ABI Research says operators must balance CapEx, sovereignty, and innovation as they build AI-ready data centers for GPU-as-a-Service
Telecom operators are entering what ABI Research Senior Analyst Larbi Belkhit calls “the very early stages of an AI infrastructure race,” as they work to monetize artificial intelligence through GPU-as-a-Service (GPUaaS) offerings. According to the firm’s latest forecasts, telcos could generate more than $21 billion in revenue by 2030 from this emerging model.
In the most recent episode of Unmuted, Belkhit explained that GPUaaS will be “foundational” to the telco ROI monetization strategy for AI infrastructure. He further noted that most current revenues — less than $1 billion — come from AI inference workloads. He added that operators are “upgrading their existing data center portfolios and even building new AI-optimized data centers, which are sometimes called AI factories.”
The “best of both worlds”
While hyperscalers have scale and cloud expertise, Belkhit said telcos hold a unique position. “Telcos can be the best of both worlds between the hyperscalers and alternative cloud providers in their value proposition,” he said. Their proximity to users and sovereign infrastructure gives them an advantage “where AI sovereignty matters just as much as compute.”
That sovereignty theme is strongest in Europe, the Middle East, and Asia-Pacific, which Belkhit identified as leading regions. “In Asia-Pacific, telcos are partnering with governments for contracts, likely driven by the fact that they need sovereign infrastructure to develop national language LLMs,” he said. “In Europe, we’re seeing something slightly different where the demand is coming from the manufacturing sector and helping the industry with industrial automation.”
Deutsche Telekom and SK Telecom set the pace
A standout example comes from Europe, where German carrier Deutsche Telekom is making a major push into AI infrastructure. “You may have seen the Deutsche Telekom announcement that came out … where they’re planning to start their AI Gigafactory, as they like to call it, which I believe will comprise 100,000 GPUs in 2026 in Germany,” Belkhit said.
Specifically, DT plans to begin construction in 2026 on a large-scale AI data center as part of its expansion into AI-driven services. The facility, which is being developed in partnership with Nvidia and investment firm Brookfield, is intended to serve as the world’s first industrial AI cloud for European manufacturers.
In South Korea, meanwhile, SK Telecom recently unveiled a GPU-as-a-Service platform built on NVIDIA B200 GPUs. “It indicates that there’s a lot of demand for the compute that SK Telecom is offering in South Korea,” Belkhit said. “Strategically, SK Telecom wants to use the data center they’ve developed for this offering and make it a core hub of the growth of Korea’s AI industry.”
Initiatives like these illustrate how telcos across regions are transforming into AI infrastructure providers, pairing sovereign data strategies with powerful, GPU-optimized facilities designed to fuel the next wave of digital intelligence.
Balancing capex and control
Still, the opportunity is paired with heavy risk. “The continuous, consistent investment cycle in the cloud space is very different … every year there’s a refresh of the best next GPU,” he warned. Telcos must also “avoid both under- and overestimating the short- and long-term revenue opportunity,” while developing their own orchestration software to reduce dependence on hyperscalers and “neoclouds.”
As Belkhit summed up: “Developing in-house expertise in terms of software development and orchestration platforms is absolutely key … and this shift will be crucial as they move beyond just providing the compute or GPU as a service to actually offering AI as a service to enterprises as well.”