Many vendors have found hope in NFV opportunity, including network vendors, software vendors, server vendors, and chip vendors. At VMworld the CEOs of Dell and VMware held a kind of NFV love-fest, and Intel has long been promoting NFV for the obvious reason that hosting anything consumes hosts, which consume chips. At the same time as all of this is going on, though, we hear that operator projects in NFV have largely focused on premises-hosted vCPE applications. Are these going to evolve to “real” NFV, or are all these vendors dreaming? Do they even have to? Could NFV be a “success” if it never goes beyond virtual CPE?
Let’s deal with that last question first. Operators’ consensus on defining NFV success is that it would have to improve their overall profit per bit by 10% or more. Except for business-service-only players, that clearly cannot be achieved with the business-directed vCPE services that are the current priority. We have to get beyond those, and the question is how we do that.
NFV is about virtualizing network functions, meaning extracting features from dedicated appliances and making them available in software form so that they can be hosted on something. The original NFV model was focused on hosting in the cloud, or at least on virtualization-equipped data centers of some sort. “COTS”, meaning commercial off-the-shelf servers, as the hosting point could credibly lead to carrier cloud deployments that my models have forecast could add over one hundred thousand new data centers worldwide. That’s the kind of opportunity that would engage Dell or Intel or HPE or anyone with a server business.
The challenge up to now has been the limited number of business prospects suitable for vCPE. While you can credibly host any VNFs in a cloud data center, the vCPE initiatives to date have largely focused on business buyers with Ethernet access. There aren’t enough of these opportunities to create a big demand for hosting or drive deployment of a hundred thousand data centers. In fact, one of the reasons vCPE has been popular as an application of NFV is that it doesn’t require data centers at all. Operators like the idea of putting a general-purpose device on the premises, a kind of mini-server-CPE box, and then deploying the software in it. The benefit is less economy of scale than it is agility in adding features for managed services. But that agility value is hard to put a number on, and it doesn’t build masses of data centers that consume servers and chips.
Operators who love the idea of vCPE will generally admit (or at least have admitted to me) that these applications don’t seem to lead to a carrier cloud. A very few hope that the repertoire of VNFs that could be hosted will expand, but they don’t have convincing candidates for the expansion or market data to validate the opportunity. Some think that consumer-level vCPE might get them there, but the benefit of deploying cloud-hosted access features to consumers when a typical broadband hub costs less than fifty bucks is limited. Particularly when you still need home termination of the broadband connection and WiFi.
If vCPE is going to build carrier cloud, it would have to extend to the consumer, and that extension would obviously depend on having a large number of new service opportunities that would justify cloud hosting rather than edge hosting. Most operators say that things like home monitoring could help, but marketing these against established incumbents is a challenge, and if you need hosting you’d have to deal with the first-cost question, which is getting the hosting out there to fulfill the opportunities marketing creates.
This is why most operators believe that it will take something other than vCPE to drive carrier cloud. What that could be divides them somewhat, but mostly in terms of the priority given to each option. The most obvious and most credible is mobile infrastructure, particularly the way that infrastructure would change leading up to 5G. IoT ranks second, and the generic hosting of application components ranks third. Let’s look at them in reverse.
Application hosting (meaning offering cloud services) has been attractive to operators for the carrier cloud for almost as long as cloud computing has existed. Verizon tried it, and by most accounts failed, and nothing much has really changed. The cloud is about marketing, positioning, and brand. Verizon shot behind the duck, or ahead of it, depending on what you thought the opportunity was. Most operators still yearn for cloud revenue, but they still seem uncertain as to how to get it. That makes the cloud computing driver the least dependable in terms of building infrastructure. However, it is a logical way to extend vCPE services if you could get that started, which is why it makes the list of drivers in the first place. My model says you can’t drive an initial deployment with cloud services.
IoT is the most interesting and probably compelling of all the drivers. My model says that IoT alone could drive a deployment of a hundred thousand or more data centers, making it the only driver of carrier cloud that could stand on its own. If it could stand at all, that is. The problem with IoT is that it’s a poster child for what’s called a nascent opportunity. IoT is to carrier cloud what a whiff of perfume is to a walk down the aisle. It’s a major commitment in the long term, but in the short term it’s just a vague promise. There are so many things that would have to come together to make IoT a driver for carrier cloud that the combination looks unlikely in the near term. In the long term, it rules.
My current model says that mobile infrastructure would likely add no more than about 30,000 data centers worldwide if taken alone. That doesn’t get us to nirvana, but it would be enough base deployment to facilitate other applications’ use of the data centers, which could then promote those applications, and that would bootstrap us to the necessary level of deployment. If you believe in 5G, which I do for “arms race” reasons, then it’s going to happen in some form. The trick would be making sure that the form that happens is a driver for carrier cloud and not just an abstract change in the RAN.
One other thing to consider is the combinatory value of the carrier cloud computing services and IoT drivers, particularly if they’re combining with a consumer target and maybe even consumer vCPE. Home control, after all, could be framed as an IoT application as long as we don’t get religious about demanding all the sensors be directly on the Internet. It’s not a major step from home control to home financial management, home photo management, and so forth. Thus, those applications kind of have a foot in multiple doors. Could you target them collectively?
You could. Verizon rolled out FiOS in a cherry-picking way, focusing on the areas where they had the highest probability of earning early return on infrastructure. You could do carrier cloud the same way, providing that you had credible services with provable opportunities and that you had a very strong marketing plan to promote them.
All my modeling and conversations with operators converge on the point that to make NFV successful, to make it into something other than a niche approach to business services, you have to build out carrier cloud at enough scale to enable a cascade of other applications that can exploit but not justify the cloud. So doing that is critical, and it’s going to either mean betting on 5G and mobile infrastructure or constructing a more complicated service set around a combination of cloud hosting and IoT. That’s why I’ve been saying that I think 5G may be the critical NFV driver.
That would seem to spell success for players who are incumbent in the mobile space, but operators tell me that none of these mobile incumbents are really swinging for the carrier-cloud bleachers. Instead, they’re bunting by aiming at very limited mobile missions. The opportunity is still there for any of the full-spectrum NFV players to step up and claim the space. Which might mean claim the market.