Weekend Wrap: July Ended Flat and Nothing Inside It Was. The Nasdaq Lost 3.2% While Amazon Jumped 15% on AWS.

Saturday, August 1, 2026. There is no live tape this weekend, so the last prints on the screen are Friday’s July 31 closes and the next scheduled event that can move anything is the OPEC+ ministerial tomorrow. Treat this as stocktaking rather than reaction. July closed with the S&P 500 at 7,489.72, up 0.70 percent on Friday and down about 0.1 percent for the month. That number is arithmetically true and analytically useless. Inside those four weeks the Nasdaq Composite lost 3.2 percent, the Dow gained 0.3 percent for a fourth consecutive monthly advance, the Philadelphia Semiconductor Index went into a bear market and came out of it, Korea produced both the worst monthly decline and the best single day in the history of its benchmark, Brent traded a range of roughly $72 to $102, and the 30-year Treasury yield ended at a 19-year high. A flat headline sitting on top of that much internal violence is not calm. It is cancellation.

Top Call: The Index Went Nowhere Because the Rotation Did the Work

The single most useful fact about July is that the market did not de-rate the AI trade. It re-sorted it. Capital left the crowded logic and consumer-hardware end of the complex and went into the physical bottleneck, into the Dow’s defensive and energy weightings, and out of the highest-beta growth expressions altogether. Amazon rose 15.3 percent on Friday for accelerating cloud revenue against a rising capital budget. Apple fell 7.4 percent for guiding down on the cost of memory it has to buy. Space names fell roughly a third across the board. Korea fell 22 percent and then had its best day ever.

What that pattern says is that the marginal dollar is now underwriting a specific claim rather than a theme: show me the customer of the capacity, and show me that the input cost is on someone else’s income statement. The two things that can break the sort are both scheduled. OPEC+ votes Sunday on September barrels that mostly cannot physically leave the Gulf, and July payrolls land Friday with consensus at 88,000 into a Fed that already has three dissenters wanting a hike. Written Saturday morning Eastern with no live market, so anything below is a read on Friday’s closes rather than on a live tape.

The Month in One Number, and Why It Is the Wrong Number

On the week the S&P 500 added about 1.1 percent and the Nasdaq about 1.6 percent, the first winning week in three, while the Russell 2000 finished flat. The tell inside the weekly numbers is that small caps did not participate in the recovery at all. A rebound carried entirely by megacap cloud earnings, with no breadth underneath and with the long end of the curve selling off at the same time, is a narrow repair rather than a broad one.

Europe is the quiet outperformer again and the reason is structural rather than tactical. The Stoxx 600 touched a record 656.67 on Friday before closing fractionally lower at 649.19, and it booked a fourth consecutive monthly gain through a month in which the Nasdaq corrected and Korea crashed. Europe carries no hyperscaler capital budget to defend, essentially no commodity memory manufacturing, and heavy weightings in energy, banks, defence and industrials. An index that never took the bet has nothing to hand back when the bet is questioned. Underneath the calm surface the dispersion was European too: Universal Music fell 25.4 percent on Friday for its worst day on record after first-half results, while Teleperformance rose 13.2 percent on confirmed 2026 targets.

Amazon Closed the Cash-Conversion Argument

The framework that explained this earnings season is that the market is not screening against capital expenditure, it is screening against capital expenditure without a nameable customer. Amazon is the cleanest confirmation available. AWS is at a $169 billion annualised run rate, the backlog is $496 billion and growing triple digits, segment operating margin expanded 650 basis points to 39 percent, and the AI and custom silicon businesses each now run above $25 billion annualised with multi-year, multi-gigawatt commitments disclosed from Anthropic and OpenAI. Against that, a capital budget rising by roughly $20 billion reads as a supply constraint being relieved rather than as money disappearing.

The Trainium disclosure deserves more weight than it received. A custom accelerator franchise at a $25 billion run rate growing triple digits, with two of the largest frontier labs on multi-gigawatt contracts, is the first genuinely credible second source in AI training silicon. It does not dent Nvidia’s 2026, but it changes the terminal-margin argument that underpins a large part of the semiconductor complex’s valuation, and it is a better reason to own Amazon than the retail margin story that most sell-side models still lead with.

The counterweight is financing, and it keeps being underweighted. Amazon carries the heaviest bond stack of the megacap group after a roughly $54 billion deal in March. The four largest hyperscalers now point to combined 2026 capital spending of $720 billion to $745 billion. Adding another $20 billion of duration to that pile in a week when the 30-year printed 5.23 percent is not free, and the investors applauding the equity are not the ones who have to absorb the paper.

Apple Paid the Memory Bill, and the Bill Is About to Stop Growing as Fast

Apple beat every line that describes the quarter it just had and fell 7.4 percent on one sentence about the quarter ahead, with supply constraints and DRAM and NAND inflation named as the cause, and services and China revenue also coming in light. The framing that matters is that this is the first time the memory shortage has been priced as a cost borne by a buyer large enough that nobody can call it idiosyncratic. Apple is the largest purchaser of NAND on earth.

Here is the part that changes the trade rather than confirming it. The rate of memory inflation is decelerating even as the level stays punishing. TrendForce now models third-quarter conventional DRAM contract prices up 13 to 18 percent sequentially and NAND up 10 to 15 percent, against roughly 60 percent increases in the second quarter. ADATA’s chairman is more aggressive at 20 to 30 percent for DRAM and 35 to 40 percent for NAND, but even the bullish read is a sharp slowdown in the second derivative. That combination is awkward for both sides of the trade. Apple absorbs a high absolute cost base into a guided quarter, so the earnings drag is real and persistent. The memory makers, meanwhile, are lapping the steepest part of their own pricing curve just as the consumer end of demand reaches an affordability ceiling. The supplier upgrade cycle is not over, but the easy part of it finished in the second quarter, and the market spent Friday pricing Micron down 5.9 percent after a session that swung from up 6.4 percent to down 6.5 percent. That volatility is what disagreement about the second derivative looks like.

The Long End Did Not Ratify Any of It

The most important unresolved fact of the week is that September hike odds fell from about 79 percent before Warsh spoke to roughly 60 percent after, and the 30-year Treasury yield still finished at 5.23 percent, the highest since 2007. The 10-year ended at 4.71 percent, up from 4.68 percent Thursday and up from 3.97 percent before the Iran conflict escalated. The two-year sits at 4.28 percent.

Read those together and the message is unambiguous. If the long end were pricing the policy path, a 19-point drop in September hike probability would have pulled the 30-year down with it. It did not. What is being repriced is term premium: the compensation demanded for holding duration against an inflation process that has an energy shock inside it, a Federal Open Market Committee that split 9-3 with three dissents in favour of an immediate hike, a new chair still establishing credibility, and a corporate calendar that is about to issue a great deal of AI-related paper. An equity market recovering to a monthly high alongside a 30-year at a 19-year high is a tension, not an equilibrium, and the resolution mechanism is either lower yields or a lower multiple.

Oil is what makes this a genuine inflation question rather than a positioning one. Brent settled up 1.2 percent at $87.93 after a July range of roughly $72 to $102 and a monthly gain near 24 percent, its strongest month since March. The national average gasoline price is $4.11 a gallon against $3.85 a month ago. That flows into headline inflation with a lag and into the University of Michigan expectations series almost immediately, and it lands on a committee whose 2026 projected PCE inflation was already 3.6 percent.

OPEC+ Votes Sunday on Barrels That Cannot Leave

Ministers meet tomorrow, August 2, and are expected to approve a September increase of roughly 188,000 barrels per day, which would effectively complete the unwind of the 2023 supply agreement once the UAE’s exit is adjusted for. The number is close to meaningless as a physical event and quite meaningful as a signal.

It is meaningless physically because the barrels cannot get out. Only five vessels transited the Strait of Hormuz in the 24 hours to 21:00 UTC on July 31, four outbound and one inbound, against a pre-crisis baseline near 140 vessels a day. Three of those five carried OFAC designations. The Kavomaleas, struck on July 20, has been stationary near Larak Island since July 21 with the crew off the vessel, and the July 31 satellite pass showed smoke returning from the aft structure, eleven days on. What is moving is moving through an improvised ship-to-ship shuttling operation outside the Gulf, frequently with transponders off. Saudi Arabia, Kuwait and Iraq are all inside the constraint, so a higher quota for those producers is an accounting entry rather than a supply increase.

It is meaningful as a signal because it tells you the cartel has decided to hold market share rather than defend price into an eventual reopening. If the strait normalises on any timeline, the world discovers that OPEC+ has already unwound its cuts on paper and the physical barrels arrive into a market with no quota discipline left to reimpose quickly. That is the asymmetry in oil at $88 that the equity market’s energy re-rating is not pricing. It also sits alongside the behaviour of the majors: Exxon and Chevron booked $26.6 billion of combined quarterly profit last week and steered it into debt reduction rather than into buybacks or accelerated capital programmes, which is a management price forecast expressed in cash rather than in slides.

Korea, Storage, Optics and Rockets: Where July Actually Happened

The Kospi rose 17.9 percent on Friday to 6,595.45, the best single day in its history, with Samsung Electronics and SK Hynix both up more than 26.8 percent, and it still finished July down 22 percent. Nothing in the memory outlook changed between the crash and the rally. Earnings estimates for the Korean complex never came down during the drawdown and have not gone up since. What moved was leverage: a China lithography scare and a Samsung print read as peak margin forced hedge fund deleveraging, Microsoft’s Wednesday night result relieved the AI spending panic, and a personal share purchase by the SK Group chairman supplied the trigger. A market that can move 18 percent in a day in either direction has lost the ability to price the asset in an orderly way, and that instability is itself the risk factor for August. Japan rode the same wave, with the Nikkei up 4.03 percent to 64,362.02 after the Bank of Japan held at 1.0 percent and flagged that higher crude and a weaker yen are broadening consumer price pressure, this a day after suspected yen-buying intervention took the dollar down to the 158 area.

Storage told the cleanest version of the shortage story. SanDisk guided fiscal fourth-quarter revenue to $7.75 billion to $8.25 billion with earnings of $30 to $33 a share, which prices in another leg of NAND tightness. Seagate guided to $3.45 billion and about $5. Western Digital reported on July 31, beat on recovering NAND pricing and cost discipline, rose 3.74 percent, and is proceeding with a structural separation into two listed businesses. Yet SanDisk fell 23 percent and Seagate 19 percent over the past month. The guides say the cycle is intact and the tape says the market will no longer pay a rising multiple for a cyclical peak it can see. That gap is the most interesting risk-reward in the complex right now.

Optics quietly had the better fundamental quarter and got less credit. Lumentum printed fiscal third-quarter revenue of $808.4 million, up 90 percent, with co-packaged optics and optical circuit switch backlog already above $400 million. Coherent grew 21 percent to $1.8 billion with datacom up 41 percent. LightCounting sees the AI cluster optics market up 60 percent to $26 billion this year. Both names finished Friday higher with Applied Optoelectronics up 7 percent and Lumentum up 5 percent, and both trade above 127 times trailing earnings, which is the entire debate in one number: the demand is verified and the price already assumes flawless execution.

Space was the month’s clearest casualty and the least discriminating selloff. SpaceX fell about 36 percent over the month, Rocket Lab about 36 percent, AST SpaceMobile about 34 percent. The two are not the same asset. SpaceX is working off post-listing index flows and the ordinary digestion of a June debut. Rocket Lab has a $266 million Space Force suborbital award booked on July 21, its largest defence contract, revenue growing above 60 percent, and a pending $8 billion cash-and-stock acquisition of Iridium at $54 a share that does not close until mid-2027 and carries genuine dilution and integration risk between now and then. Selling both by the same third is a beta trade, not an analytical one, and it is the sort of dislocation that resolves once a single quarter separates the two.

Finally, CoreWeave remains the stress test for the whole financing thesis. Second-quarter adjusted EBITDA was $1.16 billion at a 56 percent margin, down from 62 percent a year earlier, against capital spending guided to $31 billion to $35 billion this year on roughly $6.2 billion of trailing revenue. The stock touched $86 during the week and closed Friday at $71.88. Investors were happy to fund that ratio in 2025. They are visibly no longer happy to fund it at a 5.23 percent 30-year yield, and that repricing of the leveraged neocloud model is a better leading indicator for the AI complex than any hyperscaler capex line.

Geopolitical Tracker

US and Iran: The war is in its fifth month with strikes and pauses alternating rather than resolving. Tehran continues to deny any direct negotiations with Washington and maintains that the strait reopens on Iranian terms. The June memorandum of understanding is functionally dead as a ceasefire and alive only as a framework nobody has formally abandoned.

Hormuz: Five transits in the 24 hours to 21:00 UTC July 31 against a baseline near 140 a day. Three of the five carried OFAC designations. The Kavomaleas is still burning near Larak Island eleven days after being struck, an unresolved navigation and spill hazard. The functioning trade is ship-to-ship shuttling outside the Gulf, which is expensive, opaque and capacity-limited.

Oil market consequence: Brent at $87.93 with the chokepoint at a small fraction of normal throughput is a market pricing a normalisation it has no evidence for, into an OPEC+ decision tomorrow that removes the last of the group’s formal supply discipline.

Elsewhere: The Russia-Ukraine war has turned increasingly naval, Houthi threats to Saudi-bound shipping remain live, and Putin and Xi are due to discuss Ukraine and Taiwan at the Shanghai Cooperation Organisation meeting. None of these is currently priced in developed-market risk assets, which is worth noting rather than trading.

The Week Ahead

Payrolls is the one that decides August. A hot print into 3.6 percent projected PCE inflation and three existing dissents makes a September hike the base case and puts the 30-year through 5.25 percent, which is the level at which the equity multiple has to start doing the adjusting. A soft print relieves the front end but does very little for a long end that is trading term premium rather than policy. The asymmetry is not symmetric, and positioning into Friday should reflect that.