Apple killed its Xserve server product line two years ago—and while an IT pro can configure a Mac Pro with OS X Server, that’s not really the same thing as a “real” server. In a few short years, I bet that Apple will be out of the server business completely—even as an opportunistic dabbler in the small business-server market. OS X Server, which is aimed directly at small businesses, has quite a lot in common with Microsoft Windows Server Essentials. These OSes are a sideshow of sorts to both companies’ primary sales channels—retail and enterprise—and it’s difficult to get a concise description of what really goes in their respective boxes; for the most part, they’re packaged for selling to technically unsophisticated small companies by local resellers and systems integrators. For most small businesses, the current trend is “embrace the cloud.” You and your local reseller are probably talking about this “cloud thing” and how to migrate your apps and data to a remote hosting center. Both Microsoft and Apple, which clearly recognize this trend, are building bridges to their cloud services in their newest small business OS offerings. Indeed, cloud services offer them better margins—but that means small-business server sales are rapidly becoming a casualty of the cloud.

The Pivot

In a recent interview, Apple chief executive Tim Cook offered two important points about Apple’s business model: first, the company will only do a few things, and it will do them “great.” Second, it will only do things that make a significant contribution to society at large. While Mr. Cook has a reputation of being friendlier to enterprise customers than Mr. Jobs, Apple’s product focus is on the consumer, and to make a huge pile of cash in the process. Any success in enterprise comes from consumer products that people bring into their workplaces. The Mac Pro hardware is a tower chassis. Towers made their contribution to society about 30 years ago. My guess is that Apple has already defunded most future development of the platform and that the Mac Pro form factor is not long for this world.


Apple’s Xserve product line came preloaded with Apple’s X Server OS. Nobody had any incentive to write other OS drivers for the Xserve hardware platform. By the time Apple switched to Intel processors in late 2006, its server arm was selling an undifferentiated product with poor documentation (exposing internal design choices has never been an Apple strength) into a market outside of their core corporate focus. Apple’s internal problems with Xserve were a result of scale. Apple introduced iTunes and the iPod in 2001. Sales stagnated until the launch of the iTunes Store in 2003. As the iTunes Store gained popularity, Apple’s IT folks faced a huge challenge—it had to massively scale an online consumer service without the benefit of any published best practices as a guide. In other words, this wasn’t the time to build things in-house. Instead, the IT folks bought the best-in-class solutions and tied them together. I won’t be the first to note that in the enterprise infrastructure market, vendors must use their own products to build credibility—i.e. “eating your own dog food.” When I was with AMD’s Opteron server processor start-up team, AMD rebuilt their corporate datacenter using Opteron-based servers as a demonstration and gesture of commitment to customers. If you don’t trust your business to your own products, why would anyone else? But Apple couldn’t scale with Xserve, all but guaranteeing the product line’s doom.

There’s Scale, and Then There’s Hyperscale

Apple now operates hyperscale datacenters in North Carolina and California. Its North Carolina facility will consume 20 megawatts of power when it hits full capacity. Apple has stated that it’s building two new datacenters in Oregon and Nevada. There are rumors of a Hong Kong datacenter and it seems likely that Apple will build a datacenter in Europe as well. What’s running in those datacenters? Apple iTunes Store, its App Store, iCloud, and the Apple Store. In order to deliver fantastic margins on all these products, these datacenter operations folks must continually reduce the cost of purchasing, deploying and managing their underlying server hardware and systems software. To date, those cost reductions have centered on purchasing known good enterprise-class solutions in such quantities that they can negotiate great deals. But that technique has a floor: at some point suppliers simply can’t reduce costs and margins and stay in business. There is an asymptotic limit to supplier cost reductions, a point of diminishing returns. To work around this, Apple will have to start thinking like Google. Apple already has significant hardware design and contract manufacturing experience and resources. Add to that another near-term advantage over Google: Apple is already in the processor design business. It crafts ARM-based mobile System-on-a-Chip (SoC) processor solutions for its iPod, iPhone, and iPad products. As ARM invades the server market over the next few years, Apple will have an interesting opportunity to design vertically integrated server hardware solutions in-house. The company owns its own services source code, the core of its magic, and a wide range of enterprise-class open source infrastructure software is emerging to support the ARM instruction set. What would it take for Apple to design its own server processors? Apple has long been rumored to have a long-term ARM architecture license, which (if true) would give it the architectural flexibility it needs to tune the microarchitecture of an ARM core for server-class performance. While Apple has not made significant acquisitions in datacenter network fabrics, Imagination’s Flow Technology and Ensigma Multistandard Communications IP blocks could help—Apple owns a small stake in Imagination Technologies, and presumably has access to that intellectual property. It doesn’t really matter that Apple’s mobile SoC developers don’t have a traditional server hardware reliability focus. Most of that ‘RAS’ functionality happens at the system software level in a hyperscale datacenter. But it does matter that Apple picks an instruction set with a strong software development tools and ecosystem support; it is unlikely to adopt an instruction set with weak SoC design tools and immature software development tools support. Starting from scratch by using a publicly available design (MIPS, et al), or designing its own instruction set, are simply not viable options. It’s also very likely that Apple, as it undergoes a steep learning curve on rack-level hardware architecture and software drivers needed to transparently port workloads from Intel-dominated infrastructure, could end up relying on a mix of Intel Xeon server processors alongside homebrew ARM SoCs. The decision about which workloads to move first to ARM will be based more on thread-level performance demands, rather than what traditional CIOs would consider “high value” workloads. If Apple can contract-manufacture server SoCs and motherboards, and assemble those components into hyperscale datacenters, it can absorb some of the margins currently earned by Intel and its hardware partners. But there’s another interesting effect of vertical integration of their hyperscale datacenter hardware: It presents Apple with an opportunity to implement end-to-end features–from consumer device to cloud service–such as secure transactions.

So, you want to buy a real server…

The Xserve was never a mistake for Apple; it was a learning experience, one that gave the company invaluable insight into datacenter architecture. But in the end it was a distraction to their core businesses. Did Apple stop building its own in-house servers when they stopped shipping Xserve hardware? Of course it did. But that doesn’t mean it won’t build its own in-house solutions again. The Mac Pro and OS X Server have played an important role in Apple’s ability to serve small businesses. As Apple focuses on what it does well (serving consumers) and as small business moves to the cloud, the market will shrink and Apple’s attention will divert from SMBs and the tower form factor. This is inevitable and likely to happen soon. If you want to buy premium graphics capabilities from Apple in the future, I’d expect them to sell virtual Mac Pro images as a cloud service. But selling servers isn’t in Apple’s long-term interest.   Image: Apple