In what is shaping up to be a multi-part series in performing quantitative analysis of Intel's (INTC) core businesses, I now look to highlight the the firm's data-center business as what is likely the fastest growing, highest operating margin, and likely largest contributor to Intel's bottom line over the next several years. It appears that the opportunity for growth here is not fully understood, so it is my hope that the reader will find this informative.
The Data Center Group
Formally, the data center group ("DCG") provides "products that are incorporated into servers, workstations, networking, and other products that help make up the infrastructure for data center and cloud computing environments."
In terms of products, this means:
- Solid state drives
- RAID cards
- Ethernet cards
- Infiniband cards/switches (courtesy of QLogic (QLGC) asset acquisition)
In short, if it goes into a server, Intel probably sells it.
This Segment Is High Margin
Before I start talking about formal projections and net income estimates, I want to make a point.
Products in the data-center are extremely high margin. While the Street sits and worries about whether Intel will be able to compete with Qualcomm (QCOM) or Nvidia (NVDA) on a $25 smartphone chip, Intel is selling 2-8 of the following types processors per system:
Notice the prices on these processors? Anywhere from $300 at the very low end to $4616 at the high end.
Further, notice that these processors are built on the 32nm process technology node. The time that it takes to design and validate such processors is quite high, so they generally lag behind the PC processors on process node. This means that these things are built on nearly fully depreciated fabs (remember: 32nm fabs have been in high volume operation since 2009). The end result? A business with 50% operating margin.
In addition, the actual processor cores (whether they are "Atom" or "Core") are leveraged to build specialized server processors. So all of the R&D for the actual CPU logic/design is leveraged as new server-grade chips are built. There are costs to designing the "uncore" portion of the chip (i.e. everything that's not the core), since this is generally different from the PC/client versions of the chips, but the fact that a good part of the R&D is subsidized by a higher volume business is another perk of this business.
But What About Those Micro-servers?
The big overhang that many people see is that as the "cloud" infrastructure moves more towards a sea of cheaper/low power chips, that Intel's ASPs/margins will come down. A look at what Intel is charging for their server-grade Atoms can make this picture a little less rosy:
Yes. By definition the average selling price per unit will come down, but the idea is that the main selling point for these products is density (i.e. having a lot of compute nodes in a smaller amount of space). So for the kinds of workloads that don't need the monster single-threaded performance, but still need distributed horsepower, Intel ends up just selling a lot more of these per server. I believe the following slide from the Atom S1200 series launch helps to hammer home the point:
We can see that in a given space/thermal envelope for a server, the Atom actually leads to higher per-system revenue!
Additionally, since these processors are cheaper to make (smaller die), the gross margin profile on them is still, in the words of DCG GM Diane Bryant, "quite good". In fact at the launch of these product, the following Q&A occurred:
Q: This is [Gaber Moe] from BMO. I just have a follow-up question on, you had showed two systems of Intel CPU revenues of roughly $30,000 to $35,000 from the Xenon and Atom. Do you expect they have a gross margin profile?
A: Yes. I am glad you raised that, because I missed that point when I presented it, which is I often get asked, you obviously would prefer the Xeon space and not have the Atom space grow, and that point that we made on the slide is that from an Intel revenue perspective it really doesn't matter, because of the density of compute our revenue off of either one of those Xeon versus Atom it's really quite a watch. In fact the Atom is slightly greater, so across our entire server product line each of our products have slightly varying margin profiles, but holistically it is still very good margin for us and we are absolutely fine if that Atom SoC does very well in the market.
So clearly the "Atom cannibalizing Xeon" is not a problem. The micro-server segment that Xeon serves actually is not comprised of the obscenely high ASP chips that I showed you from the price list earlier. In fact, this is the price list for the products that go into these "micro-server" environments:
So the parts that go into the micro-servers ("L" for "low voltage) are not the big $1000+ chips built on depreciated fabs. They are leading edge process chips that are more-or-less the same processors that are sold as desktop chips. These chips, as highlighted in the chart, go for $200 - $300, so it's not unreasonable to expect that with greater density a bunch of $50 - $100 chips can be sold in sufficient quantity to offset the lower revenue per chip.
What About Competition In Microservers?
While the high end Intel stronghold in workstations, enterprise IT, and essentially anything requiring a higher end "Xeon" is safe from competition (the R&D costs and supplier trust that need to be built are enormous for even a large player like IBM (IBM) to successfully compete in), the low end micro-server space is apparently ripe for competition from the ARM (ARMH) startups looking to build their own ARM-compatible server products. While there may be some near term pricing pressure should any of these companies put out a product (the mere existence of these platforms will be used as a bargaining chip against Intel), Intel will ultimately be able to out-price and out-perform competing solutions for the following reasons:
- There is a significant cost involved in rewriting software that has traditionally run on Intel Architecture (X86) devices that most will probably not want to incur without a good reason (performance/$, performance/watt)
- The first "good reason" is unlikely since Intel's cost structure on tis chips is superior to any fabless chipmaker, and it infact collects the margin that the third-party foundry would take. Further, as Intel will be able to use smaller geometries, the cost per die is likely to be much lower (especially as these fabs are paid for by the "larger" chips)
- Intel will have a performance/watt lead. While the nearest competitor just recently taped out (i.e. finished the initial design of) a 40nm based micro-server product (the competitor is Applied Micro (AMCC) ), Intel is selling 32nm parts today and has been sampling (i.e. chips are back from the fab and working) 22nm parts since at least the Q3 conference call. If the stars align for the most aggressive ARM player, it will be a 40nm v.s. 22nm fight. At 22nm Intel has a bigger transistor budget to pack more performance, features, cores, etc.
Unless Intel really drops the ball here, it will simply have too many tailwinds to actually let a competitor slip past in any major way.
Examining The Streams Of Revenue
While the cloud and networking are both very high growth parts of the data-center group at Intel, they are certainly not the only revenue drivers. Take a look at the following slide to try to understand just how it all breaks down:
As we can see, the HPC, public cloud, networking, and enterprise storage growth is projected to be quite good. The traditional SMB & enterprise will also grow, but not as rapidly as these new segments.
More importantly, we see that Intel is very vehement about this segment growing to $20B in sales by 2016. Assuming that Intel can keep operating margin at ~50% (where it is today), then this represents $5B in additional operating income from this group within Intel alone.
To put this in perspective, take a look at this breakdown of 2011 revenues/operating margin by segment:
Of 2011's operating income, 25% came from the DCG while the remaining came from the PC Client Group. However, if by 2016 Intel can double the business, even if all other segments remain flat (which is an entirely unrealistic assumption), then total operating income will grow by 25% by 2016 from 2011 levels.
This would be a modest ~5% CAGR for overall operating income, but given that the stock is priced for revenue/profit declines, this kind of low-single digit growth overall is certainly better than the horrible death that many seem to predict for the company/stock.
Do not overlook the growth in the server room, in the cloud, in storage, and in networking that we will see at Intel. While PCs may be in a bit of a slump right now, keep in mind that Intel has an operating segment that is, quite frankly, a superstar. Ultimately, I expect Intel to continue to grow for the foreseeable future on both the top and bottom lines, and I expect the dividend to be raised, if anything.
This growth does not include any potential win in the "Atom" space (tablets, smartphones, etc.), nor does it account for growth in the PC space or even the software/services division.
Additional disclosure: I may initiate a short position in AMCC and/or ARMH over the next 72 hours.