Intel (INTC) is the world's largest semiconductor producer. 2011 total sales were $54 billion, with semiconductors making up $49.7 billion of that total. Since Intel's 2012 sales are expected to be flat with 2011, we can expect about $50 billion in semiconductors.
So, we know that Intel had the capacity to make nearly $50 billion worth of semiconductors in 2011. That might seem like a "duhhh" comment, but read on. That $50 billion in revenue was made with capacity that existed at the end of 2010, which was three main fab complexes. In the fourth quarter, 2010 conference call (Jan 2011), CEO Paul Otellini made some comments. One was that the company would move to 22nm feature size for microprocessors. That move by itself would have allowed the existing revenue to be made in two of those three complexes. In the next paragraph, Mr. Otellini told everyone listening that the company would not only maintain the three fab complexes, but it would outfit another fab complex, for a total of four major complexes. Clearly, Mr. Otellini and Intel management had big plans for 22nm production.
Since that time Intel has spent $25 billion+ on CapEx, suggesting that they have funded and built and equipped six fabs. Last quarter, the company took a charge for the expense of carrying unused capacity. This from a company that sensed and adjusted to the business decline of the Great Recession better than any company that I know of.
Why $50 billion in ready-to-go capacity? The line of sight for the smart mobile devices market is one billion units in the very near future. With an Application Processor SoC at $25 and an LTE modem and transceiver set at $25, the total business opportunity is $50 billion, what a coincidence.
To take a little side trip here; due to Intel's Tick-Tock strategy for moving the manufacturing technology down the road, a certain amount of extra capacity is required so that when they make the transitions to the next technology, they don't run short on capacity to handle that ongoing business. These transitions are a never-ending constant in the corporate life of Intel. That assurance of supply to customers is embedded in the very genes of the Intel business model. With regard to that assurance of supply to customers, I believe that Intel truly believes that they have the products to go from "zero to hero" in mobile a very short period of time. They will need every bit of this capacity to ensure timely delivery to nearly all mobile device customers. Of course, by the time that the already-started conversion to 14nm is complete those six fabs will have the dollar output of nine 22nm fabs.
Since business is flat and assuming that Intel management are not a bunch of idiot children, these very expensive, state-of-the-art fabs will not go unused for very long.
Here's the chess game part: For part of 2013, and maybe all of 2013, three of the existing six fabs will go empty, or each operating significantly below capacity. Why would Intel allow themselves to get into this underutilization position? One reason is that those three fabs represent nearly $50 billion worth of sales potential. That potential is at a node and soon two nodes ahead of the next best in class. If you are TSMC (TSM) seeing $50 billion worth of unused semiconductor capacity sitting at Intel is nothing if it isn't intimidating.
Consider (from the standpoint of TSMC management): If Apple (AAPL) decides to move their "A" chip fab business to Intel from Samsung (OTC:SSNLF), about one of those Intel fabs would no longer be empty. If the Apple 4G LTE and radio transceiver business came along with the application processor, it would consume the output of 1.5 of those idle Intel fabs. With that one move, TSMC can no longer justify the investment in additional fab capacity that would be obsolete as soon as it came on stream. Such a move in the Apple business would be a double whammy to TSMC because if Samsung lost the Apple business and TSMC didn't get it, Samsung immediately becomes a mega-competitor to TSMC in the standard foundry business.
Consider further that Intel comes to market with a "Super SoC" that is so far ahead of anything that Qualcomm (QCOM) or NVidia (NVDA) has now or has planned that all non-Intel smartphone SoC chips would instantly be viewed as obsolete. There is some evidence that such a device is coming down the road sooner than expected from Intel. It is called Merrifield by Intel and is dual core, new architecture Atom based full function application processor SoC. Some leaked (rumored) search benchmarks put this chip at three times the performance of the best ARM (ARMH) based application processors. The Merrifield chip was supposed to be delayed until 2014, yet we have these leaked benchmarks. Another chess move by Intel?
The Intel roadmaps indicate that a Clovertrail chip should be next to market. That chip is a dual core modified medfield with better graphics and some tweaks to the CPU section. Has Intel leapfrogged the Clovertrail and jammed the 22nm Merrifield based Redhook Bay into early production? Even if we don't see production of Merrifield until 2014, customers and competitors now know (or suspect) that this high performance core is just over the next hill. That simple fact will cause customers to delay new designs until this new super part is available.
Those three unfilled Intel fabs have the effect of freezing any potential competition in place regarding capacity additions. Without significant capacity additions TSMC and other fabs will not be able to service the next stage of mobile device growth. QCOM and NVDA will have no path beyond 28nm.
Intel is perfectly transparent. These unfilled fabs are physically there for anyone to see. The company publishes detailed roadmaps so that customers and competitors alike can make plans on how to respond.
What happens to Qualcomm if Intel introduces a Superchip and gets the Apple fab business? With no access to 22nm, moving down to 14nm, Qualcomm has to re-assess their commitment to the mobile IC business. The Qualcomm mobile IC business is about 2/3 of their total sales, but only 28% of the operating earnings, while the royalty part of the business is about 1/3 of sales and 70% of operating earnings. If I were staring at that set of facts and looking at what the obvious intentions of Intel are, I would have a plan for winding down the mobile IC business.
Qualcomm has recently bought Atheros Communications which is a play in the less congested networking business where, get your head around this, Intel might be a possibility as a foundry, much like the rumors circulating about Intel doing foundry work for Cisco System (CSCO). Qualcomm has also made an investment in the Japanese display maker Sharp. Perhaps Qualcomm will decide that the royalty business and a set of complementary joint ventures is a better use of their cash stash than competing with Intel. Qualcomm's history includes making cellular base stations and early cellphones, businesses they left for better opportunities elsewhere. That could happen again.
What happens to Nvidia ? The discrete graphic chip business in rapidly being chewed up by the increasingly more functional GPU being integrated into the Intel CPU chips. Their mobile business goes the same way as with Qualcomm.