There is nothing more frustrating than seeing a patently false myth propagate over the web, particularly regarding an investment that you have studied inside and out for quite some time. In this article, I plan to show that Intel's (NASDAQ:INTC) corporate gross margins -- which have traditionally hovered in the 55%-65% range -- are sustainable, as Intel sees an increasing portion of its revenue base coming from new devices.
Good Enough? No!
A common argument from industry analysts and spectators alike is that in a world of "good enough" computing, Intel will be unable to maintain its historically high gross margins on the back of two fundamental assumptions:
- Nobody cares about performance; a cheap $10 chip is more than enough for everybody.
- Intel will have to take a gross margin hit in order to sell into the tablet/smartphone markets, even if performance does matter.
The first point is absolutely ludicrous, and I will give simple proof by contradiction: Suppose the hypothesis that all chips available today are "good enough." If this were the case then ARM Holdings (NASDAQ:ARMH), its licensees, Intel, and Taiwan Semiconductor (NYSE:TSM) would see no benefit to advancing either chip designs or manufacturing technology because existing designs and manufacturing infrastructure could be used indefinitely to produce these "good enough" commodity parts.
However, if this were true -- and if performance really does not matter -- then why do the likes of Qualcomm (NASDAQ:QCOM), Nvidia (NASDAQ:NVDA), and even ARM itself care so much about performance comparisons to both prior-generation parts as well as competing parts? This is a clear contradiction to the notion that "performance doesn't matter" -- Q.E.D.
Now, while the argument above captures the essence of the point that I wish to convey, there is some subtlety that cannot be ignored: How is performance defined?
Performance Per Watt in Common Workloads Is the Key Metric
At firms such as Intel, Qualcomm, ARMH, and others, no microprocessor is designed independently of the workloads that it is expected to perform. That is, ISVs (independent software vendors) and device vendors tell the chip designers what programs and user experiences they wish to provide, and it is up to the chip vendors to design around what the most common customer requests are. The workloads and usage models of a smartphone are very different than, say, that of a high-end workstation.
At the end of the day, in order to hit time, power, development cost, and gross margin targets, the micro-architects simulate and model what they believe to be the best design that they can achieve within these constraints and execute to deliver maximum performance. The key here is that better/smaller transistors allow the micro-architects to be more aggressive, and as a result, the benefits of a new process are really best seen in a micro-architecture designed specifically around a given process (and, of course, IDMs have the advantage of tailoring the process to the design in a co-optimization scheme).
Now, the key metric in the mobile computing industry isn't raw performance, but instead performance per watt. It is generally much easier to optimize the performance/watt ratio at much lower performance levels than it is at higher performance levels, which is why the real trick is trying to get high performance in a relatively low power envelope, rather than simply crippling performance to hit power targets.
Why is performance important? Simple. Over time, software becomes more and more demanding. As developers are given more performance to work with, the features and richness of everyday applications and websites grow substantially. While on the PC/notebook side of things, the hardware has gotten well beyond the majority of mainstream consumer software, the reality is that on the mobile device side, there is still an insatiable need for performance as mobile devices become much less limited, even if today's application designers do a good job of keeping the hardware requirements "low" enough to make most devices seem "good enough."
Naturally, device vendors will not want to sell devices -- particularly at the high end -- without the "best" possible processing hardware that they can get their hands on. In these devices, user experience is very much defined by the software, and the software is best realized on the fastest hardware possible. So high-end vendors will pay the extra $5 to get a better chip if they can, and the low-end vendors will find a way to get as much performance per dollar as possible.
Where the Sublety Lies: Form Factor
While users love performance -- even if they are not consciously aware that they do -- users care about form factor and battery life just as much, if not more. This means that the performance race has to be within the context of high battery life in thin and light devices, preferably with touchscreens. Just winning the performance tests isn't enough -- it's also about winning the performance tests, while being in a nice form factor, and without draining the battery too quickly. This subtlety is what has most sell-side analysts and retail traders somewhat confused and this is why you often hear paradoxical claims that performance doesn't matter. It does matter, but it matters within the context of the right form factor.
Getting Paid for Leadership
There's no secret that these chip vendors are all fighting for the performance/watt crown because performance/watt almost perfectly translates into design wins, particularly if the supplier providing the "best" chip is a reliable one. With a chip that has better performance/watt, the supplier has rather good pricing power. Consider the following situation:
Suppose that we have two mobile device vendors: a tier one tablet vendor and a no-name vendor trying to establish a brand and become a top tier player. Suppose that the no-name vendor, in a bid to try to gain a competitive edge, pays a little extra for the chip with better performance/watt. This puts the tier one vendor in a tight spot. Should this vendor risk using an inferior component, which could potentially lead to market share loss, in order to save a few dollars on a device that carries quite good gross margin?
In the PC space, while AMD has tried to sell its chips at very low prices to gain share, it has continually found itself out in the cold in anything but a few token designs to keep Intel honest. Keep in mind that when AMD had the performance crown it could not meet demand, and when it lost the crown, its market share dropped like a rock. The same thing applies to the mid- and high-end smartphone and tablet markets. Device vendors will likely pay extra for better performance/watt and more features built in to the chip, and this is why Qualcomm has been incredibly successful with its Snapdragon line of processors.
So, I don't think that if Intel's Silvermont-based parts are really as world-beating as the May 6 conference call suggested, that Intel will feel any real need to -- at least on the high end -- "compete on price." That's not to say Intel can go and charge whatever price it pleases, but it certainly be able to price its best parts at the high end of the tablet/phone chip range and sell disabled/slower parts at lower prices down the product stack.
The Gross Margin "Problem"
This now brings me to the whole point of this article: demonstrating that Intel's gross margins should be just peachy in this new world of mobile devices. The Intel bears/gross margin nervous nellies tend to think like this:
- Destkop CPUs are $200;
- tablet chips are $40;
- therefore, Intel's gross margins go lower.
This line of reasoning is, unfortunately, flawed. First off, the die sizes on a mobile SoC are substantially smaller than those on a PC chip. Intel's latest-and-greatest Haswell for Ultrabooks measures in at 181mm^2, which includes beefy graphics, two very high performance cores with very sophisticated power management, a sizable L3 cache, and more. I expect that the top Baytrail Atom part with four Silvermont cores and four graphics units (Haswell ULT has two CPU cores and 40 graphics units, plus 4MB L3 cache), as well as the built in PCH (Haswell die size does not include this) to be substantially smaller at less than 100mm^2 (given that AMD's Temash 4 core is a 107mm^2 part at 28nm, Intel's should be considerably smaller at 22nm), which means that each die costs far less to make than a Core chip.
Furthermore, while Intel bears look to the company's pricing list and see that Haswel mobile chips are expensive, these prices are very likely not anywhere near what device vendors pay -- particularly as these vendors usually buy many Intel products in bulk. Finally, Intel has many structural advantages over its competitors. So, keep in mind that fabless ARM SoC vendors typically have ~50% gross margin on mid- to high-end parts, with the very bottom of the barrel players managing to maintain 35%+ gross margins. Now, take into account the following advantages Intel has on the gross margin side of things over all of the other ARM players:
- Intel does not need to pay an upfront processor and/or instruction set licensing fee as well as subsequent royalties.
- Intel does not to pay TSM's foundry margins (TSM's gross margins are usually ~40%, with leading edge margins normally smaller).
- Intel has massive scale, so it can amortize many of the fixed costs across more units.
- Intel has massive operating leverage as what's learned from Atom can be used to help Core and vice versa, and processor cores/graphics cores are used throughout the company's product lines.
So the key takeaway is this: If any of the ARM vendors can sell a chip for X dollars and make ~50% gross margin, then it is completely conceivable for Intel to be able to sell a chip for X dollars and make more than 60% gross margin.
The real problem that I suspect many pundits fail to articulate is that even if gross margins are fine, a $40 Atom chip contributes fewer raw gross margin dollars than a $110 Core chip, assuming comparable gross margins. The fear is only valid if the following conditions hold:
- For each Atom sold, a Core sale is lost.
- Atom does not expand Intel's TAM.
I believe that the Atom parts serve only to threaten Intel's lower end Pentium and Celeron parts in low-end PC clamshells, which would have been threatened by the ARM chips anyway. Furthermore, as Silvermont-based Pentium and Celeron chips hit the market with substantially smaller die sizes and tighter integration, I believe that Intel's cost structure down below improves and newer/more innovative form factors at lower costs will help to spur demand. At the high end -- where performance matters, but form factor does too -- Intel is taking Core in just the right direction. Haswell managed to help Apple (NASDAQ:AAPL) achieve a doubling of battery life in its latest MacBook Air, all while performing PC-level performance. Eventually, Atom will simply be the "low cost" all-day battery life solution for 10" tablets and below, and the Core will be the premium part that goes into 10" tablets and above. Net/Net, Intel's TAM expands considerably.
I'm not worried one bit about Atom being too good or too successful. Atom tablets will cannibalize the low-end, cheap PCs, while the high-end Ultrabooks/notebooks continue to drive performance and battery life upward. Intel is simply expanding its own computing continuum, and as a result revenues should grow, margins should stay compelling, and ultimately profitability is poised to ramp hard over the next several years. This is an exciting time to be in Intel, and I believe that buying the name on any "nervous nellie dips" will prove to be a long-term winning strategy.