Beat. Raise. Drop. | The New AI Trade | Retail Sales This Morning

Good morning, Slaters!

Something odd has been happening on the tape all week, and Thursday made it undeniable.

Companies are beating expectations. Then raising guidance. Then watching their stock get hit anyway. Cisco Systems (CSCO) beat, guided the fiscal year $6 billion above consensus, closed down 8.8%. Cerebras Systems (CBRS) beat, lifted its outlook, dropped double digits. Applied Materials (AMAT) posted record revenue after the bell, blew past the Q4 guide by nearly a billion dollars, and traded down 3%.

In the same session, SanDisk (SNDK) ran up 15% on an investor day slide deck.

The market is drawing a line. Not between winners and losers, but between companies growing into their multiples and companies still telling a story big enough to close the valuation gap. Retail sales today will tell us whether the consumer helps either camp.

SanDisk just rewrote the memory story in a single slide deck

SanDisk (SNDK) opened flat, drifted through the morning, then went vertical after management started walking through the long-term financial model at its "SanDisk In Focus" investor day. By close, the stock was up about 15%. Western Digital (WDC) added 10%. SK hynix (SKHY) rose 8%. Micron Technology (MU) climbed 6%. Seagate Technology (STX) added 5%.

The number that lit the fuse: non-GAAP gross margins of roughly 80%, sustained through fiscal 2030. Operating margins near 75%. Free cash flow margins around 50%. And 100% of excess cash returned to shareholders after reinvestment.

Those are not memory-industry numbers. Those are software-industry numbers. Memory has spent forty years being the most cyclical business in technology, with margins that swing from luxurious to negative in a single quarter. Wall Street has priced the group accordingly.

What changed the room was how management backed the forecast. New Business Model agreements have been signed with eight customers, covering roughly two-thirds of bits SanDisk expects to ship by fiscal 2028. Long-dated contracts in a category that historically has none. That is what turns a cyclical business into something the market can put a growth multiple on.

SanDisk's Q4 datacenter revenue was up 437% for the full year. The AI infrastructure trade has been mostly a GPU and networking story. Storage has been the quiet passenger. Thursday it grabbed the wheel.

The market signal?

Memory is being repriced from a commodity cycle to a structural growth category. The pricing agreements are the evidence, not the guide. If the group holds these gains through Nvidia earnings on August 27, the "AI needs everything, not just chips" thesis has a new leg.

One data point stands between the Fed and a clean pause

The Fed's September meeting is being written in real time this week, and it's writing itself dovish.

Wednesday's July CPI landed at 3.4% headline, 2.5% core, matching consensus. Thursday's July PPI was flat against a 0.2% consensus. Two-year yields fell to 4.145%. Ten-years dropped to 4.636%. The S&P 500 crossed 7,800 for the first time and closed at a record 7,799.

Odds of a September rate hike, the tail risk after Cleveland Fed President Beth Hammack's call for immediate hikes, have compressed sharply. Michael O'Rourke at Jones Trading captured it: rate hike expectations have receded, and both stocks and bonds are getting a strong bid.

The last domino is retail sales this morning. Consensus is +0.2% headline and +0.2% core. Michigan sentiment preliminary follows at 10 ET, with consensus 68.5 versus June's 66.4.

Why this print matters more than usual. The Fed's hawkish camp has one argument left: if the consumer is still spending robustly, the July inflation cool may be a one-off. A retail sales print at or above consensus lets the doves say demand is normalising alongside prices. A hot print above +0.4% hands the hawks fresh ammunition.

The market signal?

A retail sales print between 0.0% and +0.3% is the Goldilocks number that consolidates a Fed pause into a Fed cut narrative for October and December. Above +0.5% and the two-year yield reverses. Below -0.2% and the "K-shaped economy" story gets a lot louder.

Chuck Robbins said the quiet part in a very expensive way

Cisco CEO Chuck Robbins told the Wall Street Journal after Tuesday's print that the company had not experienced this level of demand across all of its portfolio in three decades.

Thirty years. Past the dot-com peak, past the smartphone build-out, past the cloud migration.

The numbers backed him up. Q4 revenue of $17.3 billion, up 18% year-over-year. Networking revenue up 28%. Product orders up 35%. Hyperscaler AI orders of $4 billion in the quarter, $9.3 billion for the full year. FY27 guidance of $72.2-$73.4 billion, a full $6 billion above consensus.

And then Cisco stock fell 8.8%, erasing roughly $29.5 billion in a single session.

The problem was one number: gross margin. Adjusted gross margin fell 210 basis points to 66.3%. Q1 FY27 guidance implied further compression to around 65%. Cisco is winning the AI networking orders. It is selling them at thinner margins than the rest of its business. As AI mix climbs, the aggregate margin profile compresses even if revenue accelerates.

Robbins is not exaggerating the demand. Every chip and infrastructure company is saying versions of the same thing: they cannot build fast enough. But the market has moved on from "who wins the AI order" to "who wins profitably." Those are very different games.

The market signal?

"Three decades of demand" is the bull case in one sentence. A 210 bps margin decline is the bear case in one number. For upcoming earnings from Dell, HPE, Marvell, and Arista, the market will be watching the second number. If margin compression is Cisco-specific, the group re-rates. If it's a category feature, the whole AI hardware trade needs a haircut.

Hormuz is closed. The oil price is falling anyway.

The Strait of Hormuz has been effectively shut since the US-Israel-Iran war erupted on February 28. Only eight vessels transited the strait on Monday, compared with roughly 130 ships per day before the war. The US Strategic Petroleum Reserve has been drawn down below 300 million barrels, the lowest level since January 1983.

And yet Brent settled at $87 on Thursday, down about 2% on the day. WTI closed at $81.25. Both benchmarks broke a five-session rally.

The reason is a demand story quietly overwriting the supply story. The IEA cut its 2026 global oil demand outlook this week, warning that prolonged conflict and elevated prices are eating into consumption. OPEC lowered its 2026 growth forecast to 580,000 barrels per day, the fourth consecutive monthly downward revision. When the demand side softens fast enough, even a chokepoint war stops setting the price.

The diplomatic track keeps stalling on the same disagreement. Iran-Oman-US negotiations for a new Hormuz maritime route are reportedly in "final stages," per Iranian Foreign Minister Abbas Araghchi. But Tehran insists Washington must lift its naval blockade before Iran will fully reopen the strait, and Trump told Axios the US is "only semi-negotiating." Military incidents keep punctuating the talks: an ADNOC vessel was targeted by a missile in the strait on Saturday, and UKMTO reported a cargo ship struck off Oman on Tuesday.

The market signal?

The Hormuz trade has quietly split into two variables: supply risk (still elevated, capped by demand destruction) and deal risk (a signed Iran-Oman route reopens 15-20% of pre-war traffic and takes Brent toward $75). Watch two numbers: Brent's $85 support and daily transit volumes through the strait. A break below $85 without a deal means the demand thesis is winning outright. A signed maritime route means both trades reset at once.

When a blowout is not enough: the AMAT problem

Applied Materials (AMAT) reported after Thursday's bell and the results were, on paper, extraordinary. Q3 revenue of $9.12 billion against $8.99 billion consensus, up 25% year-over-year. Non-GAAP EPS of $3.50 against $3.40 estimate. Record quarterly operating cash flow of $3.04 billion. Non-GAAP gross margin of 50.4%.

Then came the guide. Q4 revenue of $10.25 billion at the midpoint against a $9.54 billion consensus. That's roughly $700 million above the Street on guidance alone. Q4 non-GAAP EPS guidance of $3.82 to $4.22, versus $3.69 consensus. CEO Gary Dickerson used the word "unprecedented" to describe AI-driven demand. Management said 2027 will be another "strong growth year."

AMAT traded down 3.1% in after-hours, at $517.

This is now the pattern of the week. Cisco beat, guided high, dropped 8.8%. Cerebras beat, guided high, fell 14%. Applied Materials beat, guided very high, sold off after hours. Three separate AI-infrastructure companies delivering prints that would have been career-making two years ago, all getting punished by a market that has already priced in the good news.

The math explains it. AMAT trades at roughly 32x forward earnings after a 108% year-to-date run. Cisco entered its print up 60% for 2026. Multiples this stretched assume not just execution but positive surprise.

The market signal?

The AI hardware complex has quietly transitioned from a "growth at any price" trade to a "growth per turn of multiple" trade. Nvidia earnings on August 27 are the referendum. Anything short of a Q3 revenue guide above $60 billion could set off a compression rotation across the semiconductor space.

The leveraged ETF that lost 60% in a month while its underlying went up

Here is a story that belongs on the wall of every investor who thinks 2x and 3x ETFs are just faster versions of the underlying.

The Roundhill Daily 2X Long Memory Semiconductor ETF (DRAL), a leveraged product designed to deliver twice the daily return of the memory chip sector, cratered 60% in a single month through July 31, 2026. During that same window, the underlying memory names it tracks (SanDisk, Micron, SK Hynix, Western Digital) had a rough patch but nothing remotely close to a 60% drawdown. SanDisk itself was down about 20% for the month leading into Thursday's rip. The 2x product lost three times what its underlying did.

This is the compounding tax of daily rebalancing. Leveraged ETFs reset exposure at every close to maintain a constant leverage ratio. When the underlying oscillates, up 5% one day, down 5% the next, the 2x product does not end the week flat. It ends the week down. Over a month of choppy trading, even without a directional trend, a 2x fund can bleed a huge chunk of value while its benchmark barely moves.

Then Thursday happened. SanDisk ripped 15%. But an investor who bought DRAL on July 1 to amplify their memory bet is not thanking anyone this morning. The month of grind ate more capital than the day of glory returned.

This matters because the leveraged ETF category has doubled its assets to roughly $190 billion in the past year. Retail is increasingly using these products as directional bets, without fully understanding the volatility drag baked into daily rebalancing.

The market signal?

When the tape gets volatile, leverage is a tax, not a booster. Anyone holding these products should know exactly what they own heading into an event-heavy weekend.

• July Retail Sales today. A soft print reinforces the Fed pause. A hot print reopens the September hike conversation the CPI and PPI just closed. Last hard data before the September FOMC.
• Michigan Sentiment preliminary at 10 ET, consensus 68.5 versus 66.4. One-year inflation expectations are the sub-line worth watching. A print above 4.0% would be uncomfortable for a Fed trying to declare victory.
• The market signal across both? Every remaining variable that could disrupt the Fed's glide path to a pause is compressed into a 30-hour window ending Sunday night. Position sizing into the weekend deserves an extra look.

That's it for today's Slate.

Have a good weekend, Slaters!

Today's reply prompt: Which of this week's "beat and drop" stocks is the actual buy on this pullback?