# Edge AI May Give Telecoms a Seat at the Table — retained record > Retained news record from news:datacenterknowledge. Canonical: https://highsignal.app/data/records/cf1211b761037fa6 [← sources](https://highsignal.app/data) source: [news](https://highsignal.app/data/news) (news:datacenterknowledge) published 2026-10-30 09:00 ingested 2026-10-31 02:54 [original source ↗](https://www.datacenterknowledge.com/edge-data-centers/edge-ai-may-give-telecoms-a-seat-at-the-table) ## retained text ``` Edge AI May Give Telecoms a Seat at the Table Edge AI may give telecoms a bigger AI role, with metro sites, not towers, emerging as the clearest near-term opportunity. October 30, 2026 While the telecommunications industry has spent years absorbed in “the G’s” – 4G, 5G, and now 6G – the hyperscalers have been rewriting the broader narrative of computing. Telcos have fed every stage of that evolution through their pipes and platforms, yet by some measures they still end up cast as plumbers. More recently, the fanfare around generative AI has further eclipsed telecom’s own next act. Against the spectacle of large language models, GPU clusters, and hyperscale AI buildouts, 6G can look less like an imminent frontier than a contender stranded in an early round. Yet the next phase of AI may reopen the question of where telecoms fit in the digital infrastructure stack. As generative AI and agentic AI move into daily use, new opportunities are emerging through edge AI inference. Most of that work will still be done in hyperscale data centers. But latency, data gravity, and cost are likely to push parts of workloads toward the edge. That is where some observers see a new convergence taking shape – one that could place telecoms closer to the center of AI inference infrastructure. Nvidia has already signaled part of that opportunity. The company’s moves at the AI edge have been quieter than the daily procession of hyperscale megadeals, but the roots of convergence are visible in projects and partnerships involving AT&T, Charter Communications, Cisco, Nokia, SoftBank, Telefónica, T-Mobile, and others. Nvidia has placed particular emphasis on edge AI for factory and robotics use cases – often called “physical AI” – closely tied to AI-RAN. The GPU infrastructure connected to AI-RAN is also expected to help telecom operators improve their own efficiency through greater automation of network operations. Still, this emerging edge is by no means a monolithic thing. It is nuanced, layered, and still taking shape. How it develops is now a central question in both data center and telecom circles. The Physical Hurdles of Edge AI The edge holds real potential to offload some of the work now concentrated in hyperscaler facilities and to serve applications located far beyond regional data centers more efficiently. In practice, however, deploying AI beyond large regional facilities introduces a new set of infrastructure problems for telecom operators. David Duggal, CEO of EnterpriseWeb, said the economics and engineering can become difficult very quickly. EnterpriseWeb, which develops an orchestration platform for mapping and automating agentic AI workflows, has participated in proof-of-concept projects with Intel, Dell, and others focused on applying generative AI and automation at the edge. As Duggal explained, one of the first challenges is definitional. The “telco edge” is not the same as the “customer edge.” One refers to radio towers, base stations, and network-adjacent facilities. The other may sit inside an enterprise campus, factory, warehouse, or other operational site. That distinction matters because deploying high-performance AI inference directly at a tower or cell site is far more difficult than placing it in a better-provisioned metro or enterprise facility. “Power and thermal concerns make AI on a tower or cell site a Catch-22 situation,” Duggal said. “That GPU requires a lot more power … and more power means more heat.” More heat, in turn, means more supporting infrastructure. GPUs are expensive, dense, and power-intensive, making it difficult to justify equipping every radio tower with advanced inference hardware. As Duggal and others point out, some emerging radio access network models instead push heavy AI inference workloads back toward the data center or another better-served aggregation layer. The Many Edges of AI Today’s AI inference buildout spans several tiers of infrastructure. Since the debut of ChatGPT, hyperscale clouds have handled the most intensive generative AI training and development workloads, often running across thousands of compute nodes per region. Beyond that central cloud lies a more varied edge continuum: - Metro edge: Central offices, regional hubs, and other aggregation facilities operated by large telcos or neutral hosts. These sites can support hundreds of nodes, consolidate data flows from both the cloud and the far edge, and offer a more practical setting for higher-density GPU deployments. - Far edge: Cell towers, base stations, and local on-premises cabinets serving highly time-sensitive tasks, including machine vision and localized telemetry. These environments are the natural focus of AI-RAN and O-RAN conversations, but they also face the toughest power, cooling, and space constraints. - Endpoint AI: Devices such as smartphones, vehicles, industrial controllers, telematics units, and IoT sensors. These systems handle some processing locally and pass data or decisions upstream to the far edge, metro edge, or cloud as needed. The point is that “the edge” is not one place. It is a gradient of infrastructure, with different economics and different technical realities at each tier. New Life for Old Switch Houses? US telecom operators also hold a less discussed asset: real estate. Central offices, network huts, and legacy switching facilities may not look like the future of AI infrastructure, but many could become useful sites for edge inference, particularly as community resistance complicates large greenfield data center development. That possibility is also showing up in standards work. Refurbishing brownfield facilities is only one part of a broader push at the Telecommunications Industry Association (TIA) to define more unified practices for AI data center infrastructure. In that sense, the telecom opportunity is not just about unused space. It is about convergence among telcos, enterprises, hyperscalers, and data center operators around more modular and repeatable deployment models. “The notion of edge data centers, from our perspective, is a fully contained data center that could be created in a factory and dropped in at a location as an adjunct to an existing facility,” said Mike Peterson, lead at TIA’s Data Center Program. That kind of standardization matters because AI infrastructure is changing physical design assumptions across the board. For telecommunications, the GPU era introduces a modern twist on a classic computing constraint: the tyranny of wires. High-density GPU clusters place extraordinary demand on cabling, interconnect design, power distribution, and cooling. “When you are looking at some of the designs for the newer AI data centers, and you see the amount of cabling that goes into an Nvidia pod, for example, it’s eye-opening,” Peterson said. As AI moves further to the edge, whether through brownfield retrofits or new modular builds, those physical design questions become more pressing. TIA has been addressing some of them through AI-focused work tied to ANSI/TIA-942, the data center infrastructure standard. The supply chain adds another layer of risk. AI buildouts now pull in a diverse set of players, including contractors, component suppliers, integrators, and service providers. On Oct. 14, 2026, TIA released the first draft of its DCE 9000 quality management standard for supply chain componentry, a sign of how seriously the industry is treating consistency and specification control as telecom-led edge deployments expand. Beyond the Pipe: Can Telcos Move Up the Stack? Telecoms now face a familiar question in a new form: will the old “plumber” stereotype stick, or can operators move higher up the value chain? The answer remains open, according to Luis Fernandes, senior research manager for cloud and data centers at IDC. Telecoms may yet reinvent themselves – for example, as AI brokers serving enterprises that want to rent compute capabilities, or as AI-as-a-Service providers in their own right. A larger question is whether AI-RAN helps move that evolution forward. “AI-RAN can be of massive importance in the years to come for telcos,” Fernandes told Data Center Knowledge. But he cautions that it won’t yield an immediate payoff. “Qualified resources are in short supply right now, and budgets are being allocated to flashier projects, leaving AI-RAN as a side project for a few,” he said. Even so, Fernandes argues that AI-RAN has near-term value for telecom operators by improving network performance, reducing costs, and driving operational efficiency. In other words, the business case for AI-RAN does not depend entirely on monetizing third-party edge inference. He also cautions against overstating latency as the universal reason for pushing AI outward. In many edge discussions, latency is treated as the headline justification. Fernandes said IDC’s work suggests the picture is more nuanced. IDC collaborated with the European Commission on “The Edge Observatory for the Digital Decade,” a project examining edge workloads and deployments, including AI. Fernandes said the research found that latency mattered in some cases, but not in most. “What we found was that, although in very specific cases and areas, latency was an issue, for the most part, it did not play a major role at the edge,” he said. “Workload type, however, was a determining factor.” That distinction has practical consequences. Real-time workloads need instant responses shaped by local context and timing. Analytical workloads often do not. In those cases, data may be staged or preprocessed at the edge before moving back to a central data center or cloud environment. For enterprise buyers and operators, the key is not simply to ask how low latency can go, but which workloads truly benefit from being placed closer to the point of action. The CDN Case for Edge AI Content delivery networks fit naturally into the race toward AI inference. Their opportunity lies in smarter routing for real-time traffic and in distributed processing for applications such as computer vision, recommendation engines, and context-aware bots. To orchestrate such workloads across multiple edge locations, Akamai is deploying GPU-equipped edge nodes directly inside telecom central offices. At the same time, it is working with far-edge elements, particularly in physical AI use cases. Physical AI remains among today’s most latency-sensitive use cases, according to Jon Alexander, senior vice president of products for Akamai’s Cloud Technology Group. “You can’t wait 10 seconds to send the data all the way around to the other side of the world,” Alexander said. “Everything [would go] to a halt inside a factory.” For workloads that do not require instant on-device intervention, however, Alexander said metro-level edge infrastructure may offer the best balance of performance and cost. If GPU-enabled metro nodes can deliver response times in the 5- to 10-millisecond range, that may be sufficient for a broad swath of commercial AI workloads. “That is great for pretty much every [compute] experience that exists today. It means both good utilization and good economics,” he said. “The network is important, and economics are an issue.” ``` document fetched 2026-10-31 02:54 · canonical https://datacenterknowledge.com/edge-data-centers/edge-ai-may-give-telecoms-a-seat-at-the-table