We read a lot of market data here at The Next Platform, and we do our share of prognosticating and do even more riffing off the prognostications of others. The future is tough to predict, and the only way to know for sure what will happen is to live it. But it is fun to try to figure it out ahead of time.
Some recent data out of IDC made out eyebrow go up and also confirmed a long-held belief of ours, and it is in regard to what we are going to call “cloud outpost” spending. Which means spending on pieces of the public cloud that are running on premises or in co-location facilities rather than in the regions of the major public cloud providers. IDC calls it Dedicated Cloud Infrastructure as a Service, or DCIaaS, and that is just plain too ugly a term. So we refuse to use it. Amazon Web Services, the first major so-called “public cloud” supplier and the dominant one after establishing the market with the launch of the EC2 compute and S3 object storage services back in March 2006, called its private instances of its public cloud infrastructure Outposts, and Google and Microsoft have followed suit with varying implementations of slices of their clouds that can be run locally, and we think it is just easier to call them all cloud outposts so as to not create any more confusion.
As we like to remind everyone from time to time, Amazon Web Services – and particularly its former chief executive officer, Andy Jassy, who is now in charge of all of Amazon now that Jeff Bezos has stepped down from his CEO role – did not believe that in the future companies would have infrastructure of their own. And AWS certainly did not believe in anything resembling the Outposts that it previewed at its annual re:Invent conference in November 2018 and that became generally available in December 2019. Forgive us for not remembering precisely, but it was either in November 2013 or November 2014 when at a press conference with Jassy we raised our hand and said there was no way that all workloads would move to public clouds when it was suggested they would. (Others heard it too, it looks like in 2013.) The idea Jassy conveyed its that all workloads would move to “the cloud” and that the public cloud was the best and only reasonable implementation of “the cloud,” ergo in the “fullness of time” on-premises infrastructure was doomed to oblivion.
If IDC is correct, you forget that idea right now. Cloud outposts are coming out of the woodwork. There is the AWS Outpost offering, of course, but also Alibaba Apsara Stack, Oracle Dedicated Cloud @ Customer, Microsoft Azure Private Edge Zones, Equinix Metal, Lumen Edge Private Cloud, VMware Cloud on Dell EMC, HPE GreenLake for AzureStack, HPE GreenLake for Anthos, and Rackspace Private Cloud for OpenStack, for VMware, and for Microsoft Stack – just to name a few. It is not clear if cloud outposts are going to replace a big chunk of on-premises hardware or will be the new way that a lot of companies deploy infrastructure owned by the big clouds, but the numbers are small and are going to be a representative chunk of change within the four-year forecast period.
According to IDC, cloud outpost revenues – and specifically, annual recurring revenues for the use of server and storage gear – across all vendors accounted for a mere $138 million in 2020. That’s not much with Synergy Research saying that in 2020 spending on datacenter hardware and software was $90 billion and spending on cloud infrastructure services was $130 billion last year. (We realize these are somewhat apples to oranges numbers, but IDC does not give a clean number of cloud infrastructure services spending, but it does talk about the spending for IT gear for on-premises clouds and by the hyperscalers and cloud builders. Which is not precisely useful in this context because that shows what the hyperscalers and cloud builders spend on their iron, but not what they charge for its use at the infrastructure level – rather than the platform or application software level. In any event, between 2019 and 2025, IDC reckons that spending on cloud outposts will grow at a compound annual growth rate of 151.8 over those years to $14 billion. Importantly, IDC says that cloud outposts will be consumed by enterprises in general but also by service providers who do not want to manage their own infrastructure design and installations, either in their own datacenters or in co-location facilities.
We wonder if that hockey stick curve will continue on its steepening climb between 2025 and 2030 and if cloud outposts will be the dominant way that IT infrastructure is consumed. We can see a market evolving where a quarter of the total compute and storage capacity is in the clouds and rented like today, a quarter owned by companies and installed in facilities they control or rent, a quarter is a cloud outpost running on-premises or in a co-lo, and a quarter is not even visible because it underpins PaaS, SaaS, and serverless offerings and the price is buried in those services and not explicitly broken out.
It is not clear where to draw the line between a “datacenter” implementation and an “edge” implementation, so we are not bringing edge into this discussion, but clearly it will also be an issue and there will be even more dicing and slicing. We still believe that there will be many more orders of magnitude of compute and storage devices in the edge between datacenters and client devices, and probably anywhere from 2X to 4X the aggregate amount of compute and storage, too. No one has qualified and quantified this to our satisfaction as yet. Suffice it to say that one company’s cloud will be another company’s edge, so these lines will be fuzzy indeed.
Shift those percentages over time and over the four pillars as you see fit. But we think that is the shape of the market that will probably evolve before the decade is out. What we are calling hyperdistributed computing, in fact.
Sign up to our Newsletter
Featuring highlights, analysis, and stories from the week directly from us to your inbox with nothing in between.
dedicated cloud infrastructure as a service, although long, quite self-explanatory name