I have been thinking about this article for a week and missed the Intel (NASDAQ:INTC) guidance upgrade yesterday….snooze and you lose.
I have been talking about Intel for two plus years. Two plus losing years, I might add.
My interest was piqued in late 2010 when Paul Ottelini mentioned that the company would go from a three fab model to a four fab model, while shrinking the technology node from 32nm to 22nm. I wondered why the company would need the extra output of a fourth fab when it should have been considering shutting or slowing down one of the three fabs. Based on that capacity expansion, I have been convinced that Intel had its eye on a new business opportunity of significant magnitude.
Well, the company has continued to spend on production capacity and technology even though sales peaked in 2011 at $54 billion. The company is at least 100% over the capacity necessary to serve the traditional CPU and server chip business.
Some feel that this over building was a bad "mistake" by Intel, citing Occam's Razor to explain the most obvious and simple explanation for the extra capacity. I have a hard time believing that the capacity expansion is a simple mistake. Intel has always been quite good at adjusting to changes in the business environment. It was pretty obvious by the beginning of 2012 that the PC business was going to take a significant hit, yet the spending not only continued but actually increased in 2012 and peaked in 2013. Capex for 2014 is still twice the baseline for the early 2000s.
The mobile business has never been of the scale to explain this capacity expansion. If Intel could nail down 100% on the non-captive mobile business, the scale of that business is in the $10 billion area and the possibility of getting all that business is slim to say the least.
So, let's take a look at the scale I have been talking about for the past two years.
A dollar spent on production capacity can be expected to produce a related increase in sales for any company in any business. Paul Ottelini once mentioned that, with the cost of semiconductor fabs approaching $10 billion in the foreseeable future, a company would need a minimum, repeat minimum, of $20 billion revenue to justify that investment in a new fab. His point being that there simply were not that many companies that can make the cut to invest in state of the art fab facilities.
We can infer from Paul's comment that a dollar spent on semiconductor facility should produce $2 in revenue at a minimum.
Below is a table of Sales vs. Plant, Property and Equipment for the past 10 years. I've also included capex, gross profit margin and net profit margin for some further analysis.
What we see from the table is that Intel PP&E has grown from $17.9 billion in 2011 (which supported its record high revenue of $54 billion) to $31.4 billion today. I can further argue that, since some of the recent assets are accumulating depreciations while never being deployed for production, the actual PP&E value today is closer to $35 billion.
We can also see that the company tends to produce about 2.25 times PP&E in revenue in a steady state operating environment. In 2011 that ratio was over 3:1. I'm betting that everyone was breathing pretty hard to satisfy demand in 2011. Further, the optimum Sales/PP&E ratio seems to be about 2.5:1. At that level Intel produced the best gross profit margin and net profit margin of the past ten years.
My feeling, supported by the above data, is that Intel is at capacity "full throttle" with the brakes on waiting for some business opportunity that it is not willing to share with us "mere mortals" just yet.
So, if you haven't deleted this yet, I have to speculate on the opportunities that are possible and likely and are of the proper scale to fill the obvious extra capacity.
Mobile, of course, is a possibility, but as I mentioned above it is not large enough to fill what amounts to a $50 billion chunk of capacity.
I think the capacity has to do with memory. In the Intel white paper titled "Platform 2015," Intel talks about the desirability of moving processor main memory closer to the CPU chip. It has actually done some of this in the Crystalwell initiative. Crystalwell is a chunk of Intel-made embedded DRAM in a high end i7 processor package. This is relatively small at 128MB, but it increases performance dramatically.
Stay with me now. Imagine that Intel has come up with a memory technology that would be the equivalent to two shrinks of DRAM from 20nm. That would be DRAM at 10nm (or some other technology), a 4Gb DRAM chip at 20nm is about 60 square mm. At 10 nm it might be as small as 20 square mm. That would make an 8Gb (1GB) chip only 40 square mm in area. That piece of silicon would cost Intel about $2.50 to manufacture. Apple spends $10 for 1GB of DRAM in a package on package configuration. A DRAM chip IN the processor package would also have to be worth $10. Now once the "memory wall" is broken down compute performance would go off the chart. Or, in the alternative, the clock rate could be dialed down 30% with an attendant 50% decrease in power consumption and no compute performance degradation.
That little move could lead to sales of 500 million mobile chips where the processor with baseband would be worth $25, DRAM worth $10 and the power savings worth $5 for a total of $40 for a mobile solution that would be very compelling.
So, that could sop up $20 billion of the excess capacity.
The same DRAM put in PC CPU packages would probably also reduce power while generating $12 billion in revenue for Intel.
As long as I'm going crazy on the speculation thing, how about non-volatile memory? In the last investor presentation two senior vice presidents were gushing about "storage," in one case and "non-volatile memory" in the other case. They didn't say NAND, they said "storage" and "NVM."
What if Intel, in concert with Micron (NASDAQ:MU), has come up with a non-volatile memory technology that is effectively only two nodes ahead of today's 16nm NAND. That would put a 128GB chip on about 45 square mm of silicon at about a $2 per chip cost. That's about $16 in silicon for solid state drives in 128GB chunks. That would be SSDs at $.30 per GB with a 60% gross margin. Two million Intel wafers could make 300 million of those 128GB SSDs. That would sop up another $12 billion of the Intel over capacity.
It is kind of fun to let your mind wander on the possible homes for this $50 billion in leading edge semiconductor capacity that is sitting unused at Intel and I invite commenters to make other speculations.
In summary, Intel continues to make this "mistake" by spending billions of dollars on production capacity.
Through materials and process research, Intel alone has enabled all semiconductor progress since 100 nm.
Can we really buy the idea that Intel continues to expand its "mistake?"
I believe there is a plan and that it will come on us in a hurry when Intel is ready. I think that by the end of 2015 Intel will show that it can produce $88 billion in sales from that $35 billion in PP&E. Operating in the efficient 2.5 time PP&E range, the company will generate 26% net. When the capex slows down the share repurchase program will get turbocharged and bring shares down to 4.5 billion. I think Intel will earn $5+per share. At a 40% payout, the dividend will be $2/share, driving a 15 P/E and a $75 price, just like in 2000.
I am in Intel call options up to my ears.
Disclosure: The author is long INTC. The author wrote this article themselves, and it expresses their own opinions. The author is not receiving compensation for it (other than from Seeking Alpha). The author has no business relationship with any company whose stock is mentioned in this article.