Applied Materials: The Profit Machine Is Real, but It Runs on Memory

Applied Materials just reported the strongest quarter in its history: record revenue of $9.12 billion, up 25% from a year earlier; record non-GAAP earnings of $3.50 a share, up 41%; a 50.4% gross margin; and a record 34% operating margin. It was the company's thirteenth consecutive quarter of expanding margins. Management guided the current quarter to about $10.25 billion of revenue and $4.02 of non-GAAP EPS, both above what analysts expected. The stock fell more than 5% in after-hours trading on the news anyway. It is still up about 90% this year and roughly a third below its 52-week high.

The AI profit-machine story everyone is telling about this company is not wrong. What the story leaves out is how concentrated the machine is — and the same concentration producing record profit is what the market is now paying a record price to own.

What the profit machine is made of

Applied does not sell chips, and it does not even mainly sell more wafer capacity to chipmakers. It sells the machinery that deposits, etches, removes, and planarizes the thin films inside a chip and stacks them into three-dimensional structures. When the industry migrates to a new transistor architecture, Applied's revenue per wafer rises even if the total number of wafers does not.

That is the mechanism behind the profit line. SEMI's July forecast expects wafer-fab-equipment spending to rise about 23% this year to roughly $144 billion, with DRAM equipment spending up about 39%. Applied guides its core equipment business up more than 30% this year — an estimate management has kept raising, not cutting. It outgrows its own market because it is concentrated in the fastest-growing slice of it.

Applied's own math explains why the margin line behaves this way: gate-all-around transistors plus backside power delivery raise its revenue per wafer start by about 30%. Gross margin has expanded for thirteen consecutive quarters on that leverage. Margin expansion from unit volume runs out when the fab is full. Margin expansion from selling more process steps per wafer depends on the number of architecture migrations, which is scheduled years into the future.

The growth is not coming from everywhere

The same report that contained the record also contained the tell. China was 26% of Applied's semiconductor equipment and services revenue in the June quarter, down from 35% a year earlier. China is the second market in a company that has split in two, and it is the side that is shrinking — partly because of September 2025 U.S. export controls that Applied has said will cost it $600 million to $710 million, and partly because Chinese foundries increasingly buy domestic tools in the mature nodes that export rules still allow.

The growing side is a short list. Management describes leading-edge logic, DRAM, and advanced packaging as accounting for roughly 80% of incremental equipment spending in 2026 and 2027 — precisely the areas where Applied has chosen to concentrate. In the June quarter, DRAM revenue rose 52% year over year to a record, and the advanced packaging business is guided to grow more than 70% this year. The records are a concentrated bet on three segments, not a broad-market rise.

Why a beat produced a selloff

The drop after the report says that concentration is already in the price. After roughly doubling, Applied traded at about 41 times trailing earnings, and roughly 38 times this fiscal year's guided profit by my calculation from its reported quarters and the midpoint of current-quarter guidance. Lam Research, KLA, and ASML trade at similar or higher multiples, so the valuation is not an Applied-specific problem; it is what the entire leading equipment complex costs when the boom is visible and priced. None of that makes the record fake. It makes the entry price the risk, and the answer depends on the part of the machine most exposed to the cycle.

The engine that can break

The machine runs on memory. DRAM is Applied's fastest-growing segment, and memory is the most cyclical part of semiconductors. On the way up, the shortage produces record tool orders. On the way down, equipment is the first spending line cut, because capital expenditure is the only lever a memory company can pull quickly.

Right now the shortage is real. The three major DRAM producers have reportedly pre-sold their 2027 output, SK Hynix's chief executive warns of the worst memory shortage in 2027 with demand exceeding supply beyond, and Goldman Sachs forecasts DRAM equipment spending rising from $48 billion this year to $72 billion in 2027 and $97 billion in 2028. Applied is the largest process-equipment vendor in DRAM and has gained about ten points of share over the past decade; HBM requires three to four times the wafer starts of standard DRAM for the same number of bits. Every memory capex dollar lands disproportionately in Applied's revenue.

It is also a forecast, and memory has broken every forecast made in a shortage. Korean brokerages have already cut price targets on the memory makers by about a third on the view that the boom peaks in 2027, as new capacity from Korean cleanrooms and Chinese suppliers comes online. This is not a prediction that the cycle turns next year. It is the definition of the swing factor: the durability of the profit machine rests on a market that has never sustained a shortage indefinitely, at a multiple that pays for one.

The condition that decides it

The useful forward question for a shareholder is not whether the machine is real. It is which of two rates slows first: DRAM's growth or China's compression. The quarterly reports supply both in the same release. If DRAM growth holds while memory stays on allocation, the leverage compounds and the records continue. If DRAM growth decelerates while China stays flat or shrinks, a revenue machine growing more than 25% and priced at roughly forty times trailing earnings gets repriced quickly — because the equipment orders that create those records are the first thing cut when a memory shortage turns into memory supply.

What happened on August 13, a record quarter and a falling stock, was the market doing that arithmetic in advance. The records were real. The price was the argument.