Stacy Rasgon, veteran semiconductor analyst at Bernstein Research, talks about the «monster cycle» in the industry, the global race at the leading edge for ever more powerful chips and his bullish outlook for semiconductor equipment stocks like Applied Materials, ASML or Lam Research.
Once again, semiconductor stocks are performing superbly this year. The PHLX Sector Index has gained almost 38% in 2021 – a stellar performance after the benchmark had already advanced more than 50% in 2020.
These impressive gains beg the question of what investors can expect in the coming year. Obviously, the semiconductor industry is experiencing a boom and prices for many components have risen in the wake of global supply bottlenecks. But what happens when the current «monster cycle» turns?
We asked Stacy Rasgon. The managing director and senior equity analyst at Bernstein Research is one of the most respected observers of the semiconductor sector. He has been well versed in the chip industry since earning his doctorate at the Massachusetts Institute of Technology. Under the title «Tales from the Sell Side», he shares his rich experience on Wall Street in a noteworthy series of Twitter threads.
Mr. Rasgon, a frequent guest on financial TV networks like CNBC, first talks about the general cycle in the semiconductor sector and the outlook for the industry in the coming year. Then, he shares his view on wafer fabrication equipment suppliers such as Applied Materials, ASML and Lam Research, for which he is rather bullish.
Mr. Rasgon, tech stocks have experienced quite some volatility recently. What’s your take on the semiconductor sector?
There is a broad debate going on between «peak cycle» vs «stronger for longer». I find myself a little more in the former camp, but this cycle already lasted longer than I thought and general sentiment tends to be swinging back and forth a lot. When Texas Instruments reported earnings in late October, they were basically saying «the peak is here», and everybody started to go bananas. But then, when all the other semiconductor companies reported, it was like «blue skies». In November, we also had several analyst days and everyone was just super bullish. So sentiment swung back the other way, but recently some growth names have been under pressure more broadly.
What makes you think this could be the peak of the cycle?
There are signs that things are starting to turn, with evidence of overshipment and stockpiling. For instance, notebook CPUs – basically the stuff that Intel sells – showed pretty egregious overshipping this year and are already beginning to show a correction. What’s more, there has been some incremental talk of supply chains, while still tight, beginning to show some signs of normalization – and that’s usually a signal of a turn on the way.
Supply bottlenecks have been particularly hard on the auto industry. What is the current situation in this space?
Car manufacturers like GM are talking that they are running overtime shifts. That’s suggesting they’re starting to get parts. When there are shortages and constraints, lead times are stretching and you can’t get the parts you want when you want them. So typically, customer behavior is to order more. It’s similar to the toilet paper-situation at the beginning of the pandemic, and it’s called double ordering. For example, let’s say I’m making widgets and I need one hundred semis, but my vendor says: «Ok, I will give you 10 now, and 90 in a year». In this case, my next step is to order 1000 semis from wherever I can possibly get them. And when I get what I need, I cancel the other orders.
What does this mean for the chip industry in 2022?
This behavior is normal. It’s one of the factors that create cycles, but we’ve never had one that hit every single end market everywhere at the same time. Normally, the actual visibility of the semiconductor management teams with respect to what’s really going on at the customer level in terms of end demand is zero. They have no idea. What they see is the orders in front of their face, and right now orders are off the charts. So we don’t know how much of the demand is real or not until the lead times pull back in.
Why is that?
Because that’s usually when cancellations start. Texas Instruments basically said: «We’re seeing a change in order patterns; customers are no longer asking to expedite everything. They are just asking to expedite the specific components they need to complete their products.» Also, I talked to Analog Devices’ CFO a few days ago, and he said the lead times have stabilized. So they’re not pulling in yet, but they are not getting longer, and that’s the first step. But I don’t know. Pat Gelsinger, Intel’s CEO, was out recently saying that shortages will last into 2023. So maybe the cycle will go on for some time. All I can do is the math, and the math is making me increasingly nervous because in most cases where I can compare end market shipments with semi shipments there is a pretty sizeable gap.
To what degree are semiconductor vendors able to make sure these orders can’t be canceled?
In normal cycles, there are no penalties for overordering. You just cancel the order, but the next time you order parts, you may not get them. Right now, what semiconductor companies are trying to do is to lock customers in, making them sign order contracts with 52-week lead times. But here’s the issue with that: Your customers don’t really know what they are going to need in a year, so they are just ordering parts. And if things go wrong, we will see how non-cancellable these orders are. Are you really going to force customers to take parts they don’t need? You can, but the following year you might not be going to ship them anything.
How does this work in practice?
Broadcom for example is keeping their lead times long. They are parsing the orders, not shipping everything immediately. For instance, if you’re ordering 100 parts, they say: «We think you only need 30 parts, we’re only going to ship you 30 parts.» Microchip Technology is trying to put in committed demand. They’re essentially saying: «Look, we will put you first in line for capacity if you commit orders.» Hence, we need to watch if the demand profile still stays strong when lead times pull back in and supply and demand normalizes. But we just won’t know until then.
Sounds like a risky environment for investors.
I’m not saying «sell everything». I have some ideas people can look for if they want to be in semis considering this risk. For example, Broadcom is a little less risky because they control their semi shipments. In contrast, Texas Instruments, if they can deliver, they’re shipping. They don’t care. They figure: «We ship it today, we ship it tomorrow, what’s the difference? If you integrate over time, it’s the same thing.» They want to have a high availability for their customers as long as this boom lasts. The management of Texas Instruments will just tell you: «Our auto revenue is up 30%, but auto sales are not up 30%, so you can make of that what you will.»
TI’s management is known for excellent execution and not beating around the bush. What’s your take on the stock?
We’re neutral on Texas Instruments, but it’s hard not to fall in love with how they run the company. If they say «we think like owners» they actually mean it. They are getting ready to take CapEx up by a good amount. It’s going to probably hit their free cash flow, but they figure: «Whatever, this is a good time to build fabs if you’re in semis, so we will start building them.»
In the past, cyclical swings in the semiconductor industry have been even more extreme in the equipment space. What’s your view on the wafer fabrication equipment market?
Here’s what’s funny when we talk about all these supply chain issues: Even the semicap guys are getting hit by semiconductor constraints now, since chips go into their tools as well. Applied Materials reported a month ago, and they missed and guided down because of semiconductor constraints. It’s kind of the old saying «the shoemaker’s children have no shoes».
What does this mean regarding the outlook for wafer fabrication equipment stocks?
It’s somewhat different this cycle, because supply is tight and they have long orders. That means they have visibility, and the fundamentals for the semicap space are going to look pretty good next year; especially on the foundry and logic side where companies like TSMC and Intel are going to spend massive amounts of money. So structurally, it’s looking good, and we have all these sovereign investments that are going into these segments. And as I’ve just said, even Texas Instruments and a lot of other manufacturers are going to spend. Memory is a little more of a wild card, since this market tends to be more cyclical.
So what can investors expect in terms of growth next year?
For some time, people have been asking: «When are we going to see a year with $100 billion revenue in wafer fabrication equipment?» I think we will see it next year, and we will have $85 to $90 billion this year. In 2021, revenue is up around 40% compared to last year which is already pretty strong. So next year is going to look really good. People are nervous about 2023, and we’re modeling a little downturn. So we will see about that, but I’m very bullish long-term.
What makes you optimistic about the long-term outlook for wafer fabrication equipment?
Way back in the early days of the industry, a lot of the semi companies used to build their own tools. Today, wafer fabrication equipment is a very specialized industry. There has been a lot of consolidation over the years, both on the large scale and the small scale. But more recently, there have also been M&A attempts that didn’t work. Part of the reason is that we got down to an environment where you have three primary costumers - Intel, TSMC and Samsung - and five primary vendors - Applied Materials, ASML, KLA, Tokyo Electron and Lam Research - which all have their own niches. So they all kind of need each other, and there is a good balance of power. For instance, if a company like Intel says: «I don’t want that deal to happen», it’s hard for that deal to happen.
And what about the cyclicity of the business?
The semicap business is still cyclical, but the cycles have gotten better, just as the semiconductor industry has gotten better and bigger. Also, semiconductor manufacturing is getting ever more capital intensive. In the past, at the trough of the cycle, there was like $20 billion in yearly revenue across all the equipment vendors, and that meant they would bleed cash. More recently, we had a downturn in 2019, and the trough was like $55 billion in yearly revenue. That’s big, and the margins at the trough were way higher than the peak margins used to be. The semicap companies were still generating cash which shows the massive improvement in economics of these businesses. What’s more, they have a much stronger services contribution which is more driven by the installed base and creates more value. So even in a downturn when they are selling fewer tools, their installed base is still going up, since they’re still selling tools. And they can realize more value with services.
How do these broader improvements impact the capital return to investors?
Capital return is really being underrated. These things are cash machines, so their free cash flow margins have just gone through the roof over the years. Since they’re cyclical, cash generation varies a lot over the years. But when you take a five-year average, Applied Materials and Lam Research return around 100% of their free cash flow in the form of dividends and buybacks to shareholders. So basically, you only need to believe two things: Number one, can the underlying semiconductor industry grow over the cycle or not? I think yes. If it grows just at 5% per year over the cycle, it’s a trillion-dollar industry in about ten years. Second, do you think capital intensity over time is going to go up or not? I think absolutely, the answer is yes. That’s all you need to believe - and that’s why I love this space.
For the most advanced semiconductors, big players like Intel now are spending $20 billion or more for a fabrication plant. Can Intel catch up to TSMC and Samsung in the leading-edge chip race?
Nobody else is going to be joining the leading edge, but it’s really hard, even for Intel. Once you fall off the treadmill, I don’t know that I have seen anybody get back on because it’s a very dynamic and fast-moving industry. Intel is going to try, and look what they are doing: They are blowing up their model. Their gross margins are collapsing, CapEx is exploding, and their free cash flow is going to almost zero which is why most people don’t do this. It’s going to be painful, but they know that, they are not stupid. If the goal is to set up Intel so that the company is still here and vital in ten years, they are probably doing the right things.
Today, semiconductor manufacturing is heavily concentrated in Asia. Does it make sense economically to try to bring production back to the United States and Europe?
I don’t think all this push to build those capacities is really shortages driven. If the U.S. still had 37% of the world’s capacity, they would still have the same shortages. It’s the geopolitical agenda: We’re dependent on Taiwan which is a hundred miles offshore from China and they think they own it. That becomes increasingly problematic. But let me say this: The planned $52 billion in federal investments for the domestic semiconductor sector over five years, as proposed in the CHIPS Act, is a rounding error for the industry. It’s not nearly enough to move the needle. They need ten times that amount if the goal is to bring significantly more capacity back on shore. But it’s a start, and you have to start somewhere.
Like Intel, TSMC and Samsung are counting on government support for the construction of new factories in the United States and Europe. Does it make sense to guarantee these companies subsidies even though they are domiciled in Taiwan and South Korea, respectively?
I think so. Intel-CEO Pat Gelsinger has been running around basically saying: «Don’t give the money to the foreigners, give it to us». I disagree, because one thing that TSMC and Samsung have is customers. If they build these fabs here, they can actually start to move stuff, whereas Intel has to build a foundry business from scratch. If the goal is really to get meaningful, useful capacity built in the U.S. and in Europe, I think it’s a mistake to give all the money to Intel. You have to give it to everybody. And it would also be a mistake to not give any to Intel.
The battle for new capacity is also about the future of Moore’s Law, the observation made by Intel co-founder Gordon Moore that the number of transistors on the most advanced chips doubles about every two years. Will Moore’s Law still be applicable in the coming years?
I’m going to grossly simplify here, but a transistor is a switch which you can turn on and off. So Moore’s Law has three components: First, I can turn the switch on and off faster. Second, if I save power, it takes less energy to turn it on and off. And third, I can save costs because I’m squeezing more transistors into an area of silicon. Historically, the cost of that area of silicon didn’t really change that much. That’s why you could get every year twice as many transistors for the same costs or the same number of transistors for half the cost.
And how is it today?
When people say «Moore’s Law is dying», what they really mean is that the cost piece is going out the window. Semiconductor manufacturers can still shrink transistors, and they can do lots of other things to keep getting performance improvements and power improvements. Historically, it was fantastic: Every two years I got better products for less money. Now, if I want those improvements, I have to pay for them. Engineers are smart. If there is a business case to keep scaling, they will find some way to do it, whether it’s new structures, new materials or chiplet packaging. But now, you have to make sure you’re adding value the customer is willing to pay for. That’s the difference. That’s what the «End of Moore’s Law» means.
ASML’s extreme ultraviolet lithography technology plays a key role in producing ever more powerful chips. Is there any competitor that can pose a threat to the Dutch company in this area?
I would be really surprised, because EUV is like black magic. It’s unbelievable, it should not work at all. It actually took a lot longer than everybody thought to move EUV into production because it is so challenging. Just to make the EUV light, they hit a molten tin droplet with a laser and then they hit the same droplet again with the laser. The first hit shapes the drop into a disc, and the second hit strikes with so much power that it becomes a ball of plasma shining with EUV light. They do this 50,000 times a second. I’m just amazed this works.
In recent years, ASML’s stock has seen a phenomenal run. Is the company still an attractive investment at its current valuation?
ASML is its own kind of thing, and my colleague who covers the stock is neutral on it. The stock is extremely expensive because it’s a genuine monopoly. People like to say that TSMC is the most important company in the world. I would argue maybe it’s ASML, because TSMC doesn’t exist without ASML. And they actually know how many tools they can make over the coming years, and they are going to sell everything.
How about the other wafer fabrication equipment stocks?
For the rest of the semicap stocks, the valuations are higher than they used to be, but they are actually quite attractive. They trade at a discount to the semiconductor sector as a whole. You can also compare them to some of the industrials that trade like 50% higher. So looking at the quality of these businesses, I actually think their stocks are very attractive.
And which are your favorite picks in the space?
I cover Applied Materials and Lam Research, and I like them both. When you want to play the semicap sector, the first question is: Do you want to play company specific or the broader sector? If you want to play the sector, you can actually buy the basket. There is a little dispersion between these stocks, but they’re pretty correlated. In terms of company specific, it would be ASML or KLA. The next question is: Do you like memory or do love memory? If you love memory, it’s probably Lam Research. And if you like memory, you can pick any of them.