Storage Delivers the Best Quarter of the Cycle and Gets Sold Anyway. SanDisk Falls 9% and Western Digital 16%.

Thursday, August 6, 2026, written pre-market. Two storage companies reported the best quarters in the history of their industry last night and both were sold hard. SanDisk earned $39.25 a share on $8.97 billion of revenue at an 84.6 percent gross margin, grew the top line 51 percent in a single quarter, and closed down 5.4 percent before losing another 4.2 percent after hours. Western Digital grew revenue 44 percent to $3.75 billion with record margins, beat by nearly nine percent on earnings, guided the September quarter higher again, and finished roughly 16 percent below where it started the print. The reaction did not stay in New York. Korea opened, the Kospi fell 4.59 percent and tripped a sidecar halt, SK Hynix lost 10.4 percent and Samsung 6.3 percent, and Kioxia dropped more than ten percent in Tokyo. Meanwhile Nvidia is up about three percent because SpaceX named it the exclusive silicon supplier for an orbital data centre constellation. This is not a market rejecting artificial intelligence. It is a market that has started pricing each layer of the stack on a different question, and last night the storage layer got asked the one it could not answer.

Top Call: When Price Is the Whole Delta, the Guide Is the Whole Thesis

The number that matters in the SanDisk release is not the 84.6 percent gross margin or the $6.9 billion of net income. It is the disclosure that roughly $2.01 billion of the sequential revenue gain came from pricing. SanDisk grew revenue 51 percent quarter on quarter and about two thirds of that increase was NAND price rather than NAND volume. That is a magnificent quarter and it is also a fully cyclical one, and the entire equity case for a stock that has risen roughly 500 percent this year rests on how long the price leg runs.

So the guide became the whole report. Fiscal first quarter revenue of $10.3 billion to $10.8 billion against a consensus near $10.8 billion means the street was already sitting at the very top of the range the company was willing to give. Management did not warn, did not cut, and did not signal any deterioration in demand. It simply declined to extend the price curve beyond where the market had already drawn it. For a stock trading at $1,350 on a pricing thesis, an in line guide is not neutral news. It is the first data point that stops confirming the thesis, and the tape treated it accordingly.

Western Digital is the control experiment and it strengthens the read rather than weakening it. WDC beat on both lines, grew full year revenue 36 percent to $12.9 billion, spent only $108 million of capex in the quarter, returned $3.1 billion to shareholders across the year, and guided September revenue to $4.1 billion with a 55 to 56 percent gross margin and $4.00 of earnings. There is nothing wrong with that quarter. The stock lost about 16 percent between the regular session and the after hours print anyway. Two different companies, two different media, hard disk and flash, same reaction. When the second best result in a sector produces the same outcome as the first, the market is not reacting to a company. It is repricing a curve.

Our view is that this is a positioning event dressed as a fundamentals event, and the distinction matters for how long it lasts. Storage was the single most crowded long in global equities coming into this week. Western Digital was up more than 200 percent year to date, SanDisk roughly 500 percent, and Korea had just put in the best month in its index history in July on the same trade. Nothing in either release argues that AI storage demand is slowing. What both releases argue is that the easy part of the pricing cycle, the part where every incremental contract reprices upward and drops straight to the margin line, is now in the numbers rather than ahead of them. That justifies a lower multiple on the same earnings. It does not justify a lower earnings estimate, and we would separate those two things carefully before treating this morning as a demand signal.

The Korean response is the part that deserves genuine caution. A 4.59 percent index decline with a sidecar halt, driven by two stocks, is the third circuit breaker event on that exchange in ten sessions, and it follows a July that produced the best month the index has ever recorded. The memory complex has stopped trading as a collection of companies and started trading as a single levered expression of NAND and DRAM pricing. That works spectacularly on the way up and it removes all diversification on the way down. Anyone who owns Korean semis, Japanese memory and US storage should understand that this morning they own one position in three currencies.

Set against all of that, Nvidia is up three percent, and the reason is worth more than a passing mention. SpaceX has named Nvidia the exclusive compute supplier for Starmind AI1, an orbital data centre satellite built around Rubin GPUs and Vera CPUs, with an FCC filing seeking authority for a constellation of up to one million AI compute satellites between 500 and 2,000 kilometres. There are roughly 15,000 active satellites in orbit today. Prototype testing is targeted for early 2027. We would treat the one million figure as a regulatory ceiling rather than a plan, but the direction of the news is unambiguous. While the memory layer was being marked down on price realisation, the compute layer was handed a demand vector that did not exist in anybody’s model a week ago. That is the split to trade off, and it is a finer cut than last week’s spender versus supplier divide.

Overnight and Pre-Market Developments

Wednesday’s session already showed the rotation. The Dow rose 0.49 percent to 54,349 and closed at a record while the Nasdaq fell 0.83 percent to 26,363 and the S&P 500 slipped 0.17 percent to 7,724, just off Tuesday’s record. This morning the same shape holds and is sharper: Dow futures up 0.17 percent at 54,586, S&P futures up 0.05 percent, and Nasdaq 100 futures down 0.74 percent. Blended second quarter earnings growth for the S&P 500 is running near 47.5 percent, far above historical norms, which is precisely why in line guidance is being punished so violently. When the bar is that high, meeting it is a miss.

Macro: The Labour Market Answers Back

Wednesday’s ADP print of 44,000 was the softest since January, and we wrote that the bond market took it more seriously than equities did. This morning the labour data pushed back. Initial jobless claims came in at 199,000 against 197,000 the prior week, which is not a deteriorating series by any reading. More importantly, Challenger, Gray and Christmas reported 33,429 announced job cuts in July, the lowest monthly total in two years, down 27 percent from June and 46 percent from a year ago, with planned layoffs for the year running 41 percent below the same period in 2025. July hiring plans of 16,095 were the strongest for that month since 2022.

Two details inside the Challenger data are worth holding onto. Technology accounts for almost a third of the 477,033 job cuts announced this year, and artificial intelligence was cited as the leading reason for cuts for the fifth consecutive month, responsible for 10,970 in July alone. So the aggregate labour market is tightening at the same time the sector doing the AI spending is shedding people because of it. That is a composition story rather than a cycle story, and it is exactly the kind of divergence that makes any single payroll print misleading.

Our read is that Wednesday’s ADP was closer to noise and this morning’s data closer to signal, which raises rather than lowers the bar for Friday. July nonfarm payrolls are expected near 83,000 after 57,000, with unemployment at 4.2 to 4.3 percent. If claims sit at 199,000 and announced layoffs are at a two year low, a payroll print in the 40,000 range would be very hard to reconcile with anything other than statistical noise, and the September hike case would survive intact at around 60 percent. The ten year at 4.617 percent and a dollar index at 99.66 suggest the bond market has not yet made up its mind.

The Gold Tell, Revisited

Gold is at $4,309 this morning after a move of close to three percent on Wednesday. We flagged that rally yesterday as a labour market and debasement bid rather than a war bid, on the grounds that bullion should not rally on the morning a chokepoint deal is announced. Today the geopolitical resolution actually arrived, oil fell on it, and gold held its gains. That is consistent with the earlier read and we are inclined to give it more weight now, not less. Gold sitting at a record while the Dow sits at a record and the ten year holds near 4.6 percent describes a market hedging the fiscal and monetary backdrop rather than the Gulf.

Geopolitics: The Route Is Agreed, the Strait Is Not Open

Iran’s foreign ministry spokesman Esmaeil Baqaei confirmed that Tehran and Muscat have agreed the geographical coordinates for a proposed shipping route through the Strait of Hormuz, with a joint statement under final review. The interim structure is now specific: inbound traffic through a northern lane in Iranian territorial waters, outbound through a southern lane in Omani waters, no tolls or fees during the temporary period, and naval mines cleared from the median lane within thirty days. The route would run for roughly two to four months and does not constitute a full reopening.

Brent fell 0.5 percent to $79.08 and WTI 0.7 percent to $74.69. That is a restrained reaction to an agreement actually being reached, and it confirms what we argued yesterday. The deal was already in the price, so the news is worth a few dollars at most while a collapse is worth considerably more in the other direction. Three things still separate a signed route from an open strait. Iran maintains that no agreement guarantees safe passage while the United States blockades its ports. The Houthis claimed a strike on a Saudi tanker in the Gulf of Aden, a second corridor that no Hormuz text addresses. And eight vessels transited on Monday against roughly 130 a day before the crisis, a shortfall that thirty days of mine clearing does not close by announcement.

The fee question, which was the single hardest obstacle a week ago, has not been resolved. It has been handed to a joint working group between the two foreign ministries. That is how deals get signed rather than how disputes get settled, and it means the durability of this arrangement is untested. Separately, President Trump’s deadline for a Russia and Ukraine agreement falls on Saturday, August 8, with tariffs and other measures threatened against Moscow absent progress. Energy desks should not treat the Hormuz headline as the only live geopolitical clock this week.

Quick Take: The Earnings That Matter

SanDisk (SNDK), fiscal Q4, reported after Wednesday’s close

Our Take: The operating result is one of the great quarters of this cycle and the guide undermines the only variable the stock was trading on. With two thirds of sequential growth coming from price rather than volume, an unchanged forward range is the market’s first evidence that NAND pricing has reached the level management is willing to underwrite. We do not read this as demand weakness and we would not extrapolate it into 2027 units. We do read it as the end of the phase where storage earns a growth multiple on a commodity price move.

Western Digital (WDC), fiscal Q4, reported after Wednesday’s close

Our Take: This is the cleanest evidence that the selling is about positioning rather than results. WDC beat, raised, ran record margins on $108 million of quarterly capex, and returned $3.1 billion to holders across the year, and it still lost roughly a sixth of its value across the session and the after hours print. A stock up more than 200 percent this year needs an acceleration to hold its multiple, and a good quarter is no longer an acceleration.

HubSpot (HUBS), Q2, reported after Wednesday’s close

Our Take: The temptation is to call a 24 percent decline on a one point growth cut an overreaction, and we think that misreads which number broke. The customer add shortfall came from a deliberate shift to trials and outcome-based AI pricing that lengthened sales cycles, plus tighter software budgets pushing deals to board level. That is a transition cost of uncertain duration, and it is the second software name this week to show the AI pricing model compressing near term revenue before it expands it. We would treat elongating enterprise software sales cycles as the more important cross read here rather than anything HubSpot specific.

Constellation Energy (CEG), Q2, reported after Wednesday’s close

Our Take: On a morning when silicon is being repriced, the power leg of the AI trade raised guidance and added 920 megawatts of contracted demand. That matters because power purchase agreements are multi year commitments from the same hyperscalers whose spending everyone is questioning. If the buildout were genuinely decelerating, this is the line where it would show up first, and it did not.

Asia and Europe

Asia absorbed the storage damage almost entirely on its own. The Kospi fell 302.82 points or 4.59 percent to 6,295.44 with a sidecar suspension triggered on panic selling, SK Hynix closed down 10.37 percent at 1,495,000 won and Samsung Electronics fell 6.30 percent to 230,500 won. The Kosdaq held comparatively firm, which tells you the selling was concentrated in the memory complex rather than broad. Japan’s Nikkei fell 0.93 percent to 65,206 with Kioxia down 10.2 percent, Taiyo Yuden down 11 percent and Tokyo Electron down 5.5 percent, though cheaper oil on the Hormuz news cushioned the broader index. Hong Kong’s Hang Seng fell 1.49 percent to 25,585 on profit taking in technology and financials, with the mainland’s new 20 percent tax on offshore insurance investment gains adding pressure to the financial names. Mainland China went the other way. The Shanghai Composite rose 0.57 percent to 3,900.35 as gold miners including Zijin, Shandong Gold and Zhongjin Gold offset semiconductor weakness, while the CSI 300 was fractionally lower at 4,651.

Europe remains the quiet outperformer. The Stoxx 600 closed at another record near 657 on Wednesday for a third consecutive advance and is up roughly ten percent year to date. The DAX traded above 26,100 for the first time and sits at 26,191, France’s CAC 40 reached a record 8,725 and Italy’s FTSE MIB an unprecedented 53,540, with the FTSE 100 at 10,897. The detail worth noting is that the five best performing European stocks of 2026 are all semiconductor related: Soitec up 371 percent, AT&S up 330 percent, Technoprobe up 123 percent, Aixtron up 116 percent and STMicroelectronics up 101 percent. European equities are not a defensive alternative to the AI trade. They are a levered participant in it that happens to have avoided the memory concentration.

Sector Spotlights

Semiconductors and storage. The complex has bifurcated inside a single session. Memory and storage are being derated on pricing visibility while compute is being rerated on new demand. Nvidia up three percent on Starmind against SK Hynix down ten percent is the entire thesis in two prices. We would be slower to buy the memory dip than the violence of the tape suggests, because the correction here is to a multiple built on a price curve, and multiples take longer to rebuild than earnings do.

Software. Datadog down 20 percent and HubSpot down 24 percent on the same morning, both after beating, is the second cluster of AI-adjacent software disappointments in two weeks. The common thread is that AI is changing how these companies charge before it changes how much they collect. Consumption and outcome-based pricing lengthens sales cycles and defers revenue recognition, and the market is not yet willing to fund that transition at a premium multiple.

Energy and power. Crude is lower on the Hormuz agreement with Brent under $79 and WTI under $75, but the more interesting energy story is electricity. Constellation raised guidance and signed 920 megawatts of new long term agreements, three days after Caterpillar reported a 29 percent increase in data centre power generation. The power leg of the buildout keeps confirming while the silicon leg gets questioned, and that asymmetry has now held for a full week. ConocoPhillips reports pre-market with consensus near $2.89 on about $19.8 billion of revenue, against a quarter where WTI averaged above $100 in April and May before falling to roughly $85 in June.

Consumer and credit. Underneath the AI noise, the genuinely weak prints are consumer facing. Peloton guided revenue light, UWM Holdings missed and suspended its dividend, and Cummins missed on margin. A wholesale mortgage lender suspending a dividend in a quarter when jobless claims are at 199,000 is not a labour market signal, it is a rate and housing affordability signal, and it deserves more attention than it will get today.

Tonight and Tomorrow

Airbnb reports after the close with company guidance of $3.54 to $3.60 billion against consensus near $3.58 billion and $1.25 of earnings, with the stock at $152.49 and 2.6 percent below its 52 week high. The setup is unfavourable. A print that merely meets guidance rarely extends a rally that already embeds resilient travel demand, and the stock fell more than seven percent on a beat a year ago. DraftKings and Celsius Holdings also report tonight.

The SpaceX lock-up opened this morning, freeing up to 911.5 million shares worth roughly $101 billion, the first 20 percent tranche of a staggered nine stage release. A second tranche of 319 million shares follows on August 12, with Musk and selected holders locked until June 2027. The stock is already about 17 percent below its $135 issue price after Tuesday’s capex reaction, which means insiders considering a sale are doing so into weakness rather than strength. The staggered structure was designed to prevent exactly the outcome the capex headline has made more likely.

Friday’s July employment report at 8:30am is the week. Consensus sits near 83,000 after 57,000, and this morning’s claims and Challenger data have raised the bar for a genuinely soft print to be credible.

Key Events Today and Ahead