«This Is an Inflection Point in the Semiconductor Cycle»
Stacy Rasgon, veteran semiconductor analyst at Bernstein Research, talks about the outlook for the chip sector, the impact of artificial intelligence on the industry, and his take on behemoths such as Nvidia, Broadcom, Intel and Qualcomm.
Semiconductor stocks are celebrating an impressive comeback. The PHLX Semiconductor Sector Index has advanced 43% since the beginning of the year, leaving other industries such as software or the Internet behind.
The strong performance is in striking discrepancy to the fundamentals. According to the industry association SIA, global semiconductor sales plunged more than 20% year-over-year in May. The sector is currently experiencing its worst downturn since the Global Financial Crisis of 2008-09.
The outlook for the sector will become more visible in a few days when the second quarter results are coming in. In the lead-up to the earnings season, we therefore talked to 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 hashtag «Tales from the Sell Side», he shares his professional experience on Wall Street in a noteworthy series of Twitter threads.
«If you’re playing that back half recovery, we better get it»: Stacy Rasgon.
In an in-dept interview with The Market NZZ, Mr. Rasgon, who is known for straight talk and a frequent guest on financial TV networks like CNBC, shares his take on the state of the semiconductor cycle. He also comments on the boom in artificial intelligence and on the outlook for industry behemoths such as Nvidia, Broadcom, Intel and Qualcomm. The article has been edited for length and clarity.
Semiconductor stocks have enjoyed strong gains so far this year. How do you assess the outlook for the sector?
Investors have been playing the sector on the thesis that earnings estimates are bottoming and are looking for a recovery in the back half of the year. That’s not atypical. Semiconductor stocks tend to be anticipatory; they tend to bottom just before earnings estimates bottom. In this regard, forward estimates peaked in June ‘22, and now they are down close to 40%, which is the largest such revision in over a decade. So investors playing that bottoming theme explains some of the strength in semi stocks, and clearly there has also been some AI driven strength.
Where do you think semi stocks are going from here?
I think this is sort of an inflection point in the cycle. Investors are growing excited again, stocks have all rallied, and the space is trading at very elevated multiples. This means expectations for a year-end above-seasonal snapback are set high. Yet, inventories are extremely elevated everywhere you look. So if you’re playing that back half recovery, we better get it.
What would you advise when it comes to investments?
Investors might want to be careful where they play. There is a question which end markets have really bottomed and which have not. In general, the consumer-focused end markets that were lousy in the aftermath of Covid are probably closer to the bottom as their inventory cycles complete the transition from stockpiling to flush. PCs is a good example. Smartphones have also been pretty bad, and hopefully this segment gets a little closer to bottoming as well.
Which end markets are still further away from the trough?
On the other side, you have some of the analog chip makers where industrial demand has started to show some cracks. Texas Instruments was calling that out for a few quarters already, and last quarter Analog Devices also started to talk about industrial weakness. I still wonder about automotive. I have been making the call that this end market is peaking for a while because shipments of auto semis have far exceeded auto shipments for quite some time. To be fair, I have been wrong, but this mismatch looks even scarier now.
You know the semiconductor sector from many years of experience. How is this cycle special and different from previous ones?
This is a big one, and we never had a cycle that was kicked off by a global catastrophe before. Not all cycles necessarily hit every end market, but this one is. Importantly, this cycle is also kind of asynchronous which has the effect of lengthening, of stretching it out, since some end markets are probably closer to the bottom whereas others are earlier in their cycles. Furthermore, there are two different kinds of cycles: supply cycles and inventory cycles. We get supply cycles in the memory segment a lot, but we haven’t had a good old-fashioned supply cycle outside of memory since the tech bubble. We have one now, and we have an inventory cycle too.
Looking ahead, what is the outlook for the memory chip market?
Memory is probably undergoing its worst cycle from an inventory utilization standpoint since the tech bubble. The difference is that the companies are in better shape, so they’re not going bankrupt this time. We had bankruptcies in 2001, and we had bankruptcies in 2008-09, but today the market is more consolidated. So it’s not fun, but the largest players can handle it. Samsung is not going anywhere, and neither are Micron and SK Hynix.
What should investors pay particular attention to in the upcoming earnings season?
We have to look for positive earnings revisions now. Since multiples are up, the question is if earnings can grow into these multiples. In terms of end markets, there will also be a particular focus on the data center segment where all the AI spending is happening: Are big cloud infrastructure providers such as Amazon, Microsoft and Google going to increase their capex, or is that spending coming out of somewhere else? In Q1, they all were spending less because of the digestion cycle, but now they’re spending on AI. So if they don’t take up overall capex, what are they not spending money on? Are they not buying servers? Are they not buying memory chips? Microsoft already said that they are increasing capex. It’s going to be interesting what we hear from the others.
Meanwhile, tensions in the trade war between the US and China are increasing again. What are the consequences for semiconductor companies?
It’s hard to know what the impact is, since we don’t know the details yet. In the case of ASML, it looks like they are going to prevent exports of some older generation lithography systems now too. There are also rumors that the Biden Administration is debating whether or not to put further restrictions on Nvidia’s AI chips. So far, Nvidia was able to work around the sanctions, but I don’t know if they are still going to be able to do that.
In a countermove, China aims to restrict the supply of gallium and germanium, two materials used in semiconductors. How could this affect the industry?
China is responsible for about 90% or 95% of the production of raw gallium, and most of the world’s gallium is used for semiconductors. It’s not rare by the way, it’s a byproduct of aluminum mining, but China has put in the capacity. Germanium seems to be less of a problem. China does have a lot of germanium, but it’s not all of it and only a small portion of the world’s germanium is used for semis.
And what about the impact of China’s restrictions certain segments of the sector?
In the case of gallium, there are two important compounds: gallium arsenide and gallium nitride. They are critical ingredients in power electronics like chargers. Gallium arsenide is used mostly for radio frequency applications like power amplifiers for mobile phones or other telecom applications. It’s not large quantities, but if China all of a sudden is restricting exports, you can’t make power amps for smartphones which means you can’t sell any smartphones. I have no idea how much inventory or stock piles are out there of gallium. Also, the folks so far have been saying that it’s not too much of an issue yet. But who knows.
In general, how do you assess the risk of a further escalation in the trade war for the industry?
At this point, everybody’s just getting fatigued. This has been going on for years, essentially since the Huawei sanctions of the Trump Administration in early 2019. So it’s not really new. The last time US sanctions were tightened in the fall, the big fear was whether regular server chips and smartphone chips would no longer be allowed to be exported to China. However, according to the current provisions, this is not a problem. The only companies in my coverage that were impacted were Nvidia and the semiconductor capital equipment companies which lost some revenue and took a bit of a margin hit. The advantage they had was that the market for wafer fabrication equipment was going down anyway, so their earnings were coming down in any case.
Where are leading US semicap companies like Applied Materials, Lam Research and KLA in their cycle?
For this year, most people expect $75 billion in WFE revenue, down more than 20% from 2022. The question is what does 2024 and 2025 look like; and within that, what’s the pace of memory recovery. As mentioned, the memory market is pretty lousy and there’s uncertainty about the pace of the recovery. That’s why we downgraded Lam Research to «Market Perform» a little while ago. The Street is modeling that this quarter is the trough for Lam, but I don’t know. It doesn’t sound like Micron and Samsung aren’t thinking about taking down memory spending. But then again, memory spending is already very low. That’s why I have some sympathy for the bull case: Lam is a memory focused semicap company with almost no memory in their outlook, and yet, they’re still decently profitable.
In addition, there is increasing investment in AI. To what extent will equipment suppliers benefit from this trend?
The thing is that the AI boom is not driving a lot of wafer demand right now. Think about it: Nvidia issued a massive guidance thanks to their data center unit, and they may do even better. Since we know how big these chips are, we can use Nvidia’s outlook and make some assumptions on average selling prices, and then we know how many additional wafers they will order: It’s like 1500 wafers a month, which is a rounding error for the industry. So in my opinion, AI is not driving any material amount of uptick in semicap spending. Over time maybe, but right now the general market cycles are much bigger drivers for these companies.
And how does it look for Nvidia itself? It’s the first semiconductor company to reach a market capitalization of more than $1 trillion, which reflects a great deal of optimism.
The long-term opportunity looks enormous, and we’re early. In five to ten years, we will be talking revenue numbers that are a lot bigger than what we’re seeing right now. But I don’t know how it will look next year, because the surge right now is pretty big and there could be a pull forward in demand. So at some point we will see a digestion cycle, as we’ve seen before. But into the end of this year, I would take the over versus the under on Nvidia’s numbers. They guided $11 billion in total revenue for the current quarter - and when they’re guiding $11 billion, they think they can do better than that.
At the moment, virtually every company in the industry is trying to portray itself as a beneficiary of the AI boom. Who are really the biggest winners?
If you’re really benefiting from generative AI, you’re already seeing it, and in my coverage there are only two names that are actually seeing it: Nvidia and Broadcom. There is a narrative in the case of AMD, Marvell and all these other guys. Maybe there is some opportunity long-term, but right now there is no revenue, nothing material. So it’s mostly just a story. Everybody wants to be an AI company. I recently read that even the CEO of Kroger mentioned AI like eight times during the company’s last earnings call which seems pretty ridiculous for a grocery chain.
The bull case for Nvidia is well known by now. What is Broadcom’s AI story?
They’re actually seeing it, it’s real dollars. Broadcom does two things: One, they help hyper scalers like Google design their own AI chips. This was a $2 billion business for Broadcom last year. This year it’s going to be $3 billion, and likely $4-5 billion next year. Two, they sell networking chips for large language models: Broadcom had $200 million in revenue in the past quarter from this business, and they said it’s going to be $800 million this year. Next year, I bet it’s several billion dollars. Overall, Broadcom said that their AI revenue this year will be around 15% of their total semi business, and they see it growing to potentially 25% next year. If that’s true, and if you run the math, it suggests that their non-AI semi business next year could shrink more than 10% and they would probably still grow the overall semi business. I think that’s why Broadcom feels a little bit more comfortable about a «soft landing».
What about Intel? Has the former industry leader missed the boat in terms of AI?
They have some products for AI too, but they haven’t spun the narrative nearly as well as AMD is trying to do. Yet, for the first time in forever you can be able to make the case that earnings estimates are finally too low. That’s why we upgraded the stock recently to «Market Perform».
Why do you think Intel could surprise positively?
It’s not that I love the stock, Intel is having a hard time. But they overshipped PC chips, and now they’re undershipping them, and people never model the normalization properly. So you could say it’s hard for Intel to get any worse. They already slashed the dividend, and their gross margins are in the 30s. In the past, their gross margins used to be running well into the 60s, but now people are getting excited that they are going to get back into the 40s. There was an investor conference around a month ago where Intel actually upticked the quarter. They said that they were coming in closer to the high-end of their guidance. So maybe Intel’s numbers can go up in the second half of the year.
How is the company’s turnaround coming along?
Intel just had a foundry event. The market took it a little negatively, because they didn’t announce any customers. But I was actually mildly encouraged, because they’re finally admitting that they have a problem and exactly what the problem is - and that’s the first step.
So what is the problem?
Basically, they were never the most efficient semi manufacturer even when their leadership was unquestioned. The reason was that their business units were not incentivized to be cost efficient, since all the costs were borne by the manufacturing unit. But now, Intel is making its manufacturing unit, their foundry business, a separate business unit with its own P&L. That means the other business units are going to get charged a market price for the fabrication of their products, and it’s going to show up in their margins, and in their bonuses. On the other hand, if Intel’s foundry unit can’t fab their chips as well as TSMC, the business units are going to TSMC. So cost is now king and that’s probably the right thing to do.
In this respect, how is Intel progressing with its plans to catch up on the technological front with its Asian competitors TSMC and Samsung?
According to Intel, they are on track. But we heard that before: Everything is ok, until it’s not. The near-term processor lineup seems ok. It looks like Meteor Lake, their next-gen PC chips, are launching in Q4 as previously announced. They also indicated that their new server chips for next year are on track. What’s more, they said some good things about their process technology and data center products further down the road. So tactically, you can argue that Intel’s numbers are too low and maybe it doesn’t get worse. But it remains hard to tell if they’re on the right trajectory, if they are really getting better. So I still don’t know what the company will look like in three, four or five years.
Accordingly, Intel’s shares have widely underperformed in recent years. Are there other companies that might be underestimated?
For bottom-fishers Qualcomm may be worth a look. It’s not like Intel where there are structural issues, it’s just their market is terrible, and it’s been terrible for a long time. On their most recent earnings call, they took their numbers down, so they’re trying to guide conservatively for the second half. My estimates are down this year because Qualcomm is flushing inventory, they go up next year because you normalize that, and then they go down again in 2025 because they’re losing their business with Apple. So it’s not really a sexy story, the company doesn’t really grow, but it’s cheap, and maybe it’s washed out.
And what could give the stock a bit of a boost again?
Qualcomm seems to have at least tried to de-risk things into year-end, and there is a real adjacency story there, it just takes time. But again: The problem is that the smartphone market has been a lot weaker than people thought, and certainly I thought. I was kind of a closet smartphone bull because smartphones didn’t benefit from Covid. Unlike in the PC market, there was no pull forward in demand. And while smartphones have been continually lousy, recent monthly data out of China has been a bit more encouraging. So at least for a brief moment there’s a possibility that things finally could get better.
