The fabless model -- in which semiconductor companies outsource the production of silicon chips -- was supposed to be a good thing. It would allow design firms like Qualcomm
(NASDAQ:QCOM) to save themselves the capital expense of a foundry, and have more options for managing their supply chain. When the semiconductor industry began moving towards ARM's
(NASDAQ:ARMH) architecture, it was an extension of this logic. ARM could avoid a huge R&D burden by simply licensing its design, while clients like Qualcomm would benefit from a more competitive field, with one less barrier to entry. Breaking up the industry in this way meant less revenue for each company, but also a diversification of risk, and more flexibility. Companies could focus on what they did best, and problems at one manufacturer, or the failure of a single system-on-a-chip (SoC), wouldn't sink the whole ship.
So the theory went. Today, fabless designers are the norm, and ARM's designs are ubiquitous in smartphones and tablets -- but instead of diversification we've seen consolidation. Qualcomm alone controls half of the smartphone SoC market, with Nvidia
(NASDAQ:NVDA) in an entrenched second place. Smaller competitors like Texas Instruments
(NASDAQ:TXN) have been forced out
. The idea of an open and diverse industry has given way to the hard reality of foundry retooling costs, and the inability of smaller chip designers to command timely and affordable manufacturing services. The industry is no less aristocratic now than it was 10 years ago, when Intel
(NASDAQ:INTC) and AMD
(NYSE:AMD) controlled it. A handset maker looking to buy a SoC has few choices -- a dynamic that likely played into Apple's
(NASDAQ:AAPL) decision to produce its own line with the A4.
ARM's R&D-light model is revenue-light as well, and the company has struggled to produce a comprehensive library of designs. The turf war with Intel has manifested in an emphasis on the tablet market, where Intel's Atom is more competitive. The result is a lack of variety, which Apple responded to by abandoning ARM's catalogue
altogether and implementing a design of its own. Despite threatening moves towards servers, ARM is still at least a year away from a 64-bit offering.
If a licensed architecture hasn't exactly opened up the chip design industry, the fabless model isn’t doing any better at diversifying the manufacturing end. Large-scale independent contract (IC) foundries are few and far between. Taiwan Semiconductor
(TPE:2330) is easily the largest, with $17 billion in foundry revenue, while Samsung
(PINK:SSNLF) and Globalfoundries jockey for second at $4.5 billion a piece. UMC
(TWSE:2302) pulls in at third, and after that the field drops off quickly. A 50% annual growth in smartphones has given these companies as much work as they can take. Qualcomm wrestled with supply shortages for much of 2012, and vented publicly about its relationship
with TSMC. In desperation, the company was forced to make arrangements with Globalfoundries and even Samsung -- a competitor, with its own line of Exynos SoCs.
Qualcomm's isn't the only uncomfortable situation. Apple has a troubled relationship with Samsung, and rumor suggests that the iPhone-maker is looking for another supplier. Unfortunately, TSMC won’t give Apple priority
, and Globalfoundries has limited capacity at the 28nm manufacturing node that chip designers are demanding. So when Samsung presented it with a 20% price hike late last year, Apple responded with an obliging thank you
. Nvidia has meanwhile voiced its discontent
with TSMC and the high costs it's charging for newer, more miniaturized processes, but like Apple, Nvidia faces a dismal decision: Get in line or go home.
So although some of the barriers to entry may have been removed from the semiconductor industry, the advantages of scale haven’t; and the small club of major chip designers is just as constrained by their supply side as they would have been had they built their own foundries -- an idea Qualcomm isn't ruling out
. This hasn't prevented them from being successful, or held back the ARM ecosystem in general. Are they producing better, more efficient processors? Several studies
-- just chips optimized for a different purpose.
The principal advantage of the fabless model boils down to the fact that it's been better able to benefit from cheaper manufacturing overseas. Of the five largest IC foundries, four operate in Southeast Asia. Here, too, the result hasn't been diversification but congregation – not more but fewer choices, and not a spreading of risk, but a geographical concentration of it.
South Korea, Taiwan, and China all operate export economies, which they have maintained over the last decade through currency intervention, easy credit, and other methods. All three are dealing with the financial fallout from policies that pushed real interest rates deep into negative territory following the ’08 crisis. South Korea is struggling with an explosion in household debt, while in China and Taiwan the culprit is real estate. Their currencies are trading in lockstep as they try to reign in credit -- and move towards personal consumption economies -- without giving too much of a trade advantage to their rivals.
These countries are making an orderly exit from the export model that provided -- and still provides -- such cost advantages for fabless chip companies. The bottom line is that, at some point, Apple won’t be the only one facing higher charges. A lack of alternatives means that the semiconductor industry will have no choice but to pay up, and it may find that a good theory is small consolation for a bad price.