At the time of writing, the Direxion 3x leveraged semiconductors ETF (NYSEARCA:SOXL) is the best performing ETF in the last five years, with a total return of 1527% during that time. Even the non-leveraged fund (NASDAQ:SOXX) ranks 16th with a 355% return; very respectable considering that it's competing against leveraged funds in a raging bull market.
Clearly, semiconductor stocks have performed very well lately. In fact, the industry has grown by an average of 5% per year for the past 20 years. This performance is driven in part by digitization trends that won't be going away any time soon, so investors would be wise to learn about this red hot industry and potentially invest in it.
The trends driving this industry are projected to nearly 10x the number of connected devices in existence by 2030. They include:
- Smart home/work/city
- Internet of things
- Self driving cars
- Predictive healthcare
Unlike many investment opportunities in these spaces, semiconductor companies are generally mature. This means that they're highly profitable, return cash to shareholders, and have a long operating history, which in many ways makes them safer investments than emerging growth stocks. Even so, there are some very real risks that will be discussed later.
This article provides an overview of the semiconductor industry and the companies that comprise it. It will discuss how semiconductors work and the best investment opportunities that are available in the space. The companies discussed include: ASML (NASDAQ:ASML), Applied Materials (NASDAQ:AMAT), Lam Research (NASDAQ:LRCX), Taiwan Semiconductor Manufacturing Co (NYSE:TSM), Samsung (OTCPK:SSNLF), Intel (NASDAQ:INTC), NVIDIA (NASDAQ:NVDA), AMD (NASDAQ:AMD), Micron (NASDAQ:MU), Texas Instruments (NASDAQ:TXN), Broadcom (NASDAQ:AVGO), QUALCOMM (NASDAQ:QCOM) and many more.
The article is loosely organized around these subtypes of semiconductors:
- CPUs and GPUs: These chips are found in computing devices like computers and phones. They quickly execute arbitrary computer code like apps.
- Memory: These chips are also found in computers and phones. They store data, and some memory chips will persist their state even if the device is turned off.
- Basic/Standard Chips: These chips typically serve one simple purpose like reading a barcode, unlike CPUs and GPUs which can be reprogrammed to execute any code. These are not covered because they're highly commoditized.
- Analog: These chips read input from real world sources like a thermometer and turn it into digital 0s and 1s, unlike the above types which use digital 0s and 1s as both input and output.
We'll now look at the companies involved in this space, starting from the lowest level and working our way up the value chain.
Semiconductors are physical devices, so they require raw materials. These materials include silicon, tin, tungsten, and gold. They're sourced from thousands of suppliers all over the globe. Most of these companies aren't direct investment opportunities in semiconductors since the materials' prices are impacted by more than semiconductors. For example, gold is also used as a store of value.
Equipment: ASML, Applied Materials, Lam
To manufacture semiconductors, companies use equipment that embeds integrated circuits onto silicon chips. These circuits store and process all of the 0s and 1s that make up the logic and data in computer programs.
The manufacturing process involves multiple stages. One of the most important stages in cutting edge chips is lithography, which embeds semiconductor designs on a silicon chip. Lithography equipment is produced by three companies: ASML, Canon (NYSE:CAJ), and Nikon (OTCPK:NINOF).
Source: The Economist
Of these three, ASML has the most market share and it's been gaining share over time. That was already impressive, but in 2017 ASML became one of the widest moat companies in the world.
Prior to 2017, deep ultraviolet (DUV) technology was used to manufacture semiconductors. But shortly after ASML's public listing in 1995, the company made a bet that extreme ultraviolet (EUV) technology would be the future of the industry. This technology uses a wavelength nearly 15x smaller than DUV.
The practical difference between DUV and EUV is the size of the chips manufactured. For decades, the size of semiconductor chips has halved roughly every two years. This rate is known as Moore's Law. Today the cutting edge is an incomprehensibly tiny 3 nanometers (3nm), but 5nm, 7nm, and even 10nm+ chips are still used in most devices. Over time, more devices will start using smaller chips as they become readily available because smaller chips enable more modern/advanced applications. While DUV can be used to create 5 nm and 7 nm chips, it produces lower yield than EUV. It's also likely that EUV will be the preferred approach - if not completely necessary - for 3 nm and smaller chips.
ASML spent years working on EUV and planned to release their first EUV machine in 2007. They did in fact create EUV machines starting in 2006, but it took years to work out the kinks, and they weren't mass produced for commercial clients until over a decade later in 2017.
Canon and Nikon never attempted to develop EUV technology, so this makes ASML the world's only producer of EUV equipment. If you want an indicator of how important EUV technology is, look no further than the governments of the USA and Netherlands, which are blocking ASML from sending machines containing the technology to China citing national security. Per Malcolm Penn, it would take a Chinese company - or any other potential competitor - a decade or more to catch up to ASML's tech, just as it took ASML decades to create this technology in the first place.
It may be an understatement to say that ASML is a monopoly with a wide moat. And wide moat companies tend to beat the market over time. But just because ASML dominates the EUV space doesn't mean that it will dominate the entire industry. Demand for DUV will likely remain strong for years, and the two technologies can to some extent be used together. ASML also offers support services for its machines and it competes in this segment with a couple companies including Applied Materials.
Even so, there's no denying that EUV is the future and has a stronger growth trajectory than DUV. It's already used in the manufacturing process for many high end chips like those in Apple's (NASDAQ:AAPL) latest iPhone. ASML's growth should be quite strong this decade as more devices switch to smaller chips; it expects revenues to increase 85% by 2025 and an 11% CAGR through 2030.
Besides lithography, there are a couple other steps in the manufacturing process:
- Deposition, where the silicon is combined with other materials
- Etch, where the chip is cleaned/polished
- Diagnostics, where the chip's quality is verified
While ASML dominates the lithography stage, the deposition and etch stages have three main competitors: Applied Materials, Lam Research, and Tokyo Electron (OTCPK:TOELY). Meanwhile, KLA Corporation (NASDAQ:KLAC) has over half the market share in diagnostics.
These companies share many traits with ASML. They have wide moats based on their scale, decades of expertise, and close relationships with their manufacturing customers. They actually have less customer concentration than ASML since their offerings have more appeal to less cutting edge manufacturers. But in my view, they're not quite as interesting as ASML because they don't have such wide moats and have to compete with each other.
Despite the old saying to save the best for last, I believe that these equipment companies represent some of the best investment opportunities among semiconductor stocks. Historically, equipment companies have grown faster than the overall market. ASML, KLA, Applied Materials, and Lam averaged 9.39% revenue CAGR between 2008 and 2020. The overall industry averaged 7.98% CAGR during that time.
Granted, that number is pushed down by slower growing analog companies; these equipment companies' US-listed customers (Intel, TSMC, Micron) grew at a more comparable 9.47% CAGR. Even so, the equipment companies' comparatively smaller size, wide moats, and comparatively lower valuations (besides ASML) leave them well positioned. While most companies benefit from increasing demand for semiconductors, equipment companies also benefit from increasing complexity of semiconductors.
Manufacturers: TSMC, Samsung, Intel
The companies that buy manufacturing equipment from equipment suppliers are sometimes called foundries. They use the equipment and raw materials to manufacture semiconductors. They might produce their own designs and/or use the designs of their customers.
ASML's three largest customers account for 80.1% of its sales. These customers are TSMC, Samsung, and Intel. They represent ~31.4%, ~28%, and ~20% of ASML's revenues respectively. Thus, all three companies are quite large, and there aren't any other manufacturers of similar scale and advancement. No other foundry has over 10% market share although there are some other publicly traded ones like GlobalFoundries (NASDAQ:GFS) and United Microelectronics Corporation (NYSE:UMC).
The main difference between these three manufacturers is the integrated device manufacturer (IDM) strategy. Intel and Samsung are IDMs, which means that they design and manufacture their own chips. On the other hand, TSMC does not produce any chips under its own name, which means that it never competes directly with its customers. Thus, while Intel and Samsung have more overall revenue than TSMC, TSMC has the most revenue from the chip manufacturing businesses and accounts for over 50% of the third party foundry market share. Although each strategy has tradeoffs, TSMC's lead in market share could be interpreted to mean that its strategy is working best. Its choice not to compete with its customers has allowed it to land big name clients like Apple, Nvidia, and AMD.
Most people are familiar with Moore's Law, which I described in the previous section. However, fewer people are familiar with Moore's Second Law, which states that the cost of a foundry doubles every four years. This law intuitively makes sense because as chips get smaller, they get more expensive and complex to create. For example, it now costs $150 million just to purchase an EUV machine from ASML. Not many startups have that kind of cash lying around, so the largest foundries have a narrow moat from that especially as chips continue to get more advanced.
Taken to its logical conclusion, Moore's Second Law means that there will be fewer and fewer foundries over time. This seems to be the case historically. An innovation called FinFET eliminated all but six foundries from mass producing chips in the aughts. Then in 2015, GlobalFoundries and UMC gave up on sizes smaller than 14nm, leaving only Intel, Samsung, and TSMC capable of mass producing cutting edge nodes. Intel's recent struggles leave only two companies still on the cutting edge at 5nm: TSMC and Samsung.
A huge issue facing Intel - and to a lesser degree Samsung - is this lack of cutting edge tech. TSMC is already mass producing 5nm chips for products like the iPhone and has plans for 3nm in Q1 2023, 3nm Enhanced in 2024, and 2nm in 2025. Meanwhile, Intel is still struggling to get 7nm working. It won't be ready until at least 2023, with delays caused by issues such as low yield from the manufacturing process. In fact, Intel seems to be giving up - at least partially - and will actually use TSMC to manufacture some of its more advanced chips going forward.
Despite this, Intel retains the ambitious goal of becoming the largest third party foundry by 2025. The company has already landed significant deals where it will manufacture chips for the likes of Qualcomm and Amazon (NASDAQ:AMZN) starting next year. But it's difficult to see those wins allowing Intel to reach anywhere near TSMC scale. Apple alone accounts for 25% of TSMC's revenue and Intel itself will use TSMC. So it's a bit difficult to take their plan to become larger than TSMC in just four years seriously.
For its part, Samsung is not directly listed in the USA which makes it more difficult to follow. Most investors probably can't even buy it except indirectly through a Korea ETF like EWY, which has Samsung as a 24% holding. But it appears to be somewhere in the middle of Intel and TSMC. They are an IDM but also design chips for other companies and are the second largest third party foundry. They've recently launched their 5nm chip and plan to start mass producing a 3nm chip in the first half of 2022 - even before TSMC. However, there have been reports that Samsung's 5nm yield is lower than TSMC's. Like Intel, the foundry's apparent conflict of interest with Samsung's own hardware has prevented it from landing deals with big names (besides itself) and validating its process at a TSMC-level scale.
From most perspectives, TSMC is thus unrivaled as a foundry. It has the best tech, the best margins, the most valuable customers, and the most market share. Both Samsung and Intel remain competitive for now although the trend certainly seems to be moving slowly against them as TSMC continues to push the cutting edge. Since 2008, TSMC grew revenue at 10.7% CAGR while Intel grew at 6.26% and United Microelectronics grew at 6.38%. Global Foundries went public more recently but has been losing revenue since 2018.
If TSMC delivers on the ~15% growth it plans for the next five years then it will only increase its size advantage in the future. TSMC has also exhibited less cyclicality than its smaller competitors; it's the only foundry to not experience an EPS decline of more than 20% in the last decade.
Among these companies, I believe that TSMC is in the best position for future growth, innovation, and stability especially when considering Moore's second law. It is possible that there will eventually be only one foundry still pushing the cutting edge, and I would be shocked if that's not TSMC. Although it trades at a higher P/E than Intel and Samsung, sometimes it's worth paying more for a better positioned company and the growth that comes with it.
Designers: Nvidia, AMD, Intel
When it comes to designing logic chips for tech like PCs, smartphones, data centers, cryptocurrency mining, etc., the three biggest players are Nvidia, AMD, and Intel. These companies design chips for all of the aforementioned use cases and more, then sell them to companies like Apple and Dell (NYSE:DELL) for use in their devices.
Increasingly, the big tech mega caps are also designing their own chips for specific purposes like Amazon's Graviton processor for AWS and Apple's chips for its hardware. Because these companies do much more than design chips, they're not really semiconductor investments. But they're worth mentioning as competitors to Nvidia/AMD/Intel.
Intel was/is the largest designer but it's fallen on hard times recently. One reason for this is because of the aforementioned issues with 7nm production. Nvidia and AMD are both fabless, meaning that they rely on other manufacturers (more specifically, TSMC) but do the design work themselves. This has allowed them to produce more cutting edge products given TSMC's previously discussed manufacturing lead over Intel.
When it comes to CPUs - which are the chips that run code on computers, smartphones, etc. - there are two main types: x86 and ARM. Intel owns x86 and licenses it to only one other company: AMD. ARM is owned by an eponymous private company possibly being acquired by Nvidia. In general, x86 offers better performance while ARM uses less power. Thus, x86 is dominant on desktop/PCs while ARM is dominant on mobile, although ARM has been gaining market share on PCs lately thanks to Intel's struggles. Both types of chips are used in data centers although x86 is more popular there.
Recently, another type of chip called a GPU has experienced significant growth. This type of chip is optimized for graphical tasks like playing video games, and for some other less intuitive purposes like AI, data centers, and cryptocurrency mining. GPUs can either share memory with the CPU or have their own dedicated memory. Intel is the leader in the former type while Nvidia is dominant in the latter with over 80% market share. The latter type has better performance (but consumes more energy) which makes it more sought after for cutting edge use cases. Analysts are projecting a 33.6% CAGR for GPUs through 2027, which helps to explain Nvidia's industry leading growth rate.
Intel is hoping to catch this train before it leaves the station with a discrete GPU offering launching Q1 2022. This chip will contain its new Arc graphics card, which will be manufactured by TSMC. However, like Intel with CPUs, Nvidia has significant intellectual property advantages as the first mover in the GPU space and even licenses some of this IP to Intel. When also considering that AMD has offered discrete GPUs for years and never captured much market share from Nvidia, it's highly questionable whether Intel or anyone else can meaningfully compete with Nvidia here.
However, GPUs are more of an optimization than a strictly required component, so improvement in CPUs or other alternatives could reduce demand for GPUs even if Nvidia remains dominant in them. This seems unlikely to happen any time soon based on the projected growth rate for GPUs. But there are new types of chips that are worth monitoring such as DPUs, which are more optimized for data centers especially as it relates to security. Nvidia estimates that about 50% of data center tasks could be run on a DPU and it's already a market share leader there as well, but faces stiffer competition from Intel and smaller players like Marvell (NASDAQ:MRVL), which was one of the first movers in the space.
AMD is involved in both the CPU and GPU spaces, and has done very well lately despite being at an intellectual property disadvantage against both Intel and Nvidia. Among other key customers, AMD develops graphics architecture for well known platforms like Xbox and PlayStation. Perhaps most importantly, AMD's position in data centers is very strong because it has a manufacturing edge over Intel (due to its TSMC partnership) and unlike Nvidia or any other non-Intel company, it has the IP rights to produce x86 chips instead of ARM ones. However, with Intel moving to strengthen its x86 offering it remains to be seen whether this advantage will last.
To summarize, Intel is the leader in CPUs while Nvidia is the leader in GPUs. AMD is also a player in both spaces and has had a lot of success recently despite not having a market share lead. All three companies are trying to take market share from each other and will continue to attempt this in the coming years. This is an oversimplification because there are other types of chips that are manufactured by all of these companies, but there's only so much that can be covered in one article.
The intense competition in this part of the industry combined with a wide range of valuations makes this space more difficult to invest in.
While Intel is cheap, its IDM model will probably make it difficult to drive strong returns going forward. If it outsources manufacturing to TSMC, its design business might do well, but it will harm its manufacturing business. If it keeps manufacturing in house, its design business will continue to struggle. Either way, one of its segments gets left behind. One potential solution is to spin off its manufacturing business like AMD did, but I haven't heard any talk of this.
Meanwhile, these struggles have allowed Nvidia and AMD to be some of the best performing (and most expensive) companies in the index this year, and even over the last 10 years. As mentioned above, their projected future growth rates (33.6%) are almost unbelievable relative to the industry's historic 5% CAGR. Between 2008 and 2020, these companies grew at an average of just 6.48% CAGR, slower than most of the other companies previously discussed.
Thus, I am not convinced that this level of growth will be sustainable. But with them trading at 50-100 P/E, it would have to be to justify an investment in them. So while I wouldn't want to buy their competitors, I wouldn't necessarily want to buy them at this price either. I consider them holds.
Memory: Micron, Samsung, SK Hynix
Memory chips are similar to the CPUs and GPUs discussed above, but they're built to store a lot of information (0s and 1s) rather than execute a lot of instructions. A main differentiating factor in memory chips is whether their storage is non-volatile (such as NAND) or volatile (such as DRAM). Volatile chips don't maintain their state when powered off, but generally have faster read/write times. Thus, both types of chips are often used together alongside CPUs, GPUs, and other chips to form a complete computing system.
The leading companies in the memory space are Samsung, SK Hynix, and Micron. Micron is number three in market share and is the only one of these companies that's easily investable from the USA, so investors' options are somewhat limited.
Like CPUs and GPUs, memory chips are manufactured using processes from ASML, Lam, and Applied Materials, including EUV lithography. In particular, Lam generates over half its revenue from memory customers. However, memory companies are generally IDMs and do most of the manufacturing themselves.
The common perception is that memory chips don't have as specific of requirements as logic chips (like x86 and ARM). This gives them less room for differentiation, leading to less pricing power and more risk of oversupply.
Despite this, Micron has done quite well over the past 10 years (11.49% CAGR) as demand for memory increased dramatically and solid state took share from traditional disk drives. Demand for memory is likely to continue increasing in the future due to new use cases like data centers and AI, which could allow Micron to continue growing near this rate if it fends off the competition.
The company recently started paying a dividend and trades at just 16 P/E. Although I'm somewhat ambivalent about investing in the low moat memory space, I still view Micron as a better buy than the other designers/IDMs mentioned in previous sections.
Analog: TI & Many More
While most people think of digital products like CPUs and memory when they think of semiconductors, there's an entire separate field called analog semiconductors. While digital semiconductors process 0s and 1s in computer programs with a focus on size, speed, and efficiency, analog semiconductors process signals from the real world with a focus on reliability.
Analog products like amplifiers, switches, clocks, and converters are used for a wide variety of applications ranging from factories to toothbrushes to cars. These products don't usually require the smallest or most cutting edge tech which leads to most analog semiconductor companies being IDMs (designing and manufacturing their chips).
Some additional differences between analog and digital semiconductors include:
- The digital industry does redesigns every year or two, while analog designs can last for decades because the real world signals they process don't change much. Thus, analog tends to spend less on R&D and use less cutting edge manufacturing.
- The digital industry has a relatively small number of designs (CPU, GPU, memory) whereas the analog industry has thousands of unique products due to the wide variety and location of signals that can be processed.
- The analog industry has higher switching costs since chips are more specialized. Analog chips tend to be cheaper which also disincentivizes switching.
- Some (but certainly not all) companies in the analog industry have lower revenue concentration since they don't sell as much to the mega cap smartphone/cloud companies that exert pricing power.
- The digital industry is generally exposed to high growth through AR/VR, data centers, computers/smartphones, and AI, while analog is more exposed to growth areas like electrification, self driving cars, robotics/industrials, internet of things, and communications infrastructure. There's quite a bit of overlap from each side though.
These differences result in analog semiconductor firms generally being slower growing with more diversified revenue, more shareholder returns, and wider moats than their digital counterparts. Thus, analog semiconductor companies are sort of the value stocks of this industry. The two largest analog companies - TI and Analog Devices (NASDAQ:ADI) - averaged just 5.1% CAGR between 2008 and 2020.
While they haven't generated Nvidia-level returns, don't mistake them for boring investments; many analog companies have still easily beaten the overall market over the last 10 years. These companies' focus on industrial, auto, and communications end markets could also help to decouple their performance from the performance of the more tech-focused companies discussed in previous sections and accelerate growth in the future.
The analog industry is highly fragmented compared to digital. TI is the largest analog company with 19% market share, and no other player reaches 10% or $100B in market cap. Other players include Analog Devices, STMicro (NYSE:STM), NXP (NASDAQ:NXPI), Microchip (NASDAQ:MCHP), ON Semi (NASDAQ:ON), and Monolithic Power (NASDAQ:MPWR). Although there's a bit of a gray area between analog semiconductors and other niches like microcontrollers, mixed analog+digital, and power semiconductors, I lump these areas together for simplicity based on the growth areas and technical requirements they're most exposed to.
TI's revenue is split among thousands of products sold to a variety of industries: 37% industrials, 20% automotive, 27% electronics, 8% communication services, 6% enterprise systems, and 2% other. This is ironic because most people know TI for their calculator brand which falls into the "other" group that accounts for only 2% of revenue.
Other analog firms have different exposure which I view as one of the main differentiating factors between them. Some firms like STMicro have significant exposure to smartphone manufacturers like Apple and Samsung. Others - like Analog Devices - have noteworthy exposure to communications infrastructure like 5G.
But the wide variety of products sold by most analog companies could make it difficult to accelerate growth going forward. Even if a company gets a huge design win, it might still be a small part of revenue. A great example of this is NXP's near field communications chips. These enable mobile payments with Apple Pay, Google Pay, etc. In a decade of smartphone and NFC growth, NXP's revenue grew at just 4.5% CAGR. Even today, NFC doesn't make up much of NXP's revenue, with a near majority coming from automotive.
Moreover, despite their slower growth, these companies don't come with better starting valuations than their digital counterparts. For example, ADI at 52 P/E, having traded at 14 P/E as recently as 2019. I'm not here to argue about exactly what P/E this company should have. I'm only pointing out that there doesn't appear to be a better margin of safety on the analog side despite what's likely to be slower growth. I believe that investors are better off staying overweight on the digital side, although analog companies have certainly proven themselves capable of beating the market too.
Connectivity: Broadcom and Qualcomm
I included this mini-section because it's difficult to place Broadcom and Qualcomm, as well as a couple other smaller connectivity chip players like Skyworks (NASDAQ:SWKS), Qorvo (NASDAQ:QRVO), and some parts of Marvell's business.
These firms are focused on semiconductors for connectivity solutions like ethernet, WiFi, and 5G. Many connectivity chips such as filters and switches share traits with analog semiconductors. They process real world signals like those sent through 4G/5G networks, and do things like filtering noise from other networks and devices. As 5G expands, more filters will become necessary because most devices will also remain backwards compatible with 4G/3G and will need filters for each network.
But despite this analog behavior, these companies share many characteristics with digital chip designers:
- They outsource some of their manufacturing work to TSMC.
- They have somewhat high revenue concentration, namely among device manufacturers like Apple and Samsung.
- They have limited presence in industrials.
- Their designs are usually refreshed fairly frequently, about once per year.
- They operate in comparatively less fragmented markets, giving them grater scale.
Despite these traits, they don't really compete with analog or digital companies covered previously. Technically, Qualcomm makes a CPU called Snapdragon that's based on ARM. It's the leading chip for mobile devices since it's used in popular Android phones including some Galaxy devices. However, none of the aforementioned designers (Intel, AMD) compete in this space, since their energy intensive x86 processors work better with computers, data center, etc. Broadcom also sells an ARM-based embedded CPU specifically focused on switching and routing that only competes with smaller players like Marvell.
Most of Qualcomm's profit actually comes from licensing patents to other companies including smartphone makers like Apple and Samsung. In fact, Qualcomm gets a bit of money pretty much every time a smartphone is sold. While smartphones are plateauing in growth, the addition of new patents for 5G and the explosion of new devices in IoT should continue to drive growth for Qualcomm.
The rest of Qualcomm's businesses - plus the main businesses of Broadcom, Skyworks, and Qorvo - are focused on other types of connectivity like filters and switches. Both Skyworks and Qorvo have extremely significant revenue concentration with Apple and other smartphone makers; Apple accounted for 59% of Skyworks' revenue in the past year.
Broadcom is much more diversified with literally thousands of products including filters. About half of its revenue comes from wired networking solutions, where it's the clear market leader. This space is slower growing than wireless though. Broadcom also generates about 28% of its revenue from software and SaaS in addition to selling semiconductors.
These companies - maybe minus Broadcom - are probably the best ways to get exposure to the 5G rollout and to a lesser extent internet of things/automotive, all of which will continue to drive growth in the coming years. However, I view them as lower quality than many other semiconductor options. Broadcom carries noteworthy debt and appears to be in a comparatively slow growing business, although it's historically done well growing through acquisitions. Qualcomm always has regulatory uncertainty around its patents. And the smaller players deal with significant revenue concentration that isn't mitigated by high switching costs. Of this group, only Broadcom has outperformed the semiconductor index over the last 10 years despite the huge 4G/smartphone boom that occurred during that time.
Although the above sections cover every US-listed semiconductor company with over $100 billion market cap and many smaller companies, there are still other firms.
Another mini-sector is electronic design automation, which focuses on selling design and testing tools to semiconductor designers like Intel. Synopsys (NASDAQ:SNPS), Cadence (NASDAQ:CDNS), and Mentor dominate this sub-sector. This business model is more similar to a SaaS company (and in fact these companies aren't included in the semiconductor index) so it may prove less cyclical. Both Synopsys and Cadence have grown revenue in every year since 2008, which cannot be said for most of their customers (Intel, Nvidia, etc.). These companies may also benefit from chips becoming more widespread, complex, and difficult to design/test.
There are also smaller specialty designers/manufacturers that don't fit neatly into the previous sections. For example, Universal Display Corporation (NASDAQ:OLED) makes the well-known OLED displays in TVs and phones. Apparently these are considered enough of a semiconductor to make it into the index. Wolfspeed (NYSE:WOLF) makes wide bandgap semiconductors, which are made of a different material than traditional silicon semiconductors. They are primarily used for power, transportation, inverters, and wireless systems.
There are more companies not mentioned but I'll end it here for the sake of brevity. The full list of companies in the index is available here.
Having discussed all of the amazing and innovative work that the companies in this industry are doing, it's now time to take a look at risks facing them. This section highlights five key risks.
The most frequently mentioned risk is cyclicality. Historically, the semiconductor industry has alternated between periods of oversupply and undersupply. While things are good today with the well-publicized shortages facing the industry, there will inevitably be periods of oversupply in the future where many semiconductor companies struggle to drive growth. Over the long term, demand for semiconductors has only increased, so this risk is more of a consideration for short term traders and long term investors trying to decide when to add to their positions.
As the industry continues to consolidate, manufacturers can coordinate better and startups will find the environment more difficult, leading to less cyclicality. But the cyclicality issue is fundamental to the sector because it primarily drives growth through growth in its customers' end markets, which are unpredictable. Moreover, consolidation brings its own set of risks if it increases a company's customer concentration. Many companies in the sector already deal with significant revenue concentration, especially from tech giants like Apple and from other semiconductor companies.
As a result of many computing devices (and products in general) being manufactured outside the USA, most semiconductor companies generate most of their revenue internationally. For example, Micron generates 89% of its revenue internationally, but notes that only 56% of its sales were to companies headquartered outside the USA. Aside from the typical risks associated with international sales like currency fluctuations and trade wars, this technicality has allowed many semiconductor companies like Nvidia and Micron to have tax rates well below 15%. Potential tax reform could force these companies to pay more in taxes.
Unlike many tech companies (in particular software), semiconductors are physical objects that depend on a supply chain including raw materials. These risks have come to the forefront recently as a result of well-publicized supply chain issues. For example, Applied Materials has suffered from an inability to get semiconductor supplies, which are ironically manufactured by Applied's own customers.
As a high innovation sector, semiconductor companies often have to spend more than other sectors on research and development. This can lead to lower margins and quickly changing market position, although the largest semiconductor firms nevertheless boast strong profitability/operating margins and wide moats. Another risk stemming from constant innovation is a talent shortage; it seems like these companies often struggle to find qualified tech talent to fill all of their R&D roles. This issue is amplified for analog semiconductors because they're more specialized, but the issue of needing design wins every year is much more prominent on the digital side.
Lastly, there's always valuation risk. While this article is more focused on the companies than the current market conditions, many of the highest quality stocks are also very expensive right now. The sector as a whole has one of its highest P/E ratios ever, although that's also the case for the entire market. It's up for debate whether the high growth the sector is likely to experience in the future would be enough to allow a company with 100 P/E like Nvidia to outperform the market/index.
Coming full circle, the group of equipment companies discussed at the beginning of the article represent what I believe are the best investment opportunities in the sector at this point. Equipment companies have historically outperformed the semiconductor index, which makes sense because they benefit from both Moore's Law and Moore's Second Law. If semiconductors are the picks and shovels for the tech industry, then equipment companies are the picks and shovels of the semiconductor industry. Not a bad place to be.
The five dominant equipment companies are:
- ASML (53 P/E, 1.89 PEG)
- AMAT (23 P/E, 2.13 PEG)
- LRCX (23 P/E, 1.43 PEG)
- KLAC (23 P/E, 1.78 PEG)
- TOELY (28 P/E, 1.63 PEG)
Aside from ASML, all of these companies have lower P/E than the overall sector. Although that's also been the case historically, these companies' P/Es are not particularly stretched relative to their historical levels. Moreover, all of these companies have lower PEG ratios than more popular tech stocks like Apple (3.44), Microsoft (2.86), Tesla (3.02), and Nvidia (3.56).
Thus, equipment companies appear well positioned to outperform based on both their valuation and their position within the industry. Among these five, I expect relatively similar performance, with ASML standing out as the highest risk/highest reward option due to its higher valuation and faster growth rate. All five of these companies are high quality with wide moats, low debt, solid margins, shareholder friendly policies, and high return on capital.
While there are many high quality companies in other parts of the industry as well, these parts of the industry don't appear as well-positioned for outperformance.
- Foundry companies have also failed to outperform historically, likely due to ever-increasing expenses as a result of Moore's Second Law. TSMC has proven itself the exception here due to its big technological lead.
- Analog semiconductor companies have grown more slowly and are too diversified to change that. There have been exceptions that beat the index like NXP and Monolithic Power. But I prefer not to try and pick which will be the future exceptions among the many available options.
- Connectivity companies also generally failed to outperform, likely due to high customer concentration and company-specific issues.
None of these factors look like they're going to change going forward, so I am inclined to stick with the past winners. However, a key risk to this thesis is continued consolidation among manufacturers. This will lead to increased revenue concentration for equipment companies, which could be a factor that caused other parts of the industry to underperform in the past. Again, TSMC would be the primary beneficiary in this case.
Also, while equipment companies have done well, by far the biggest winners have been non-IDM designers, in particular Nvidia and AMD. This is a trend that could continue and I wouldn't fault investors for having exposure to these companies. However, I see two risks to their continued outperformance: increased competition from Intel (especially for AMD) and valuation, with the companies having P/Es of 103 and 50 respectively; well above their historical norms.
All told, I expect semiconductors to continue outperforming over the long run as long as digitization trends continue. And I think it's safe to bet that digitization will continue.