KLA printed a record quarter, beat on earnings, guided above consensus and raised its market forecast — and the shares are down 44% from their June peak in under five weeks. The drawdown is real and the multiple has not moved. Estimates rose as fast as the price fell, which means nothing has yet been discounted.
The author holds no position in KLAC as at the date of publication, and held none while the research was produced. No position will be taken for 72 hours after publication.
The author has received no compensation from KLA Corporation or any affiliate, holds no position exceeding 0.5% of the issuer’s share capital, and has no business relationship with the company. This report was not shown to the issuer before publication.
A 44% fall that de-rated nothing. A third-party screen put KLA at 34.2x forward earnings in February. It is roughly 34x today, after the crash. The price fell and the estimates rose by almost exactly the same amount.
The moat is not the question. Roughly 56–60% of process control with no rival above 8%, twenty-year tool lives, and a $3.2bn service book that is about 80% contracted. We score it 7.3 out of 10 — the widest in wafer-fab equipment after ASML.
Cash conversion is the one number that got worse. Free cash flow grew 0.5% on 11.7% revenue growth. The FCF margin fell from about 31% to 27.8% and conversion against net income dropped to 78%. That resolves within two quarters.
KLA closed at $301.71 on 30 June. It was quoted at $169.07 in pre-market trade on 29 July, a fall of roughly 44% in under five weeks, against a Philadelphia Semiconductor Index down more than 20% from its own June record. Nothing in the results explains it: revenue was a record $3.66bn, non-GAAP EPS of $1.05 beat the $1.00 consensus, September-quarter guidance came in above the sell side on both lines, backlog reached an all-time high of about $12.5bn, and management raised its calendar-2026 wafer-fab equipment forecast for the second time in three months.
Here is the detail that should govern how you read the chart. A screen in February put KLA at 34.2 times forward earnings. After a 44% drawdown it trades at roughly 34 times. The de-rating that the price chart appears to show has not actually happened — forward estimates rose almost exactly as fast as the price fell. What looks like a crash is, in valuation terms, a flat line.
Semicap multiples do not compress because earnings fall. They compress months before earnings fall, when estimates stop being raised. The July unwind has been attributed to a hawkish rate backdrop alongside doubts about the fundability of AI capex — which is to say the market has spent the month repricing the discount rate, not the business. At 34x forward, a 100bp move in long rates is worth roughly 15–20% of the equity value before a single estimate changes.
Read the number below as a floor, not as our answer. A ten-year discounted cash flow on a cyclical is dominated by a terminal assumption nobody can defend — terminal value is roughly three quarters of enterprise value here — so we weight the multiples work above it and use the DCF as the sanity check that keeps the band honest. On KLA it is pulling downward, not upward, and we have published it rather than quietly dropping it.
The single most important input is the discount rate. At 7%, the sort of number implied by how quality compounders have actually been priced these past few years, the shares are roughly fair. At 10–11%, which is what a 90%-realised-volatility business arguably deserves, they are not close. Move that slider first.
Terminal value is 73% of enterprise value. This is the low anchor, not the verdict — our band is below.
Cash tax held at 14.5%, the guided September-quarter rate, and constant across cases. Net debt held at $1.0bn — $5.9bn of debt against $4.9bn of cash and securities. The model is a five-year constant-growth form, so it smooths the trough in the bear path and reads slightly generous at the low end. All inputs are derived estimates and should be reconciled against the fiscal-2026 Form 10-K when it files.
A multiple on two-year-out earnings anchors to something observable, which is why it carries more weight here than the terminal assumption above. Consensus has fiscal-2027 EPS at about $4.98.
| Method | Assumption | Per share |
|---|---|---|
| Multiple on FY27E EPS | 22x — a mid-cycle KLA multiple | $110 |
| 25x — a quality premium, roughly the ten-year average for the group’s leaders | $125 | |
| 30x — what the market paid before the July unwind | $149 | |
| 34x — today’s multiple | $169 | |
| Management-target scenario | 15% revenue CAGR to FY30, the mid-point of the company’s own 13–17% target, at a 45% operating margin on 1.22bn shares. Exit at 25x, discounted back four years at 10%. | $128 |
| Same, exit at 30x | $153 | |
| DCF, base case | The model above — 10% WACC, 3% terminal growth, a cyclical down-leg in FY30 | $75 |
| Street | Consensus Moderate Buy. Published targets ranged from roughly $139 to $325 before the print; medians reported between $215 and $280 depending on source and vintage | $215–280 |
At $169.07 the shares sit roughly 9% above the top of that band and 28% above the mid-point. That is not a short thesis. It is a reason to want a better entry. The gap is small enough that a modest overshoot in either direction closes it.
The ten-year model behind the $75 line runs a slightly different revenue path from the five-year form in the widget above, which is why the two differ by a few dollars.
Same model, different inputs. Click any card to load it. The spread is enormous — the bull case is more than three times the bear — and that is not a failure of analysis. It is what happens when the two least defensible assumptions in a cyclical model, the discount rate and the terminal growth rate, are both genuinely contested.
More useful than arguing about what the shares are worth is inverting the model. Holding a 10% cost of capital, 3% terminal growth and a 33% terminal free-cash-flow margin, today’s price implies a 19.7% ten-year revenue compound growth rate — fiscal-2036 revenue above $80bn, roughly six times today, and comfortably above management’s own 13–17% target. Alternatively, holding the base-case cash flows constant, the price implies a 6.3% cost of capital for a business whose shares have moved more than 5% on twenty-five separate occasions in the past year. Neither is impossible. Both are demanding. That is what we mean by expensive.
In a modern fab a wafer passes through somewhere between 700 and 1,500 process steps. At 2nm, a defect measured in nanometres anywhere along that path can scrap a wafer worth tens of thousands of dollars. KLA’s tools sit between process steps, imaging wafers and reticles, measuring film thickness and pattern placement, and feeding the results back into yield-management software. It does not deposit, etch or pattern anything — that is Applied Materials, Lam Research and ASML. KLA measures.
That distinction is the whole commercial story. Process control is a smaller slice of wafer-fab equipment spend than deposition or lithography, but it is the slice where one vendor has accumulated near-monopoly share, because the barrier is not manufacturing scale. It is decades of defect libraries, optics know-how and algorithms — plus the fact that once a tool is qualified into a customer’s recipe, it does not come out.
| Revenue stream | What it is | Scale & character |
|---|---|---|
| Semiconductor process control | Optical and e-beam wafer inspection, reticle inspection, metrology, yield-management software. The core franchise. | The large majority of revenue. Cyclical, tied to leading-edge fab capex. September-quarter mix guided to roughly 73% foundry/logic, 27% memory, of which DRAM is about 90%. |
| Services | Maintenance, spares, upgrades and refurbishment across the installed base. | $820m in FQ4, up 17% — roughly a $3.2bn annual run-rate, about 80% under long-term contract. 57,000-plus installed tools, 4,000-plus customers. Average tool life has stretched from about 10 years to over 20. |
| Specialty semiconductor process | Etch, deposition and packaging tools for power, RF, MEMS and compound semiconductors. | Smaller, and more exposed to automotive and industrial demand — the soft spot in this cycle. |
| PCB, display & component | Largely the Orbotech business acquired in 2019. | The most consumer-cyclical and lowest-margin part of the portfolio. |
| Advanced packagingcross-cutting | Process control for HBM stacks, chiplets, hybrid bonding and panel-level packaging. | Guided to approximately $1.1bn in calendar 2026, up more than 70%. The fastest-growing exposure and the cleanest AI read-through. |
The service line is the underrated half. It is the difference between selling printers and selling ink, and it is what keeps trough earnings power well above where it sat in previous cycles — the load-bearing claim in the bull case, and the one that has not yet been tested through an actual downturn at this revenue scale.
| Metric | FQ4 FY26 | Prior | Why it matters |
|---|---|---|---|
| Revenue | $3.657bn | $3.415bn | A record. Up 7% sequentially and 15% year on year. The absolute level was never the issue. |
| Non-GAAP diluted EPS | $1.05 | $0.94 | Beat the $1.00 consensus, at the upper end of guidance. |
| GAAP diluted EPS | $1.04 | $0.912 | A one-cent gap to non-GAAP. About as clean as large-cap technology gets. |
| Non-GAAP gross margin | 62.4% | 62.0% gd. | Near the top of guidance despite two named headwinds. But guided flat at 62.5% next quarter. |
| Non-GAAP operating margin | 43.7% | — | Best-in-class for capital equipment, and close to a cycle high. That is the valuation problem. |
| Services revenue | $820m | ~$701m | Up 17%, but management placed it near the low end of its own 13–15% long-term range. |
| Free cash flow | $817m | $622m | 22.3% of revenue in the quarter; 28% on a trailing basis, down from 31%. |
| Capital returned | $876m | $875m | $571m of buyback plus $305m of dividends — more than the quarter’s free cash flow. |
| Backlog | ~$12.5bn | record | Roughly three quarters of revenue. The best single argument that fiscal 2027 is already largely booked. |
September-quarter revenue guided to $4.0bn ±$200m against a $3.92bn estimate, and non-GAAP EPS to $1.16 ±$0.10 against $1.14. The midpoint implies roughly 25% year-on-year growth and 10% sequential growth — an acceleration. Calendar-2026 wafer-fab equipment was lifted to the low $150bn range from above $140bn, second-half revenue guided about 20% ahead of the first half, and management said momentum accelerates through the second half and continues into 2027. It was met with a further double-digit decline.
Flat gross margin into a volume ramp. 62.4% achieved, 62.5% guided, on 9% more revenue. On a business with this much operating leverage, that is the number to watch in December. And cash conversion. FY26 free cash flow of $3.77bn was flat against FY25’s $3.75bn while revenue grew 11.7%; conversion against GAAP net income fell to 78%. The benign reading is a working-capital build ahead of a $4bn-plus quarterly run-rate, consistent with the long-lead-time material constraints management flagged. The hostile reading is that inventory is being built against a forecast rather than an order. This resolves within two quarters, one way or the other.
Management tone was candid rather than defensive, which is to their credit. On a beat-and-raise call they volunteered four separate headwinds: DRAM input costs feeding directly into their own cost base at roughly 100bp of gross margin, tariffs at roughly another 100bp and up to $350m a year, long-lead-time material constraints, and the admission that their ability to raise prices to offset increased costs is limited. That last one is the most surprising line on the call. An admission of limited pricing power from a company with 60% segment share suggests either that customers have more leverage than the share data implies, or that contracts are longer-dated than the cost cycle.
The structural point about wafer-fab equipment is that the majors mostly do not compete with each other. ASML owns lithography, Lam and Applied own etch and deposition, Tokyo Electron owns coat/develop, and KLA owns process control. Applied is the only one with meaningful overlap, and despite being the larger company overall it is estimated to hold roughly one-seventh of KLA’s share in the segment. That is not a competitive dynamic. It is a settled outcome both parties stopped contesting seriously.
| Advantage | Score | Our read |
|---|---|---|
| Switching costs | 9/10 | Tools are qualified into a customer’s process recipe at a specific node. Requalifying means re-baselining yield on a production line worth billions. The strongest single element, and why share is sticky through cycles. |
| Distribution & service | 9/10 | 57,000-plus installed tools and applications engineers embedded in fabs worldwide. A competitor must replicate a global field organisation before it can sell the first tool. |
| Customer captivity | 9/10 | Tool-of-record status compounds. Close to structural at the leading edge; weaker at mature nodes, which is precisely where China competes. |
| Scale | 8/10 | In a segment where the input is R&D rather than factory capacity, share is the moat. Self-reinforcing. |
| Intellectual property | 8/10 | The unpatented know-how — defect libraries, illumination, detector design — matters more than the patents. Very hard to copy, impossible to buy. |
| Cost advantage | 6/10 | Scale in R&D and service amortisation, not in manufacturing. Management’s admission of limited pricing power against input inflation is evidence the cost position is not impregnable. |
| Regulation | 5/10 | Export controls entrench KLA at the leading edge by denying Chinese fabs the alternatives — and simultaneously destroy a quarter of its addressable market. Genuinely double-edged; net negative for three years. |
| Network effects | 4/10 | Weak. Some ecosystem effect through fab-wide yield software. Not a meaningful part of the moat, whatever anyone tells you. |
Weighted average approximately 7.3 out of 10 — a wide moat by any reasonable standard, and the second-widest in wafer-fab equipment behind ASML’s EUV monopoly. Our verdict on direction is that it is expanding at the leading edge and eroding at the trailing edge, and stable overall. That is a good moat. It is not a moat that justifies paying any price.
| Metric | KLAC | AMAT | LRCX | ASML |
|---|---|---|---|---|
| Price29 Jul, pre-market | $169.07 | $465.00 | $262.00 | $1,552.63 |
| Year to date | +39.5% | +81.2% | +53.2% | +45.6% |
| From 52-week high | −45% | −37% | −40% | −22% |
| Gross margin | 62.4% | n/v | n/v | 53.0% |
| Trailing P/E | ~45x | n/v | n/v | ~53x |
| Forward P/E | ~34x | ~38x | ~32x | ~39x |
| Process-control share | 56–60% | <8% | n/a | niche |
‘n/v’ means not verified for this report — we would rather leave a cell blank than fill it with a number we cannot stand behind. Peer forward multiples are third-party data of varying vintage and are indicative only; in a tape moving 3–11% a session they are stale within hours. AMAT’s multiple is on a fiscal year ending in October, so it is closer to trailing than forward.
Two readings, and they point opposite ways. Against peers, KLA is fair and arguably slightly cheap — 34x forward sits below ASML and Applied and marginally above Lam, for the best margin profile in the group. If you have decided to own wafer-fab equipment, this is not the expensive way to do it. Against its own history, it is expensive. KLA spent most of the last decade between the mid-teens and high-20s on forward earnings. Thirty-four times is a top-decile multiple, applied to margins that are also top-decile. Cheaper than three expensive things is not cheap.
We ran a deliberately hostile read of the filings, the call and the market data. The result matters less than its shape. Everything in the accounting, disclosure, leverage and dilution columns is clean — genuinely clean, not clean-because-we-did-not-look. A one-cent GAAP-to-non-GAAP gap, a falling share count, an A-rated balance sheet and no litigation overhang is a better governance profile than most of the S&P 500. The serious flags are all versions of the same two risks.
China fell from about 43% of revenue in FY24 to 33% in FY25 and 26% by the December-2025 quarter. Commerce ordered KLA and peers to halt certain shipments to Hua Hong in April 2026. Beijing is pushing chipmakers to source at least half their equipment domestically. On 27 July a Shanghai state-backed firm was reported to have begun producing immersion DUV tools. Every incremental restriction is permanent rather than cyclical.
43.7% operating margin and 62.4% gross margin are at or near cycle highs, capitalised at 45x trailing and 59x trailing free cash flow. This is not an accounting flag — it is the flag. Cyclicals are most dangerous when margin and multiple peak together, because the two compress in the same direction.
78% of GAAP net income converted to free cash flow in FY26, against a 27.8% FCF margin that was ~31% a year earlier. Defensible if the ramp arrives and expensive if it does not. Watch FQ1 and FQ2.
Foundry/logic is guided at roughly 73% of process-control system revenue, and leading-edge foundry is effectively three customers — whose own demand traces back to a handful of hyperscaler capex budgets. The flag is raised partly because management felt the need to pre-empt the question in prepared remarks.
Roughly nineteen sales and zero open-market purchases over the trailing twelve months, all under 10b5-1 plans, which defuses most of it. The flag is the asymmetry: nobody bought the 44% decline. Alignment here runs through compensation, not co-investment.
$571m repurchased in the June quarter at prices between roughly $190 and $300, on a $7bn authorisation approved in March near the cycle peak. Comfortably affordable. The concern is judgement, not solvency: buying stock at 45x trailing converts cash into a multiple bet.
Neither of the two serious risks is a reason to distrust the numbers. Both are reasons to distrust the price. This is a cyclical business priced as a secular one, and a quarter of its historical market is being taken away by policy rather than by competition — the single most likely way KLA loses a quarter of its revenue is a government decision, and roughly that has already happened.
Process control is where China is furthest behind, and KLA’s leading-edge position is genuinely defensible. But Chinese vendors do not need to beat KLA at 2nm. They need to be adequate at 28nm behind a 50% domestic-content mandate. That is a decade-long threat, not a quarterly one — and it is being priced on weekly headlines.
Two consecutive quarters of free-cash-flow conversion above 85% as the inventory build ships; gross margin breaking above 63% in the December quarter rather than sitting at the 62.5% guide; TSMC guiding 2027 capex higher in January, which is the single most important external datapoint for KLA; advanced packaging sustaining 40%-plus growth into 2027; services accelerating to the top of its 13–15% range; or falling long-term rates, which at 34x forward does more work than any of the above.
Any hyperscaler trimming a capex plan; a memory capex reset after the HBM pull-forward; gross margin stuck at 62.5% for a third consecutive quarter; conversion below 80% twice more, which would establish the earnings-quality case outright; further export restrictions or a credible Chinese qualification win at an advanced node; or full-year China revenue landing below 25% in the fiscal-2026 10-K, due within weeks.
We rate KLAC expensive — constructive on the company, patient on the stock. This is a high-quality business with honest management and clean accounts at a genuinely improved entry point, still trading above our fair-value band. We would want to see either the price in the $110–$140 range, or two consecutive quarters of conversion above 85% alongside gross margin breaking 63%, before the risk-reward turns clearly favourable. Neither is a long wait.
Where we differ from the Street is narrower than it looks. Consensus is a Moderate Buy with targets embedding 28–35x multiples on out-year earnings. We are not arguing the analysts are wrong about the business — they are largely right. We are arguing that their targets assume no cyclical down-leg between here and 2028. In a business that has had three revenue declines in the past decade, that is an assumption, not a base case. Our band is what we get when a normal cycle is allowed to happen.
BULLTwenty-five percent year-on-year revenue growth guided for September, backlog at a record $12.5bn, advanced packaging up more than 70%, and management raising its market forecast twice in three months. Consensus has fiscal-2027 EPS up 34%. This is a business accelerating, and the share price is behaving as though the opposite were true.
BEAREvery number you just cited is a 2026 number. My concern is 2028. Backlog of $12.5bn is three quarters of revenue — that is visibility into next June, not next decade. And notice what the raised forecast implies: KLA is telling us the industry is spending at a record rate in the same month the sector index fell 20% because the market doubts that spending is fundable.
BULLThen look at the price. Down 44% in five weeks. At 34x forward it is cheaper than ASML and Applied for a better margin profile and a wider moat. You are being handed the highest-quality asset in the group at a discount to two lower-quality ones.
BEARCheaper than three expensive things is not cheap. Look at the absolute number: 45x trailing, 59x trailing free cash flow, 16x revenue. And here is the detail that should trouble you — a screen in February put the forward multiple at 34.2x. It is 34x today, after the crash. The multiple has not de-rated at all. Nothing has been discounted yet.
BULLSixty percent share, no competitor above eight, twenty-year tool lives, eighty percent of a $3.2bn service book under contract. This is a toll booth on every leading-edge wafer manufactured on earth, and toll booths deserve premium multiples.
BEARI agree with all of it, which is why I am not short. But a toll booth whose road is being rerouted by legislation is a different asset. China was 43% of revenue two years ago and 26% by last December. And then explain the cash: revenue grew 11.7% and free cash flow grew half a percent. One of your two numbers is telling the truth about earnings quality, and it is not the one on the income statement.
CHAIRThe bear has the stronger evidence; the bull has the stronger asset. On the facts placed in front of the committee the bear wins valuation, financials and management, and does so with specific checkable numbers. The bull wins growth and business quality, and wins them convincingly. Catalysts are a draw, because both sides are describing the same events with opposite signs. The committee’s position is constructive on the company and patient on the stock.
What is genuinely uncertain, and we are not going to pretend otherwise: whether the fiscal-2026 working-capital build is a ramp or a misjudgement, which is not knowable today and resolves within two quarters; whether AI infrastructure capex is a decade-long build or a two-year pull-forward, which nobody knows; and the correct discount rate for a business with a 60%-share moat and 90% realised volatility, which moves fair value by a factor of three on its own.
Think of a chip factory as the world’s most demanding bakery. A wafer — a thin disc of silicon — goes through something like a thousand steps before it becomes chips. Get one step slightly wrong and the whole batch is ruined, and each batch can be worth more than a house. KLA doesn’t bake. KLA inspects. Its machines sit between the steps, photographing wafers at unimaginable magnification and looking for flaws too small to see. If something goes wrong at step 300, KLA’s tools catch it before you waste money on steps 301 through 1,000.
Two ways, and the second is the underrated bit. Selling machines — expensive, lumpy, and dependent on whether chipmakers are in a building mood. And servicing them: KLA has over 57,000 tools out in the world, they run for more than twenty years, and roughly 80% of the servicing is on long-term contracts. That brought in $820m last quarter alone and grows whether or not anyone is buying new machines. It’s the difference between selling printers and selling ink.
They didn’t fall because anything went wrong. Revenue was the best ever at $3.66bn, and the company said the next quarter would be around $4bn. But the price had already been bid up so far that a very good quarter wasn’t good enough — and the whole chip sector was falling at the same time. The stock is now down about 44% from its high five weeks ago.
A great company and a great investment are not the same thing. The price you pay decides which one you got. KLA is very clearly a great company. Whether it’s a great investment at $169 depends entirely on what you think the next few years look like — and on that, reasonable people genuinely disagree.
One last thought. Notice how much of this report is about things nobody can know — what AI spending does in 2028, what the right discount rate is, how fast China catches up. Published price targets on this exact stock ranged from about $139 to $325 before the results. That spread is not a failure of analysis. It is an honest reflection of genuine uncertainty, and anyone who tells you they know what this stock is worth to two decimal places is selling you something.