SK Telecom lays out a 15GW AI data center buildout for Korea

SK Telecom said on Sunday that it will pursue a 15 gigawatt AI data center buildout across Korea, with five gigawatts scheduled to come online in stages from 2029 and the full footprint reaching 15GW by 2035. The plan links the Ulsan AI Data Center under construction to a national AI G3 strategy, positioning SKT as the architect of Korea's third national infrastructure revolution.

The announcement came on July 5 from SK Telecom's newsroom via the PR Newswire press release dated 08:34 CST, and a same-day Korea Times writeup on the buildout timeline. The carrier framed the project as a preemptive response to a global AI infrastructure shortfall. McKinsey, cited in the release, forecasts 19 to 22 percent annual growth in global data center demand and an estimated fifteen gigawatt shortfall in the United States alone by 2030. SKT's argument is that Korea, with stable nuclear and LNG power, GW-class semiconductor fab operating know-how, and the world's strongest HBM supply chain, can absorb part of that demand and become Asia's hub for hyperscaler AI data centers.

The 15GW plan and the five gigawatt first phase

The headline number is 15GW, but the operational buildout is staged. The Ulsan AI Data Center in the southeastern Gyeongsang region will scale to more than 2GW, with an additional 1GW in the southwestern Jeolla region, for a combined 5GW that the carrier says will open in stages starting in 2029. The remaining 10GW is scheduled to come online across the 2030s, reaching the full 15GW target by 2035. The release describes the 5GW phase as the base for attracting global big tech tenants into Korea, and the second phase as the broader regional build that positions Korea as an Asia-wide AI infrastructure hub.

The economics are sized accordingly. SKT says a typical 1GW-class AI data center may now cost approximately KRW 70 trillion, or roughly 50 billion US dollars, once high-performance compute and rising memory prices are factored in. The carrier says the 15GW capex will be financed through a mix of company capital, strategic partner investment, long-term customer contracts, and project financing, rather than from a single balance sheet. The release names a June 8, 2026 partnership with NVIDIA, announced in the same newsroom, as one of the strategic anchors, and a separate 2025 SK Group and AWS cloud computing agreement as a second.

SKT framed the 5GW first phase as an anchor-tenants-first plan. The carrier will select sites, secure power supply, and pre-commit to hyperscaler customers before breaking ground, in coordination with the government's AI G3 strategy and the country's balanced regional development agenda. The first AI Factory, a next-generation AI data center that the carrier described in earlier releases, is scheduled to begin operations in 2027 and to scale to GW class in the years that follow.

SK Group's full-stack AI infrastructure bet

The release leans on SK Group's full-stack AI infrastructure capabilities as the structural reason the project is feasible. The carrier notes that AI data center infrastructure depends on three core elements: semiconductors, energy solutions, and data center construction and operation. SK Group already holds all three across its affiliates, and the release frames SKT as the architect that brings them together. The language echoes SKT President and CEO Jung Jai-hun's AI infrastructure roadmap from the SK AI Summit 2025 last November, where he argued that the carrier would lead the evolution of AI infrastructure as the nation's leading AI operator.

For the broader AI infrastructure market, the most consequential part of the announcement is the HBM and memory angle. Korea is the only country with a domestic HBM supply chain at scale, anchored by SK hynix, and the release leans on that fact as a locational advantage. The buildout will pull on Korean HBM, Korean memory, and Korean power generation simultaneously, and SKT is signaling that the cost curve for AI data center memory is now a national industrial variable rather than a corporate procurement line. A separate recent article on private capital into AI data center buildouts, including the CPP and EQT $2.4B commitment to EdgeConneX, has tracked the same convergence from a different angle.

The buildout is also a play for the energy and grid layer. Korea's stable nuclear and LNG power supply is a recurring point in the release, and the 15GW target implies multi-gigawatt new power demand on top of the country's existing grid. The 5GW first phase alone, if it materializes on the 2029 timeline, would put SKT's data center demand in the same league as a major industrial baseload, and the second phase to 15GW would push that further. The release does not name a power purchase agreement or a specific energy supplier, but it positions Korea's power stability as a competitive differentiator against hyperscaler builds in markets with grid bottlenecks. The build pattern, end to end, is a case study in the choices laid out in the AI Infrastructure in 2026 resource page: chips, cloud, and capacity under one national roof.

The signals to watch in the next 18 months

The most concrete next milestones are not in the second half of 2026. The first AI Factory is scheduled to begin operations in 2027, and the first 5GW of the broader buildout is staged to open in stages from 2029. Between now and then, the carrier has to land anchor tenants, finalize power purchase arrangements, and select sites in the Jeolla and additional Gyeongsang regions. The release says the company will pursue AI data center construction in connection with the government's balanced regional development tasks and strategic supply plans, which is a signal that site selection is partly a public-private negotiation rather than a private capital decision alone.

The procurement signal is also significant. The carrier named NVIDIA as a June 2026 partner for the AI Factory layer, and the release leans on the existing AWS relationship for cloud infrastructure. The implication is that the project is a multi-vendor stack, with SKT providing the site, the power, and the carrier-grade data center operations; NVIDIA providing the GPU and system layer; and AWS providing the public cloud adjacency. That is the same pattern that has emerged in US AI data center buildouts, and the SKT release is one of the clearest statements of the pattern from a non-US carrier.

The longer-term claim is more contested. SKT argues that AI data centers will be Korea's third national infrastructure, on the same plane as the Gyeongbu Expressway in 1968 and the high-speed internet buildout in 1998. The 15GW target is large enough that, if delivered, it would make Korea a tier-one AI infrastructure market by capacity. If the project slips on the 2029 first phase, the 15GW headline number is more aspirational than operational. The release is short on site-level engineering detail, and the next eighteen months of anchor-tenant and power-contract disclosures will be the test of whether the 15GW target is a buildout or a statement. The Korean AI infrastructure market is not large enough to absorb fifteen gigawatts of AI data center demand on its own, and the cross-border customer signal will be the load-bearing fact to watch.