I rarely enjoy doing this, but this piece will be a direct follow-up on a recent piece I published, "Intel's Mobile Strategy: 60% Gross Margins Are More Than Likely". In the comments thread of this particular piece, I was left this vitriolic - but perhaps important - comment:
In this article, I would like to prove that my assertions are, in fact, quite well grounded in reality with some math (I have a degree in the subject, actually) as well as some qualitative analysis (I promise it will be "actual" and not "mythical").
What We Need To Prove
In my previous article, I claimed that Intel could maintain 60% gross margins (this is at the midpoint of the long term 55-65% range that Intel has given investors at its investor meetings) on tablet/smartphone chips. Further, I made the claim that the mid/high end of these processors sell for quite a bit more than $10. I will justify both of these claims now.
Let's Talk Chip Selling Prices
Intel's (INTC) PC Client Group (which sells both PC micro-processors and PC chipsets) reported the following financial results (straight from the latest form 10-K (I can use this for "research", right?) that Intel filed with the SEC, so this is most certainly fact):
Note that net revenue (this, for those of you who need a refresher, is defined as how much Intel took in from selling products) is $34.3B. Now, keep in mind that this net revenue figure represents the sale of both a PC microprocessor as well as the accompanying chipset that is sold with the microprocessor. Now, if we look at the latest IDC numbers, which forecast 348 million PC units to be sold in 2013 (up 3.2% Y/Y), then we see that 2012 saw shipment of 337 million units.
Now consider that the most recent market share data (it's actually for Q3 2012, but it should be reasonably representative) for PC microprocessors suggests that Intel has 83% market share of the PC chip market. Some basic algebra tells us that Intel, then, shipped 279 million PC processor + PC chipset pairs in 2012. Basic division tells us that the average MPU + chipset pair selling price is $121, which is a far cry from Mr. stu_s's assertion that Intel's PC ASP was in the neighborhood of $200 (this, for the numbers folks, is called being off by 65%). This also somewhat damages the thesis that AMD (AMD) bulls present that Intel will be undercut by AMD since Intel "charges $200+" for their chips.
PC chips aren't as "expensive" as many investors believe, so the theses that Intel will see an ASP reduction of 20x (from $200 -> ~$10) are likely incorrect.
Now, this leads me to my next point...
Mid-Range And High End Mobile Chips Aren't $10
The next assertion that Mr. stu_s makes is the following, re: chip ASPs for mobility:
So, there's this misconception that all mobile chips sell for <= $10/unit. While this is certainly true of the low end players such as MediaTek and Spreadtrum (SPRD), it is absolutely not true for the mid to high end space in which Intel, Qualcomm (QCOM), Nvidia (NVDA), and the like play. I now prove the claim.
In 2012, ~117.1M tablets were shipped. The tablet apps processor market as a whole saw revenues of $2.7B according to Strategy Analytics, so simple math says that ASPs for tablet chips were $23/unit, or roughly a 130% premium to the assumptions that were used to attack my previous article. Now, note that many Android tablets overseas use very low end chips which likely has a very outsized drag on ASP. Also note that Apple does not purchase apps processors in a merchant-vendor like fashion (it designs its own parts which are then fabricated at Samsung), so ASPs on the Apple products likely underestimate the selling price of chips for merchant vendors.
So, we have debunked both the PC chip cost myth and the tablet chip cost myth. We now need to talk again about gross margins.
Gross margin, as the majority of individuals reading articles about individual equities should be well aware of, is defined as:
Gross Margin = Net Revenue - Cost of Goods Sold
In the semiconductor world, gross margins are primarily affected by the following two factors:
- Wafer cost
So, as many of you may be aware, processors are made by taking highly refined silicon wafers and etching the chips (Intel has a nice, simple visual explanation of the basics here) onto the wafer (this is enough to understand the basic economics; a semiconductor professional may cringe). A final "wafer" after all of the very sophisticated voodoo happens looks something like this:
Anyway, remember the two factors that affect gross margin? Well, the first factor - wafer cost - is simply how much it costs to buy the processed, finished wafer. The next factor is yield, which basically asks, "from a given wafer, how many usable chips do I get?"
The yield is dependent on many things, but a good set of rules to keep in mind are:
- Smaller dies are much better for margins than larger dies, since not only do you get to stick more die per wafer, but the larger your chip, the more likely that a single defect will show up and spoil the party
- The "quality" of the manufacturing process as well as how elegantly the chip designers "followed the rules" to make sure things were more likely to work
Now, when it comes to gross margin, the fabless companies are usually at a disadvantage for the following reasons:
- There aren't a lot of wafer fabs out there, especially at the leading edge technology (TSMC (TSM) has had a monopoly on 28nm for quite a while since no other fab could really get it right), so the foundries typically charge per wafer which generally means that it's up to the fabless vendor to eat the cost of non-functioning dies or dies that do not run to spec (the yield with respect to the former is called "functional yield" and the latter is known as "parametic yield")
- The wafer fab usually runs a business in which it is collecting >= 40% gross margins, so not only are the fabless vendors paying for the cost of the wafers/die, but they are also paying for TSMC's handsome profit margin
So, now we get to why Intel isn't going to be at any real disadvantage (and will in fact be at a huge advantage) with regards to gross margins. While a fabless vendor such as Qualcomm or Nvidia can make ~50% gross margins on the chips after paying for the wafer and for TSMC's profit, Intel is collecting both. That is, any gross profit that Intel would have made if it were fabless plus any gross profit that Intel would have made as a foundry - all on one chip!
To put it bluntly: if Qualcomm or Nvidia can make ~50% gross margin on these mobile chips (which contain dies that are much smaller than those found on PC chips which is great for # of chips per wafer as well as yield), and if TSMC is raking in ~30-40% gross margin, then why is it farfetched that Intel can command its traditional 55% - 65% gross margins? There is nothing "magic" about the ARM (ARMH) vendors that makes their die sizes inherently smaller (no, the "CISC" decoder does not take up more than ~5% of the die, so put that argument away).
It's not, and that's been the point that Intel bulls have been trying to make the whole time.
Cost Per Die Gets Worse For All ARM Vendors At 20nm And Below
Something that does not come up often from the sell-side is that the ARM vendors are, for the first time, projecting that the 20nm and 16nm nodes at the foundries will not be providing the traditional reduction in cost per transistor that is usually seen at each node, and that the cost is actually set to increase. This is a huge deal; it means that the ARM vendors looking to compete at the high end either have to substantially increase unit volume shipments at the same ASP to keep profitability constant, or that ASPs rise. Intel, which has repeatedly publicly stated that cost per transistor continues to decline at the 14nm and 10nm generations, will continue to benefit on the cost/margin side of this.
I reiterate my view that Intel will be able to maintain gross margins in the 55-65% range (60% at midpoint), and further believe that the tablet/phone markets are mostly additive to the PC/Ultrabook/convertible sales. I further believe that Intel's continued push into the data-center both in the big-iron space formerly dominated by proprietary RISC-inspired architectures such as IBM's (IBM) POWER and Oracle's (ORCL) SPARC, as well as the cloud/scale-out workloads will further help to support the company's blended gross margin profile, even if Intel decides to get extremely aggressive and undercut competitors in mobile to gain share.