The debate about sovereign AI often focuses on models and software. Underneath it sits a more physical question: where will national AI compute actually run, and who owns the buildings, power connections and fibre that support it? In October 2026, telecom operators gave a clear answer. Taiwan’s Chief Telecom broke ground on a new AI data centre. Bell Canada accelerated an AI data centre near Regina with backing from the Saskatchewan government. Globe Telecom set out plans to invest at least $1 billion in data centre capacity in the Philippines. Elisa reported around €100 million in data centre connectivity orders in Finland. And SK Telecom’s SK Hyper unit began promoting South Korea as a destination for global AI model training.
These moves describe operators acting as landlords for sovereign AI: providing the land, power, connectivity and trusted local operations that national AI capacity needs, whether or not they also sell the AI services that run on top. It is a role that plays directly to the assets telcos have spent decades accumulating.
Why Telcos Own the Physical Layer of Sovereign AI Infrastructure
Sovereign AI infrastructure has to meet requirements that favour established national operators. Data must stay within borders, facilities must be operated by trusted entities, and capacity must be connected to enterprises and public bodies across the country. Telcos bring several advantages to that brief:
- Real estate: data centres, exchanges and central offices in or near major cities, often with land available for expansion.
- Power connections: existing grid connections and relationships with utilities, increasingly the scarcest input for AI capacity.
- Fibre: national backbone and metro networks that connect AI facilities to customers, clouds and each other.
- Distributed footprint: regional and edge sites that can host inference close to users and devices.
- Regulatory trust: long experience operating critical national infrastructure under government oversight.
Hyperscalers and specialist AI cloud providers remain central to AI infrastructure. But where governments and regulated industries want domestically owned and operated capacity, telcos are natural partners and, increasingly, owners.
Four Sovereign AI Infrastructure Models for Telcos
Operators are taking different routes into the physical layer, depending on their balance sheets, assets and appetite for risk.
| Model | What the operator provides | Risk profile | Examples |
| Build and own | AI data centres, power and cooling, sometimes GPUs | Highest capital and utilisation risk | Chief Telecom, Globe Telecom, Bell Canada, TELUS |
| Connect | Fibre and high-capacity links to AI facilities | Lower risk, contracted revenue | Elisa, Lumen, Zayo |
| Distribute | Edge sites with GPUs for local, sovereign inference | Moderate, depends on enterprise demand | TIM, Airtel |
| Monetise | Infrastructure partly sold or listed to release capital | Shares risk with investors | PLDT Vitro data centre REIT (planned for 2027) |
Building National Capacity: Sovereign AI Infrastructure Projects
The most direct model is building and owning AI data centres. Chief Telecom is investing around NT$3.3 billion (about $102 million) in a new AI data centre at the Central Taiwan Science Park, scheduled to open in the second half of 2028 to serve enterprise AI and high-performance computing demand. Globe Telecom’s planned investment of at least $1 billion aims to attract AI workloads and hyperscale tenants as the Philippines pursues a regional AI hub role, supported by a national AI plan targeting a thirty-fold increase in AI data centre capacity.
In Canada, Bell’s AI data centre near Regina has moved quickly with provincial government support, adding regional AI capacity and local processing. TELUS has gone furthest in that market, reporting that its first sovereign AI factory in Rimouski, Quebec, is fully sold out and planning a British Columbia cluster of more than 60,000 GPUs and 150 MW of AI compute by 2032. In South Korea, SK Telecom’s SK Hyper unit is positioning the country’s infrastructure as a location for international AI training workloads.
Operator-linked groups are building at scale elsewhere too. NTT DATA committed $1.5 billion to AI-ready sovereign data centres in India, and Ooredoo Group and Indosat’s Zankore platform targets 1 GW of AI capacity in Indonesia, as covered in our September Digital Infrastructure Insights.
Connectivity as the Lower-Risk Sovereign AI Infrastructure Play
Not every operator wants to own GPUs. Many see stronger, steadier returns in connecting AI facilities. Elisa’s €100 million in cumulative data centre connectivity orders in the first nine months of 2026 came alongside an expanded partnership with Google covering facilities in Hamina, Kajaani, Muhos and Vaala. Finland’s cool climate and clean power make it attractive for large data centres, and Elisa’s role is to provide the high-capacity links that tie them together and to the rest of Europe.
In the United States, Lumen is combining Eminence Grey’s sovereign AI platform with its physical infrastructure and programmable network to connect edge and cloud AI environments. Zayo is planning more than 8,000 miles of new fibre for AI data centres. Connectivity contracts are typically long-term and less exposed to questions about GPU utilisation, making this a lower-risk way for operators to participate in the sovereign AI buildout.
The Edge Layer of Sovereign AI Infrastructure
Sovereign AI will not live only in large campuses. Inference for public services, healthcare, manufacturing and finance often needs to run close to users and data. TIM is deploying edge cloud nodes equipped with NVIDIA GPUs across Italy as part of the IPCEI CIS programme, combining them with its central cloud to offer sovereign AI with data localisation and access governance. Airtel presented multi-availability-zone sovereign cloud infrastructure at India Mobile Congress to help enterprises meet data residency requirements, alongside AI-driven energy efficiency in its Nxtra data centres.
This distributed layer is where telcos are hardest to replicate. Few other organisations operate hundreds or thousands of connected, powered sites across a country, with the staff and processes to run them.
Landlord Economics for Sovereign AI Infrastructure
Acting as a landlord changes the financial profile of AI investment. Instead of carrying the full risk of filling GPU clusters, operators can earn lease, power and connectivity revenue from tenants that bring their own compute, or anchor projects with government and hyperscale commitments before building. Financing structures are evolving to match. PLDT has moved the listing of its Vitro data centre REIT to 2027, India’s market is using structured debt and sale-leasebacks to fund AI growth, and government backing is helping projects such as Bell’s move faster.
The landlord model also lets operators separate infrastructure from services, a pattern already familiar from towers and fibre. Data centre assets can be partly sold to infrastructure investors to release capital, while the operator retains the customer relationship and the AI services layered on top. Operators with large digital platforms, such as Jio, discussed in our analysis of the Jio Platforms IPO, show how infrastructure and services can be valued together or apart.
Challenges for Telcos as Sovereign AI Landlords
The role brings real constraints. Power is the biggest. Grid connections for large AI facilities can take years, and operators competing with hyperscalers for the same capacity need strong relationships with utilities and regulators. Community acceptance is another factor. Governments and local authorities from New York and Ireland to Denmark and Thailand are introducing new rules on construction, energy and water use, and Indonesia and Thailand have paused some major data centre developments for regulatory review.
There is also a density challenge. Average rack power has risen to around 26 kW, and AI systems are moving towards 250 kW racks with liquid cooling. Most legacy exchanges and central offices were designed for far lower densities, so converting them for AI requires significant electrical and cooling upgrades. Operators will need to choose carefully which sites to upgrade, which to redevelop and which to reserve for lighter edge inference.
What to Watch in Sovereign AI Infrastructure
- New government-backed AI data centre projects anchored by national operators, particularly in Asia-Pacific, the Gulf and Canada.
- Telco data centre REIT listings and infrastructure partnerships as operators seek capital for AI builds.
- Growth in data centre connectivity revenue reported by operators such as Elisa.
- Upgrades of central offices and edge sites to support GPU-based inference.
- Power and planning decisions that determine which sovereign projects can proceed on schedule.
Sovereign AI will need a physical home, and telcos hold many of the keys to it. Operators that combine their land, power, fibre and trusted status with disciplined financing can become essential landlords of national AI capacity, earning steady returns while governments and enterprises build the AI services on top.
For more insights on advanced connectivity, explore the TeckNexus Monthly Insights, and to share your perspective with our audience and generate qualified leads, partner with TeckNexus through an Intelligence Program.
