Edge computing at MWC 2025: AI is the trigger

Download

Edge computing was present across the Fira this year, though not as the headline act. Instead, it appeared in its rightful place as a key enabler, deeply woven into the discourse surrounding AI monetisation. 

The trigger point for the resurgence in hype around edge computing is AI, and particularly in how localised compute infrastructure can serve as a platform to meet the surging demand for AI inferencing. The key debate centres on where this edge infrastructure should be deployed (on-premises, base stations, CU hotels, telco central offices or further afield?), the hardware requirements at these sites and which anchor use cases this will satisfy.

Key announcements

• Fujitsu announced the addition of agentic AI capabilities onto its on-premises edge solution, named Private GPT. The solution is positioned as addressing data sovereignty concerns and to help convince enterprises of the upfront investment, Fujitsu offers a ‘test-before-you-invest’ initiative.
• Nokia launched MX Context, a solution built upon its on-premises edge platform to harmonise sensor data and provide a fundamental contextual awareness to bring more sophistication to industrial use cases, such as worker safety.
• Khasm Labs announced the Telco Cloud AI Pilot, a framework for deploying real-time AI application at the network edge. In a PoC, AT&T’s network edge was used to identify pedestrian safety risks and suggest traffic adjustments accordingly in the US city of Bellevue, Washington, DC.
• MEF, in collaboration with Infosys, Nvidia and IronYun, demonstrated how its Lifecycle Service Orchestration (LSO) APIs can enable a fully automated process for enterprises to obtain pricing and consume GPU resources at the edge – in this case to power IronYun’s computer vision applications.

Key takeaways

AI and RAN: Voulez-vous or Déjà vu?

The AI-RAN Alliance laid at the heart of telco edge discourse at MWC this year. This initiative, originally announced at the last year’s congress, has now grown from 11 to 82 members (including eight operators). At MWC 2025, it held several demos related to its ambition of running enterprise AI workloads at the RAN. One such example was presented by Arm which used Nvidia Grace Hopper chip to demonstrate how network workloads at the RAN could run on the CPU (with some L1 functions offloaded onto the graphics processing unit, or the GPU), while the remaining GPU capacity can be used to run a GenAI assistant, in this case James.

Source: STL Partners

While these demos evidenced compelling progress in the technical feasibility of the ‘AI and RAN’ ambition, we were ultimately left feeling that this may be another cycle of the same discussion the industry was having five years ago when operators similarly expressed a substantial interest in telco edge, typically under the label of multi-access edge computing (MEC). Another clear parallel between these two cycles of discussion is that in both instances, at least one large tech player has acted as a central champion of the solution – in the case of MEC, it was driven by AWS and Microsoft Azure, while AI-RAN is being driven by Nvidia.

Our previous network edge survey indicated enthusiasm for the convergence of network and enterprise workloads

Despite the initial enthusiasm in this previous cycle of hype around telco edge, operators opted away from convergence. We question whether this latest cycle will lead to a different result.

In addition, it remains unclear whether enterprises want to deploy applications across such an extremely distributed footprint – taking into account the tepid demand to date for more centralised telco edge sites. There are early suggestions that members are leaning towards a more centralised deployment model, with T-Mobile’s president of technology Ulf Ewaldsson expressing that it is not yet clear whether “the GPUs will be at the [base station] or further upstream” (likely referring to CU hotels). The real question is not about the technical feasibility but the commercial strategy and attractiveness of AI-RAN.

AI in an OT world: where private networks and on-premises edge intersect

In contrast to AI-RAN, an area where there is a more proven appetite for the convergence between network functions and enterprise applications is with private networks. An emerging theme at MWC this year was the intersection of private networks and on-premises edge to satisfy the need for AI in the OT world. Many players, including Nokia, Ericsson and Capgemini, demonstrated how private networks form an essential platform for Industry 4.0 applications; although they have adopted markedly different approaches.

The proliferation of GenAI at the edge

There is a growing demand from enterprises to deploy GenAI at the edge, driven largely by security, resiliency and cost concerns. Therefore, many players have developed on-premises platforms to support SLMs – which, in comparison to their LLM counterparts, have been shown to be remarkably capable and offer lower hallucination factors. The differentiator across these solutions is increasingly in the way these platforms interact with local data and less so on the underlying language model that has been used – which is becoming a commoditised space (there are now 1.5 million models available on Hugging Face). SLMs and LLMs are primarily trained on internet data which limits their use – it is only when these models can ingest enterprise data (typically achieved via retrieval-augmented generation, or RAG) that compelling use cases are unlocked.

Progress in European telco edge federation at last

A clear highlight from the conference lays in the progress demonstrated by Telefónica, Telecom Italia and Deutsche Telekom in establishing a federated European edge-cloud network. Public funding (via the IPCEI-CIS EU funding vehicle) has provided the impetus for European operators to progress in this domain. As we state in our ‘Telco edge manifesto’, federation is a critical prerequisite to convince enterprises and developers to deploy applications at the network edge. Indeed, Telefónica demonstrated how this process will function via the platform it has developed in collaboration with Nearby Computing. Demonstrating that these efforts will continue beyond the closure of the IPCEI-CIS project in 2026, the 8ra initiative was founded in 2024. Likewise, it has been suggested that the EU will launch two more IPCEI projects this year that will focus on developing edge cloud and AI infrastructure – a process likely encouraged by the current geopolitical climate.

Telefónica MEC accessed via the Nearby One platform

Source: Nearby Computing and Telefónica

Notable absence: Hyperscalers take a backseat in telco edge

Whereas in previous years, hyperscalers (led primarily by AWS with Wavelength) played a commanding role in driving the expansion of the telco edge, the renewed interest in this domain often does not position hyperscalers as core partners. Operators that previously deployed Wavelength, such as SK Telecom and KDDI, have presented visions of expanded network edge footprints that no longer include AWS as the key partner.

One possible reason for this shift in partnership strategy is that many operators are now entering the latter stages of their 5G investment cycles and, as a result, have greater financial flexibility to take ownership of the full solution. In addition, the inherent similarity between AWS’s Wavelength and Local Zones had made it difficult for operators to differentiate their services. By taking greater ownership of the stack, telcos can drive more areas of differentiation.

From the hyperscalers’ perspective, their attention is shifting to solutions with a more defined ROI, as large-scale network edge deployments such as the partnership between AWS and Verizon have struggled to deliver meaningful returns. A signal of AWS deprioritising telco edge is the fact that it is no longer labelled as a founding member of the AI-RAN Alliance.

What next?

STL Partners will continue to closely monitor the evolution of the edge computing market, and how telecoms operators fit within it. To view STL Partners’ broader perspective on MWC 2025 – covering topics such as network innovation, enterprise platforms and private networks – click here.