Some sentences on earnings calls are not planned as news and turn out to be the most important thing said all evening. Andy Jassy delivered one on Thursday. Amazon would not spend roughly $200 billion on capital expenditure in 2026, he said, but roughly $220 billion. And then, in the same breath, the reason: the higher cost of memory is pushing that number up.
For two years the market has read a rising capex number as a demand signal. More capex meant more data centres, more data centres meant more customers, more customers meant profit eventually. The extra twenty billion dollars Amazon added on Thursday night means none of that. It buys not one additional server. It is the surcharge on exactly the same components that cost half as much six months ago.
This is a different finding from Wednesday night, when the market began to distinguish between two kinds of capital spending. The question here is not whether the billions have a meter attached. The question is that part of the billions is no longer investment at all — it is inflation. And because the same price increase showed up on the same evening at the other end of Apple’s income statement, it is possible for the first time to say with some precision where that money is going.
What Amazon actually said
The headline numbers were strong, and they were not the interesting part. Amazon reported second-quarter revenue of $200.6 billion, up 20 percent, crossing two hundred billion dollars in a single quarter for the first time. Operating income rose 43 percent to $27.5 billion. The reported earnings per share of $5.75 should impress nobody: it contains $53.4 billion of non-operating pre-tax income, primarily from the revaluation of the Anthropic stake. That is a mark, not a business, and it is the same mechanism that confused readers of Alphabet’s and Intel’s results earlier in July.
The real event was the cloud division. AWS grew 37 percent to $42.2 billion — a fifth consecutive quarter of accelerating growth, an annualised run rate of $169 billion, operating income of $16.6 billion at a 39.4 percent margin against 32.9 percent a year ago. The backlog stands at $496 billion. Anyone looking for evidence that demand for compute is real received it.
Which is precisely why the aside about capital spending stands out. Jassy did not say Amazon was investing more because it wanted to build more. He said the number was rising because the inputs had become more expensive. And he added something that most summaries dropped: even at $220 billion, the company will not have enough capacity to meet all the demand we have in 2026, and he believes that dynamic will hold in 2027 as well.
You can read that as confirmation of the boom, and the market did: Amazon traded up roughly twelve percent before the open. But it can also be stated more soberly. A company raises its investment budget by a tenth, receives no additional output for it, and still has to turn customers away. In any other industry that would be the description of a cost shock.
The proof sits inside a single set of accounts
A claim like this is normally hard to test, because buyers and sellers are different companies reporting on different bases. In this case there is a shortcut, and it is so clean it almost looks constructed. Samsung Electronics reported for the same quarter on Thursday morning.
Group operating profit came in at 89.5 trillion won against 4.68 trillion a year earlier — an increase of about 1,814 percent, roughly nineteenfold. Revenue rose from 74.57 trillion won to around 171 trillion. Practically all of it came from the semiconductor division: 127.5 trillion won of revenue and 89.2 trillion won of operating profit. That is an operating margin around seventy percent, a figure normally associated with software licences rather than with a physical good manufactured in cleanrooms.
And in the same report, a few lines further down, sits the actual evidence. Samsung’s mobile and networks division posted an operating loss of 0.7 trillion won on revenue of 33.2 trillion. The reason given: rising component costs that wiped out the margin. They are the same components. Samsung sells the memory to itself — and records in one set of accounts what those prices do to a customer who does not own a fab.
Anyone who wants to understand what memory pricing is currently doing to the world economy does not need a model. They need to put two segment lines from the same company side by side. On one, a margin the semiconductor industry has rarely seen. On the other, an established, globally leading handset business pushed into a loss by the very same prices.
How fast prices rose — and the fact that they are already slowing
The magnitude is striking even when you already know it. According to TrendForce, contract prices for conventional DRAM rose 90 to 95 percent quarter over quarter in the first quarter of 2026, with NAND flash up 55 to 60 percent. In the second quarter DRAM added another 58 to 63 percent and NAND 70 to 75. First-quarter DRAM industry revenue jumped 81 percent sequentially. Gartner expects DRAM prices to rise around 130 percent across the full year.
Tim Cook found the phrase for it on Thursday evening that will stick. He described a 100-year flood on the memory pricing, with exponential increases. And he added an observation Apple rarely states that plainly: the DRAM market is currently dominated by just three primary suppliers, and the entry of additional suppliers would be a welcome development, offering potential relief for both the supply chain and pricing structures. Translated out of chief-executive language, that is a customer describing an oligopoly he is captive to.
The continuation of the curve matters more, though, because it is the strongest argument against the obvious investor reaction. For the third quarter TrendForce expects DRAM contract prices to rise only 13 to 18 percent and NAND 10 to 15. That is still a great deal, but it is a fraction of the initial momentum. The increases are cooling because end customers are hitting the limits of affordability. Buying memory makers at record levels today means buying a second derivative that is already falling.
Apple pays the same bill at the other end
At Apple the same number appeared not in the capex line but in the margin — and in the guidance. The June quarter itself was the best in company history: revenue of $109.4 billion, up 16 percent, earnings per share of $2.02 after a 29 percent increase, a record iPhone quarter of $54.25 billion on 22 percent growth, and a Mac business of $10.35 billion against an expected $8.74 billion. Both beat comfortably.
The stock still fell roughly seven percent before the open. Three reasons were cited, and they deserve different amounts of attention. Services came in at $30.74 billion against an expected $31.22 billion. Greater China delivered $18.8 billion against roughly $19.5 billion expected. Those are real soft spots, but they do not explain a seven percent decline after a record quarter.
The third reason does. Apple guided to revenue growth of nine to eleven percent for the September quarter — about $113 billion at the midpoint against a consensus of $114.9 billion. More importantly, it guided to a gross margin of 47 to 48 percent after a reported 50.1 percent, naming memory costs as the primary driver. Cook said the company projects even steeper memory costs into September. Apple has already raised prices on iPad and Mac. And supply of iPhone, Mac and iPad will be constrained during the current quarter.
One detail belongs in the count, because it softens the headline. The reported 50.1 percent gross margin included roughly two percentage points from tariff refunds, as did eleven cents of earnings per share — a consequence of the Supreme Court decision in February that struck down the blanket tariffs. On a comparable basis the margin was closer to 48.1 percent, which makes the guided 47 to 48 percent a considerably smaller step than a comparison of the two headline figures suggests. It was also Tim Cook’s ninetieth and final earnings call as chief executive; John Ternus takes over on September 1.
Why Korea had the biggest day in its market’s history
Where the money is going was answered on Friday morning with unusual clarity. South Korea’s Kospi closed 17.91 percent higher at 6,595.45 — a gain of 1,001.89 points and the largest single-day advance in the index’s history in both points and percentage terms. Buy-side sidecars triggered on the Kospi and the Kosdaq at 9:06 a.m. Samsung Electronics rose 26.81 percent and SK Hynix 29.95 percent, which is precisely the daily limit. Advancers outnumbered decliners 772 to 127.
In the United States the move had already begun on Thursday. Memory names rose by double digits: depending on the timestamp and source, SanDisk gained 18 to 26 percent, Western Digital 15 to 18, Micron 11 to 18 and Seagate 14 to 16, while Lam Research added 17 percent. The broad market followed. The S&P 500 closed 1.66 percent higher at 7,437.63 and the Nasdaq 100 gained 3.36 percent, having slipped into a correction on Wednesday. Microsoft rose 15 percent, its largest one-day gain since 2008 and the largest single-day increase in market value ever recorded by any company.
The Korean move should not be over-read. On Tuesday the same index had fallen 10.84 percent to 6,023.66, with trading halts triggered and both Samsung and SK Hynix down more than fourteen percent. On Friday foreign investors bought a net 7.22 trillion won while retail investors sold a net 8.26 trillion. A meaningful share of the move is short covering and mechanical rebalancing by leveraged funds rather than fresh conviction.
The case against this reading
The strongest objections come from the numbers themselves. First, Amazon is evidently passing the cost through: the AWS margin rose from 32.9 to 39.4 percent. For a business with pricing power the memory bill is a pass-through item. The cost shock is not hitting the cloud providers where it counts.
Second, Cook explicitly characterised the constraint differently from the headlines. It is not a regular supply issue, he said, it is a demand forecast issue — iPhone and Mac are both doing considerably better than expected. A bottleneck created by excess demand is not a macroeconomic warning.
Third, Amazon’s third-quarter revenue guidance of $197 billion to $202 billion is genuinely below the consensus of around $204 billion. The stock rose double digits anyway. When a guidance miss does not matter, that says more about investor positioning than about the business.
Fourth, as described above, the price momentum is already decelerating sharply. Anyone concluding from this analysis that now is the moment to buy memory manufacturers is buying a move that has been running for a while and is losing speed.
What it means for a portfolio
The useful sort is not by sector but by which side of the memory bill a company sits on. Micron is the only remaining American memory manufacturer and therefore the purest expression of the trade; SanDisk, Western Digital and Seagate sit alongside it in storage. On the paying side stand the obvious hardware names — Apple, Dell and HP — whose bills of materials inflate directly and whose end prices adjust with a lag, if at all.
The name most often placed on the wrong side of this line is Nvidia. High-bandwidth memory sits on every accelerator it ships, which makes Nvidia one of the largest memory buyers on earth, not a beneficiary of memory pricing. Its gross margin absorbs the same increase Apple just guided to. Broadcom sits in a similar position on custom silicon. Meanwhile the equipment makers — Lam Research and Applied Materials among them — benefit from the capacity response rather than from the price itself, which is a longer and more durable exposure than the spot margin of a memory maker.
Two practical points for taxable American accounts. Positions of this kind tend to be volatile enough that the difference between short-term and long-term capital gains rates is not a rounding error but a quantified cost of reacting quickly, and the pricing cycle described here is measured in quarters rather than days. And because the memory complex has already moved this far this fast, position sizing matters more than direction: a business earning a seventy percent operating margin on a commodity is, by construction, at a point in the cycle that does not persist.
The bill is moving into consumer prices
The point that outlasts earnings season was made almost in passing. Apple has raised prices on iPad and Mac because of memory costs. Samsung’s handset division is already loss-making. Which means the artificial-intelligence bill, until now a capital-markets story, starts appearing on shelves.
It arrives at a central bank that is not relaxed. The Federal Reserve held at 3.50 to 3.75 percent on Wednesday, but on a nine-to-three vote — three dissents in the same direction calling for an immediate increase, the first such alignment since September 2016. Core PCE came in at 3.3 percent in June, down from 3.4, with a monthly increase of only 0.1 percent. The advance estimate of second-quarter GDP was 1.5 percent against an expected 2.1. The thirty-year Treasury yield sits near 5.24 percent, its highest in decades.
Into that configuration arrives a cost impulse that raises the price of consumer electronics, vehicles and data centres simultaneously. It is not an energy price that falls back as quickly as it rose, and it is not a tariff that can be lifted by decree. Samsung itself warns the shortage may persist into 2028.
What to watch now
One practical rule survives this Friday: read capital spending in units, not in dollars. As long as a capex number rises because more is being built, it is a demand signal. The moment it rises because the same thing costs more, it is a price index with a corporate logo in front of it. At Amazon the difference can be quantified, because management named it. At most other companies in this wave, you have to ask.
So a third figure joins the two this earnings season has already forced into view — backlog and free cash flow. It is this: how much of the increase in capital spending is quantity, and how much is price? The first number tells you whether anyone has ordered. The second tells you what is left after the building. The third tells you whether anything was built at all.
Three things are worth watching specifically. Whether the deceleration in contract prices is confirmed in the third quarter or whether forecasts are revised upward again. Whether more companies follow Apple and raise end-user prices — that is the moment the effect becomes statistically visible. And whether the memory manufacturers convert their extraordinary margins into new capacity. Because that is the mechanism that has historically ended cycles like this one: never falling demand, never policy, but always the fabs approved at the peak of prices that came online two years later.
Try TradingView Free for 30 Days
Plus get a $15 discount on your first subscription through this link.
