Intel's (INTC) server division could be in a lot of trouble. Now I know this sounds like another doomsday Intel article, but before disregarding the information, please consider the following.
Over the past five years, data center technologies and efficiencies have improved significantly. I wasn't able to fully quantify improving server efficiency until I came across Facebook's (FB) technology transcript, which implied that the product refresh cycle in the server segment could drag out for substantially longer than what Intel is willing to reveal to shareholders. Because of this fact, investors should factor in further downside from Intel's server segment going forward.
Intel's financial breakdown
Intel has had a pretty weak fiscal year, but in the most recent fiscal quarter, Intel came ahead due to strength in its server segment. However, the strength in servers should be temporary, as I will reveal further on in my article.
Intel has had a pretty strong quarter: gross margin was up 5% year-over-year, operating income was up 29% year-over-year, and earnings per share was up 49% year-over-year. Most of the gains in the business were driven by stronger pricing from the data center segment and unit shipment growth within that segment.
Looking closely at the PC client group, the company reported year-over-year declines in revenue. This implies that Intel cannot price any of its consumer products any higher to earn additional revenue/profit. On the other hand, the data center group was able to report 12% year-over-year revenue growth. Investors have held onto Intel's stock because the data center group generates earnings and revenue growth.
To be specific, Intel increased the average selling prices of its data center products by 8% year-over-year. In comparison, the PC client Platform was only able to increase its average selling price by 1% year-over-year.
Intel provided guidance of $13.7 billion revenue for the fourth quarter of 2013. In Q4 2012, Intel generated $13.5 billion in revenue, implying that Intel will only grow revenue by around 1.4% year-over-year in Q4 2013. I don't think Intel will have any problems with meeting its own short-term guidance. However, the long-term looks to be a very tricky picture, especially in the server side of its business.
Server segment decline
Intel has two main product lines: solid state drives for servers and server-class processors. A server/mainframe computer is a computer that tends to perform tasks for either a website or closed network. Network infrastructure is built by combining a large number of microprocessors, RAM, and hard drives. By combining a large number of components, the combined components can process a large number of commands, and retrieve large volumes of data.
Data storage needs are expected to grow from 3 doggabytes to 20 doggabytes between 2013 and 2020 (one doggabyte is equivalent to one sextillion gigabytes). However, most of this growth will come from cold storage (memory stored to never be read again). Intel only sells solid state drives, so there's limited upside from rising data storage for Intel. This is because the cost per GB can be a little prohibitive with SSDs, so it's highly unlikely that data centers will adopt SSDs for quite a while. When dealing with cold storage, data centers will prefer buying the lowest cost solution per GB, which would mean buying disks. Therefore, Intel's business in solid state drives will not significantly offset the decline in server chips in the foreseeable future.
Intel's server segment is in a lot of trouble as a result of what Facebook has done to optimize its data center. According to Jason Taylor (Facebook's Director of Infrastructure):
So what you get from there is you get volume pricing, which is huge when you are operating at scale. You also get re-purposing. So if we have ten services that all have a forecast and there is some uncertainty in the forecast, being able to take unused servers from one service and reallocate them to another is a huge efficiency win. If we had a different kind of server for each service, then repurposing wouldn't be possible and we'd always be dealing with the worst case of all forecast. So it's a surprising win but it's kind of analogous to I guess managing a mutual fund, some things are up, some things are down but across the board you do every well.
So you also get easier operations. So in other facilities, not our facilities, you might have a server to technician ratio of about 1 to -- 450 servers to 1 technician. Because all of our servers are the same we're able to attain about a 20,000 to one server to technician ratio. So these servers are all very easy to maintain, very easy to operate.
Now I know this sounded like a ton of gibberish. In fact, I'd probably lay out the mattress for you so you can fall asleep by the time you're finished reading this article, but there's something really important that I want you to understand.
What Facebook believes that it can do with its disaggregated rack technologies is that it can better optimize resources by upgrading specific parts of the server without having to replace or add new servers. This means that if the server only needs more RAM, then buy the RAM, but don't install additional CPUs. If the server needs more CPU power, buy the additional CPUs without buying additional server racks.
Second, the disaggregated rack allows for certain server resources to be pooled from other servers that are not being put to use. For example, Facebook explains that it can take processing resources from activities that require little processing and put them towards activities that require more processing.
Earlier in the article, I mentioned that a mainframe computer is just a bunch of computers combined. Well, Facebook is taking server racks with low CPU usage and putting them into tasks with higher CPU usage. Likewise, it is taking server racks that use little RAM memory and putting it towards resources that have need for a ton of RAM memory.
To put this in perspective, my computer currently runs at around 11% CPU usage and uses about 2GB of RAM memory out of 6GB of RAM memory. Facebook with its disaggregated rack is basically trying to utilize near 100% CPU usage and memory usage across all of its servers. To do this has been a challenge, but it basically involves sharing resources by decentralizing resources and upgrading the servers in specific areas. This helps to eliminate waste and results in huge cost savings. Because significant financial resources are dedicated to IT at many Web 2.0 companies, there's a lot of financial incentive, plus expertise to maximize server load and reduce server refresh cycles.
According to Jason Taylor (Facebook's Director of Infrastructure):
The other thing that you can get out of this disaggregated rack is a longer useful life. So everybody everywhere it's right of a computer when do you think that they need is no longer fitting. I don't have enough CPU, I don't have enough RAM, I don't have enough flash around disk base. In general most people refresh their servers every three years. But with disaggregated rack, because it's just one resource we think we can keep compute for more like 3 to 6 years. RAM doesn't go bad, right, it just RAMs, all the devices last forever. And so those could be five years or more, at this slate, easily four to five years and the flash slate could be easily six years or even 10.
Now the approximate win estimates - so when I say OpEx I really mean depreciation of power. Conservative assumptions show a 12 to 20% OpEx savings and more aggressive assumptions between 14% and 30% OpEx and this is just a reasonable savings. So this kind of approach of disaggregating the components is something that of course we can do but pretty much anybody in the industry can do, it just really requires being able to have your software adjust to this different model of compute.
Okay so let's re-examine those last two paragraphs. What Facebook is trying to say is that it can lower its operating expenses by 12%-30%. This implies that every bit of savings that Facebook experiences on Opex is lost revenue for Intel. This is because Intel doesn't get to sell all those extra CPUs and Solid State disk drives as a result of efficiency gains on the part of well-managed data centers.
Facebook also mentions that almost anyone in the IT industry can do what they're doing. This implies that many of the data centers in the world have a lot of excess capacity. With server resources being moved to cloud based services like Amazon (AMZN) Cloud and IBM (IBM) Cloud, the amount of waste will drastically decline.
Facebook believes that it can extend the lifecycle of its refresh from 3 years to 6 years. This is a key assumption for my five year forecast on Intel's data center group.
The cloud will grow significantly, which hurts Intel
According to IDC:
Infrastructure as a service (IaaS) broadly includes compute resources, storage resources, and system infrastructure software delivered via the cloud services model. Revenue in the IaaS market was about $12 billion in 2011 and is expected to increase at a CAGR of 29% through 2015.
The specific segment that Intel operates in (providing hardware) is diminishing in size. This is because Intel generates profit regardless of how well the clients optimize the hardware that it sells. In fact, waste is beneficial to Intel; because it can ship more server chip units than what businesses actually need.
According to IDC:
The most significant benefit comes from re-hosting applications on AWS (Amazon Web Services) infrastructure due to lower capital and operational costs. This reduction in capex and opex accounted for a savings of nearly $276,000 per application per year. Companies were able to consolidate, integrate, and standardize their infrastructure.
So in summary, by housing servers on the cloud, various companies have been able to save about $276,000 on hardware per application per year. Of course, the study determines that an application ranges from a small internally developed application to an application that has over 20 million customers. So the average cost savings applies to a very broad sample population.
That being the case, Intel could be in significant heaps of trouble. Because if more data centers move to the cloud, Intel's primary customer will be data centers that are managed by companies with a dedicated need (Facebook) or intermediaries like Amazon. Both Amazon and Facebook are very effective at managing data center resources.
Currently, server vendors have reported 6.2% revenue deceleration on a consolidated basis. Declining demand for servers should not be a surprise as server resources are being used more efficiently.
In the next section, I am going to update my assumptions for Intel's server segment over the next five years to factor in a five-year regression of its current unit shipment volumes.
Five year assumption for Intel's server business
Facebook states that it can reduce its server refresh from 3 year to 6 years. Better management of data centers paired with the transition to the cloud will cause chip shipments to fall by 50%. In response to falling shipments, I expect Intel to increase the average selling price of its server chips.
Over the course of five years, shipments will fall by 50% resulting in a 13% annualized decay factor. Second, I expect 8% annualized increases in pricing. Therefore, by summing a 13% decline in shipment with an 8% growth rate in pricing, we can arrive at a 5% annualized decay rate for server segment revenue.
My 2013 revenue and operating income figure come from Intel's investor meeting. From that point onwards, I projected a 5% decline in annual revenue from that segment, and I also assume a 47% operating margin for the segment through 2013 to 2017. I believe that Intel will sustain its operating margins, because falling volumes will have limited impact on both fixed and variable costs.
Currently analysts on a consensus basis estimate that Intel will generate $53.21 billion in annual revenue in fiscal year 2014. When I factor in falling data center sales, Intel will generate $52.61 billion in annual revenue during fiscal year 2014. This implies that revenue growth will be flat year-over-year in 2014.
The transition to the cloud may result in Intel's data center revenues falling from $11.5 billion to $9.4 billion through 2013 to 2017. On the other hand, web-based enterprises should benefit from this transition by reporting falling operating expenditures.
If consumer PC demand falls beyond my expectations, then Intel stock could go from reporting flat revenue growth to negative growth in fiscal year 2014. Intel shareholders may stomach one business segment reporting declining revenues, but if two of its income generating segments report year-over-year declines, investors may end up exiting the stock in droves.
So to summarize my outlook: investors should avoid Intel for all of 2014.