What it does · why you have never heard of it · how big it quietly became
Let me start where I always start: can I explain how this company makes money to a ten-year-old? Here is the whole thing. Everybody knows the companies that design chips — Nvidia, Apple, Intel. Rather fewer know the companies that make them — Taiwan Semiconductor, Samsung. Almost nobody knows the company that makes the machines that the chip-makers use. That is Applied Materials. It sells the equipment that lays films of material a few atoms thick onto a silicon wafer, carves patterns into them, shoots ions into them to make them conduct, and then inspects the result for flaws too small to see. A modern chip is built up in something like a thousand of these steps. Applied has a machine for most of them.
This makes it something I have always liked the shape of: a toll booth. It does not have to guess which chip designer wins. Whether the world's transistors end up inside a Nvidia accelerator, a Samsung memory stack or a car's engine controller, they were almost certainly built on a line that contains Applied's tools. Founded in 1967 in Santa Clara, California, it has grown into the largest supplier of semiconductor manufacturing equipment in the world by revenue. In the fiscal year ended October 2025 it booked $28.4 billion of revenue and $7.0 billion of profit. And two days before this letter, on 13 August 2026, it reported the best quarter in its fifty-nine-year history.
"The single most important decision in evaluating a business is pricing power." — the test I apply here, and the one this company passed in public two days ago
On the call for those results, the chief financial officer said something that would have made me sit forward in my chair: "Three years ago, we implemented a systematic approach to value-based pricing, and today you can see the benefits reflected in our strong revenue growth and gross margins, which have increased to over 50% for the company and over 55% in semiconductor systems." Read that plainly. A company that supplies capital equipment to five or six of the most sophisticated, most powerful buyers on earth decided to charge more for the value it delivers — and its customers paid. That is not a slide in a deck. That is the moat showing up in the gross margin line, which is exactly where a moat is supposed to show up.
Note the date, and note the price. At $507.18 the shares have risen roughly 167% in twelve months, from a 52-week low near $154. They also fell about 31% from a late-June high near $740, and then fell another 5.1% on the day after those record results. The business and the share price are telling two very different stories, and this report is mostly about the space between them.
| Founded | 1967 · Santa Clara, California · listed on NASDAQ |
| Sector / Industry | Technology · Semiconductor manufacturing equipment |
| CEO | Gary Dickerson (since 2013) · CFO Brice Hill |
| What it sells | Deposition, etch, ion implant, thermal, planarisation, metrology & inspection systems — and the service contracts on all of them |
| Revenue (FY2025 · TTM) | $28.37 B · $30.84 B |
| Market capitalisation | ~$402.7 B · ~794 million shares |
The product is explainable · the decade is not
"Never invest in a business you cannot understand." Now, I want to draw a distinction here that matters more for this company than for almost any other on our shelf, because it is the distinction people get wrong. Understanding what a business does and being able to say where it will stand in ten years are two entirely different exercises. Applied Materials is easy on the first and genuinely hard on the second. Here is the chain, end to end:
I want to be candid about the boundary of my own circle, because this is where a reader is most likely to be led astray by a confident tone. I can read this company's accounts and tell you it is a fine business. I cannot tell you what its customers will spend in 2029, and neither can its customers, and neither can the analysts who have raised their price targets by more than $190 in the last year while the business itself did roughly what it said it would do. The demand for these machines is the capital budget of about a dozen corporations. Capital budgets are the most volatile line in commerce. That is not a criticism of Applied — it is the weather this business has always lived in.
Three lines · one installed base · a very concentrated map
Read Applied as one machine business with a service annuity bolted underneath it. The machines are cyclical and magnificent; the annuity is dull and grows every year. FY2025, with the most recent quarter beside it:
Pause on the second of those numbers for a moment, because it is the tell. In December 2025, SEMI forecast 2026 wafer-fab-equipment spending at $126.1 billion. Six months later it forecast $143.9 billion — an upgrade of fourteen per cent to a single year's number, in half a year, on no new physics. The industry's own equipment makers did the same: Applied lifted its calendar-2026 systems growth guidance from "over 20%" in February to "over 30%" in May to "stronger than that" this month. That is what a boom looks like from the inside, and it is exactly the sort of upward-revision spiral that precedes every capital-goods hangover in history. It may run for years. It has never run forever.
Two observations an owner should hold onto. First, the pond is growing and consolidating at the same time — five companies supply most of the world's chip-making equipment, and the technical difficulty of each new generation makes it harder, not easier, for a sixth to appear. Second, and less comfortable: the pond has tides. Applied's revenue went from $16.7bn in FY2018 down to $14.6bn in FY2019 — a 13% fall in a single year, with no recession and no scandal, simply because a dozen customers decided to spend less. Anyone reading today's +25% quarter as a run rate should write that number on the inside of their spectacles.
The verdict, and the reasoning behind it
Now to the question that matters more than any other, and I want to answer it precisely rather than with an adjective. My verdict: the moat is real and it is wide, but it is a moat of breadth, not of monopoly — and that distinction sets both its strength and its ceiling. Let me show my work.
Start with what it is not. Applied does not own a single irreplaceable machine. ASML does — nobody else on earth can make an extreme-ultraviolet lithography system, which is why ASML is not really a competitor at all but a neighbour selling a different, unattainable thing. Applied's position is the opposite shape. In etch it meets Lam Research, which holds more than half that market. In deposition it meets Lam and Tokyo Electron. In inspection and measurement it meets KLA, which dominates process control the way ASML dominates lithography. Step by step, Applied is challenged nearly everywhere. A moat scored on any single product line would come in at six.
| Who | Where they are strongest | How they meet Applied |
|---|---|---|
| ASML | EUV lithography — a genuine monopoly | Not a competitor. Different step, and an unassailable one |
| Lam Research | Dry etch — over 40,000 conductor-etch chambers installed | The direct rival — head to head on the largest steps |
| Tokyo Electron | Coat/develop track (over 90% share), deposition, etch | Broad Japanese rival; strong incumbency at Japanese and Korean fabs |
| KLA | Process control — roughly 63% of metrology & inspection | Applied holds under 8% here, down from a ~13% peak. A step it has already lost |
| Naura, AMEC (China) | Mature-node tools, state-backed | Not yet a technical threat at the leading edge — but they are aimed squarely at the mature-node business that export rules have already pushed Applied out of |
So where does the width actually come from? Four springs, and they compound on one another.
1 · Breadth is itself the product. Applied is present in more distinct process steps than anyone. That lets it sell something no single-step rival can assemble: an integrated solution across several consecutive steps, tuned as one system. When the difficult part of chip-making stops being "make each step better" and becomes "make these five steps work together," breadth converts from a merchandising convenience into a technical advantage. Trend: widening, because the industry is squarely in that phase.
2 · Switching costs that are measured in yield, not in dollars. This is the strongest spring and the least visible. A chip recipe is qualified tool by tool over years. Once a machine is in a production flow that works, swapping it for a competitor's risks the only thing a fab genuinely cannot afford — yield. So the incumbent keeps the step for the life of the node, and usually gets the first look at the next one. Trend: stable, and structurally durable.
3 · The installed base, which quietly became an annuity. Applied Global Services is now running at $1.8 billion a quarter, growing 22%, at a 30% operating margin, with 37,000 chambers in the field connected to Applied's own software. This is the part of the business a cyclical downturn cannot easily take away — a fab that has stopped buying new tools still has to keep the old ones running. Trend: widening, and it is the single most under-appreciated line in the company.
4 · Research at a scale only four companies can fund. Close to $25 billion of R&D over nine years, 12.6% of FY2025 sales, plus an industry-scale research facility in Silicon Valley — the EPIC Center — where customers and now universities co-develop the next generation. You cannot enter this industry with money. You enter it with twenty years of accumulated process knowledge, or not at all.
And the proof, which arrived on 13 August: gross margin above 50% for the company and above 55% in Semiconductor Systems, with the CFO attributing the gain openly to a deliberate programme of charging for value. A supplier without a moat announces price rises to its customers and loses the business. Applied announced them, and the customers paid, and the margin went up 150 basis points year on year. That is the moat, audited.
And the counterweight, which is where the growth is. The physics is currently moving Applied's way, and it can be measured. As transistors move to gate-all-around structures with power delivered from the back of the wafer, management puts the combined transistor-and-wiring opportunity at $14 billion per 100,000 wafer starts, up from $12 billion — roughly 30% more Applied content for the same number of wafers — and expects to take more than half of it. The high-bandwidth memory used in AI accelerators needs three to four times the wafer starts of ordinary memory for the same output; Applied's business there went from about $1bn to $1.5bn in FY2025 with a stated ambition to double it. Advanced packaging is guided to grow more than 70% this calendar year. None of that requires the industry to build more fabs. It only requires chips to get harder to make, which they reliably do.
Where it could crack — and why I stop at 8, not 9. Three places. First, no step is safe individually, and one has already been lost. Applied's share of process control has fallen from a peak near 13% to under 8% while KLA took roughly 63% — a fifteen-year erosion in a segment whose importance is rising. That is not a hypothetical; it is a completed demonstration that this moat can be breached step by step. Second, a large part of the customer base is a political variable — Applied's moat protects it from competitors, not from a rule written in Washington, and in FY2025 two customers alone were about 19% and 15% of revenue, some 34% between them, up from about 23% the year before. (Applied stopped naming them in the FY2025 annual report.) Third, and most important to keep straight: a moat protects share and price. It does not protect volume. Applied can hold every step it owns, keep raising prices, and still see revenue fall 13% because a dozen customers cut their capital budgets — which is precisely what happened in FY2019. Coca-Cola's moat keeps people drinking through a recession. Applied's does not.
The record with the shareholders' money — and what the insiders did this summer
I want managers who are honest, able, and who think like owners. The way to test the third one is not to read the letter; it is to watch what they do with the cash when the stock is cheap and again when it is dear.
Weigh those last two cells against each other, because together they say more than any strategy slide. On the company's money, management showed real discipline — buying back heavily in the cheap years (nearly $6.1bn in FY2022) and pulling back sharply to $440m a quarter now that the shares cost 43× earnings. I score that highly; most managements do exactly the reverse. On their own money, the executives were sellers all summer at $590 to $736. Neither fact is scandalous — the sales were spread over months and Mr Dickerson retains a fortune in the stock — but an honest reader should notice that the people who know this business best were converting paper into cash while the market was paying up, and that the treasurer stopped buying at almost exactly the same moment. I score management 8: an excellent allocation record, and a summer of selling that I decline to explain away.
TTM to 26 Jul 2026 unless noted · from the 15 Aug 2026 data pull
| Metric | Value | Read |
|---|---|---|
| Revenue (FY2016 → FY2025) | $10.8B → $28.4B | ▲ ~11%/yr through two downturns |
| Revenue (TTM) · latest quarter | $30.8B · $9.12B (+25%) | ▲ record quarter, Q4 guided to $10.25B (+51%) |
| Gross margin (decade → today) | 41.7% → 49.4% | ▲ +150bps YoY on deliberate pricing — the moat, audited |
| Operating margin | 30.5% (33.7% in Q3) | ▲ record; 19.9% a decade ago |
| Return on equity (ROE) | 40.4% | ▲ thoroughbred — flattered somewhat by investment gains |
| Return on invested capital (ROIC) | 23.6% | ▲ far above any sane cost of capital |
| Balance sheet | Net cash · Altman-Z 15.4 | ▲ fortress — interest covered 34× |
| R&D / revenue · nine-year total | 12.9% · ~$25B | ▲ the entry barrier, paid for annually |
| Capex / revenue · capex vs depreciation | 7.0% · 5.7× | ◆ spending far above replacement — a growth bet, not upkeep |
| Free cash flow / share vs EPS | $7.87 vs $11.60 | ▼ cash runs BELOW earnings — see valuation |
| Cash conversion cycle | +245 days | ▼ inventory 154 days, receivables 91 — it finances its customers |
| Dividend payout | 12.4% of earnings | ▲ nine straight raises, ample room |
That is a thoroughbred's scorecard, and I do not want to bury the compliment: 40% on equity, half-a-dollar of gross profit on every dollar of sales, more cash than debt, and a company that could pay off every borrowing it has out of one year's operating cash flow. There is no financial risk here worth the name.
But look at the two amber rows near the bottom, because they contain the one thing a casual reader will miss. Applied's free cash flow is running well below its reported earnings — $7.87 a share against $11.60 — and the reason is the +245-day cash conversion cycle. Inventory sits for 154 days; customers take 91 days to pay. To grow revenue 25% you must first fill warehouses and extend credit, and that money leaves before the profit arrives. Now, contrast this with Amazon, whose cycle is minus 51 days — its customers finance the machine. Applied finances its customers. That is not a defect; it is simply what a capital-equipment business is, and the working capital comes flooding back at the top of a cycle when growth stops. But it means that in a boom, the reported earnings are the flattering number and the cash is the honest one — which is the exact opposite of Microsoft, where heavy growth capex made cash flow understate the truth.
Three lenses · the cyclical's trap · and a model that returns a negative number
| Yardstick | Today | On consensus | Read |
|---|---|---|---|
| P/E — reported earnings | 43.7x | ~41x FY26 · ~30x FY27 · ~23x FY28 | the whole case rests on FY27–28 arriving |
| Price / Owner Earnings ◆ | ~58x | — | the honest middle |
| P/FCF — free cash flow | 64.5x | — | harshest — working capital counted |
| EV / EBITDA | 35.4x | — | rich for capital equipment |
| Price / Sales · Price / Book | 13.1x · 15.7x | — | a software multiple on a machine-maker |
| PEG (trailing) | 1.13 | — | the one metric that looks reasonable |
We have run this exercise four times now — Microsoft, Alphabet, Amazon — and every time the lesson was the same: free cash flow understated the truth because growth capex was being charged as though it were upkeep. Applied stands the lesson on its head. Here the cash number is lower than the earnings number, and it deserves to be, because the gap is working capital lent to customers in order to grow. Nothing is being hidden and nothing is being distorted — but a reader who arrives expecting the usual "the P/E overstates the price" conclusion should notice that at Applied, the P/E understates it. Forty-three times earnings is the kindest way to describe this valuation, not the harshest.
Here is the trap that has emptied more pockets than any other in this industry, and it is worth stating without decoration: a cyclical business is most dangerous when it looks cheapest, and most expensive when its earnings are at their best. Applied's earnings are, right now, at the best they have ever been — record revenue, record margins, a quarter guided up 51%. Consensus takes that and extrapolates: $12.29 of earnings this fiscal year, $17.05 next, $21.95 by FY2028 — which is to say revenue rising from $28bn to $52bn in three years. At $507, the shares cost 41× this year, 30× next, and 23× the year after that.
Two honest things can be said about those numbers. The first is that they may well be right: the AI infrastructure build is real, it is contracted, and management says leading-edge logic, DRAM and advanced packaging will supply about 80% of industry growth in 2026 and 2027 — which happens to be exactly where Applied is strongest. The second is that you are being asked to pay 43× for today and to accept the 2028 estimate as the reason it is reasonable. If FY2028 arrives as forecast, today's price is fair-to-good. If the cycle does what cycles do — and this company's revenue fell 13% in FY2019 with no crisis at all — then the market will be applying a much lower multiple to a much lower number, and both halves of that sentence hurt at once. There is no version of this price that contains a margin of safety. That is not a prediction. It is arithmetic.
The rulebook · the cycle · the courtroom that closed · verified 15 Aug 2026
Weigh them in that order, because the first one is not really a risk — it is the price. Nothing about this company's operations looks fragile. What looks fragile is paying a record multiple for record earnings in an industry whose customers' capital budgets are the most cyclical line in commerce.
⚖ THE EXPORT FILE — and it is heavier than the reputation. On 11 February 2026 Applied settled with the US Commerce Department's Bureau of Industry and Security, agreeing to pay $252.5 million. The conduct: 56 separate shipments of ion implanters, manufactured in Massachusetts, routed through South Korea for final assembly, and forwarded to SMIC in China during 2021 and 2022 — roughly $126 million of goods sent where they were not permitted to go. ✅ Importantly, the Department of Justice and the SEC have both notified the company that they closed their related investigations without action, and subpoenas had been outstanding since 2022. So this is resolved, and the penalty is immaterial against $30bn of revenue. What is not immaterial is the pattern it reveals: this company's largest concentration of revenue sits in the one jurisdiction where its own compliance has already failed once, at scale, over an extended period.
🇨🇳 AND THE RULES KEEP MOVING. After Washington widened the blacklist to cover majority-owned affiliates of sanctioned firms, Applied guided to roughly $600 million of lost fiscal-2026 revenue, and Mr Dickerson has said plainly that the company can no longer supply China's memory and older-generation chip-making markets. China's share of revenue duly fell from 37.2% in FY2024 to 30.1% in FY2025, and to about 26% of Semiconductor Systems plus Services revenue in the latest quarter. ⚠️ There is a nuance the headlines miss, and it cuts the other way: management now expects China revenue to rise this calendar year, led by 28-nanometre foundry logic, which remains permitted. Both facts are true. The point for an owner is that neither of them was decided by a customer.
🏭 THE SLOWER THREAT is that the mature-node business Applied has been ordered out of is precisely the business China's state-backed toolmakers, Naura and AMEC, are being funded to take. They are not a technical threat at the leading edge today. They do not need to be: they need only to make the vacated ground permanently unrecoverable.
🧾 A NEAR-TERM OPERATIONAL NOTE that helps explain the 5% fall after record results: management added more than 1,500 people in manufacturing and support in one quarter, opened a new plant in Singapore, and committed to double quarterly system output by 2028. Those ramp costs hold gross margin roughly flat next quarter. That is a good problem — but the market had priced a company with no problems at all.
I spent most of my life avoiding technology companies, and the reason was never that I thought them bad businesses. It was that I could not tell you where they would stand in ten years. I want to open with that confession because Applied Materials tempts a man to forget it. Everything on the surface of this company reads like the sort of enterprise I have spent sixty years looking for: a toll booth that collects from every chip made anywhere, half-a-dollar of gross profit on every dollar of sales, forty per cent on equity, more cash than debt, and a management that has quietly bought back and cancelled twenty-eight per cent of the shares in nine years. If someone described that to me without naming the industry, I would ask what the price was before they finished the sentence.
So let me give the business its due properly, because it has earned it. Two days ago this company reported the best quarter in its history — nine and a tenth billion dollars of revenue, up a quarter on last year, with gross margin above fifty per cent — and then told the market to expect ten and a quarter billion next quarter, up fifty-one per cent. More interesting to me than any of those figures was a sentence from the chief financial officer: three years ago they set about charging deliberately for the value they deliver, and the margin has gone up ever since. Think about what that means. This company raised its prices on the most powerful and most sophisticated buyers in the world, and they paid. There is no better evidence of a moat anywhere in a set of accounts.
And what is the moat, exactly? I want to be precise rather than admiring, because the difference matters. It is breadth, not monopoly. ASML makes a machine nobody else on earth can make; that is a monopoly, and Applied does not have one. Step by step it is challenged — by Lam Research in etch, by Tokyo Electron in deposition, by KLA in inspection. What it has instead is presence in more steps than anyone else, at a moment when the hard part of chip-making has become making consecutive steps work together; switching costs measured not in dollars but in yield, so that once a tool is qualified into a recipe it stays for the life of the node; twenty-five billion dollars of research over nine years, which is the toll for entry and is payable annually; and — the part almost nobody talks about — a service business now running at one and four-fifths billion a quarter, growing at twenty-two per cent, on thirty-seven thousand chambers wired into Applied's own software. That last one is the ballast, and it is why the next downturn will be gentler than the last. I score the moat eight. But I want to say clearly what a moat of this kind does and does not do: it protects your share and your price. It does not protect your volume. In 2019 this company's revenue fell thirteen per cent in a single year — no recession, no scandal, no lost customer — simply because a dozen buyers decided to spend less that year. A moat is a defence against competitors. It is no defence at all against the weather.
Which brings me to the price, and here I must be blunt. The shares are five hundred and seven dollars. Twelve months ago they were a hundred and fifty-four. That is a hundred and sixty-seven per cent, and the business, fine as it is, did not become two and two-thirds times better in a year. At this price you are paying forty-three times reported earnings, about fifty-eight times what I would call owner earnings, and sixty-four times free cash flow — and note the order of those numbers, because it is the reverse of what we found at Microsoft and Amazon. There the cash figure understated the truth; here the earnings figure flatters it, because a boom in this business consumes cash into inventory and customer credit before it delivers a cent of profit. Forty-three times is the kindest way to state this valuation. And it is forty-three times earnings that are themselves the highest they have ever been. The oldest way to lose money in a cyclical business is to pay a high multiple for a high number, and to call the combination growth.
Now, the case against my caution deserves a fair hearing, and it is not weak. Analysts put fair value at six hundred and seventy-one dollars, and the earnings path they are drawing — twelve dollars this year, seventeen next, twenty-two by 2028 — would make today's price twenty-three times the profits of the year after next, which is not extravagant for a company of this quality. Management says leading-edge logic, memory and advanced packaging will supply four-fifths of industry growth for two years, and that is exactly where Applied is strongest. That case may well prove right. But two facts sit beside it that I cannot put down. The first is that those same analysts averaged a three-hundred-and-twenty-seven-dollar target across their whole history and four hundred and seventy-five over the past year: the targets followed the price, they did not lead it. The second is that our own discounted-cash-flow model, fed this data, returns minus three hundred and fifty-three dollars a share — a number so absurd that I publish it only to make the point that nobody, including us, can value this business by formula right now, and anyone who tells you otherwise is selling something.
Two smaller things, and then I will stop. On the credit side, I want to praise something that almost nobody praises: management stopped buying back stock as the price ran. Four hundred and forty million dollars of repurchases last quarter, against an average of one and a fifth billion a quarter in the year before, while the shares doubled. Most managements do precisely the opposite, buying eagerly at the top out of a desire to look confident. This one did not, and paid more out in dividends than it spent on its own shares. That is the behaviour of people who understand what a share is worth. On the debit side, and in the same spirit of not explaining things away: the insiders were sellers all summer — some hundred and sixty-nine million dollars' worth between April and July, none of them buying a single share on the open market, the chief executive alone selling a hundred and five million dollars between five hundred and ninety and seven hundred and thirty-six dollars. He still owns roughly eight hundred million dollars of it, so I would not call that a signal of despair. But when the treasurer stops buying and the officers start selling in the same quarter, an owner should at least write it down.
So where does that leave a thoughtful reader? Exactly where the two halves of Mr Graham's old question always leave one, and I will not pretend to collapse them into an instruction. Is this a wonderful business? Yes — by any test I know, and more so today than five years ago, because the service annuity has made it steadier than its reputation. Does today's price offer any cushion? None whatsoever. At forty-three times peak earnings and sixty-four times cash, the price does not merely assume the boom continues; it assumes the boom continues, the margins keep rising, the politicians stay their hands, and the cycle — which has turned in this industry every few years since 1967 — does not turn. Each of those may happen. All four happening together is what you are paying for. Whether that combination is worth five hundred and seven dollars is not a question I can answer for anybody but myself, and I have deliberately not answered it here. What I can tell you is what the numbers show, what the moat is made of, and where the cushion would have to come from — and that at this price, it is not there. The rest is yours.