The semiconductor, or chip, industry is perhaps one of the most important and complex sectors in the stock market to understand. Chips are in everything from our smartphones to cars and to all facets of computing, from PCs to massive data centers used for the cloud. Put simply, they are the bricks and mortar of the digital world. We don’t see them, but we know they magically make stuff work. Semiconductor companies are also in the news a lot lately, whether it’s the U.S. government cracking down on chip exports to China or innovations in connected cars and artificial intelligence. But that doesn’t make the industry any easier to understand, even for those familiar with the well-known players. Making things trickier: The market tends to bucket anything chip-related into broad-based investment vehicles, such as exchange-traded funds (ETFs) and mutual funds. These companies, however, are quite specialized. Despite the industry being out-of-favor right now – and our ongoing effort to scale back our once over-weighted chip holdings – we believe that investors’ portfolios should have at least some exposure to the industry. For starters, they are crucial to nearly all secular growth trends today and into the future. And trying to time the booms and busts of the cycle is incredibly difficult if not impossible to do consistently. In our portfolio, we have small positions remaining in Advanced Micro Devices (AMD), Nvidia (NVDA), Marvell Technology (MRVL) and Qualcomm (QCOM) — all of which design chips for varying uses. ( We’ll get into the specifics later ). Here then is our guide to all things semiconductor. Our goal is to help you better understand where different chip-related companies sit in the supply chain, to differentiate these companies by their end markets, and to gain more insight on how money flows through the industry. After all, one company’s capital expenditure is another company’s revenue. What is a foundry model? The term foundry is industry jargon, referring to the factory where chips are made. There are four main categories in the supply chain under this model: Capital equipment makers, pure-play foundries or fabs, fabless designers, and integrated device makers. Capital equipment companies — such as Applied Materials (AMAT), Lam Research (LRCX), ASML Holdings (ASML) and KLA Corporation (KLAC) — make the machines used in the manufacturing of semiconductors. Pure-play foundries , or fabs, are companies with factories that fabricate (manufacture) chips designed by other firms (some designers have their own factories but we will touch on that below). In these monstrous facilities, you will find the capital equipment made by the capital equipment makers. That alone should provide some insight into the money flow, because when a foundry is being updated or built from scratch, a good deal of that spending is going to go into buying equipment from the capital equipment makers. When you hear commentary on a call from a foundry player about capital expenditure plans, just remember that what you’re also hearing is commentary on demand and revenue for capital equipment makers. Players in the foundry space include companies such as Taiwan Semiconductor Manufacturing Company (TSM), often referred to as just TSMC, and GlobalFoundries (GFS). Fabless designers generally outsource the chips they draw up to foundries. Again, we can start to see the flow of funds, as strong guidance from a fabless designer means high demand — and therefore, a need for a greater number of chips. Of course, the more chips a company needs, the greater the order value it will place with a foundry player such as TSMC. Fabless designers include our Club holdings Nvidia, AMD, Marvell Technology and Qualcomm. Since these companies don’t need to invest in expensive fabrication facilities, they can run a more agile asset-light business model. They don’t have to concern themselves with the need to lay out significant capital in order to ensure their ability to manufacture the most cutting-edge chips. Finally, integrated device companies design and manufacture their own chips in-house, essentially designers with their own foundries. Players in this group include Intel (INTC), Micron Technology (MU), ON Semiconductor (ON) and Texas Instruments (TXN). Semiconductor supply chain Understanding the structure of the semiconductor sector can help to better understand how money flows through the industry — and therefore, help members leverage earnings and corporate updates to make more informed decisions when investing in the industry. Here are some examples. If we know that capital equipment is used in foundries, what we really know from a financial perspective is that the capital expenditures (capex) made at the foundry level are directly tied to capital equipment maker revenue. So, when TSMC discusses capex on its conference call, it’s our jobs as investors to read those comments through to demand for Applied Materials and Lam Research. When Nvidia discusses supply and demand dynamics, it provides insight into what it will need from a TSMC-type foundry or manufacturing facility. If there is an inventory glut, then the last thing Nvidia wants is more chips; that means fewer orders with TSMC and vice versa. Of course, in that same line of thinking, when cloud providers such as Club holdings Amazon (AMZN), Microsoft (MSFT), and Alphabet (GOOGL) discuss investments, it means they need chips. These companies’ capex is linked to revenue at fabless designers like Nvidia and AMD. We recently saw this dynamic at play when Meta Platforms (META) announced that 2023 capex spending would increase versus 2022 due to data center investments. The news tanked Meta stock due to what Wall Street believes to be undisciplined spending. However, it provided a boost to the semiconductor stocks that would realize Meta’s spending as revenue. If you hear the economy is slowing and the cloud providers say they want to “digest past investments,” they are really saying that they spent a lot of money recently and are going to pump the breaks on additional investments in cloud capacity. In your head, you need to say, “OK, cloud providers are slowing spending, which means less demand for Nvidia. If Nvidia is going to see less demand for the next few quarters then it may need to reduce orders from TSMC. And if TSMC is going to receive fewer orders, it may need cut capex — and as a result, demand for Lam Research’s equipment may decline in the near-term.” Of course, there are a lot of moving parts and timing the flow of funds is incredibly difficult. But that is how one needs to think at a high level. Remaining on the cutting edge requires always looking ahead, and companies must balance spending slowdowns with continued investment in the long term, which is why gauging the stages of the semiconductor investment cycle is so difficult. Nvidia may be seeing a glut of 30-series chips, but the company stills need to work on ramping production of its 40-series chips to be ready once the inventory glut is worked through. TSMC may not be seeing as much demand for 7-nanometer chips, but it needs to be in a position to start producing 5-nanometer and 3-nanometer chips after that. The company can’t stop spending entirely. Without going into it too deeply, the smaller the chip size, the more densely packed the transistors. This results in more speed, less power consumption, and less heat, which would also mean less effort/power consumption/cost to control temperature levels. Types of semiconductor chips Just because two companies may be lumped together based on their places within the foundry business model does not mean they should be viewed as peers or even direct competitors. Despite both being fabless designers, Qualcomm, which focuses heavily on connectivity solutions, should not be likened to fellow fabless designer Nvidia, whose main focus is on graphics processing units (GPUs). Here is a high-level overview of the different kinds of chips that should help members better understand some of the terms thrown around when discussing this industry and a jumping-off point for those that want to research more on their own. Memory : The two main categories of memory chips are NAND and DRAM. Both markets are essentially oligopolies — meaning a few players control the supply. Samsung, Micron and SK Hynix pretty much own the DRAM market. In the NAND market, Samsung and Micron are also big players, in addition to Kioxia (formerly Toshiba), Western Digital (WDC), SK Hynix and Intel. NAND generally refers to a type of flash memory whereas DRAM stands for dynamic random access memory. As you can tell, this industry can get very technical, confusing, and frustrating very fast for those not familiar with the terms. Flash memory (again, think NAND) refers to a type of non-volatile storage medium. Non-volatile simply means that your data won’t be lost once the power goes off. The most common use for flash memory that you may have seen the last time you purchased a personal computer is in solid-state drives (SSD), where all the files are stored on your computer. To really simplify things, when you hear NAND or flash memory, just think of the solid-state drive storage on a consumer PC. (Now you might be asking yourself what an SSD is. Basically, it’s the storage device that is fast replacing traditional hard disk drives (HDDs) seen in older computers. Whereas HDDs have a spinning disc on which data is written, SSDs have no moving parts, making them faster and more secure — though you will pay up for that of course. DRAM is what’s known as a volatile memory, meaning that it will retain data only so long as there is power. Oftentimes when you hear the term RAM or see memory listed on your PC specs, this is what is being referred to. Unlike flash, which stores data and files over long periods even when the power is off, DRAM is “working memory” that is called upon only when needed by a computer processor to perform a given function. The more intense the function, the more DRAM you might need, which is why a computer used for intense video editing or gaming will require more DRAM than one used to simply surf the web and check emails. Microprocessors : The three main processors to know are the central processing unit (CPU), the graphics processing unit (GPU), and the newer data processing unit (DPU). The CPU like those made by Intel and AMD, which essentially have a duopoly — two players control the supply — is basically the brains of a computer. It is responsible for retrieving instructions/inputs, decoding those instructions, and sending them along in order to have an operation carried out to deliver the desired result. As Nvidia put in a past blog post , if the CPU is the brain, then the GPU is the soul. GPUs are more specialized than CPUs and are good at taking on many tasks at once. Whereas a CPU will process data sequentially, a GPU will break down a complex problem into many small tasks and perform them at once. This is why we are seeing their prevalence grow in data centers. While the CPU remains essential, adding a GPU allows for an acceleration in data processing. With more data being transmitted, stored, and processed than ever before — as cloud computing is increasingly adopted and work on deep learning and artificial intelligence advances — speed is crucial. The GPU is essentially a duopoly owned by Nvidia and AMD. A DPU like those made by Marvell Technology and Nvidia is a newer type of processor that is becoming increasingly relevant as data centers become more complex. Nvidia CEO Jensen Huang said in a blog post , “This is going to represent one of the three major pillars of computing going forward. The CPU is for general-purpose computing, the GPU is for accelerated computing, and the DPU, which moves data around the data center, does data processing.” Everything else : To be sure, there are many different types of chips that fall outside of the memory or microprocessor classifications. Examples include those used for 5G, WiFi, Bluetooth, radiofrequency chips, near field communication chips (NFC), application-specific integrated circuit chips (ASICs), and so on. These chips are made by companies like Qualcomm, Marvell Technology, Broadcom (AVGO), ON Semiconductor, NXP Semiconductor (NXPI), and others. Rather than a deep dive into each one — something beyond the scope of this foundry industry analysis — we simply want to highlight that the term semiconductor applies to a broad array of chips designed for different purposes and exposed to different end markets that each have their own demand drivers. Bottom line When investing in the chip industry, it is crucial to understand a target company’s exposure. You don’t necessarily need an engineering level of understanding of how the chips work or are designed, but you do need to have an idea of what end market the company sells into and who the customers are. From there, you can begin to study the relevant end markets to better understand demand trends. Remember, at the end of the day, your primary question and the goal of your research is to understand where the money is flowing. That’s true for all investments but especially so when it comes to semiconductors. Because they are literally everywhere but almost never seen, it’s not as easy as saying, “Apple is going to sell a lot of iPhones this quarter.” Chances are nobody will be sitting around the Thanksgiving table talking about how excited they are for the next-generation memory and microprocessors to drop. However, when you do hear about that new gadget everyone can’t wait to get their hands on, ask yourself, “What semiconductors are sitting inside it?” Despite being a boom/bust industry, sales do tend to increase over the years, and demand throughout the cycle increases thanks to the growing prevalence of semiconductors in our daily lives, with more chips jammed into each device in smaller and smaller form factors. The iPhone, for example, didn’t always have light detection and ranging (LiDAR) technology, but it does now to support new features and that means another chip packed in the handset. While we do see secular growth on longer timeframes, the industry still suffers from brutal boom and bust dynamics. Supply and demand drives all industries, but the semiconductor industry in particular is incredibly sensitive to it. Chipmakers have significant pricing power when demand outstrips supply, as we’ve seen in recent years; new cars will literally sit in the lot collecting dust as manufacturers wait for a key chip. But we see that pricing power turn on a dime once supply exceeds demand, leading to less pricing power and inventory gluts that must be addressed before the next cycle can kick off. (Jim Cramer’s Charitable Trust is long AMD, NVDA, MRVL, QCOM, AMZN, GOOGL, META and AAPL. See here for a full list of the stocks.) As a subscriber to the CNBC Investing Club with Jim Cramer, you will receive a trade alert before Jim makes a trade. Jim waits 45 minutes after sending a trade alert before buying or selling a stock in his charitable trust’s portfolio. If Jim has talked about a stock on CNBC TV, he waits 72 hours after issuing the trade alert before executing the trade. THE ABOVE INVESTING CLUB INFORMATION IS SUBJECT TO OUR TERMS AND CONDITIONS AND PRIVACY POLICY , TOGETHER WITH OUR DISCLAIMER . NO FIDUCIARY OBLIGATION OR DUTY EXISTS, OR IS CREATED, BY VIRTUE OF YOUR RECEIPT OF ANY INFORMATION PROVIDED IN CONNECTION WITH THE INVESTING CLUB. NO SPECIFIC OUTCOME OR PROFIT IS GUARANTEED.
The logo of Nvidia Corporation is seen during the annual Computex computer exhibition in Taipei, Taiwan May 30, 2017.
Tyrone Siu | Reuters
The semiconductor, or chip, industry is perhaps one of the most important and complex sectors in the stock market to understand. Chips are in everything from our smartphones to cars and to all facets of computing, from PCs to massive data centers used for the cloud. Put simply, they are the bricks and mortar of the digital world. We don’t see them, but we know they magically make stuff work.