Briefly, Intel may be a good investment due to:
- Cheap. The P/E is around 10.3.
- In the long run, Intel will be able to consistently grow its revenues as we find more uses for computers (e.g. smartphones, tablets, cloud). 8% revenue growth will likely continue into the near future.
- Its economic moat. Its dominant market position lets it enjoy economies of scale in R&D and in manufacturing. On the fabrication side, Intel is almost always a step ahead when it comes to process size and process technology.
Processor design: Usually Intel will have a better processor design than its archrival AMD, though this is not always the case. Intel and AMD were going neck to neck in terms of performance around 2003-2006. Intel’s Prescott processors for example were disappointing since heat and power consumption was a much bigger problem than anticipated. In the short run, there is some variation in how well Intel’s processor designs turn out. In the long run Intel’s size will be an advantage when it comes to outspending competitors on R&D.
Process size: Process size refers to how small processor parts can be made. Being a node ahead in process size means that a semiconductor manufacturer can cram twice as many transistors onto the same area of silicon. Usually the extra transistors can be used to double performance.
Process technology: Intel was ahead of its competition in implementing high-K dielectrics and finFET technology (Intel refers to their finFET variation as “Tri-Gate”). What this boils down to is that these technologies make processors better in performance and/or power consumption.
My tech background is mostly in programming programming Photoshop plug-ins. I am not an expert in semiconductor technology by any means (I just Google this stuff). However, it seems obvious to me that Intel has an ‘unfair’ advantage when competing against its competitors. Historically, it has almost always been a step ahead in terms of process size and process technology. Currently, Intel is shipping 22nm parts (with Tri-Gate and high-k dieletrics) while competitors are shipping 32nm (no Tri-Gate/finFET, some fabs have high-k dielectrics). Some companies are shipping 28nm parts, which are a half-node between 22nm and 32nm. Intel may be 3-4 years ahead in implementing finFET technology.
Here is a chart showing Intel’s share price versus AMD’s… this is Intel’s moat at work:
Where Intel may experience growth (or lose market share)
Desktop/laptop/netbook: Right now, Intel has a historically higher market share against AMD than in the past. Reversion to the mean could occur. In the long run, I would expect that the trend in the chart above will continue.
Servers, high performance computing: This market is a little different than the desktop market since servers tend to be single purpose. This means that the x86 instruction set (which only Intel and AMD control) is not that important. Unlike the desktop market, the x86 instruction set does not protect Intel and AMD in these specialized markets.
Intel management should be congratulated as it has taken away a lot of market share from RISC-based processors (IBM’s Power, SGI, Sun’s SPARC, etc.). In the past, the RISC technology/approach to processor design was superior since it led to increased performance. (As I understand it) Intel’s processors actually implement RISC internally and have to waste transistors on converting x86 instructions to RISC(-like) instructions. Nowadays this overhead is no longer an issue since transistor counts are so high that the overhead is negligible.
Smartphones: Intel has practically no market share in smartphones right now. This market is dominated by products based on the ARM instruction set. I don’t believe that the instruction set will be a big problem for Intel as it is not impossible for handset manufacturers to switch to Intel’s x86 instruction set. (In the past, Apple switched from powerPC processors to Intel processors. The technical barriers are minor and easy to overcome. The switch went smoothly, though some Mac users had to wait a few months for updated software.) The Xolo smartphone demonstrates that it is not difficult for Android phones to use an Intel x86 processor. There are a few games where the Intel processor is slower due to the emulation of the ARM instruction set. However, I don’t believe that this will be an issue in the future as developers will compile a version of their games for x86/Intel (very easy to do).
There is a misconception that Intel processors are inferior to ARM-based products due to power consumption. This review of the Xolo smartphone shows that Intel’s Medfield chip has around the same power consumption (and also performance) as its competition. I believe that Intel will dominate this field as its chips will have advantages in power/price/performance due to Intel’s advantages in fabrication. Intel’s lead in process size and in implementing finFET will allow its processors to have lower power consumption. The Medfield chip was manufactured on a 32nm process without finFET. Intel will gain ground once its moves production to its best fabs.
However, the smartphone market might only turn out to be a $10B market*. In the year ended 2011 Intel made $54B. Smartphones may not add substantially to Intel’s profits. (*This Strategy Analytics blog post states that the smartphone processor market was $7.9B in 2011. I believe that it is reasonable for the smartphone market to grow to $10B in the future.)
Intel may have some difficulty getting Apple to adopt its smartphone processors. Apple’s strategy has been to restrict the variety of hardware in Apple smartphones. This makes it much easier for developers to make smartphone apps. (Some game manufacturers no longer make games for Android as it costs more to develop for, Android users buy less games, and Android users require more technical support.)
Tablets: This will likely be a similar situation to smartphones.
Graphics, GPGPU: In budget applications, Intel competes with AMD as they both sell processors with integrated graphics. Intel’s products tilt towards lower cost and/or faster CPU performance at the expense of inferior graphics (e.g. worse at playing 3D games). Intel may have difficulty doing higher-end graphics as it is not so easy to make good software drivers to go along with the hardware.
In the past, Intel had an experimental Larrabee project (Wikipedia) which tried to make a high-end graphics chip/GPU that could also be used for GPGPU purposes (Wikipedia). This project was mostly a failure. The outcome of the Larrabee project has morphed into the Knights Corner / Xeon Phi coprocessor (which I don’t think will be a huge market for Intel) and will not be a GPU.
Intel may have difficulty breaking into the high-end graphics market, which is the traditional bread and butter of ATI/AMD and Nvidia.
FPGAs: This is a market dominated by Xilinx and Altera (89% market share combined). Intel has tried to enter this market in the past and failed. It will not be easy for Intel to break into as Intel would have to develop really good FPGA design software (this is Xilinx and Altera’s moat). GPGPUs will eat away at the FPGA market though as GPGPUs are better at some applications than FPGAs.
Back to the bigger picture
Historically, Intel has grown revenues at a rate of several percent a year. Gurufocus shows Intel’s growth to be 8.6%/yr per share over the past 10 years. Going forward, I expect revenues to grow alongside the growth of the computing industry. On top of that, Intel will likely slowly expand into new markets. Historically, it has taken Intel a while to expand into new markets. Keep in mind that it had to focus on fighting AMD for dominance of the desktop PC market and faced very strong competition in the server market. Intel has failed in the FPGA and high-end graphics markets.
I don’t think that one can be too accurate in predicting Intel’s future growth. However, my best guess would be that Intel will continue its historical rate of growth. To me, it is reasonable to assume that Intel will slowly encroach onto new markets. A lot of these markets exist in the first place because they serve a niche that the x86 desktop does not fill; that’s why Intel/AMD haven’t wiped out these markets in the first place. Intel’s existing markets will also grow due to developing countries increasing the penetration of computers (desktops, laptops, servers, etc.). It will also grow as people figure out new applications for computer hardware as it becomes more powerful, the Internet gets faster and more pervasive, and software tools continue to lower development costs.
Smartphones, tablets, and cloud computing are hot new areas but they may not be large markets for Intel. If I am wrong, then Intel shares will have additional upside.
Moore’s Law reaching its limits
Moore’s law predicts that transistor counts will double every 18 to 24 months. Historically, this has meant that performance doubles around every 2 years. However, processors will not get faster at their historical rate.
- Heat limitations mean that we have hit a wall in terms of clock speed. This is forcing processor manufacturers to use multiple cores to increase performance. Unfortunately, doubling the number of cores does not always double performance. One of the problems is Amdahl’s Law. It is likely that Amdahl’s Law will kick into place for commercial software. Making programs completely parallel will often be too costly (parallelism is hard).
- Eventually, we will not be able to double the number of transistors every 2 years. Historically, we have only been able to see a decade out and people always predict the imminent end of Moore’s Law. Yet Moore’s Law keeps on continuing. However, there are some very real limitations to the current approach. We cannot have processor parts that are smaller than one molecule. Before that, there are other technical challenges imposed by the laws of physics (e.g. electron tunneling). Moore’s Law may end as early as 2022.
I believe that the rate of performance increase will start to slow down in the future. If that rate slows down to a trickle, Intel will have a problem as its competitors will start to catch up in terms of performance. Eventually, the business will likely morph into a commodity business where the lowest-cost players have an advantage. (The x86 license will give Intel and AMD a very minor advantage in the desktop market.) This will be terrible for Intel as its 20%+ profit margins will become a fraction of what it is.
Financial management at Intel may be merely average
During the tech bubble, Intel was spending cash on buying back shares at astronomical prices under Craig Barrett’s tenure. (Barrett named Paul Otellini, who is the current CEO.)
Intel has a defined benefit pension plan. I’m not a fan of these types of pensions as the liability can push companies into bankruptcy (and even the pensioners get screwed, e.g. Nortel).
Intel’s acquisition of McAfee was arguably a bad one. The main reason for the acquisition is McAfee’s Deep Defender product (review/overview), which works alongside Intel processors to detect rootkit viruses (these viruses are very difficult to detect with a virus scan). Intel bought McAfee not knowing if there is merit to Deep Defender’s approach (it wasn’t tried on a commercial scale when Intel made the purchase and there are other methods of detecting rootkits). Another caveat is that Deep Defender ties users to purchasing Intel chips with the premium VPro feature. IT departments may not want to be tied to a single vendor.
On the good side, Intel is using its excess cash to buy back shares and pay dividends. Considering Intel’s current P/E, the share buybacks make a lot of sense. Also, Intel has conservative levels of debt and I like that it has taken on lower risk debt that is due far in the future with reasonable interest rates (e.g. some of Intel’s debt is in 30 year bonds).
Intel’s P/E may be slightly inflated
The cost of Intel’s cross-licensing agreement with Nvidia will be amortized over 17 years even though the cross-licensing agreement expires in 6 years. From the 10-K for YE2011:
In January 2011, we entered into a long-term patent cross-license agreement with NVIDIA. Under the agreement, we received a license to all of NVIDIA’s patents with a capture period that runs through March 2017 while NVIDIA products are licensed to our patents, subject to exclusions for x86 products, certain chipsets, and certain flash memory technology products. The agreement also included settlement of the existing litigation between the companies as well as broad mutual general releases. We agreed to make payments totaling $1.5 billion to NVIDIA over six years ($300 million in each of January 2011, 2012, and 2013; and $200 million in each of January 2014, 2015, and 2016), which
resulted in a liability totaling approximately $1.4 billion, on a discounted basis. In the fourth quarter of 2010, we recognized an expense of $100 million related to the litigation settlement. In the first quarter of 2011, we recognized the remaining amount of $1.3 billion as licensed technology, which will be amortized into cost of sales over its estimated useful life of 17 years.
(*I am no lawyer and may have read this incorrectly. The patent cross license agreement is also on Intel’s website.)
Berkshire Hathaway buying IBM and selling Intel
I don’t know why Berkshire did this. I do not consider IBM to be in my circle of competence as I don’t understand the IT business. (In the processor business, I can tell you that IBM lost the Mac business and some of its server and supercomputer business to Intel.) Warren explains his IBM purchase in this CNBC transcript. At ~$10B, Warren’s position in IBM is rather sizable.
Maybe the smarter move is to buy IBM instead of Intel. I do not know.
What I plan on doing
I will hold onto Intel on the expectation that its share price goes up on gaining market share in smartphones and tablets. I don’t want to deal with Moore’s Law coming to an end so I don’t see myself holding onto Intel for the really long term.
*Disclosure: Long Intel. No position in IBM.