It seems like 2024 is already the year of busted myths. One of particular interest is the idea that telcos would cede “carrier cloud” hosting applications for their services, including 5G, to public cloud providers. As a Light Reading story recounts, this just doesn’t seem to be happening. Of 38 operators who offered me comment on this so far this year, only 3 said they were “considering” public cloud hosting of critical elements of network services, and 24 said they’d rejected the idea completely. I don’t have consistent comments from this same group going back to early 2023, but in 1Q23 over half of operators operators were at least “considering” some public cloud hosting. What happened?
It would be wonderful to report some set of changed conditions that drove this shift, and probably interesting too, but I can’t say that. Among the 38 operators, all but 5 said that the views I might have heard early in 2023 likely came from people who were not actively involved with planning for cloud hosting of features at the time. In other words, the early view was misguided.
OK, then why is the current view so negative, forgetting the past? The article cites regulations and security, but the 38 operators’ number one answer was simply “reliability and availability”, cited by 33 of the 38. Security was a distant second with 17 votes, and only 2 cited regulatory issues. Governance and policy compliance, though, got third place with 9 votes.
5G features are currently largely focused on the RAN, and these features tend to be hosted at least fairly forward, though not usually proximate to the towers. The fact is that most of the hosting isn’t even “carrier cloud” because resource requirements way out there are sparse and so often stuff is run on dedicated devices with some local hot standby, rather than from a resource pool we could even loosely call a “cloud”. This makes these features something akin to the enterprise IoT edge applications, which are typically close to the operation they support, running on dedicated devices.
Could you cloud-host this stuff? In theory, say 28 of the 38 operators, but even discounting the question of whether you’d really gain much resource economy of scale, there’s concern about creating too much separation. It’s not about latency as much as it is multiplying the points of failure that could disconnect the features from the rest of the service. Operators point out that redundancy in connections gets sparse toward the edge, so the more paths you need to transit, the more stuff you’re hoping stays up.
Another issue that’s hard to pin down quantitatively is that of the exploding “one-source” mindset operators have. The majority of the operators say that they have a dominant 5G vendor, and the majority who have one say that’s a policy to reduce integration and support problems. They see adding in a public cloud provider as not only adding another player who has to be tuned into the installation and operations processes, but one who may have some latent tension with the major mobile equipment vendors. If you have Ericsson or Nokia 5G, you probably have a lot of solid reasons to use them for everything, including hosting, or at least use their recommendations. Nobody wanted to be pinned down on whether their dominant vendor was telling them not to use public cloud, but I have a strong sense that’s the case.
Another issue that seems to bear on the public cloud decision but doesn’t get much discussion from telcos is the attitude of telcos outside the US toward the public cloud providers, the most credible of which are US companies. Some operator contacts I know well enough to have a candid exchange with admit that there would be some political pushback on a decision to use a US cloud provider even via a zone local to the operators’ countries. There’s usually some political/regulatory tension with these providers, and there’s also the fear that regulatory action taken in another area than telco might spill over and impact a telco feature-hosting relationship.
What might change the current situation? There are three factors that operators cite. First, it’s possible that if 5G hosting became more 5G-Core-centric, they’d consider public cloud. Second, it’s possible that if the public cloud providers offered some compelling service value proposition (for many, “compelling discounts” seems to be the undertone) they’d reconsider. Finally, if “feature hosting” exploded beyond core network features, it’s possible that if these new features were best handled by public cloud, they could pull the rest of feature hosting in there too.
None of this means that public cloud partnerships with telcos are coming to an end. What it does show is that telcos are behind enterprises in terms of cloud planning skills, and enterprises admitted to me that they’d only started to realize, in early 2023, that they were being led astray by hype. Telcos seem to be coming to that same realization, but the delay may be as much due to their stubbornly clinging to 5G myths as to lack of cloud skills. Even in the last quarter of 2023, the majority of telcos seemed to believe that there would be widespread “feature hosting” arising from 5G. This month, half of those who believed then no longer believe.
You might well ask what this could mean for carrier cloud. Will operators end up building out their own cloud infrastructure after all? Maybe, but (not to sound like a broken record) I think that will depend on the evolution of their metro strategy. As I’ve said before, metro appears to be the feature-hosting sweet spot, but in the higher-opportunity-density major metro areas (about 40 of them in the US and another 230 worldwide) even metro could be distributed, making it less a single resource pool than a collection of very tightly coupled pools spread out over a radius of 40 miles or so.
There’s also a question of how you build these pools. Is carrier cloud, feature hosting, and metro a combined reprise of public-cloud models of separate servers and network gear, or do we see something that starts to look like server coupling to networks? This model would actually make sense for network feature hosting, and DriveNets for example seems to be aligning features to support it. They’ve added optical interfaces for tighter coupling at higher speeds, and their architecture of distributed white boxes could easily accommodate the introduction of a mixture of white boxes and servers.
For vendors overall, I think that the telcos’ growing awareness of the problems they’d encounter with public cloud hosting could be a positive, perhaps a strong positive, but telcos aren’t seeing aggressive positioning along that line (most still say vendors aren’t talking up a mixed server model, even DriveNets). It’s early in the telco awakening (or reawakening) on carrier cloud, though, and we may see progress through the year.
