Blog Post

How I learned to stop worrying and love commoditization

Customers love commoditization. It improves their capital economics, allows them to standardize their operational procedures and reduces their operational expenses. History has shown that cost and performance curves favor the commodity, and more to the point, key customers have come to believe this as an article of faith. While this is bad news for certain classes of startups — the vertically integrated system, the proprietary platform — it opens the door for another: the “commoditization accelerant.”

Commoditization accelerants are not commodities themselves, but make it easier for customers to do what they already want to do: consume commodity technology. The trick here is to identify a potential commoditization wave that is somehow held back by a missing capability — and then to supply it.

Here are three examples:

The network commoditization accelerant

Networking is the last bastion of proprietary IT. It is here that the mainframe-style business model has been most persistent and most deeply entrenched. If the major infrastructure disciplines were all cast as Cold War dictatorships, networking would be North Korea. The big opportunity is to bring the commodity economics of the server industry to networking. This means cleaving the vertically integrated networking business model into an open structure where customers choose their hardware and software separately.

Data center switching is the segment most ready to take this leap. There are at least two reasons for this. First is the arrival of standardized top-of-rack switch designs, available from a growing number of ODMs and all based on the same highly performant merchant silicon. Second is the influence of the giant webscale operators (e.g. [company]Amazon[/company], [company]Facebook[/company], [company]Google[/company]), who have the massive scale and technical know-how that allows them to substitute their own R&D dollars for vendor margin dollars and come out way ahead. They have already begun the move to commodity hardware. A quick look at Open Compute Project attendee lists reveals that this agenda has crossed over into the enterprise market as well.

[pullquote person=”” attribution=””]If the major infrastructure disciplines were all cast as Cold War dictatorships, networking would be North Korea.[/pullquote]

So what is holding back the complete commoditization of data center switching? Software. The operating systems available to run on the new breed of hardware have been sorely lacking. Prior efforts like FutureSoft, Level7 and Vyatta have all fallen short and were probably too early. These days a new cohort of startups such as Cumulus Networks (see disclosure) and Pica8 is taking a fresh run at it, aided by better market timing, a more robust technical approach and the runaway success of Linux in the data center. Stakes are high: the company that provides a capable networking operating system will be lighting the match on a multi-billion-dollar commoditization bonfire. This is as big a threat to proprietary networking incumbents like Cisco as Linux-on-x86 posed to Sun (and we all know how that turned out).

The storage commoditization accelerant

Years ago I was having lunch with Dan Warmenhoeven, then CEO of NetApp. One of the companies I was working with at the time, Acopia Networks, was trying to create a market for “network file virtualization.” I mentioned Acopia to Dan and while he was gracious as always, I could tell he wasn’t a fan. “Why?” I asked. “They suck our brains out,” he stated plainly. NetApp and EMC did everything they could to marginalize Acopia and their ilk. Through this experience I learned a lesson that seems obvious today: nobody likes to be virtualized. If you are virtualized, you are commoditized. Circling back to our thesis, all forms of “virtualization” are commoditization accelerants. And in storage there is a lot to commoditize.

For years, storage systems vendors like NetApp and [company]EMC[/company] have gotten away with marking up commodity disks to high, proprietary-vendor margins. This has always been obvious to customers, who can easily compare the $-per-GB of their integrated storage arrays with that of the disks themselves. While customers haven’t liked the price disparity, they have had no other way to achieve enterprise-grade storage capabilities.

The dramatic arrival of Flash on the enterprise storage scene exposed the vulnerability of the incumbents, who were slow to exploit its potential. New competitors such as [company]Nimble Storage[/company], Nutanix and Pure Storage took advantage. But these companies all operate on the same vertically integrated business model as the incumbents. They are not commoditization accelerants, just superior evolutions of the traditional strategy.

Hadoop taps into customers’ desire to use commodity in-server disks, and brings a new storage architecture to the table that radically reduces $-per-GB. But this storage architecture is tightly linked to batch analytic workloads and Hadoop’s own protocols and processing frameworks. It was never intended to be a horizontal primary storage technology.

The next and ultimately more hazardous shoe to drop will be a software-based approach that squarely targets existing enterprise storage requirements — a phenomenon sweepingly referred to as “software-defined storage.” Some of the more relevant offerings in this camp include [company]VMware[/company]’s VSAN as well as products from startups such as Coho Data, Maxta, StoreVisor and PernixData. All these companies are following a classic “innovator’s dilemma” strategy by attacking the incumbents from below and gradually expanding their addressable market as their products become more capable. They have abandoned the vertically integrated model, and in particular the egregious markup of commodity disks and SSDs that constitutes such a large proportion of total vendor margin. Once again, conditions are ripe for a software accelerant of this pent-up commoditization wave.

The two examples above, in networking and storage, take advantage of the commoditization of hardware. Commoditization of software is trickier. Software products are less likely to be “perfect substitutes,” are more likely to sport differentiating capabilities that enable divergent use cases and often pose higher switching costs. These caveats firmly on the table, let’s explore the new commoditization dynamics around one of today’s most critical software platforms: server virtualization.

The virtualization commoditization accelerant

We’ve seen that all forms of virtualization can act as commoditization accelerants. However, it is also true that virtualization itself is vulnerable to commoditization. The main historical threat to proprietary server virtualization has come from open source hypervisors such as Xen and KVM, which made inroads in key segments like the public cloud. But recently a very old alternative has started to make very new gains. Linux containers, which have been around for more than a decade, are suddenly all the rage.

One important reason for container-mania is the impact of a commoditization accelerant: Docker. Docker didn’t invent containers, but it has made them vastly more usable and useful to developers. In doing so, Docker is tapping into a pent-up desire for a lighter-weight, commodity alternative to virtual machine technology, and helping unleash a powerful commoditization wave that poses a real problem for hypervisor vendors VMware, [company]Microsoft[/company] and [company]Citrix[/company]. Aided by Docker, the “new” container-based approach has crossed the “good enough” threshold on enough dimensions that a tipping point has clearly been reached.

The case of Linux containers and virtualization is more nuanced than the networking and storage examples. Virtualization proponents quickly point out that there remain important properties of virtual machines (such as superior isolation) that are still not delivered with containers, and that existing enterprise workloads running on virtual infrastructure (especially Windows workloads) are not going to be redeployed into containers anytime soon. Meanwhile, Linux container proponents argue that their most important implication is not VM-replacement, but the enablement of a new way of building applications based on “micro-services.” And then there is the inconvenient truth that many containers will actually be deployed inside VMs, not instead of them! Valid points all, but the underlying trend remains: many new applications that might have been deployed directly in VMs will now be deployed in containers. The effect of this on the programming, systems management and operations model will be profound. The virtualization platform’s primacy will be eroded even in those environments where it is not completely eliminated.

There are certainly other accelerants besides Docker that will be required for Linux containers to expand into additional enterprise use cases. Innovations in security, storage and networking are a few that come to mind. There will be others. Some may even be substantial enough to sustain independent businesses riding the containerization wave.

Accelerants are not commodities

Note that the commoditization accelerant should never be confused with the commodity itself. The networking OS enables white-box switching hardware but is itself quite distinctive. Software-defined storage products are themselves proprietary but allow the use of standard in-server disks. Docker may not yet have found its revenue model but has staked out a unique piece of prime real estate atop the Linux container movement. All three accelerants hold the potential for defensible high-margin businesses buoyed by the flood of commoditization they have unleashed.

Chart by Peter Wagner
Chart by Peter Wagner

What are some other pent-up commoditization waves?  What accelerants can be injected to unlock their potential?  We believe such opportunities also exist in databases, the “cloud stack” and other primary elements of data center infrastructure. The gravity of commoditization is too powerful to be held in check, and great companies will be built by channeling this potential. They will be the ones that succeed in translating the gravitational pull of commoditization into high velocity, without straying too close to the black hole and ending up crushed inside the event horizon themselves.

Peter Wagner is founding partner of Wing, an early stage venture capital firm focused on building companies that matter in business technology markets. He was previously managing partner at Accel Partners, where he helped build companies such as Nimble Storage, Opower, Fusion-io, Riverbed and Blue Jeans Network over a 15-year career with the firm. Follow him on Twitter @peter_wagner.

Disclosure: Wing is an investor in Cumulus Networks.

10 Responses to “How I learned to stop worrying and love commoditization”

  1. Excellent article, but I do want to call attention to the huge misclassification of Nutanix as a storage competitor. Nothing could be further from the truth. Unlike storage products, Nutanix utilizes no proprietary hardware at all. The Nutanix Distributed File System moves all of the intelligence out of the hardware and into software. Nutanix converges compute, flash and disk to bring the same type of web-scale technology pioneered by Google in the cloud provider space to the enterprise.

    Indeed, one of Nutanix’s three cofounders was the lead Google scientist behind the development of GFS, MapReduce and NoSQL. Our engineering department is staffed by engineers from Google, Facebook and Twitter – they’re used to building out massively scalable environments.

    Nutanix is widely recognized as the pioneer of hyper-converged/web-scale infrastructure in the enterprise, and is certainly the industry leader. As the fastest-growing infrastructure company of the past decade, Nutanix is disrupting both established server and storage products across the globe.

    By your definition, there cannot be the slightest doubt that Nutanix is a commoditization accelerant.

    Note: I recently launched a website dedicated to how datacenter disruption is impacting the partner channel:

    • Peter Wagner

      Thanks for the comment, Steve. Its true that Nutanix offers more than just storage, having helped pioneer the “converged infrastructure” category for virtualization. And the company is clearly performing very well.

      That said, Nutanix sells vertically-integrated systems and applies large markups to the commodity components inside its boxes. So from a business model point of view, I can’t really agree that its a “commodity accelerant” as the paper defines the term. Its a good product, and a successful company, but its still operating on the traditional model. This shouldn’t be regarded as a criticism! Some of our most beloved companies from the Wing portfolio (such as Nimble Storage which I mention alongside Nutanix in the paper) are also operating on the traditional model, and in a very compelling way that brings a lot of value to customers.

      • Peter,

        Thank you for the kind comments about Nutanix, but I need to make another attempt at clarifying why we should not be lumped in with storage manufacturers, beloved or not, such as Nimble. These manufacturers are not truly software-defined; they utilize proprietary hardware components (i.e. storage controllers) along with commodity disk and flash. You can’t, for example, take Nimble’s software and utilize it with another array.

        Nutanix could, if we chose to do so, sell our product as a software-only option. But we want to ensure that our customers have an “iPhone-like” datacenter experience where the commodity hardware components we select always work extremely well with our software. There is not, however, a single proprietary component in the hardware. This is evident by Dell’s launch today of the Dell XC Series Web-scale Converged Appliances – powered by Nutanix software. It could easily be argued that Nutanix does not mark up the commodity hardware at all – but we do charge for our software.

        By your logic, VMware’s EVO:Rail would not be an accelerant since it is sold in vertically integrated systems of various manufacturers. And yet, EVO:Rail is the exact same product as VSAN which you specifically spell out as an example of an accelerant.

        • Peter Wagner

          I think you make my point for me, Steve, with the phrase “if we chose to do so”. There are many examples of products in which all the technology developed by the company is in the form of software, but the company chooses to sell an integrated system. This is a choice of business model, and is the essence of the case here. Commoditization accelerants disrupt the business model of incumbents by adopting a new one. That is one reason they are so hard to compete with!

          I understand why Nutanix has chosen to persist in the old model. Its very lucrative!

  2. anshublog

    Excellent way to summarize how we are essentially de-constructing and re-building the 3 biggest pieces – storage, compute and network – to move to a ‘cheap boxes’ with ‘smart software’ model (as opposed to expensive hardware+software co-mingled.) Reminds me of how Insurance salesmen (yes, they are a legend) used to sell “whole life insurance” to unsuspecting buyers – by co-mingling term life and investment products they could charge way more for both. The customer wins when you separate out the two.

  3. Robert Perez

    Great piece which I enjoyed. However, it seems as if you are redefining what “commoditization” even means. Commoditization is the shift from design as a value proposition to cost. By saying that the accelerant “should never be confused with the commodity”, you are retaining design as your value prop. That’s like saying that your custom built home is still a cookie-cutter house, except for the fact that it had custom blueprints. I still agree with what your’re saying, but I think it might be confusing to some who don’t understand that you are talking about the whole system as the product vs the just the hardware.

    Robert Perez

    • Peter Wagner

      Hi Robert,

      Thanks for the comment. You bring up a central point which I am glad to further elaborate on. The accelerant is indeed NOT a commodity at all, and so as you say its value proposition is “design”. The accelerant IS closely coupled with a commodity, though. In fact the accelerant is only interesting precisely because it allows people to get more value out of a commodity than they would if they had to consume that commodity in its “raw” state (i.e. without the accelerant). The end result is that the customer captures the value of riding a particular commoditization wave, in a way that they may have been precluded from had the accelerant not been there. That value might be better hardware economics, or it might be leverage of an open source software ecosystem. The addition of the non-commodity accelerant does not detract from these very real commoditization benefits, and in fact works to make them more broadly available.

      This all sounds very theoretical but there are fortunately real world examples to be found. I discussed three in the paper but there are more. Looking forward I expect this to be an increasingly important phenomenon as larger parts of the market turn away from highly proprietary platforms and towards commodities (like standardized hardware and open source software) for large parts of their infrastructure.