SK Telecom targets 15GW of AI data centers to export compute

Fifteen gigawatts of AI data‑center capacity is an audacious number—on par with the combined AI buildouts some US hyperscalers are contemplating. SK Telecom is signalling that it wants Korea to jump from AI consumer to AI compute exporter, selling both colocation and GPU cloud capacity to global customers. The plan hinges on securing land, power, anchor tenants and capital over a decade, but if even a fraction materialises it would materially expand the global pool of frontier‑grade compute. ([v.daum.net](https://v.daum.net/v/20260629171202089))

This is also a bet that big tech demand for external capacity will remain insatiable. By courting partners like AWS and Nvidia, SKT is positioning itself as a regional infrastructure ally to US platforms that are compute‑constrained at home. For the AI‑model ecosystem, more geographically diversified infrastructure could mitigate some political and grid‑related risks, but it also concentrates bargaining power in a small club of telcos and hyperscalers with access to cheap energy and permissive regulation.

From an AGI‑race perspective, plans of this scale underscore how quickly the hardware footprint is catching up with ambitious model roadmaps. If realised, they would help ensure that compute, not capital, remains the primary limiting factor—pushing timelines forward as engineering and safety challenges become the dominant brakes.