Operators Lean In On AI Grid Location

Earlier this week I argued that the AI Grid debate needs to move on from where you place a GPU to whether geographically dispersed compute can behave as a single fabric. I stand by that. But a story that has been building across the press this week is a useful reminder that the location question, the one I have been answering the same way for two years, is now being settled in public by the people who actually own the radio networks. And they are settling it against the tower.

The reporting is consistent. Light Reading describes Nokia and Nvidia's AI-RAN proposition running into telco resistance. Verizon, Vodafone, Orange and, notably for me, Telus have all raised doubts about putting graphics processing units into the radio access network. AT&T's chief technology officer has cast public doubt on the case for AI compute at the far edge. The enthusiasm for GPU-in-the-RAN comes from two operators, T-Mobile US and SoftBank, and almost no one else. Much of the rest of the industry is looking at Intel's newer CPUs for its open RAN rollouts rather than filling cell sites with accelerators.

I want to be precise about what this does and does not prove. It does not prove that AI in the RAN is a bad idea. Applying machine learning to scheduling, link adaptation and energy management inside the baseband is real, it is shipping, and Ericsson's AI-in-RAN software subscription is a reasonable way to bring it into existing hardware. What the operators are rejecting is narrower and more specific. They are rejecting the proposition that the cell site should become a general-purpose AI inference venue, stuffed with GPUs, monetised by hosting third-party workloads at the edge of the network. That is the proposition I have said for two years does not survive contact with power, cooling, space, security and, above all, the absence of a monetisation model.

My position has been that AI Grid deployment begins at the central office and the mobile switching office, not the cell site, because every physical and commercial constraint favours the aggregation point over the tower. The reasoning was never controversial to anyone who has stood in both kinds of building. A central office has power feeds, environmental control, physical security and fibre already in place. A cell site has a cabinet, a limited power budget and a landlord. When Verizon, Vodafone, Orange and Telus decline to put GPUs at the far edge, they are not making a new argument. They are confirming an old one, and they are confirming it with capital allocation decisions rather than conference slides, which is the only confirmation that counts.

There is a workstream reason this caught my eye. Telus appearing on the skeptics' list is consistent with what I see in the market: operators that are serious about autonomous operations are also the ones being disciplined about where AI compute physically lands. Those two forms of discipline are related. An operator that thinks clearly about the economics of edge inference tends to think clearly about the economics of everything else in the network.

The AI-RAN enthusiasm gap also matters for how we read vendor claims. When a technology has two vocal operator champions and a longer list of vocal operator skeptics, that is the signature of a capability that has been field-validated in specific conditions but not commercially validated across the market. I have made this distinction before and it applies cleanly here. SoftBank's agentic AI-RAN demonstrations and T-Mobile's Nvidia-backed edge trials are real engineering. They are not yet evidence that the model generalises to operators with different cost structures, different energy prices and different enterprise demand. Treat a two-operator enthusiasm as a pilot signal, not a market verdict.

So where does this leave the fabric argument I made last week? Exactly where I left it, and stronger. The operators are removing the least defensible node from the AI Grid, the cell site as inference host, which clears the ground for the argument that actually matters. Once you accept that heavy inference will not live at the tower, the interesting question becomes how you knit central offices, regional data centres and a small number of genuinely latency-bound edge sites into one addressable pool. The industry spent this week deciding where the compute will not go. That is progress. The harder decision, who owns the fabric that arbitrates across the places it will go, is still open, though.

AI Grid: Fabric vs Location Considerations

The debate about where AI compute belongs in a telecom network has been framed as a location question from the start. Do you put the GPUs at the cell site, the central office, the regional data centre, or the hyperscale campus? I have argued consistently that the honest answer begins at the central office and the mobile switching office, because power, cooling, fibre, physical security and latency sufficiency all favour those sites over the tower. That position has not changed. But two announcements from Asia this week suggest the more consequential question is no longer where the compute sits. It is whether the compute behaves as one pool regardless of where it sits.

NTT Docomo disclosed a nationwide testbed it calls GPU over APN. It pools graphics processing units spread across eight locations in five Japanese cities and presents them to a workload as a single platform, connected over the all-photonics network that NTT Group has been building under its IOWN programme. Docomo describes it as the realisation of its AI-Centric ICT Platform concept, part of what the group now labels AIOWN, its AI-native infrastructure. Strip away the acronyms and the claim is precise and significant: distributed GPUs, addressed as if co-located, over deterministic optical transport.

KT made the point from the other direction. Its new chief executive committed 18 trillion won, roughly 11.7 billion dollars, over three years, including 3.26 billion for one gigawatt of AI data centre capacity and a plan to connect that centralised infrastructure with edge sites serving low-latency workloads such as autonomous vehicles and industrial robotics. One operator is making dispersed compute act centralised. The other is extending centralised compute out to the edge. Both are describing the same thing from opposite ends, which is a compute fabric rather than a compute site.

This matters because it decouples two decisions the industry keeps conflating. Where you place a GPU is a question about power, land and cost. Where you run a workload is a question about latency, data gravity and sovereignty. As long as placement and execution are the same decision, every AI deployment becomes a real estate argument. Once a photonic fabric can make placement invisible to the workload, the two decisions separate. Training and heavy batch inference go where power and space are cheap. Latency-bound inference lands close to the user. The fabric arbitrates between them. This does not contradict the case for the central office, it absorbs it: the central office still wins for latency-bound edge inference, but that win is now a node in a graph rather than an isolated site.

I have some history with this problem. In 2018 I wrote about building at Telefonica what was probably the industry's first fully programmable multi-access edge computing platform, and the hardest part was never the compute. It was making distributed compute addressable, governable and billable as a coherent resource rather than a scatter of isolated sites. The technology around it has moved on considerably, but the unsolved problem is the same one Docomo is now attacking with photonics.

A practitioner's caution is in order. A fabric that makes national-scale GPUs behave as one pool is a testbed today, not a product. Docomo demonstrated it in a lab-grade programme. KT's edge connection is a plan, not a live deployment. Deterministic optical transport carrying commercial service level agreements under contended traffic is a materially harder thing than a controlled demonstration, and I would treat "as if co-located" the way I treat vendor energy savings figures: directionally real, quantitatively unproven at scale. The distance between a testbed that works and a fabric that carries production workloads is precisely the distance Open RAN spent five years crossing.

Still, the framing is the takeaway. The AI Grid conversation needs to move from siting to fabric. The operators that win will be the ones who can treat geographically dispersed compute as a single addressable resource, orchestrate workloads across it against real constraints, and price and settle access to it. That is a transport, orchestration and settlement problem before it is a property problem. The question is no longer which building holds the GPUs. It is who owns the fabric that makes the buildings irrelevant.