Main image of article Will Wimpy Cores Power Your Future Cloud?
Apple and Google dominate the mobile-app market with their respective iOS and Android ecosystems, at least when it comes to consumer wallets. Waiting in the wings are Amazon, which offers apps via its online storefront, and Microsoft’s Windows RT ecosystem. It’s a very real possibility that only three of these app ecosystems will survive in the long term, meaning we have some very exciting years in front of us. (While Research In Motion and WebOS have their fans, it’s unlikely those systems will challenge the big players in the space, at least in the next while.) The popularity of these apps represents a gift and a curse to ecosystem owners. On one hand, they face the infrastructure pressures that come with serving millions of users simultaneously. However, they also have the cash to tackle those pressures in interesting ways: namely, designing vertically integrated server hardware solutions in-house. There are sizable benefits in doing so. X86 enterprise servers and infrastructure aren’t cheap by today’s standards, not when measured by both acquisition costs and emerging hyperscale operational metrics. X86 servers now account for about 98 percent of all branded server chassis shipped, worldwide. RISC servers entered the datacenter market in the 1980s alongside of PCs and IT repurposing PCs for server use, but now only account for most of the remaining ~2 percent of shipments. Competition killed off RISC, as AMD pushed Intel into a race for competitive product development. Today, Intel has essentially killed off AMD, at least in the server market. Xeon processors have outrun the thread-level performance requirements of most mainstream IT workloads; consolidating and virtualizing a lot of threads onto one server is an admission that the server has much more capacity than a single workload instance demands. Mobile-app ecosystem owners control their own services’ source code. Combine that with the emergence of enterprise-class open-source infrastructure software that supports the ARM instruction set, and you have a situation where those owners can start handling more of their infrastructure in-house.

One Vendor to Rule Them All?

New architectures are created via experiments. Companies that do not have dominant market share, seeking an advantage over an incumbent vendor, often end up performing stranger and more interesting experiments than their stronger rivals. When many companies are experimenting at once, then competitive innovation will create evolutionary, and in many cases revolutionary progress. Today, a huge amount of investment is now aimed at crafting viable non-x86 server SoCs and datacenter architectures:   Apple invented the end-to-end app ecosystem when it invented the “App Store” concept. Apple is now designing its own client processor Systems-on-a-Chip (SoCs) to absorb margins that would have gone to a branded SoC design company such as Intel or Qualcomm. Meanwhile, Bloomberg has speculated that Apple could use ARM processors to replace Intel parts in its Mac product lines. At least in theory, it would be a short step for Apple to repurpose a “desktop” processor into a server SoC as it scales its hyperscale datacenter operations from a handful of locations to many more. To those who believe that all Apple wants is a lower price from Intel: a) there is no more credible threat than going into production with an alternative, and b) Apple has identified that lack of second source for critical components is a major risk it needs to address. It is a major theme in their SEC risk statements. Google is the only currently viable end-to-end ecosystem competitor to Apple’s vertically integrated marketing machine. They have the capability to design their own SoCs. Google has much more incentive to design server SoCs than Apple, given the size of their globe-spanning hyper-scale infrastructure. Google is very secretive about its internal operations. Its SEC filings identify only seven specific factors contributing to cost of revenue derived from their datacenter operations, one of which is energy costs. How many companies identify datacenter operations as a cost of revenue to begin with? While processor costs don’t appear to be a top concern for Google, increasing reliance on Intel as a single supplier increases risk, and that could take Google down the same path as Apple. Google could let rapid evolution among ARM server SoCs play out for a couple of years, after which it may invest in bringing a winning architecture in-house though M&A. Amazon’s value is content–they own the online market for the printed word and can rival Apple in negotiations with audio and video content owners. Amazon has the potential to dominate the tablet market by leveraging its e-books and video in the context of an online store. Amazon already owns its Lab126 platform design subsidiary (which produces Kindle devices), and it’s been rumored to be investigating the purchase of an ARM SoC design team (presumably to power e-readers and tablets). While the TI OMAP rumors didn’t ultimately happen, it could still acquire ARM-related assets through mergers and acquisitions, similar to Google. Amazon’s datacenter scale is closer to Google than to Apple, providing incentive for any new SoC design team to lower costs there as well. The company’s challenge is to break away from Google with their own mobile OS and app ecosystem–the company either owns or can easily acquire all of the other pieces. Amazon will need to balance their ongoing sales of x86 cloud instances with on-boarding new architectures to efficiently power their other services at scale. Microsoft has placed a substantial bet in mobility with Windows Phone and Windows 8 RT—both of which run on ARM SoCs. Microsoft designs and contract-manufactures game consoles and human interface peripherals, in addition to the branded “reference” Windows 8 RT tablet platform, so a selling a branded tablet platform isn’t much of a stretch. Microsoft already has an ARM Architecture License, presumably covering ARM’s v8 64-bit architecture; it could easily assemble an in-house server architecture and SoC design team. Microsoft’s Bing, Maps, and other services are rivals for many Google services, and Microsoft’s Azure operating infrastructure is proof that they can adapt and scale. The challenge for Microsoft is that it must become the third viable mobile ecosystem as a declining PC market diminishes the value of its PC OS and personal productivity franchises, and thatAzure seems to be lagging its cloud competition. Facebook identifies strong risk factors with respect to customer usage shifting from PCs to mobile form factors in their October 2012 US SEC 10-Q filing; but, unlike the other four companies, it hasn’t indicated any direction for manufacturing or selling branded mobile form factors. Facebook has an app ecosystem, but it doesn’t have an ecommerce backend to sell native apps for client devices. Facebook started the Open Compute platform and is maintaining very close ties to them. In light of all that, it’s a fair argument that Facebook wants to enable an ecosystem that will deliver the efficient hyperscale hardware infrastructure it needs to control their operational costs; at the same time—and this is where Open Compute comes in—it’s likely not interested in keeping such an infrastructure to themselves as a closed internal resource or secret competitive advantage, a la Google. It’s unlikely that Facebook will design its own SoCs for mobile clients or for servers. If these hyper-scale ecosystem owners can contract-manufacture server SoCs and motherboards, and self-assemble those components into hyperscale datacenters, they can absorb some of the margins currently earned by Intel and its OEM hardware partners. And there’s another interesting effect of vertical integration of their hyperscale datacenter hardware: it presents these ecosystems with an opportunity to implement end-to-end features—from consumer device to cloud service—such as secure transactions.   Image: Sashkin/