — a genuine thirty-year extreme. United Kingdom ten-year gilts moved above 5 percent, the highest since July 2008. Germany's thirty-year auction cleared at the highest yield since May 2011. Only Japan is at a multi-decade high in the strict sense; the point is not that every market broke a record but that the bid for sovereign duration was tested simultaneously in all of them. And in the one market where the issuer stepped in to support it directly, the support did not hold for forty-eight hours.

Bagehot's line at the top of this edition is the whole of it. An institution that must announce an operation to demonstrate that its paper is wanted has already told the market something about how much it is wanted.

III. The Chips Fell on Rates, Not on Chips

The temptation last week was to read the semiconductor decline as a fresh instalment of the artificial-intelligence demand-verification argument that has run through this publication since July. That reading does not survive contact with the cross-section.

Settled-to-settled, two-source confirmed. Nasdaq Composite −2.05 percent, Dow Jones Industrial Average −0.85 percent, Russell 2000 −1.65 percent over the same week.

Micron, the memory name at the centre of any genuine artificial-intelligence demand question, fell half a percent. Taiwan Semiconductor, which actually manufactures the units in dispute, fell 1.74 percent. Meanwhile Marvell rose almost seven percent on the disclosure of a custom-silicon agreement with Google, carrying a warrant for 58,970,907 shares struck at $206.58 — a signed, dollar-denominated commitment rather than a projection. If the market were adjudicating whether artificial-intelligence demand is real, the names closest to the physical units would be the ones under pressure and the name with the new contract would not be the week's outlier to the upside.

What actually happened is more ordinary and, for positioning purposes, more useful. Market-implied odds of a September hike had already collapsed from roughly sixty-five percent in the first week of August to roughly thirty percent by mid-month, on a weak July payrolls print and a benign consumer price index. The market had, in other words, settled comfortably back into the easing path. Wednesday's minutes then told that market that three voters at the July meeting had wanted an increase, and that many participants thought tightening would likely be necessary if inflation did not decline. Long-duration equity is the most rate-sensitive asset class in the index, and the names that fell hardest are the ones whose valuations sit furthest out on the cash-flow curve.

Intel is the one name on that list that does not belong to the pattern, and it is worth separating out rather than folding into the story. Its twelve percent decline is the digestion of an upsized $20 billion equity offering priced on August 10 at $95 — roughly 210.5 million shares, about four percent dilution. The stock spent last week breaking below the offer price and closed at $90.07. Broad semiconductor weakness made it worse, but the overhang is why Intel underperformed its own sector by roughly a factor of two, and attributing that move to the rate story would be over-reading the evidence.

The confirming evidence for everything else is in the sector table, and specifically in the utilities.

Sector Select SPDR exchange-traded funds, settled-to-settled. ETF returns, not the underlying GICS sector indices; the two differ modestly by weighting and dividend treatment.

Utilities fell 3.48 percent, essentially in line with information technology. For eighteen months utilities have traded as the electricity leg of the artificial-intelligence buildout, which should have decoupled them from technology weakness. They did not decouple, because utilities are also the market's most reliable bond proxy, and last week the bond proxy dominated the growth story. A sector that is simultaneously a rate instrument and a capital-expenditure instrument gets to be hit twice, and last week it was.

The honest caveat: health care's 4.33 percent gain is materially distorted by a single name. Moderna rose roughly 177 percent on Wednesday alone on Phase 3 melanoma-vaccine data, and finished the week 129 percent higher. Stripping that out, the defensive rotation is real but considerably less dramatic than the headline sector number implies.

IV. Credit Did Not Blink

The equity market spent last week repricing the discount rate applied to artificial-intelligence capital expenditure. The credit market, which is the actual source of the money funding that capital expenditure, did not participate. Investment-grade option-adjusted spread finished at 82 basis points and high yield at 275, both effectively unchanged. August was the third consecutive record month for investment-grade issuance, and S&P Global counted roughly $225 billion of hyperscaler and artificial-intelligence-related supply through midyear alone.

This divergence is the single most interesting unresolved fact of the week, and it is worth stating both readings without picking one.

The first reading is that credit is correct. Bondholders sit ahead of shareholders, the assets being financed are contracted and depreciating on a known schedule, and a discount-rate move that halves an equity multiple does very little to a five-year investment-grade coupon. On that reading equity is having a valuation argument and credit is having none, which is exactly what should happen.

The second reading is less comfortable. NVIDIA announced on August 10 a partnership with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilise more than $500 billion of third-party capital for infrastructure buildout, structured to keep the obligations off NVIDIA's own balance sheet. Reporting since — not the announcement itself — indicates NVIDIA may provide residual-value support on up to roughly a quarter of an individual opportunity, assessed project by project, with the mechanism undefined. Separately, a filing disclosed on August 17 shows NVIDIA providing SB Energy, the developer of the Ohio campus that OpenAI will occupy, a residual-value guarantee capped at $105 billion against the initial 4.25 gigawatts, triggered only on specified default events. When the equipment vendor is also a contingent guarantor of the assets its customers are buying, spreads that do not move are not necessarily spreads that have assessed the risk. They may simply be spreads on paper whose ultimate credit support has not yet been tested by anything.

Which of the two readings above is right will be settled by the first quarter in which a hyperscaler misses a payment schedule, and not before.

V. Two Energy Shocks, Not One

Crude had a strong week on renewed disruption at the Strait of Hormuz, now in its sixth month. Brent settled Friday at $94.39, up 6.63 percent; West Texas Intermediate at $87.06, up 5.66 percent. Both are lower before Monday's open. Vessel traffic is the constraint that matters: Lloyd's List Intelligence counted 73 transits of the strait in the week of August 10 to 16, down from 91 the week before — roughly ten a day against a pre-crisis baseline near seventy. The retail transmission is already visible: the national average for regular gasoline is $4.099 per gallon and for diesel $5.613, the latter far above where it sat a year ago and the more consequential of the two, because diesel is the freight cost that shows up in every downstream good.

What makes the shock analytically interesting is that it is not one shock but two, running in opposite directions on the same commodity.

American natural gas is a domestic commodity in a global crisis, priced off pipeline capacity that cannot reach the tightness. European gas is priced off a storage position that is the worst on record for the date, going into a heating season that begins in roughly eight weeks. That is a structural asymmetry rather than a trading observation, and it has a direct industrial consequence: for any energy-intensive process that can be sited on either continent, the marginal cost gap between the two is now wide enough to be a location decision rather than a margin decision.

Precious metals had the better week and gold is having the better morning. Gold settled Friday at $4,624.10, up 5.56 percent, and is bid to $4,706.30 before the open. Silver settled at $69.47, up 6.89 percent. Both remain far below their January records — gold roughly sixteen percent below the $5,608 peak set in January of this year — which is worth keeping in view, because the rally that has run since May is a recovery within a large drawdown, not a series of new highs. Copper, by contrast, was flat to slightly lower on the week after the squeeze that ran through the middle of the month eased on Wednesday: roughly 20,000 tonnes arrived on warrant in a single day, the largest one-day build since April, and the cash-to-three-month backwardation collapsed from a peak near $545 a tonne. Eased, not resolved — on-warrant stocks remain far below their April levels, and the underlying tightness is unchanged.

VI. What to Watch

This is an unusually loaded week, with three separate items capable of resetting the thesis in Section II.

What would prove the Section II thesis wrong. If the buyback expansion that begins on September 9 is met by real demand and the thirty-year settles below 5.10 percent on a sustained basis, then last week's round trip was a positioning event rather than a credibility event, and the long-end pairs are carrying a thesis the market has already stopped holding. A second falsifier: if Chair Warsh delivers a genuinely hawkish message on Friday and the long end rallies on it, that is a market that has decided the Federal Reserve's credibility is intact.

What would prove the Section III thesis wrong. If NVIDIA misses on Wednesday and the semiconductor complex sells off with long-end yields falling, then the decline was a demand event after all and the discount-rate reading in this edition is the wrong one. The distinguishing evidence will be available within about eighteen hours of the print.

The two items already live. United States tariffs of 50 percent on approximately $20 billion of Canadian goods took effect on Saturday, August 22, after talks collapsed, with Canadian retaliation scheduled from September 8 — a two-week window in which the second leg is known and not yet priced. And sentiment is stretched in a way that cuts both ways. The Bank of America Bull and Bear indicator was raised to 9.7 in the reading published on August 7 — its highest since 2021, and a contrarian sell signal on its own construction. That reading is now more than two weeks old and has not been refreshed publicly through Friday, which is itself worth noting before leaning on it. Against it, Vanda Research data reported on August 19 show retail put buying across the twelve most popular retail names running near 110 percent of net cash buying, up from roughly 26 percent in the first quarter. The indicator says euphoria; the flows say the same people are hedged. Both cannot be the operative fact.

The Synthesis — Pre-Market

Nothing in the NRT column is settled, and the regular session can move meaningfully from these levels in either direction before the close that actually locks a mark. What the week just past does establish, on settled data, is a single proposition: the selling was in the discount rate, not in the demand. The evidence is that the memory and foundry names barely moved while the long-duration equity names fell hard, that the one chip name with a newly signed contract was the week's best performer, that utilities fell in line with technology despite being the electricity leg of the same buildout, and that credit — which funds the capital expenditure the equity market was supposedly doubting — did not widen a basis point.

Underneath all of it is the round trip in Section II. The Treasury announced that it would double the size of its long-end support operations and the market sold the rally within two sessions. That is a credibility question, not a rates question. The market gets its first real chance to answer on Friday morning, at the foot of the mountains in the photograph above.