The headline reads "Intel mothballs new Arizona factory." That sounds like a bad thing, yet the stock price has reacted positively since the information has been available. What is going on here?
Intel (NASDAQ:INTC) has mothballed Fab 42, its largest, newest production factory that was supposed to be the shining star for leading edge production.
The article also comments on the Hillsboro D1X "development" fab. The Oregon development fabs normally develop the manufacturing processes necessary for the next advanced "nodes." They do this in close concert with the design and development groups in Hillsboro, generally in smaller facilities since, after all, these are what the rest of the world would call "prototype" and "pilot production" facilities.
Not this time. The Hillsboro facility is enormous, apparently two 1.1 million sq. ft "development" buildings. One of these was completed, and then the decision was made to build a second identical building in early 2013 and is still under construction. The timing is interesting because by early 2013 Intel had to know that the PC business was going to be slow for an extended period of time. Intel management had to also know that they would not be a player in the mobile device market for the foreseeable future. Intel's path to revenue in mobile produced utter domination in the servers that made the data centers possible, which supports all the mobile devices. Intel is still not a player of magnitude in the mobile device part of the business
Ok, the normal flow is a smallish development fab surrounded by the best technical staff that Intel can assemble that defines and de-bugs the new process. When that is complete the process is transferred to production fabs, as the article says, in Arizona, New Mexico, Ireland and Israel.
Fab42 in Chandler, Arizona, is one of the production fabs. That campus has other production fabs as does New Mexico, Ireland, and Israel that, without D1X and Fab42, seem to be imminently capable of supporting the manufacturing requirements to produce the low $50 billion sales for Intel.
Intel management tells us (Stacy Smith in the recent analyst meeting) that the company is running with an 80% manufacturing utilization rate. Of course, that would be without including D1X and Fab 42.
Here's the rub:
Fab 42 (and D1X) are by far the largest fabs ever built by Intel. The next largest fab for Intel is probably capable of producing 80,000 300mm wafers per month. Fab 42 is at least twice that size, but to be reasonable, I've put the number at 150,000 WSPM (Wafer Starts per Month) at 300mm or about 80,000 WSPM if configured for 450mm wafers.
At 14nm, the average size of a client PC CPU chip probably drops to something around 100 sq. mm and the average AP (Application Processor) for mobile applications should drop to about 50 sq. mm. That means Intel will get 500 14nm CPU chips per wafer and over 1,000 AP chips per wafer.
Intel is the market for CPU chips at about 300 million per year give or take. All of those CPUs would take about 600,000 wafers per year. Remember, Fab 42 should be able to make 1.8 million wafers per year.
100% of the mobile AP chip business, including Apple (NASDAQ:AAPL) and Samsung (OTC:SSNLF), would be about 1 billion of those 50 sq. mm chips at 1,000 per wafer. All the mobile AP business would take another 1 million wafers.
If you're still with me, Fab 42, running a 14nm process, could make all the CPU chips and all the mobile AP used in the world each year. One facility, all those parts. Of course Intel will not have all the mobile business for a long, long time.
Intel obviously knew how big Fab 42 would be. They also know better than all of us mere earthling shareholders how small chips become on a 14nm process.
Intel would NEVER put all its CPU production is a single fab in a single geographic location. As we have seen recently at HYNIX, there still are fab fires.
So, why would Intel build such a huge fab facility as Fab 42 if not for CPU chips?
Why would a mothballed Fab42 need a "Development" fab the size of D1X?
If these two fabs were included in the factory utilization rates, the number would be closer to 40% than the 80% that Stacy Smith says it is.
This all sounds like I am an Intel bear and presenting bad news when actually I am a first degree unrepentant Intel bull.
I have been scratching my head for over a year about the obvious and massive Intel overcapacity and associated capital spending of $40 billion over four years.
The conclusion that I have come to is that Intel has, right under our nose, built out the fab capacity to double the top line sales of the company. Intel hasn't told us what all the capacity will be used for, but as we can see, it won't be used for CPU chips alone.
All the available non-competing foundry businesses in the world would use up about a broom closet in Fab 42. The only sizable customers for foundry are Xilinx (NASDAQ:XLNX), Altera (NASDAQ:ALTR) and Cisco (NASDAQ:CSCO). They have Cisco and Altera in the fold and it will produce less than $1 billion in revenue. Since Altera is an exclusive deal, Xilinx will be left out.
I have speculated that the only product required in large enough volume to fill these cavernous fabs is NAND memory or some other kind of non-volatile memory for use in Solid State Drives ((NYSE:SSD)). A complete and rapid conversion from Hard Disk Drives (HDD) to SSD, using today's memory technology, would require a few Fab 42s. I question whether Fab D1X is really just for process development. Intel has several other fabs on the Hillsboro campus that could serve as 14 nm or 450mm development fabs.
If Fab 42 and D1X were committed to SSD with some materials breakthrough from Intel that could build a 10nm 256Gb NAND chip at 100 sq. mm, those two fabs combined could put out the memory to make a complete industry conversion from HDD to SSD.
Obviously, this is a whole lot of speculation, but the fact remains that Intel has sufficient unused fab capacity to double the sales of the company. They have a plan and we will know it soon because the questions about Fab 42 and D1X will become more and more probing. I'm sure that Intel has good reason to not tell the world about their plans, but it is pretty hard to hide those giant fabs.
In my opinion, Intel is a buy and 2014 is the year.