Investment Gems

Briefing

• On July 23 I published five specific, falsifiable things to watch on Lam’s July 29 print, and I published them because Lam reads the AI memory cycle about four steps upstream of the neoclouds that report a month later. Four of the five resolved, all bullish, including the one I named as the trip-wire. One is still open.
• The print itself: revenue $6.72B against a $6.60B guide, EPS $1.82 against a $1.65 guide, gross margin 52% which is the highest quarterly level in twenty years, FY2026 revenue $23.2B and a fourth consecutive record quarter. Industry equipment spending for 2027 was raised to $140 to $145 billion. Advanced packaging came into the year guided above 40% growth, got raised to above 50%, and is now tracking above 70%.
• The line I actually cared about was none of those. Customer advance deposits had fallen three quarters running, from $2.7B to $2.25B to $2.22B, and I wrote in the July 25 piece that this is the first thing that softens if an order book is quietly rolling over behind a strong revenue headline. It rebuilt to $2.43B, up $213 million.
• And the stock fell 7.04% in that regular session, to $250.63, before recovering 5.31% after hours. Eighty three semiconductor and optics names sat at 20-day lows the same day, with a median of about -19% on the week.
• So the question this piece exists to answer is not “was the quarter good.” It was good. The question is why the equity did not care, and whether that gap is an opportunity or a warning. My answer is that Lam is being sold with a group it does not belong to, and that is the argument for the upgrade.

What I published, and when

I want to be precise about the timestamp, because the whole value of this exercise is that it was written before the answer existed.

On July 25, four days ahead of the print, I published a piece arguing that Lam’s July 29 report was the earliest available read on over-build risk across the whole AI compute chain, and therefore a free look at the one thing that would de-rate the neocloud complex independently of any individual operator’s execution. In it I listed five things I would be watching, with the bullish and bearish resolution of each written out in advance, and I named which one was the trip-wire.

Doing it that way is uncomfortable on purpose. A watch list published after the event is a summary. Published before it, it is a bet you have to settle in public. So here is the settlement.

The grade

#	What I said I would watch	The baseline going in	What it did	Grade
1	HBM and advanced packaging tone. The trip-wire. Any language that HBM capacity was getting ahead of accelerator demand, which is Lam’s own stated break	Packaging guided above 50% growth for 2026	Raised again, now tracking above 70% year over year. DRAM hit a record 27% of systems revenue on HBM. No over-build language anywhere I could find	✅ Holds, and strengthened
2	Japan acceptance-pending inventory. Shipped, not yet recognised	$131M, then $226M, then $434M at March. A widening wedge	Unresolved. The figure was not in the release or in any coverage I could reach	⏳ Open
3	Customer advance deposits. The early-warning line	$2.7B to $2.25B to $2.22B. Bleeding three quarters	Rebuilt to $2.43B, up $213 million	✅ Holds, and this is the one that matters
4	2027 equipment spending framing	$135B raised to $140B “with a bias to the upside”	Raised to $140 to $145B, with 2027 framed as another year of compelling growth as greenfield fabs begin tooling	✅ Holds, upgraded
5	NAND conversion timing	About $40B pulled into calendar 2027	Pulled forward further. The majority of that spend is now expected before the end of calendar 2027	✅ Holds, strengthened

Four of five, all in the same direction, including the trip-wire. One open.

Let me be honest about what that does and does not prove. Four of these were, in effect, “does the good thing keep being good,” and momentum in an upcycle is not a hard test. The one that was a genuine test was item 3, because it was the only line already moving against me. It had declined for three straight quarters and I wrote down in advance that it was the thing I would treat as the tell. It reversed. That is worth more than the record revenue headline, because a record revenue headline is a report on the last ninety days and a deposit rebuild is a report on the next ones.

The one I have not settled

Item 2 is open and I am not going to paper over it, because it is the item with the most interesting downside.

Lam holds Japanese shipments in inventory at cost until the customer formally accepts them. That inventory went $131M to $226M to $434M across three quarters. Tools delivered, revenue not yet recognised. Two readings: demand so strong that the acceptance queue is backing up, or a queue that is backing up because customers are in no hurry to accept.

The June figure was not in the release and I could not source it. So the fair grade is incomplete, and the FY2026 10-K, which should land within a few weeks based on last year’s August 11 filing, is where it settles. If that number kept climbing while everything else improved, it is the one wedge in this print worth worrying about, and I will say so when I see it.

I am flagging this partly because it is the item nobody else tracks and partly because a watch list where every item conveniently resolves in your favour should make a reader suspicious of the person who wrote it.

The tension: the quarter was excellent and the stock fell 7%

Record revenue. The best gross margin in twenty years. Equipment spending raised for next year. Packaging raised twice. The early-warning line reversed. And the shares traded down 7.04% in the regular session, then recovered a little over 5% after hours once the release was actually out.

The sequencing there is worth noticing. The drop preceded the release. The recovery followed it. Whatever sold Lam that day was not reading Lam’s numbers.

What it was reading is visible if you stop looking at the ticker and look at the cross-section. On the same date, my own screen across roughly 940 names had 83 semiconductor and optics names sitting at 20-day lows with a median of about -19% on the week, and 17 semiconductor equipment names down about -37% on the month. Meanwhile software was up on the month. Both of those groups sell into the same AI build.

A demand scare does not sort a market that way. A demand scare takes the customers down with the suppliers.

The half of this I think is being missed

Here is the distinction I keep coming back to, and it is the actual reason for the rating change.

What is being liquidated across this complex is not weak fundamentals, it is dependence on the primary market to finish a build. Sort companies by whether they still need to go out and raise capital to complete what they have started, and the July tape sorts almost perfectly. The neoclouds, the datacenter builders, the power developers, pre-revenue biotech, anything mid-construction with a funding gap: all at lows. Companies that fund themselves out of their own cash flow: fine. Railroads, which are the most asset-heavy business that exists, are up on the month. It is not capital intensity that is being punished. It is the need to go and get more capital.

Now apply that test to Lam.

Capital expenditure was about 5.7% of revenue last quarter, $332 million on $5.84 billion. Nine month capex was $778 million on $16.5 billion of revenue. The company carries roughly $1.0 billion of net cash, retired its near-term debt (the current portion went from $754.3 million to $4.1 million), and over nine months returned $3.6 billion in buybacks and $945 million in dividends against $4.4 billion of operating cash flow. Lam does not build fabs. It builds chambers and hires field engineers.

Lam is not a capital consumer. It is being sold as one, because it is priced off the same theme. It sits in the same ETF, the same screen, the same mental bucket. When a sector is de-rated on funding conditions, the self-funder inside it gets marked down with everything else, and that mispricing is the entire opportunity here.

I hold the loose version of this view about the complex generally and the strict version about Lam specifically, because Lam is the one name where the balance sheet argument is not a judgement call.

Valuation: same estimates, lower multiple

FY2027 consensus is $8.11 in earnings per share, on a Street range of $7.25 to $10.03 across 32 analysts. That band is 38% wide, which tells you the disagreement is about the amplitude of the cycle rather than about whether Lam can execute.

At $319.29 on July 22, Lam traded at 39.4 times that consensus number. At $265.75, it trades at about 32.8 times the same number. At the intraday low of $250.63 it was about 30.9 times.

Nothing in the estimate moved. In fact the June quarter beat its own guide by seventeen cents, industry spending for 2027 got raised, and packaging growth got raised, which makes an upward revision to that $8.11 more likely than a downward one. If FY2027 consensus drifts up, the multiple falls further without the stock doing anything.

That is a materially different payoff shape than the one I wrote about a week ago:

Case	Earnings basis	Multiple	Implied	From $265.75
Aspirational	FY27 Street high, $10.03	44.7x, the sector’s top rung	$448	+69%
Base, own multiple restored	FY27 consensus, $8.11	39.4x, its own multiple a week ago	$320	+20%
Base, own multiple as-is	FY27 consensus, $8.11	32.8x, where it trades now	$266	0%
Bear	FY27 Street low, $7.25	32.6x	$236	-11%
Floor	FY27 consensus, $8.11	23.5x, Lam’s own five year average	$191	-28%

The comparison I trust without a refresh is the middle of the table, because it is Lam against itself. A week ago the honest read was up 40% if everything broke right, zero if consensus was simply correct, and down 40% on a pure multiple reversion. Now consensus being simply correct, with nothing more than the multiple returning to where it stood eight sessions ago, is worth about 20%, and the floor case costs 28%.

That is what changed my rating. Not the record quarter. The record quarter was mostly expected. What changed is that the market handed back roughly a sixth of the price while the evidence for the earnings moved the other way, and it did it for a reason that is about the sector’s funding conditions rather than about this company’s cash flows.

The bear case that survives all of this

I would rather state the strongest version against me than the convenient one.

The multiple is still not cheap in absolute terms. Something above 30 times forward earnings for a cyclical toolmaker is not a value proposition, it is a growth proposition, and it needs the cycle to keep going. Lam’s own five year average multiple is around 23.5 times and its 2022 trough was near 13.8 times. Both of those were observed inside the last five years. On unchanged estimates the first is 28% below here and the second is a different conversation entirely.

A confirmed backdrop does not stop a de-rating. This is the part I got wrong in the framing a week ago, and I will own it: I treated the print as the resolving event. It resolved, favourably, and the stock fell. Being right about the fundamentals and wrong about what the tape was pricing is a specific kind of error and I made it. If the whole complex keeps being sold on funding conditions, a self-funder inside it can stay marked down for longer than the argument says it should.

And the awkward one. Equipment spending guidance keeps rising while the equipment equities keep falling. Those two things cannot both be information. Either the equities are wrong, which is my position, or the guidance is a lagging indicator of orders that have not been cut yet. I hold my view, but I do not hold it comfortably, and if the next quarter’s spending framing gets trimmed even slightly that is the argument arriving.

Plus the China vector I wrote about last week, unchanged: 34% of revenue on the demand side under export controls, and on the supply side a rare earth export licensing regime suspended only in part until November 2026 unless extended. Nothing in this print altered either.

What would change my mind

Written on the thesis, not on the price, because the price is the symptom.

1. The Japan acceptance-pending line keeps climbing in the 10-K. That is the one open item and it speaks directly to whether shipped tools are converting.
2. The deposit line rolls back over next quarter. One rebuild is a data point. Two consecutive would make it a trend and one reversal would make the rebuild a timing artefact.
3. 2027 equipment spending framing gets trimmed. Not cut, trimmed. The bias-to-the-upside language disappearing is the early version.
4. Any HBM-ahead-of-demand language. This was the trip-wire going in and it stays the trip-wire going forward.

Note what is not on that list: the share price. A 15% bounce in Lam tells me nothing about any of the four, and neither does another 15% down.

The next watch list

Same discipline, dated forward, so this piece is gradeable too.

1. FY2026 10-K, expected within weeks. The Japan acceptance-pending figure against $434 million at March. Down means acceptance is flowing. Up again means the wedge across the whole chain is still widening.
2. The 10-K’s deposit and deferred revenue detail, to confirm the $213 million rebuild is fresh orders rather than reclassification.
3. The neocloud prints in late August. This whole exercise existed to read the backdrop ahead of them. Backdrop confirmed means their risk is now their own commissioning and customer acceptance, not the chip cycle. If one of them misses on acceptance while the chip cycle is visibly accelerating, that is the cleanest available proof that the two risks are separate, which is the argument I have been making all month.
4. The next hyperscaler capex guidance round. If it comes in higher, the funding story that de-rated this whole group was misread, mine included, and Lam re-rates fastest of anything in the group because it never had a funding problem to begin with.

Rating

I am moving to Buy from Hold.

The July 23 piece was an explicitly conditional Hold with a stated trigger, and I named two levels at which the conditionality would resolve toward buying: around $290, which was Applied Materials’ multiple applied to Lam’s consensus, and around $259 on the mild bear row. The stock printed $250.63 intraday and closed the after-hours session at $265.75. Both levels came, the trigger event landed, and the trigger event resolved four of five items in the constructive direction with the fifth still open.

Holding on after all of that would mean the conditions were decoration. So the rating moves, and the honest framing of the upgrade is this: the business got better, the price got worse, and the reason the price got worse is a sector-wide funding constraint that this particular balance sheet does not have.

The risk I am accepting in exchange is that a de-rating does not care what I think about a balance sheet, and can run considerably longer than the argument justifies. Size accordingly. I would rather be adding into that than waiting for the tape to agree with me first, but I am not pretending the tape agrees with me now.