It's no secret that the move to next-generation semiconductor manufacturing technologies is becoming more expensive per generation. Not only are the actual capital requirements per wafer on an uptrend, but they are -- for the first time in a long time -- increasing at a rate that actually exceeds the benefits that come from the density improvements that come with next-generation technology. This means that the cost per transistor heads in the wrong direction: up.
Why cost per transistor matters
To implement a feature in a chip, a chip designer must use a certain number of transistors, which are incredibly tiny semiconductor devices that serve as the fundamental building blocks for all chip designs. In a given manufacturing technology, each transistor occupies a set amount of physical area (all else equal, more area means higher manufacturing cost) and consumes some amount of power.
The goal that chip manufacturers aspire to, then, is to reduce the area that a transistor takes up to try to lower the cost of each transistor while also making the transistors more efficient, which allows chip designers to use more of them in a product while keeping the total power consumption of the chip in check.
Cost per transistor is heading in the wrong direction
After years of these great cost/transistor improvements generation after generation, the fabless semiconductor companies have begun to complain that cost/transistor is beginning to head (perhaps violently) in the wrong direction. Here's Broadcom's (a leading fabless semiconductor company) take on the matter:
It gets worse, too. Not only do these new manufacturing technologies driving a higher COGS per-transistor, but the actual manpower required to design a chip around these new manufacturing technologies is actually growing at an alarming rate, as shown here:
Why Apple is important
In the fabless semiconductor world, there are only a few foundries with leading-edge technology in real quantity: Intel (NASDAQ:INTC), Taiwan Semiconductor (NYSE:TSM), and Samsung (NASDAQOTH:SSNLF). Now, what's interesting is that Intel already has some pretty substantial leading edge business via its PC chips, so it doesn't have to worry as much about filling its leading-edge plants and getting that next generation of technology paid for.
Intel, of course, needs mobile to grow its business (and eventually will need those volumes to be able to sustain its ability to own and operate cutting-edge chip plants), but the issue isn't pressing today. TSMC and Samsung, on the other hand, need clients willing to pay handsomely for advanced chip manufacturing capacity to support the large initial investments that it makes in both technology development as well as in the factories themselves. For Samsung, that client has traditionally (and ironically) been Apple (NASDAQ:AAPL), and for TSMC, that customer was Qualcomm (NASDAQ:QCOM).
Why even Qualcomm isn't enough
Qualcomm is well-known for powering high-end, flagship smartphones with its cutting-edge Snapdragon parts. However, while these high-end design wins are nice, the real volume is in the mid-range to low-end. In this space, cost matters, and so for the vast majority of Qualcomm's parts, the company is likely to want to stick to an optimal cost/transistor node since a mid-range smartphone doesn't need the bleeding edge (unless, of course, it's cheaper).
This means that while Qualcomm will help drive things at the very high end, the volumes here – particularly since Qualcomm already dual sources from Samsung and TSMC – probably won't be enough longer-term (especially since the high end of the smartphone market is slowing down). Apple, on the other hand, gets paid a real premium for its devices (and it prides itself on advancing performance, even with large die-sizes), so it will likely be the "first" to new, more expensive nodes going forward.
Apple will use its importance to its advantage
Apple knows that it is strategically important to the high-end foundry business, so investors will likely see the foundries bending over backwards to try to win Apple's business (which means that Apple will get the upper hand with respect to wafer pricing). The company that can deliver the best performing transistors to Apple for the best price will likely win the majority of the orders. TSMC is making a lot of noise about how its 16 FinFET node is superior to that from Samsung (and Global Foundries), but the tale will ultimately be told over the next few years.
At any rate, Apple's business – since it will likely be the single largest fabless consumer of ultra-high end apps processors – will be vital longer term. The only question now is how the chips will fall – something that the next generation iPhone launches will shed some serious light on.