The AI divide isn’t belief vs. skepticism — it’s data-center AI vs. network AI
Ericsson is betting on distributed, deterministic infrastructure over hyperscale LLM gravity
Uplink growth and agentic AI could shift long-term value toward autonomous networks
I’m in London for Ericsson’s pre-MWC media day, which takes place each year ahead of Mobile World Congress in Barcelona and follows a familiar formula: Erik Ekudden, Ericsson’s CTO, appears as if from the future and outlines what’s about to happen in the communications industry.
The assembled analysts nod as they take notes, then later pretend they thought of it. The rest of us try to keep up.
This year feels different. Not because of another generational upgrade cycle — but because the market has split.
And to be clear, this is not AI vs. no AI. It’s data center AI vs. telco and edge AI.
On one side sits the hyperscaler LLM core camp — disaggregated, cloud-native and economically tied to massive training clusters. Vendors like Nokia and Cisco are increasingly exposed to that world: optical interconnect, IP routing, switching silicon and the near-term revenue surge driven by centralized AI factories.
If hyperscalers remain the economic center of gravity, that strategy delivers immediate growth. But it depends on centralized, probabilistic AI staying dominant. Ericsson is making a different bet.
Ekudden framed it plainly: “AI will not reach its full potential unless you combine it with advanced connectivity and distributed cloud.”
That’s a network-first thesis.
Rather than positioning AI as something that happens in a distant data center, Ericsson describes the network as becoming an “intelligent fabric” — a distributed mesh connecting not just devices, but agents: in sensors, cars, glasses, edge nodes and cores.
And crucially: “This fabric cannot be managed manually. It has to be done with autonomy.”
That line matters because it reframes the network from the transport layer to the control layer.
Ericsson remains anchored in standalone 5G cores, programmable RAN, network slicing, open APIs and automation — what it calls AI-powered 5G evolving toward AI-native networks (and, at the end of the decade, AI-native 6G networks). It’s laser-focused on building deterministic infrastructure designed for reliability and control.
But the strategy goes further: hosting AI inference inside the network itself. Not just carrying AI traffic, but becoming part of the execution environment — whether at the core, the edge or deeper into the infrastructure.
The operational signal supporting this view is uplink. Ekudden highlighted that uplink traffic could triple over the next five years, driven by AI glasses, persistent voice interaction, sensors and real-time video. In roughly a third of operator networks today, uplink growth is already outpacing downlink growth by 50%.
That is not LLM training traffic flowing into hyperscale clusters. That is distributed, agent-driven AI interacting with the physical world.
Ericsson’s bet, then, is that enduring value may not sit solely inside centralized AI factories. It will also sit in the autonomous, SLA-driven, programmable networks that connect and host AI across the real economy — from enterprise and mission-critical sectors to consumer wearables.
The divide isn’t about whether AI matters; it matters everywhere. It’s about where control, economics and long-term strategic leverage will settle: inside probabilistic hyperscale cores — or across deterministic, autonomous networks that form the intelligent fabric of the AI era.
Who let the dogs out?
I brought my dog Bowie to Ericsson’s media day — fully kitted out with Ericsson-enabled 5G video streaming tech, of course. Here he is doing a spot of high-level networking with Chris Lewis of Access Partnership and his four-legged associate. Also pictured: Ray Le Maistre of TelecomTV, visibly shamed by his lack of dog.
Tsk. Tsk.
Next year, Ray. Next year.
Read next: What Fierce Network is watching for at MWC 2026: AI, 5G and network transformation