We often return to the issue of the role of the hyperscalers in 5G, because it is such a significant one for the future shape of the telecoms networks and the business model of operators. The spotlight is currently firmly on Amazon AWS, after a string of announcements and alliances that highlight the rapid evolution of its telecoms strategy.
Following US-based Dish Network’s decision to deploy the RAN functions, in its forthcoming greenfield 5G network, on the AWS public cloud, several Open RAN vendors have tightened their ties with AWS, including Mavenir, Nokia and Viavi.
And Telefónica said last week that it had validated AWS to run its 5G core in Brazil. It is keeping its virtualized RAN plans tied into its own telco cloud for now, but appears to be getting close to AWS on a group-wide basis – last September its German arm announced that it would run its enterprise-focused mobile core on AWS. All that places question marks over the future role of Telefónica’s own Unica cloud, which has been in development for several years with Ericsson as the primary contractor.
The Spanish group’s shift towards an external cloud is common to many large operators. Many telcos, such as AT&T and Verizon, have retreated from the telco cloud in stages – first abandoning the idea of competing with the hyperscalers for enterprise or consumer cloud business; then starting to move IT functions to the public cloud, in AT&T’s case with Microsoft Azure; and in many cases, then evaluating public cloud solutions for network functions such as the 5G cloud-native core.
It is not a one-way process. Some operators are concerned about being locked into a particular public cloud in the same way they were locked into a certain RAN vendor. There are conversations about groups of telcos, such as the major European groups, cooperating to build a shared cloud that would be optimized for their needs and dilute the influence of the big US players. There will be hybrid models and cooperations, as Orange’s initiative to create its own cloud, but based on Microsoft Azure, shows (see separate item below). Many operators will be eager to ensure they use as many clouds as possible so that they remain the seat of power, and will work with specialist providers in some scenarios (see item below about BSO).
The factors that may halt this progression to the public cloud could be the virtualized RAN (vRAN) and the edge cloud it will require. Operators have different views of how distributed their vRANs will be, but most are moving towards an architecture in which non-real time functions are run in a central data center (centralized units or CUs), in order to tap into the scale economics of the cloud; while real time and Layer 1 network functions such as Massive MIMO beamforming or spectrum sharing are carried out closer to the cell site and the user, on distributed units (DUs). These may start life as appliances, but the aim of architectures like Open RAN is to deploy them in a high performance distributed cloud.
That means investing in edge cloud infrastructure that is capable of supporting very high performance RAN functions, while keeping cost and power consumption under control (see separate item on Nokia and Open RAN). Some operators believe they will need 20-30 DUs, others that they will have thousands (as Rakuten does in Japan), potentially locating them at large cell sites.
In either case, there is significant hesitancy to commit these real time functions, which underpin most of the operators’ performance differentiation, to a third party cloud, at least in the early stages. As we have argued before, if telcos embark on a new wave of cloud building, specifically to support vRAN, they will also find themselves with a very high performance edge cloud that could not only run their own distributed RAN, core and transport network functions, but also enable services for enterprises that require similarly high levels of processing.
These could be offered directly, in competition with hyperscalers or, more probably, the two groups will work together. Hyperscalers would gain an optimized edge cloud without having to built it out, while operators would have access to the cloud giants’ developer and customer bases, with some sharing of the revenues and an enhanced role in the value chain. This is already seen, to some extent, in AWS’s partnerships with some operators, such as Verizon and Vodafone, centered on its Wavelengths edge offering.
And the hyperscalers could also improve their share of the value of the high performance edge, without having to invest in the build-out, by applying their R&D efforts to the thorny issue of processors that are capable of supporting demanding functions such as Layer 1 vRAN, while also being comparable in cost and power efficiency with dedicated chips.
Already, Dish is moving some vRAN instances from Intel processors to AWS’s own Graviton platform; Microsoft has been active in allowing access to high performance processors as part of its enterprise cloud offering and is likely to extend this to its telecoms strategy; Google is aggressively chasing the telco edge as a means of gaining ground on its two larger cloud rivals and its processor developments, especially the tensor processing unit (TPU) for AI applications, have gained it considerable influence in the ecosystem.
With some vendors and operators nervous of an Open RAN platform that is too dominated by Intel, and with the future of ARM uncertain as Nvidia seeks to acquire it, hyperscaler silicon is likely to play an increasingly important role in the technologies and politics of the vRAN.
For telcos, the key will be to strike a delicate and difficult balance between taking advantage of the resources, technologies and ecosystems of the hyperscalers without being forced to accept the dreaded bit-pipe role in the edge cloud, as they largely have in the central cloud.
They need to ensure they keep control of their most important processes, even when these are offloaded. Nick McQuire, head of enterprise research at analyst firm CCS Insight, said: “As telcos go more cloud-native and they start to expose those services into the cloud environments, will AWS disaggregate that and then actually control that API layer? In which case, will it then put the telcos in this compromised position where they become a back end service and they don’t control the customer?”
So telcos must improve the balance of power in the enterprise services market by differentiating themselves with multicloud support and souped-up edge capabilities; while working with the hyperscalers to accelerate their own progress to a fully cloud-native 5G architecture that is deployable and efficient.