Semiconductor fabs (fabrication facilities) are expensive things. They cost billions of dollars to build and equip. $5-6 billion will buy a fab capable of 100,000 300mm wafer production per month of logic products, more output if the product is memory. So what happens to the state-of-the-art fabs when they are no longer state-of-the-art? Do they go to "fab heaven"? No, they are sometimes upgraded to be SOTA again, sometimes they are mothballed. Sometimes, as happened in the recent Elpida/Micron (MU) acquisition, the fab (and company) is run through bankruptcy to come out the other side as a fab with written off assets. That happened with Elpida. Micron bought Elpida with Elpida's own money for a fraction of the value of the production assets embedded in the company. Now the Elpida operation of Micron is the world's lowest cost memory producer because there is no (very little) depreciation in their cost of goods sold. Micron really held up the stage coach on the Elpida deal.
Depreciation can be anywhere from 20% to over 50% of the cost of producing an integrated circuit depending on the technology used.
For example: Samsung (SSNHY.OB) is spending $7 billion to build a NAND facility in china. A memory facility of that cost will be capable of producing about 150,000 wafers per month. These facilities are often depreciated over very short period to protect profits from taxes and because the technology moves along very rapidly. So, Samsung will probably write this facility to zero over five years. In five years that facility will make nine million wafers, so $7 billion will have be amortized over nine million wafers. Pulling out my trusty iPhone calculator and pushing a few buttons, gives me a number of $778 of depreciation embedded in every one of those nine million Samsung wafers. How would you like to be Samsung competing against Elpida/Micron with … umm, zero depreciation at Elpida?
OK, now I'm going to switch gears on you. NAND memory is expected to grow substantially over the next three years. This will be due to the adoption of NAND based solid state drives in virtually all PCs. So, Samsung is doing the right thing by expanding production of NAND. Intel will ship, in a bad year, 300 million+ processor chips. Each of these processor chips will need either a hard disk drive or a solid state drive. The Intel sponsored Ultra book pretty much dictates solid state drives.
The world of NAND memory is about in balance between supply and demand today. Apple (AAPL) is almost 100% SSDs; PCs will be next. The balance will not last.
Scale of the issue: 300 million SSDs at an average density of 192GB, 12 128Gb chips times 300 million SSDs each year. The newest processes will produce a 128Gb NAND chip on about 150 sq. mm of silicon. We will need 3.6 billion of those 128Gb chips per year. A single wafer will produce about 300 of these 128Gb NAND chips. So, 3.6 billion divided by 300 returns 12 million wafers per year or about seven of those Samsung fabs. That's $49 billion in capex that has to happen soon in order to satisfy the burgeoning and near certain demand for SSDs and the NAND chips to make them.
Where are those fabs going to come from? Where is the money to build them going to come from? These SSDs need to be cheap. Wouldn't it be nice if we had a few fully depreciated semiconductor fabs in which to build these NAND parts? Any available fabs would be too small and too ancient to use for production of these sophisticated parts. Except for one place. Intel (INTC). Here's a list of the Intel manufacturing facilities by location and technology.
Including 2010 through 2013, Intel will have spent $36 billion on new and upgraded fab facilities. This activity has taken place, primarily, at their Chandler AZ and Hillsboro, OR locations. Each of these locations is a campus of fabs. Either of them could produce all the wafers currently required to support the Intel business level at 14nm. The other campus could supply the world's requirements for mobile products, again at 14nm.
In the process of all this spending and construction, Intel has generated a whole bunch of what? You guessed it, fully depreciated but still high tech fab capacity that could go to fab heaven, or be put to work in a lights out environment making the lowest cost NAND memories on the planet.
A little review: It take NAND memory to make SSDs, it take SSDs to make Ultrabooks. If SSDs become scarce, no one can sell Ultrabooks, if no one can sell Ultrabooks, Intel can't sell 65% gross margin Haswell chips. Intel has a very direct interest in sufficient NAND supply and virtually no interest in being directly in the memory business.
So what's an answer? Intel is in a joint venture with Micron that involves all memory technology. Intel has excess, high quality fab capacity, and money. Micron has a nice business now and they DO know the memory business. A smart thing to do would be to lease some or all of the soon to be unused Intel facilities in Ireland, New Mexico, and Israel to IMFT, the Micron/Intel joint venture, for the production of NAND memory. These plants could be throttled by the demand for NAND, something the traditional memory business hasn't been able to do, because with multi-billion dollar investments, you better produce around the clock regardless of the market price for the parts. That's the old depreciation thing I started to talk about at the beginning of this article.
I sincerely believe that some of the Intel excess fab capacity will be utilized to make sure there is sufficient NAND product available to support the ultimate goal of unrestricted Haswell sales. And Intel and Micron will make some nice money in the process.