The Spectrum of Hardware for the Cloud

When I joined Cisco in 1996, I was sitting in new-hire orientation and distinctly remember one of the VPs tell us that, "we are a software company".  I remember being surprised at this, because we had just finished touring an incredible lab filled with huge boxes of blinking lights. But after working with the products for a few months, it became obvious that the value to our customers was the stability and flexibility of the IOS software (no kids, not the stuff on your iPhone).

Ten years went by and I found myself in a spin-out that was using the ODM hardware manufacturers used by Linksys. We expected those products to be very simple, since they were targeted at very low price points. But as we began testing them, we were shocked at the performance levels - they were 20% of the cost of "enterprise" products with at least 20-30% better performance,  including 80-90% of the same features. Wow!  The previous 10 years had been all about custom-built hardware and ASICs, and now we had crossed-over to where "merchant silicon" was a legitimate alternative.

Within a decade, our industry had transitioned from software to hardware and back to software being the key value element in IT infrastructure. Seeing those price/performance economics, you might think that this would quickly become the defacto standard for all vendors, just like other electronics in other industries (TVs, Kitchen Appliances, etc.). But that's not the case. In fact, the IT industry has actually spread out the way they deal with hardware in a wide spectrum.

Custom ASICs
Cisco continues to lead the market in custom high-end ASIC development, making it a critical part of their future strategy in the data center. Juniper is another company that relies heavily on custom ASIC development to control their ability to deliver features across their platforms. Both of these companies are also starting to create programs to allow programability or API-level interaction with their systems. On the other end of custom hardware are companies like SeaMicro (acquired by AMD) creating unique hardware density to link low-power CPU chips, driving density and cost/watt costs down dramatically.


Programmable FPGAs / Merchant Silicon
Dozens of networking companies leverage merchant silicon from Broadcom, with Intel potentially growing share in this space with their acquisition of Fulcrum Technologies in 2011. While merchant silicon levels the playing field for vendors in terms of most networking features, it does deliver lower costs of acquisitions, faster delivery of new configurations or density, and the ability to focus their differentiation on software. Arista is one of the leaders in trying to drive this software-centric approach to merchant silicon, without introducing radically new architectures.

Standard x86
With Intel continuing to creating greater core-density on Sandy Bridge and Romley chips, combined with virtualization technologies (VMware, Xen, KVM, etc.), many companies are moving to an appliance model with high-speed PCIe interconnects. By layering on open-source software, many functions (storage, networking, security, load-balancing, etc.) are now readily available in x86 form-factors. But it's important to remember that x86 architectures do have certain pros/cons considerations that are different from purpose-built designs for networking, storage, etc. Some companies are also using the strategy of basing their new designs on x86 architectures, but still attempting to patent certain elements above and beyond the standard hardware.

x86 is also central to the Software Defined Networking (SDN) movement, where the intelligence in the network is either moved to the edge (typically the hypervisor or virtual-switch in a server), or a centralized controller (typically running on x86 hardware).

Open Hardware Projects
As we see the growth of Cloud Computing and massive data centers, the ability to optimize the internals of 21st-century bits factory is driving a range of open hardware projects (Open Compute, etc.) Companies like Google and Facebook are driving these movements because of their scale, but the the byproduct of their community movements will begin to shift into broader usage by companies in all industries. We're even beginning to see the ODMs setup distribution divisions to eliminate some of the supply-chain.

New Value Creation, or New Value Distribution?

Just like other technology movements within the IT industry over the last 10-15 years, value creation and value capture will adjust and adapt over time. New opportunities will arise for those that identify this value. Will it be in consulting, packaging, unique design add-ons or software + hardware integration?

How much will the ability to influence our industry move from vendors to cloud providers to ODMs? Or will it be the power of the communities that collectively own the power, but distribute the value (and profits) to new companies that don't exist today?