A recent research report continued on a theme that’s become a bit of a cloud computing mantra—we’re exiting the “early adopter” phase of the cloud and heading into the main event. In some sense this is true, because we are certainly in the “early” phase of the cloud. In another sense it’s misleading because almost all big companies whose spending dominates IT overall have been cloud adopters for a year or more.
The biggest issue with the early adopter model, IMHO, is its intrinsic presumption of homogeneity of the cloud market model itself. We assume that adoption is working against a static goal, that users are on a learning curve that will take them to the same place at the end. Suppose, though, that it’s not the user that’s changing but the cloud? If in fact the cloud is a moving target, then how would you know if users were “early” to the cloud or perhaps not even there yet if measured by the evolving cloud paradigm? What is the actual evolution of the cloud model, and where are we in that process?
If we start at the top (as I’m always inclined to do) then we can divide cloud computing into two main benefit models—cost-based and feature-based. The cost-based model of the cloud says that it can do the stuff we do today at a lower cost. The feature-based model says the cloud can do stuff we don’t do today at all.
There has always been a problem with the cost-based benefit model. If you look at the essence of cost-based cloud adoption, it is this: “I can create an IT service so much cheaper than you can that I can earn a respectable ROI on my service and still offer you enough savings to offset your concerns.” That’s obviously possible, but also obviously something that gets harder as you move out of “early adopters” with special cost situations and into the mainstream. Economy of scale isn’t linear, as I’ve pointed out before. The curve plateaus, and that means that every win makes the next one harder to come by.
If the cost-based cloud is indeed under pressure you’d expect warning signs, and I think they’re there for all to see. If IaaS margins and value propositions are under more threat, we should see more focus on PaaS and SaaS because these two cloud models displace more costs and thus build benefits for a broader community. That’s exactly what we do see. You’d expect to see cloud pricing fall as providers try to get volume adoption kicked off, and we see that too. You’d expect to see cloud providers looking more to augment basic cloud revenues as their own margins shrink, and new “platform service” enhancements to the cloud seem to be announced every day.
IaaS, in my model, runs out of gas short of 10% penetration into overall IT spending. If you add in SaaS and PaaS you can get cost-based cloud up to less than 25%, but beyond that the model says that savings are too low for buyers and profits too low for sellers. By 2020 you need a transition to a feature-driven model of the cloud or you hit the wall.
Feature-based cloud opportunity seems simple and obvious, even inevitable, on the surface. It’s easy to identify things (like mobility) that are opportunity drivers. The problem is that these drivers have been around for a year or more and nothing of any consequence has happened. At least, so the enterprises say. Almost all of them recognize that there are special application planning and development techniques associated with building cloud-specific apps that could drive a feature-based cloud explosion. They just don’t know what they are.
Some startups do. Many, in fact, have build custom applications for the cloud and are exploiting cloud capabilities extensively. Most social-network companies fit this model, for example. The problem of course is that the skills involved are in great demand and enterprises can’t afford them, nor would there be enough to drive a buyer-side revolution. It would have to be up to vendors…
…who in the main are happy to stay the course with regard to application design. It makes sense to milk your R&D as far as you can, and if everyone else has that same mindset you have little to fear. To be fair, though, it’s easier to think cloud revolution in a narrow social-network application range than for a broad enough enterprise market to make things interesting.
The other issue with evolving the cloud to being feature-driven is that the architecture of the cloud today is all wrong. Most cloud providers have few data centers designed to be (you guessed it) cost-efficient, there is little attempt to integrate big data or contextual resources. The latter of these conflicts with the former; contextual mobile apps demand fairly local processing for short control loops and response time. We need a rich intra-cloud network and that’s hard when you don’t even have an intra-cloud infrastructure outside a data center.
The feature-driven model of the cloud seems to need the very thing that the cost-driven model doesn’t want—distribution of resources. There are two paths that might be taken to that goal—one by exploiting the opportunities of mobile contextual services and the other by exploiting Network Functions Virtualization. In both cases there’s an incentive to push processing more to the edge, and to distribute processes and data co-equally. That distributable-process notion seems to be the major difference between a cloud-specific application and one that just runs in the cloud.
Users of the cloud aren’t in the early-adopter phase, then. The cloud is in the early-adoption phase, and we get out of that when we can enlighten enterprises and enrich cloud infrastructure at the same time and at a reasonable pace. I don’t think any evolution of the traditional cloud model is going to get us to that point, but I think that either mobility or NFV might well do it. That would make these two network technologies the best friends the cloud ever had.