Intel (INTC) is primarily a manufacturing company. They make and sell silicon chips. Like any manufacturing company, they need to keep the factory running at full capacity to maximize profits. Last year, Intel only used about half their capacity. This article addresses two issues: What can they do with the other half, and how much can that improve Intel's earnings?
Intel's full capacity is the equivalent of about 441,000 wafers (300 mm equivalent) per month. A 300 mm wafer is about 7x105 mm2. The total capacity can also be measured in square millimeters; it is 3.7 1011 mm2/yr. A 300 mm wafer is about 7x105 mm2, but in this article it will be easier to think of things in terms of 100 mm2 dies. One 300 mm wafer has about 700 of these 100 mm2 dies.
A silicon foundry makes money by selling processed silicon wafers to others. A single wafer contains many copies of the silicon chip being produced. An individual chip is called a die. After packaging and testing, this die will be the product that is sold to a customer. The same die, or piece of silicon, can be engineered into products with vastly different values. At the high end are field programmable gate arrays (FPGAs), which can be sold for over a thousand dollars per 100 mm2 chip (but not many of these are sold per year), and at the low end is memory, which goes for about $10 per 100 mm2 chip.
Intel needs to maximize their profit per wafer. Intel's incremental cost to manufacture a wafer is not disclosed, and it will depend upon the details of the particular design (number of masks, number of metal layers), but I estimate it to be about $500-$1000 for a 300 mm wafer. I'm going to use round numbers and call it $700, then I can use the incremental cost of a typical die (100mm2 of silicon) to be about $1.
You will notice that even memory seems highly profitable. You can sell it for $10, but it only costs $1 to produce. That's because I'm leaving out the fact that the factory costs $5 billion to set up and is out of date in 5 years. Plus, it might need $1 billion/year in maintenance. But those are sunk costs, and I'm only considering incremental costs here.
A single die is about 100 mm2. Of course, there is a large variation. Intel has sold some single core processors with die sizes as small as 26 mm2, and an Itanium processor that had a die size of almost 700 mm2, the Tukwila. Many other chips: Apple's (AAPL) A6, Qualcomm's (QCOM) MSM series, and even the largest current NAND memory 64 Gb, have die sizes close to 100 mm2.
Intel's strategy should be, and seems to be, to use their fabs to only produce the most expensive designs. What are the most expensive chips per square mm? Take a look at this chart: the numbers are approximate, because I'm just trying to get an idea of how valuable things are.
5 108 mm2
15 1010 mm
Qualcomm MSM modems (Snapdragon)
6 1010 mm2
Now, if you look at this graphic depiction of the problem, where I've indicated the volume (number of units per year) by shaded areas, the problem looks like this:
Intel had so much excess capacity last year; they could fit Apple's A6 business and Qualcomm's MSM business into that capacity along with Xilinx (XLNX) FPGAs.
How much would this add to Intel's profit?
Assume that Qualcomm pays $15 to Intel per chip (this is about how much they pay TSMC (TSM) to manufacture the chips), this adds about $9 billion in revenue, with a very small incremental cost. Apple's A6 business would add about $4 billion in revenue. Since this is excess capacity, we are only considering the incremental costs, which would be 5-10%. Although the Xilinx FPGAs are expensive, the volume is so small that the numbers are almost immaterial.
Last year, Intel had an operating profit of $14.6 billion that translated to earnings per share of $2.13. If Intel had the Qualcomm and Apple A6 business, the operating profit would increase by $12-13 billion, putting the earnings per share into the $4 range.
The best case
Of course, the ideal case would be that the market for Intel's CPUs would grow and the excess capacity would be filled with the much more valuable CPUs. This would add about $40 billion to Intel's revenue, increasing operating profit from $14.6 billion to the $40 billion range. This would push earnings per share from $2.13 to well above $5 per share.
Intel's earnings per share could double if they run their factories at full capacity.