Shares of several semiconductor companies soared after reporting year-end results. If you are taking a current look at this sector, the key challenge is valuation. I see too many Johnny-come-lately buy recommendations. For example, Marvell Technology Group (MRVL) is getting deservedly good press. But the shares were far more attractive one year ago. After doubling in share price over the calendar year, Marvell's stock now trades at almost 40x its forward earnings estimate (i.e., a forward P/E ratio of 40x). You should start to be paranoid at these levels. A good way to lose money is to buy into high-growth stocks after they have been pumped up. On the one hand, several chip makers have terrific short- and long-run prospects. On the other hand, many of them now have aggressive growth expectations already built into current prices.

The industry is notoriously cyclical. According to the Semiconductor Industry Association, "from its inception, the semiconductor industry has been cyclical. Cycles typically included two strong years of 20 percent growth, one year of slow growth, and one year of flat or declining growth." (source: SIA 2005 Annual Report). Against this remarkably dependable pattern, 2005 was supposed to be the "slow growth" year-after +18% growth in 2003 and a record-setting +28% growth in 2004-and 2006 was supposed to be a roughly flat year. Against cycle theory and high valuations, this should be a bad time to buy into chips.

But things change and so too does the semiconductor industry. Cycle theory may be less helpful going forward: better inventory management will reduce recovery times and consumer markets are overwhelming corporate spending cycles. Corporate information technology (IT) budgets used to be the primary growth driver, but the mantle has switched to consumers. Consumers now account for at least one-half of semiconductor sales, and if the fourth quarter was any indication, they will account for most of the new growth. At least one of these consumer markets, the market for DSL broadband, is growing without discernable cyclicality or seasonality!

The popular story this year is how Advanced Micro Devices (AMD) is stealing market share from Intel (INTC). AMD's Computation Product Group (CPG) produced fourth quarter sales of $1.3 billion, a 79% increase over the prior year's quarter. CPG makes chips for desktops, notebooks, and servers; this group breached the 20% level of market share against Intel. As finely detailed in a recent BusinessWeek cover story, Intel clearly recognizes the need for a new future. But don't get distracted by history: PC chips are not the important market here. Therefore, competition with AMD is not really the issue. For Intel, the enemy is thyself and the long-term investment thesis turns on whether this titanic chipmaker can, in fact, change directions to capture new consumer markets.

One way to understand the semiconductor landscape is to compare growth to returns. Specifically, in the chart below I mapped recent earnings growth (on the x axis) to return on invested capital (ROIC, the y-axis). ROIC is a useful metric for semiconductor companies because semiconductors require massive investments into fixed assets. When analyzing semiconductor companies, it is especially dubious to consider only earnings growth because you don't know how much capital is used for the growth. It is entirely possible, and not altogether uncommon, for a growing company to destroy shareholder value if it consumes too much capital along the way.


This chart illustrates the natural tension between growth and return in a capital intensive industry. Intel (INTC) leads the pack in return on invested capital (ROIC) at 30%. They also grew earnings year-over-year by 21%. These are phenomenal economics for their size; in few industries would such earnings growth trail behind so many competitors. While Intel is highly diversified, Marvell is the reverse: more concentrated, lower ROIC, but unworldly earnings growth. Two companies that standout in terms of balance-that is, balancing diversified growth with ROIC-are Broadcom (BRCM) and, to a lesser extent, Texas Instruments (TXN).

Broadcom (BRCM) jumped 24% last week after beating their own raised guidance and sending several shareholder-friendly signals including a stock split, accelerated share buybacks, and accelerating guidance going into 2006. Broadcom shows strength in several sizzling markets. A key sales driver this year was their Blutonium family of Bluetooth chips. Bluetooth is a wireless standard that enables so-called personal area networks (PANs) by connecting personal devices across short-ranges. Also coming on stronger and sooner than expected are mobile multimedia chips. Consumer devices will continue to need low-power, high performance manipulation of graphics, music/video streaming, and gaming applications.

Texas Instruments (TXN) did not beat expectations and the shares suffered. But I like them for the long-run because they dominate in digital signal processor (DSPs) and analog chipsets. Most of our "real-world" information is analog (e.g., temperatures, visual images, sound waves). Analog chips convert these signals into digital format. This is already a huge, competitive market but Texas Instruments has a dominant position and this market is going to outpace most of the other semiconductor segments. Many new applications will require analog-to-digital processing. Most of TXN's DSP revenue comes from the cell-phone market, which is rapidly growing and not yet mature. The advantage for TXN is that they tend to sell DSP chips that are more complex and higher-up the value chain.

If I were you, I would focus on the emergent, fast growing markets in semiconductors. And they abound! The Federal Communications Commission (FCC) will soon mandate that television stations must broadcast in digital format. This is one driver of the convergence toward a "digital living room" that will contain a multimedia PC and a high-definition television (HDTV). Demand for cable and satellite set-top boxes will increase and all of these devices (e.g., home-based networks, HDTV, gaming consoles, digital cameras) require semiconductors.

Two other interesting markets are voice over IP (VOIP) and RFID. VOIP is growing fast and requires semiconductors; e.g., IP phones, cordless phones, and IP adapters. Radio frequency identification (RFID) systems will provide a pervasive platform for the application of semiconductors. RFID systems put physical tags-or labels-on objects (e.g., inventory, produce, people) such that RFID "readers" can locate and identify their position. Despite technical glitches (and the occasional challenge from privacy advocates) Wal mart is already proving that RFID improves inventory processing. But this is just the obvious application. In his book RFID essentials, Himanshu Bhatt predicts a provocative future era that he calls "the internet of things." He predicts everything will be tagged-that RFID will enable digital sensor networks that are so pervasive that, like electricity, they will barely be noticeable: "in this era, physical objects will be tied to the Internet through their digital identities."

The exponential increase in productivity exhibited by Moore's Law is certainly one of technology's greatest marvels. Transistor thickness has reached 90 nanometers, which is less than one-hundredth the length of a micron, which in turn is one-millionth of a meter! In other words, words don't help us comprehend the magic performed by these scientists in unimaginably small laboratories. Put another way, semiconductor manufacturing is now nanotechnology and will look to nanotechnology for the next paradigm. Intel has two production facilities using 65 nanometer processes and, last week, they announced the development of a 45 nanometer process. Only a few years ago, many people thought such shrinkage was impossible. Light itself becomes clumsy in such small spaces. Amazingly, the Software Industry Association now predicts that Moore's Law can be prolonged under the current paradigm until around 2020. Stay tuned!