At one point in the Q&A for Intel's (NASDAQ:INTC) Q4 2013 earnings call, an analyst asked whether Intel could be "sustainably profitable in the mobile space". Of course, Intel execs answered in the affirmative, but then cited the planned Sophia series of processors as an example of a chip they were sure would be profitable. Sophia is planned for "low-end" tablets in 2015, and so is a long way off. Was this an admission that Intel's mobile processor business, contained in the "Other Intel Architecture" Group wouldn't be profitable in 2014? Probably.
Overall, Intel continues to be very profitable, but these profits are fueled mostly by its legacy PC Client and Data Center Groups. PC Client group revenue was $8.6 billion with flat growth compared to Q4 2013, while Data Center Group had revenue of $3.0 billion and 8% year over year growth. But we all know mobile is where the growth is, and here the profitability story has been a mess for some time, as the chart below shows.
Other Intel Architecture Group revenue did grow by 8.5% y/y, but clearly Intel's battle for dominance of the mobile space is costing it, and the headwinds will continue into 2014. Intel is ramping up production of it next generation 14 nm process, and this is expected to depress gross margin back into the upper 50s percent range.
Subsidizing Intel Inside
Furthermore, as Intel disclosed in its Investor Meeting presentation on 11/21/13, Intel intends to make what it calls contra revenue payments to tablet manufacturers in order bring the Bill of Materials cost of Intel tablets into line with ARM based tablets. Clearly, Intel was forced into these subsidies by the tablet makers, who weren't going to offer Intel based tablets without them.
These subsidies were also the subject of numerous analyst questions, one of whom asked if there were any tablets that weren't receiving subsidies. Intel admitted that most were, while claiming that the subsidies would be phased out over the course of the year.
The need for the subsidies points to an issue that I have raised in previous articles, which is the question of whether Intel mobile processors can be cost competitive with ARM. This has been largely overlooked by Intel evangelists such as Arnold Frisch and Ashraf Eassa who have tended to focus on processor performance as the key metric determining who would win the mobile processor battle. As technologists, it was all very cut-and-dried to them: Intel had the superior processor technology, so it would win easily.
Looking back at predictions made early in 2013 is revealing. In his first article for SA, Frisch wrote in May 2013:
Without further ado, I can state with certainty that the new mobile oriented processors coming from Intel later this year will capture the processor space for the leading smartphones and tablets.
There are two fundamental, related, reasons for this:
These processors are faster than anything that ARM/TSMC (or AMD) can offer at the same level of power consumption.
Their power consumption at the same level of performance is much less than anything ARM/TSMC (or AMD) can offer.
Frisch then goes on to explain that Intel's superiority is all just a matter of physics, thereby lending an even greater sense of inevitability to Intel's triumph. In a comment, Eassa dubbed the article "superb" and warmly welcomed Frisch to SA. In June 2013 Eassa wrote an article "Yes, Intel Will Take Over Tablets".
Where did Frisch and those who followed him go wrong? I could spend several articles critiquing all the conceptual problems in his and subsequent articles, but let me just focus on two that are most relevant to investors looking at Intel's prospects for 2014.
1) First and foremost is cost. Frisch never even mentions it. Intel evangelists have gone to great lengths to argue (correctly) that Intel processors are not disadvantaged compared to ARM in processing speed and power efficiency, and that Bay Trail is arguably superior to anything in the ARM world, except possibly Apple's A7. They're right, but they never talk about cost.
Dealing with the issue is difficult, since Intel hasn't been very forthcoming lately with details that might help us estimate the cost of a Bay Trail processor, for instance. Intel has stopped publishing transistor counts or die sizes, and the numbers I've seen posted on the Web don't appear to come from reliable sources.
The issue of transistor count is particularly important since Intel argues (correctly) that cost per transistor keeps going down for each process node: a 14 nm process transistor will be cheaper than a 22 nm process transistor, etc. This tends to leave people with the impression that an Intel Bay Trail processor (22 nm) must be cheaper than an ARM processor built on a 28 or 32 nm process. Not necessarily, if the ARM processor requires fewer transistors and therefore less area per chip, it might still be cheaper. ARM's processor architecture is quite a bit simpler than Intel x86 Silvermont, so there's reason to believe that this is the case.
I realize the discussion above is heuristic and fuzzy, but this is due to the lack of specific data. I admit that this is speculation, but there is plenty of circumstantial evidence that current Intel mobile processors are cost disadvantaged: the loses in the Other Intel Architecture Division, the tablet subsidies, and the more complex x86 architecture.
2) The Apple (NASDAQ:AAPL) A7 processor. I get the impression that Intel, its many supporters, and the mobile device industry in general were really blind-sided by the A7, the worlds first 64 bit processor for smart phones. I certainly wasn't expecting anything that good from Apple, and I'm an Apple developer. In Anandtech's review of the iPad Air, the A7 bested in most tests the Intel Bay Trail Z3740 in the Asus Transformer Book T100. In fairness, the T100 was running a 32 bit version of Win 8.1, so the 64 bit Bay Trail couldn't really perform at its best. A 64 bit version of 8.1 for Bay Trail, due soon, should turn the tables in favor of the Z3740. But just the fact that the A7 could even get close to Bay Trail performance was a surprise. And then there was the fact that it could be put into the iPhone 5s unmodified. And then there was the fact that it only cost $19, according to IHS/iSupply.
Some may question whether the $19 includes all development costs. It probably does, since the iSupply cost estimate for the quad core Qualcomm processor in the LTE Samsung Galaxy S4 is $20, which Samsung would have to purchase outright. The iSupply numbers should give some feel for the cost target that Intel has to meet in order to be competitive with ARM.
The A7 achieved performance comparable to Bay Trail, despite having 2 fewer cores, and being fabricated on a 28 nm Samsung process (non-FinFET). This also suggests that ARM has an advantage in simplicity and lower transistor count than Intel which can, to some degree, overcome Intel's process advantage.
The Question for Investors
I suppose it's possible that Intel could still crush its competition while taking losses on its subsidized mobile processors, but Apple, Google (NASDAQ:GOOG), and Samsung (OTC:SSNLF) all have plenty of cash in the bank, so they could probably outlast Intel in a subsidy war. Especially with Apple's $130 billion in cash, it's hard to see how Apple would allow itself to be crushed by subsidized Intel processors.
So the question for Intel investors is when does Intel close the cost gap. The implied answer of 2015 contained within the assertion that Sophia would be profitable is probably the right answer. But it's the wrong answer for Intel investors. Apple has clearly demonstrated that it's a moving target. As much as Intel evangelists like to harp about the superiority of Intel's forthcoming 14 nm process, the ARM fabricators such as TSMC, Samsung, and Global Foundries are all working on 14-16 nm FinFET as well, with optimistic predictions of entering production this year. Even if these predictions are wildly optimistic, by 2015 they probably will be in production, especially if cash rich companies like Apple help them out. Right now, I can't think of a better use for Apple's cash horde.
It's inevitable that ARM foundries will eventually achieve their own 14-16 nm processes, the only question is when. If they manage it by 2015, then Intel never closes the cost gap.