(Note: The Bay Trail die-size data that I will be using in my analysis was first brought to light by InvestorsHub forum member, 'wbmw'. While I believe that I am adding additional value and analysis here, my policy is to always give credit where it is due. See wbmw's post here)
A question that has been on the minds of many Intel (INTC) investors is whether the company can compete successfully in the low cost tablet system-on-chip market while at the same time keeping gross margins healthy enough. While many Intel bulls such as myself have believed that the answer to this question is "yes", the first die-size data is in and we can now start talking hard numbers instead of educated speculation.
The Bay Trail Die Size - 102mm^2
Intel recently released the datasheet for the Bay Trail-T part (and can be found here), and it included the following diagram (annotated by yours truly):
This is a single-chip solution (for traditional PC chips, the solutions are either two chip or two-die per package) built on Intel's 22nm FinFET process. The burning question(s) on investors' minds are the following:
- Can Intel do this profitably?
- Isn't this just a one-for-one replacement for the PC sales that are lost?
I'll address both of these questions (and more) here.
Let's Talk Chipsets
For those of you unfamiliar with how Intel segments its businesses, everything that has to do with PCs goes under the "PC client group". In CY2012, this division did about $34B in sales. Assuming that Intel had 80% of the PC market, and assuming the PC market TAM was about 352M in 2012, this means that Intel shipped about 281M PC units. This suggests a platform average selling price of $120. Now, note that I said "platform" and not "chip" selling price. This goes back to my claim that a PC platform actually involves two chips. Let me borrow a diagram from Intel's recent Ivy Bridge-E slide deck to illustrate my point:
Notice how there's one block that says "Core i7". This is the processor itself, which includes the CPU cores, memory controller, PCIe lanes, caches, and all of the goodies that determine the "performance". The second chip, known as the "chipset" includes non performance sensitive things such as USB connectivity, SATA (this is how hard disks/solid state drives are connected), and more.
Typically speaking, Intel builds the processor itself on the latest process node that it can, and to save on cost, it builds the chipset on an older process node (the X79 chipset above is built on the 65nm process). So, when you see the average selling price of the PC platform coming to $120, this not only includes a CPU but a chipset.
So, just how much silicon is that?
Looking At The Cost Of A 'Cheap' Platform
It's clear that Intel's gross margins at the mid/high end are in good shape, and quite frankly, these are the markets that tablets aren't really cannibalizing to any great extent (since the performance, software, and capabilities just are too disparate), but these markets within client are, at the same time, saturated/not really growing (unless you could enthusiast/gamer PC, but this is just a drop in the bucket).
So, what are the platforms that are under serious fire? The low end/mainstream parts. That's Celeron, Pentium, and Core i3, with particular emphasis on the mobile flavors of these parts. These parts are, to the best of my knowledge, currently based on the "Ivy Bridge M-2" die (disabled to various degrees). Now, I also want the size of the PCH, so for that, I'm going to use the recent image of "Broadwell" that Anand Shimpi of Anandtech managed to snag to try to do some estimates (image here). In particular, it looks like the PCH is 1/3rd the size of the actual CPU/SoC complex, so I'd wager that the silicon content of that PCH comes in at 60mm^2 on the 32nm process.
So, for the vast majority of "cheap" notebooks (think $199 - $499), Intel is selling 93mm^2 of 22nm CPU silicon plus another 60mm^2 or so of 32nm silicon, for a grand total of 153mm^2 of silicon. Also, while the average selling price for all PC platforms is $120, don't think that a cheap notebook's platform commands anything close to this (remember, this average is skewed by some pretty expensive, high end SKUs). This means that Intel is probably selling this two-chip platform (that means that Intel has to fab, package, and test two separate silicon die) into the lower end markets for about $50 - $60 total. In fact, if one takes a look at Newegg.com, we see that Intel sells dual core desktop processors for as low as $50 (this is including the box, bundled heatsink and fan, manual, and e-tailer markup) and full motherboards featuring Intel's lower end chipset go for as little as $50. Clearly, Intel is taking in significantly less than $120/unit for these lower end platforms (as would be expected).
Is 'Bay Trail' A Neutral, Or A Positive?
The first question is whether Bay Trail ends up being a neutral to the top and bottom lines or a positive. Well, it depends. Objectively speaking, I think that if Intel is able to sell Bay Trail-M/D parts into low end notebooks/all-in-one desktops at the same prices that it is currently selling its 2-chip platforms, then this is a very clear gross margin win. However, the real question is whether Bay Trail-T (the tablet version of the chip) will prove to be an issue.
Well, the base assumption is that the tablet editions of Bay Trail will be much cheaper than the notebook/desktop editions. The official Intel price list claims that the lowest end Bay Trail-M based "Celeron" part lists for $132, while the official price for the Atom Z3770 (highest end quad core tablet SKU) is a mere $37 (and I am confident that the OEMs are paying significantly less than this as they buy in much larger quantities). My guess is that the "Celeron" part doesn't actually sell for anything near that (again, volumes), but we can surmise that the notebook/desktop part is more expensive than the tablet part.
So, clearly the best case scenario is that the new "Bay Trail-M" based designs are much slicker and sell better, stemming the decline in the total "PC" market (and improving gross margin dollars per unit at the low end), and maybe even re-energizing growth in emerging markets. Then, on top of this, Bay Trail-T drives incremental volume growth, with slightly lower gross margin dollar content than the Bay Trail-M parts, but likely on par with the die-harvested Celeron/Pentium parts based on the "big" core.
But I'm not going to suggest that we bank on a "best case" scenario. What I see as fairly reasonable to assume is that the Bay Trail-M/D parts optimize the low end PC space for margin improvements (which may not necessarily mean a renewal in volume growth, but helps gross margin/profitability along), while volume growth comes from the Bay Trail-T in tablets. I also assume that the "Core" products remain mostly flat to down (high end isn't necessarily being cannibalized, but upgrade cycles are slowing) until the corporate/enterprise refresh cycle starts up again (which could be catalyzed by the end of Windows XP support in early 2014).
As far as I see it, though, Bay Trail is a product that will have substantially better gross margins than the parts it is replacing in the PC space, it will enable sleeker, cheaper designs (when you don't need a fan, system costs go down), and it will help to drive significant share gains in the tablet market (which will likely be larger than the PC market soon enough). Further, the product isn't fast enough to really "cannibalize" a higher end Core product, so I'm not too worried about that. Bay Trail, in my view, is an unequivocal win.
How About The Competition?
While Bay Trail is a good product that really does a lot of good things for Intel from a business standpoint, it doesn't exist in a vacuum. While I'm confident that competitors will have a hard time displacing Intel in the low end PC space as well as the Windows tablet space (no matter how hard Microsoft wants to push Windows RT), the Android space is very competitive. The other companies (chiefly Nvidia (NVDA) and Qualcomm (QCOM)) have solid products, too. I heard from various folks at IDF that Android design momentum is strong, although as always, we'll need to see real sell-through data before declaring victory.
What's The Bottom Line?
The bottom line here is that for the very first time since Intel even started talking about its ambitions to be a premier mobile SoC player, we are seeing a truly leading mobile system-on-chip. As far as the CPU goes, it's at the head of the pack in terms of performance per watt (benchmark chart courtesy of Anandtech):
In graphics, Intel's mobile parts are finally not a laughing stock, as shown in 3D Mark (courtesy of Anandtech):
The die size of 102mm^2 looks about right for a top-tier tablet chip, and given that Intel has some real cost advantages here (no need to pay ARM for a license and subsequent royalties, no need to pay TSMC's margins, and no need to pay the margins of an OSAT (outsourced packaging and test) player), I'm not too worried about gross margins. From an operating margin perspective, the Atom core itself is reused across the entirety of Intel's organization, the graphics is also leveraged across both high end PC all the way through tablet, and the R&D costs for the Atom core are significantly lower as the team is smaller and uses industry-standard EDA tools.
Intel is here with a competitive tablet/low end PC part today at a good cost structure. The only two open questions remaining are when Intel will become a competitive force in smartphones and whether the PC market will bottom and eventually rebound or not.