2021 WFE Playbook
An overview of the Big 5 Semicap companies
Hey! I originally wanted to write slowly about semicap in 2021, along the lines of deeper primers like I did on lithography or packaging, but for other segments such as Etch and Deposition. Sadly due to TSMC’s aggressive spending plans, I don’t have the time to slowly walk through each of the businesses and the history, science, and future for each segment. We are going to have to focus on the top-down right now and give the SparkNotes overviews of many players in the industry.
What is going on right now is very “hot” information, and the market is pricing in quickly what this means. I want to give you the tools to make your own decisions on how large the market could be given initial top-down guides, and how each category of WFE spend could grow. I will also attach my workbook so you can flex the assumptions yourselves. I think this will be the most value-added part of the analysis - so you can make your own conclusions.
Link to download the WFE Model yourself is here.
Total WFE Estimates and Sensitivities
First, just for simplicity, I am going to break the WFE market into 3 segments. NAND, DRAM, and Logic/Foundry.
Pretty much ignore 2022 estimates, because no one knows what the next year looks like. I just chose ~7% as that's the long term average. You can flex the numbers yourself to get a “turbo WFE” of 75 billion like the likes of Evercore’s recent call (~25% Logic, 35% DRAM, 15% NAND). I am probably biased bullish to ~20% (I think it ends up 22-25%) but want to see 1) Intel and 2) Samsung’s capex numbers to get me there.
Some other reference notes I put in the spreadsheet is a bottom-up sanity check of the big fabs and semiconductor companies. I feel pretty confident walking to ~19% capex growth for the year (in dollars) and there is an upside if Intel and Samsung move capex up.
The biggest things that could change is Intel’s eventual capital spending plan, and Samsung’s report later this month. Samsung initially looks like they will be conservative on NAND spending for the year, but the real question is what will foundry do. My estimate of the foundry is pretty rough - just using Goldman’s memory capex (thanks Twitter!) and assuming the balance is the foundry.
Segment and Semicap Estimates
This gets us to the semicap and top-down estimates segment. There is a *bit* of conservative bias to these, as I used ~15% for service revenue growth and ~10% for display. Note these are all calendarized with the exception of Tokyo Electron.
I tried to put everything into a December calendar year and in dollars. Note the slight conservativism on most of the numbers. I think that is because right now the bottom up and the top down don’t quite match, with the top-down estimate being a bit more aggressive of the two.
I think the thing that you’d have to believe for accelerating growth is that the services revenue gets the same amount of growth (historically this is not the case) as WFE. But broadly on the bottom-up numbers, I feel pretty good about these estimates. I also want to note that I think that for once, Lam might not grow like gangbusters through the cycle. This is largely due to this cycle being driven by DRAM spend, not NAND spends. Lam Research is the clear winner in a strong NAND environment, but logic and most importantly DRAM is the best segment to be long right now. In order of preference in my opinion is DRAM spending, Logic spending, and lastly NAND spending.
This is imperfect - but everything being equal (valuation) you would prefer to invest in ASML, KLAC, AMAT, TEL, and LRCX in that order this cycle. This is odd because if you compare that to history, LRCX has been the super share gainer. Meanwhile, AMAT has been a laggard for the last cycle. This is the basis of the recent Needham downgrade on LRCX and I actually agree with. Many investors (myself included!) lazily assumed LRCX has good memory positioning and thus DRAM positioning and will continue to win share, but this cycle will likely not be NAND driven.
Every cycle is different, and this cycle is driven by the largely appreciated EUV share gains. Secondary to that I would say Advanced Packaging is the real second leg of share gains, but right now in particular that is being expressed mostly in DRAM and Logic. This cycle will likely favor that - if you can get access to Needham, I really liked the thought process of the analyst there. Highly recommend.
I am really looking forward to some of the Advanced Packaging segments within the large companies as well as smaller pure plays like Camtek. PVD for example in AMAT is a great winner but AMAT’s portfolio is so large that there are many puts and takes. Anyways I will briefly discuss the big 5 and what each does.
So like the other company overviews, I want to do a quick overview of what exactly does Applied Materials do? If you want a longer-term primer and history book - I found this one very helpful.
Applied Materials pretty much invented Chemical Vapor Deposition back in the day. Applied was founded by Michael McNeilly who worked at Union Carbide supplying chemicals to the semiconductor companies before striking it out on his own. He believed that they needed new technology to deposit films on wafers, namely through a process called CVD (Chemical Vapor Deposition) and excellent customer service, AMAT entered and created the modern-day semicap industry. They raised money and invented products. Anecdotally the very first CVD was designed by Walter Benzing in a Chinese restaurant over dinner. The rest is history (and a lot of hard work).
The book I have follows Applied from 1968 to 2002, but the core portfolio seems very similar. First, it starts with the AMS tool that focused on the CVD, and then eventually an entrance into the Etch market via the AME 8100 tool. Additionally, they would eventually enter the PVD (Physical Vapor Deposition) Market, and to some extent that is their core business today. Bolt-on a few acquisitions and multiple more product launches and voila! It’s modern-day Applied.
For thinking about this company as an investment, I believe it’s pretty hard to really follow what is going on at a deep level and product by product. I believe it is not worth your time to try to track each product in detail. Applied literally has meaningful market positions in Deposition, Etch, Process Control, Material Removal/Cleaning, CMP, RTP, Ion Implantation, and probably more. This is not even including their display business and likely other segments.
But the thing that probably moves Applied more than anything is Deposition. It makes sense, they invented CVD and are a leader in PVD. In my eyes that is the simplest way to think about Applied. Applied is the company that deposits materials onto a wafer during the critical parts of making a semiconductor.
Valuation and Numbers
Applied today has traded at a slight discount to Lam for the last cycle, and ever so slowly that gap seems to be disappearing. I think that there really is no reason for this.
Frankly, any valuation difference between the companies with the exception of ASML doesn’t seem very justified to me. I am definitely aware that they are now the most expensive they have been in a long time, and unlike say August of last year when you could point to it obviously being way too low on next year numbers - that news to some extent is priced in. More on that in the conclusion.
I think right now it seems that you can still walk to a decent-sized beat on revenue and operating income, but that has to be expected by the buy-side. A reminder this is changed to a calendar year from fiscal years.
Applied actually seems to have the biggest ability to beat in my opinion, just based on relatively naive exposures to their end markets.
Refer to my ASML and Lithography Primer.
I have to say that on top of that - ASML has been pretty robust especially in the incremental margins territory. If they continue to impress I think that is where the upside comes from. ASML’s actual guidance seemed so sluggish considering that EUV will take share and WFE should grow high teens to 20% this year. My very simple model is below.
The real game-changer is gross profit and of course, incremental margin. I assume that incremental margin slightly decelerates - kind of reasonable for most projection periods. This however gets me revenue that is ~3% ahead of the street but EBIT that is ~10% ahead! Everything that could shock to the upside is based on that gross margin and operating income margin.
So let’s go wild - assume earnings are 20% ahead of the street, and the problem becomes clear. That is 44x / 36x 2021 and 2022 earnings. The problem to me on ASML is for it to really warrant a very good long term return, you have to assume that multiple expands. Let’s just say that it does trade at ~40x forward, or 2.5% FCF (once again simplifying) yield. Assume it can grow ~12-15% forever, but your terminal multiple stays the same, you can possibly underwrite a ~high teens IRR but it is very sensitive on that single assumption of the terminal value. I like the business, but I am not super in love with the price. If it even hits ~30x forward think it’s likely a buy. There usually also is almost always an opportunity sometimes for it to be owned at a better price.
Lam Research is best thought of as the leader in Etch. Etch can best be thought of as the process of removing a material that is laid onto a wafer. This is compared to Deposition which is putting material onto a wafer.
In particular, Lam can also be viewed from the point of 3D NAND that shocked the industry capex in 2017. This happened in particular because the single hardest and most challenging part of the race to higher 3D NAND structures was etching deep silicon vias through the multiple layers of NAND.
Look at the stark difference in Lam’s 2017 share gains relative to competitors, and this is the reason they mopped up. Additionally within the supercritical part of 3D NAND etch - they feel confident saying things like this in public.
“We sit here today with more than 90% market share in the critical etch and deposition applications in 3D NAND, and those applications define the road map of the industry.” - Credit Suisse 22nd Annual TMT Conference 2018
Additionally, they have ALD (ASM is a very good name here but the premium price - I like it a lot) and other strong and concentrated bets for share gain. What I like about Lam relative to AMAT, is that there are very specific bets for how they will gain share. In particular their dry resist strategy for EUV, or continuing to benefit from even higher levels of NAND intensity, I think unlike AMAT where they “own everything” - Lam gets the benefit of focus.
Valuation and Numbers
The same graph applies! I think what is interesting is look at the slight discount becomes a premium for Lam, and that is part of what (other than revenue growth) has driven last cycle’s outsized returns. Now Lam is perceived as a share gainer, despite being exposed to arguably worse trends this time (NAND compared to DRAM).
There are still beats here - but it is very driven by their incremental operating profit power.
If they can do ~45% (possible!) then the stock has more juice. But this isn’t the huge outsized beats of yesterday. I think it’s time, to be frank about Lam. Lam likely doesn’t have as good of a cycle this time as last. Longer-term it’s likely positioned fine, as NAND capex intensity will increase as NAND layers increase. There are many ways they can win still, mostly by using their strong footprint to drive higher ALD penetration or other new technologies forward.
KLAC is a bit different than the other WFE players because I think that something they have that others don’t have is a bit less cyclicality. This helps their consistency of results and of course their multiple. The correct way to think about KLAC is that they are more levered to the process control and total throughput of fabs. Process control is mostly the measuring and making sure that the previous steps of Lithography, Deposition, and Etch are done correctly. Below is a list of products, and mostly it is about inspection and metrology, especially for defects.
In addition, there is something a bit more dominant about KLAC than AMAT, Tokyo Electron, or Lam. They really own their end market and it’s pretty sizeable. For a little bit more about the smaller members of metrology, refer to the packaging stocks review. NVMI, CAMT, and ONTO in particular.
So inspection and metrology are mostly levered to throughput and not large capex expansions. It’s likely that this year that they will not grow as much as WFE does broadly, but may tie or even beat WFE over a blended multi-year period.
This is what happened in the last cycle to some extent, but this is partially covered by the acquisition of Orbotech. While many of their competitors struggled during the 2019 downturn if you notice they actually were flat YoY for revenue ex-Orbotech! This points to the durability of the model in particular.
A kind of generally accepted wisdom is that you buy KLAC when you want to be defensive, and AMAT/LRCX (or even UCTT) when you want to be offensive. For longer-term long-only it’s very easy to understand why you’d want to own KLAC over the rest of semicap because volatility should be less than average.
Additionally, their incremental margins and ROE is actually best in class, partially because of the lessened volatility and ability to repurchase more shares than competitors.
Right now KLAC seems “just right” - and frankly will do well over any longer-term time period given its dominant position. I own some and hope to own some more whenever I decide to play full defense out of AMAT / LRCX.
Lastly, let’s talk Tokyo Electron. Tokyo Electron to me is an odd one in the space. First - they are the preferred service provider because they are the easiest to work with. They skew overwhelmingly to Logic exposure, and of course, they are Japanese. For some reason, they seem to never be able to generate the same FCF conversion the American companies do and often keep a decent cash balance. For a Japanese management team, I do give them high marks. For many, this is enough reason to keep away. I think that’s unfair - but I am also in a hard spot feeling particularly bullish on them right now.
The way to think about their exact product portfolio is this slide
I actually ended up talking very extensively about Coating / Developing in the ASML primer. That is how to best think about their exposure - additionally with solid franchises in deposition, etch, and cleaning.
Valuation and Numbers
First - here is my simplistic model for Tokyo Electron. Their logic exposure is pretty good this year, and will likely help drive them to have above-average results.
I think as most semicaps you can walk to pretty easy beats, but everything of course is driven by the incremental operating margin. Notably, I made the assumption that it’s one of the lowest of the group, but this fits their long term average.
Right now Tokyo Electron looks expensive relative to itself. Which is fine - because it’s well-positioned. I have the weakest view on this company relative to the others. If you had a Japanese mandate I could understand how it fits, but to the likes of the rest of semicap, I think it’s not the best risk/reward.
Some Thoughts on Services and Chambers
Something I didn’t really mention but am going to pass on today (this has taken quite a bit of work as is) is the home field advantage of having chambers. Pretty much the largest semicap companies win for a reason (they have taken share) and that’s partially acquiring and then cross-selling into their huge installed base.
If you’re TSMC or the like, you will probably try to not increase your number of relationships and support staff unless you really need something new. The likes of Applied and Lam can likely provide anything a modern fab needs, but often at the expense of convenience over the perfect technical fit.
These service businesses tend to be pretty good too and are much less cyclical than system sales. Often growing even during downturns, and thus muting the cyclicality of the businesses somewhat.
Hence the players with the most chambers and installed base often win. This is the huge strategic benefit that the likes of AMAT, LRCX, ASML, and TEL have over smaller cap companies. And often why if they are acquired the combination is a homerun. Maybe more on this later whenever I update this spreadsheet in the future.
The numbers are good. Revenue is accelerating, there seems to be only incremental good news. But I am hesitant and actually more than ever am looking at the door for the semicap space. My simplistic thought on how to manage the cycle is from the lens of the better framing s-curves piece I wrote before.
We want to be long the companies while they are adding more dollars in revenue this LTM period over last LTM period and then want to not be long the companies when this number peaks out with either a quarter to quarter flat guide or a deceleration. The second they decelerate you need to be out. Below is a chart with dollar revenue increases each quarter for AMAT and share price returns quarterly on a chart. Notice how good it is to buy into a trough - say at around ~2Q19/3Q19. Orange bars rpresent forward estimates - and if you believe they can add more revenue this cycle than last the forward numbers look conservative (I believe that).
Pretty much whenever the amount of revenue slows down, you want to be not long, and whenever the revenue is increasing you want to be long. I think that the rest of the year’s guide is kind of scary, especially if they don’t post strong results in Q3 or Q4 or have a tepid guide. Also, notice how insane the 80% price run is in comparison to periods of the past.
I think there is even line of sight to where I would be short - and that comes with a plateau. Remember that right now we are in no man’s land because these are based on the consensus that likely completely wrong. I would guess 2Q21/3Q21 are materially higher and prove to add more net new dollars.
But the second - and I mean the second that the net new dollars of revenue added flatten out I believe it is time to move on or think about being defensive. This can be shorting your notional via puts (probably what I’d do), selling your stock, or rotating into more defensive names (KLAC). We will address that when we get there - but for now I expect that initial guides will move these numbers much higher and thus remove the possibility of decelerations for a bit. It still looks stretched with this amount of performance this last quarter.
Also for those of you who have no idea what I’m talking about - please comment below and read the piece. It is actually one of the things I’m most proud of writing - as it’s a relatively clever way to chart the 2nd derivative of growth. Anyway, that’s all for today’s update! I’ll update this chart after the actual earnings for AMAT - but most of the large semicap companies look very similar since they all share the same cycle.
Thank you again for this exceptionally lucid overview of an otherwise challenging industry to understand. I wanted to ask about your framing of timing the cycle. Is this much like the cycles you've seen before as far as length and breadth go (ignoring the share prices for a moment)? Is 3Q '21 being the peak potentially very wrong if maybe things are different this time? What do you make the argument that the industry being more consolidated in various sub spaces (like DRAM) so it won't be like previous horrible cycles? Thank you.