Summary
- SK Telecom has approved the creation of SK Hyper, a wholly owned AI data centre subsidiary.
- The company is targeting 5GW of capacity from 2029 and 15GW by 2035.
- The move shows how telecoms groups are reorganising around power-backed AI infrastructure rather than treating compute as an incremental network service.
SK Telecom has approved the creation of SK Hyper, a wholly owned subsidiary focused on AI data centre development, with a capacity target of 5GW from 2029 and 15GW by 2035.
The South Korean telecoms group plans to invest KRW750 billion by 2030 through the new platform. SK Hyper gives the company a dedicated structure for AI data centre expansion, separating the asset-heavy compute infrastructure programme from ordinary network or enterprise service development.
The target is ambitious even by current AI infrastructure standards. A 15GW capacity plan requires access to power, land, grid reinforcement, substations, cooling technology, construction partners, equipment supply, and long-term customers at a scale closer to national energy planning than traditional data centre rollout.
Telecoms assets meet AI density
Telecoms groups have long operated critical infrastructure. They understand networks, fibre routes, interconnection, resilience, regulated services, enterprise customers, and distributed operations. Those capabilities give them a natural route into parts of the data centre market, especially where connectivity and service continuity are central to customer demand.
AI data centres bring a heavier physical brief. Accelerator clusters concentrate power demand, create more difficult thermal loads, and require high-speed internal networks that differ from conventional enterprise hosting environments. Dense racks may need liquid cooling, revised electrical distribution, new commissioning practices, and faster hardware refresh planning.
By creating SK Hyper, SK Telecom is treating AI capacity as a platform business. That can make capital allocation cleaner, improve partnership options, and create a clearer structure for working with power providers, investors, customers, and equipment suppliers. It also concentrates accountability for delivery, rather than spreading responsibility across existing telecoms divisions.
The model reflects a wider industrial shift. AI infrastructure is no longer being organised only by hyperscalers and colocation developers. Telecoms operators, energy companies, sovereign funds, chip suppliers, and specialist infrastructure investors are all moving closer to the capacity layer.
Europe has room for similar structures
The SK Hyper plan is centred on South Korea, but the structure will be watched by European infrastructure groups. Europe has large telecoms operators with fibre networks, enterprise customer bases, data centre assets, and operational experience in regulated infrastructure. Some may look again at AI data centre platforms as demand for domestic compute capacity grows.
The barriers are physical and commercial. A telecoms brand does not create grid headroom, transformer supply, land, or cooling capacity. European markets also face sustainability reporting, planning controls, public scrutiny, and in some places direct political resistance to large data centre loads.
Dedicated platforms can still help. They allow investors and partners to assess the AI infrastructure business separately from legacy telecoms operations. They also make it easier to define the technical scope of the assets: whether they are designed for AI training, inference, enterprise colocation, sovereign cloud, edge workloads, or a mix of those uses.
SK Telecom’s 15GW target underlines the power intensity of the next phase. The main question is not whether AI demand exists, but which companies can turn that demand into powered, cooled, resilient, and financeable facilities. Dedicated data centre vehicles are likely to become more common as the infrastructure burden grows.
The European version may be smaller and more fragmented than SK Telecom’s plan, but the same direction is visible. AI capacity is being pulled into platform structures that combine power strategy, real estate, cooling, connectivity, customer contracts, and capital. Telecoms operators that can organise those pieces may regain a stronger role in digital infrastructure than they held during the first hyperscale buildout.
