Nokia makes its AI-RAN play at #MWC25

• There’s growing interest (as well as scepticism) in the AI-RAN concept
• Nokia, one of the founders of the AI-RAN Alliance, looks determined to position itself at the forefront of AI-RAN developments
• It has rocked up to MWC25 with news of advances with fellow AI-RAN Alliance founders Nvidia, T-Mobile US, SoftBank, as well as with KDDI
• But how might AI-RAN be deployed in commercial mobile networks? That’s still a work in progress

BARCELONA – #MWC25 – Nokia has rocked up to this year’s Mobile World Congress with the clear intention of positioning itself as the leading mobile infrastructure vendor in the emerging AI-RAN space, touting advances with three influential telcos – T-Mobile US, SoftBank and KDDI – as well as AI tech giant Nvidia. 

That the Finnish vendor has been heavily engaged with Nvidia, T-Mobile US and SoftBank on AI-RAN matters is not a surprise, as all four companies are founding members of the AI-RAN Alliance, which was unveiled a year ago at MWC24 with a mission to develop technology and use cases focused on three key areas: AI for RAN, exploring how AI can enhance radio access network (RAN) operations and capabilities; AI and RAN, developing infrastructure that supports shared AI and RAN workloads; and AI on RAN, focused on the development and deployment of AI-enabled applications at the edge of the network that can be delivered over 5G connections. 

And that Nokia is also heavily engaged with KDDI, which is often found at the leading edge of innovative, next-generation telco infrastructure developments  – see here how the Japanese operator is embracing and benefitting from the deployment of disaggregated routing systems to help cope with AI traffic – is a feather in its cap. (Though it’s worth noting that KDDI is also exploring AI-enabled next-gen RAN options with Ericsson too.)

So what is Nokia up to? 

Together with T-Mobile US, it has been evaluating AI-RAN network architecture in general and, specifically, “the feasibility of using accelerated computing for L1”, as well as getting to grips with the co-existence of AI and RAN workloads on the shared infrastructure using Nokia Cloud RAN and Nvidia processors. And notably, added Nokia in this announcement, “the companies are also exploring monetisation opportunities and techno-economics of the AI-RAN multi-purpose cloud infrastructure.”

Ah yes, the financials. This is one of the big barriers for AI-RAN – the deployment costs, the operational costs (including energy consumption) and the return on investment (RoI) model, which is a particularly sore point for mobile operators wondering if they’ll ever break even with 5G. 

SoftBank and Nvidia have already tried to persuade the industry that AI-RAN can deliver some great returns on investments – see SoftBank and Nvidia tie revenue model to new AI-RAN solution.

There’s no doubt that T-Mobile US is at least invested in the idea and ready to promote it. At Nokia’s pre-MWC presentation in Barcelona on Sunday (2 March), the US operator’s president of technology, Ulf Ewaldsson, rocked up to say that T-Mobile US is “excited that the RAN edge could be used for AI” and that it’s “super excited” about the potential of the AI-RAN Alliance. 

And what of SoftBank, which, along with Nvidia, is the driving force behind the AI-RAN Alliance and its propositions? (Remember, Alex Jinsung Choi, principal fellow of SoftBank ’s Research Institute of Advanced Technology, is chair of the Alliance). 

Nokia notes that it has worked with the Japanese giant to show that “multi-purpose, optimised AI workloads” can run on an integrated platform running Red Hat OpenShift and managed using Nokia’s MantaRay NM solution for network management and SoftBank’s AITRAS Orchestrator.” AITRAS is the AI-RAN solution developed by SoftBank and Nvidia.  

The vendor adds: “This collaboration illustrates how both RAN and non-RAN AI workloads can efficiently share the computing resources, significantly enhancing resource utilisation. This not only leads to improved operational efficiencies but also accelerates the return on investment for network operators.” There’s the RoI model reference again – fundamentally important but currently (at least to this reporter) unconvincing. 

And what of KDDI (which is currently not a member of the AI-RAN Alliance)?

Nokia says it has forged a “strategic partnership with KDDI to research the practical applications of AI-RAN, including use cases and architectures, to make it commercially viable in the future. The companies will explore how AI applications can enhance the user experience, enhance network quality, reduce 5G network-related costs and power consumption through automation, and create monetisation opportunities that leverage GPUs and generative AI. The companies will conduct a commercial trial using AI-enabled RAN hardware and research AI utilisation to optimise network performance.” 

Separately, KDDI has been undertaking some interesting intent-based AI network management trials that would certainly chime with some of the AI-RAN concepts.

The AI-RAN concept seems like a good idea and it is gaining industry traction, with the AI-RAN Alliance announcing last week that it now has 75 members, though it’s notable that only seven of those are network operators and the rest are vendors, industry groups, academic institutes etc. After a year that’s not a ringing endorsement from the carrier community. Sure, KDDI is interested and Indonesia’s Indosat Ooredoo Hutchison (IOH), which is pitching itself as an AI-native telco, was also at the Nokia presentation to extol the general virtues of AI-RAN, but it seems to be interested from a slight distance currently. 

That’s because, right now, this is a tough sell for telcos. The big question for mobile operators is – how does this concept fit in with mobile network architectures and investment plans? For most operators, not very easily.

And it’s not like it’s crystal clear from the RAN vendor side either – there are plenty of permutations. 

From Nokia’s perspective, it seems like cloud RAN offers the way forward (though right now, cloud RAN doesn’t appear to be the primary radio access network architecture approach for most mobile operators). The vendor notes that its ongoing work with its partners “will lay the foundations for developing platform-as-a-service (PaaS) business models for CSPs, which helps them unlock new monetisation opportunities by offering scalable computing infrastructure and capabilities for processing AI and other services. Under its anyRAN approach,” where the vendor offers a range of RAN system options to meet its customers’ needs, “Nokia is evolving cloud RAN solutions to include AI computing in the shared infrastructure to maximise resource efficiency for operators,” it adds. 

That, in turn, suggests that AI-RAN might work in a distributed architecture, though not necessarily one that involves deploying shared AI-RAN compute stacks at every base station. And this is where there’s uncertainty: T-Mobile US’s Ewaldsson noted at the Nokia presentation day that “we’re not sure when the architecture will be ready” and “whether the GPUs will be at the [far] edge or further upstream,” hinting at a more regional node architecture that actually fits in with the kind of regional distributed telco datacentre architecture that SoftBank has been talking about recently – see Japan’s SoftBank provides AI strategy update.

So there’s a lot of work to be done and Nokia is keen to play a defining role because at the end of the day, the vendor won’t want the AI-RAN concept to manifest itself in such a way that its traditional RAN business is all but wiped out. 

To further explore the AI-RAN concept and potential, Nokia is establishing an AI-RAN Centre at its offices in Dallas, Texas. The facility “will enable Nokia’s partners to develop and test AI-RAN solutions in real-world network conditions with a focus on creating innovative use cases, prototypes, and to validate AI-RAN reference architecture,” states the vendor. 

AI-RAN is an idea, a concept, that could evolve in many different ways and currently it’s a big unknown, so expect to hear a lot of ideas and statements about it during MWC25 (and beyond). 

- Ray Le Maistre, Editorial Director, TelecomTV