While eyes remain fixated on the architectural battle between Intel and ARM, a second front is soon about to open up that will determine mobile supremacy for the rest of the decade. Whereas yesterday’s story on the collapse of Wintel and the anointing of Google and Samsung are repeated endlessly, now tables are being set for a significant turning. Thanks to Intel introducing a very competitive Silvermont architecture and the reality that a process gap is growing vis-à-vis Samsung’s Foundry, a new hierarchy is forming that will allow the x86 monopoly to extend its power by pulling in DRAM and NAND into an extended SOC family of solutions. Shortages and rising prices in the memory market are actually about to work to the favor of Intel relative to the ARM merchant players and perhaps sets the stage for a Samsung alliance that one could say resembles that of the IBM PC win in 1981.
It was twenty years ago that DRAM was entering its peak valuation stage with the blow off occurring at the launch of Win95. Now, with ten turns of the Moore’s Law crank we are down to two players who form a tightening oligarchy on the one commodity, outside oil, that could bring the world economy to a halt. Samsung’s dreams of cornering the mobile market are based on leveraging DRAM, NAND and their ARM processors into becoming the lowest cost manufacturer. Furthermore, if DRAM were to become constricted, the empire could expand their tentacles into the PC space. Thus threatening Intel’s dominance…if only Samsung could build an advanced ARM processor on a leading edge process. My bet now is that Intel will flip this hierarchy.
Intel’s launch of Haswell comes at a time when many analysts are skeptical of a rebound in mobile PCs. It is understandable as the refresh cycle that came around every three years like clockwork has stretched in most cases to 5 years and beyond. The significance of Haswell lies not just in its lower power but also in the addition of graphics DRAM and I/O onto a multi chip package signaling a directional change. By sourcing memory and extending the definition of the SOC, Intel will simplify the design for OEMs and thus move more of the platform value away from them. Next year’s 14nm shrink of Haswell will likely signal the start of a cost down curve for Intel as ULV parts head below $200 and towards $100, thus removing the low power premium that has in the past been bankable ($) but at a cost of low volume stagnation.
With the introduction of the Silvermont architecture, Intel is able to accomplish two significant goals. The first is to have a Fab filler immediately after the launch of the newest mobile PC processor thus improving its ROI on its massive Capex, which should lead to an improvement in Intel’s financials. Second, though, I think Intel finally figured out that the <$200 tablet and smartphone market is open to a packaging strategy that combines all digital electronics under one roof – including memory. There will be hundreds of millions of mobile devices that will use the minimum DRAM and NAND configuration. Intel can supply this to a broader set of OEMs residing in China Inc. and set a low end pressure point on Samsung that can't be ignored.
The interesting part though comes next. By pushing Atom to the front of the process line, it has created an enormous cost gap with Samsung’s own mobile ARM processors. If Intel embarks on a cheap, couple sizes fits all mobile SOC product line for China Inc., then Samsung will be compelled to enter the Atom ecosphere in order to remain cost competitive. Samsung’s announcement that it will use Intel in its Windows tablet was a foregone conclusion based on the fact that a play in the business world requires that an OEM be compatible with Office. Microsoft took an additional step to solidify its corporate monopoly by recently offering Free versions of Office for the 10” tablet market. Clearly this is a strategy to block Google Chrome and slowdown the march of Apple’s iPAD.
But whom will Intel source its DRAM and NAND from? For now the answer is Micron for DRAM and a combination of Micron and Intel for the NAND. A switch by Samsung to Intel’s x86 architecture with a memory sourcing agreement is likely to be in the best interest of the former and would allow the latter to play off memory suppliers to drive down platform costs. Samsung sits in the middle of the value chain between Apple on the high end and China Inc on the low side. Cost reductions stave off market share losses at the low end and allow it to target Apple for the top 30% of the market where the majority of the profits reside.
On Intel’s part, the resurgence of its core x86 business, especially the Silvermont based mobiles, now appears to bring greater benefits than just the one on one battle with ARM as it becomes a forcing function to align mobile OEMs with its long-term strategy of owning all the silicon content in the system in a single package. Furthermore, Intel is necessarily working on solutions to reduce the processor-memory performance gap that is stunting performance/watt gains in the data center. Control of memory then becomes crucial in both the high-end data center and in mobiles.
One can argue that Samsung will never back out of its mobile ARM development, however economics in the short term and long term appear to be taking over. With the exit of Apple from its Foundries, Samsung desperately needs something to fill its fabs and the only candidate would be an internally designed mobile processor synchronized to come online as TSMC hits the gas on its 20nm production of the A7 processor. Samsung’s semiconductor group has a difficult decision to make on its foundry equipment investment just as Intel courts the smartphone and tablet groups with cheap x86 Atoms. The likely scenario is that Samsung hedges by building a wide range of products that in some cases uses Intel and in others uses Qualcomm. The end game will be in motion as the train leaves the station and there will be no waiting on the semiconductor group if it means that a $20 part holds up the sales of hundreds of millions of mobiles containing a rich content of Samsung Inc components.
Winning Samsung, while eliminating its foundry as a competitor is the best of all worlds for Intel as it reduces the game down to Intel vs. TSMC. I expect Apple to stay aligned with TSMC as they receive greater control and benefit in a relationship that they can partially invest in with their overseas capital. In addition, the Apple’s iOS ecosystem can afford to pay a little more for processors that are one node behind Intel. I expect Intel to offer very aggressive pricing to Samsung later this year on its 22nm Bay Trail processors that enables broad adoption and is quickly followed by low cost 14nm Atoms and a significant partnership on memory. If this comes into being, then oddly enough Micron may align much closer with Apple. The semiconductor mobile market is attempting to settle into a Bipolar world.
A semiconductor friend of mine sent me a quick note the other day as we were discussing the above scenarios open to Intel in 2014. In essence he wondered if the Intel win in the Galaxy 3 tablet was similar to IBM selecting the 8088 for the original PC. In a way it is similar in that the IBM PC essentially eliminated all other personal computer architectures except those under the direction of Apple. If Intel is able to capture Samsung and China Inc, then there is a possibility to grab 70% of the mobile market. This still leaves open the question of Qualcomm and Broadcom’s dominance with its communications footprint.