Industry reporting indicates planned capacity across five locations could reach 5 GW, approximately two to three times Korea’s current domestic data center power consumption
Decision Focus
SK Group, the South Korean conglomerate that controls semiconductor memory leader SK hynix and gas distributor SK Gas, is moving to build multiple gigawatt-scale AI data centers — described internally as “AI Factories” — across South Korea and Japan. The trigger is competitive: Chairman Chey Tae-won has publicly stated that data center capacity should expand to at least 10 to 20 GW, framing power access as the defining industrial variable of the AI era.
The organizational move that accompanied that statement is the more immediate signal for energy operators. On June 1, SK appointed Jung Seung-il — former vice minister of trade and energy and former CEO of both Korea Gas Corporation and Korea Electric Power Corporation — as president of SK Inc. in a dedicated energy leadership role. When a conglomerate of this scale elevates a former national grid chief to run energy strategy, the implication for global power markets is worth reading carefully.
90-Second Brief
As the week closes, sK Group is preparing to announce a large-scale AI data center buildout anchored by the Ulsan project, where Amazon Web Services is reported as a major customer, with additional sites reportedly spanning multiple South Korean regions. Industry reporting indicates planned capacity across five locations could reach 5 GW, approximately two to three times Korea’s current domestic data center power consumption. Chairman Chey has publicly referenced GW-scale data centers operating at the power consumption level of a major city. Power, not compute, is the stated bottleneck.
What Is Really Happening?
The AI compute buildout is creating a structural shift in what it means to be a data center developer. SK’s strategy reveals the new competitive logic: to operate at GW scale, a company must control generation assets, not just purchase from them. SK’s portfolio — SK Innovation for power generation, SK Gas for LNG supply, and an equity stake in TerraPower’s small modular reactor program — is being redeployed as core AI infrastructure, connecting HBM production through SK hynix to data center construction to energy supply within a single conglomerate.
The IEA figure that global data center electricity consumption reached 415 terawatt-hours last year — roughly 1.5 percent of global power demand — already understates the trajectory. AI workloads run at materially higher power density than conventional cloud compute, and the infrastructure to support them takes years to commission. Korea’s specific constraint is that power demand concentrates in the Seoul metropolitan area while transmission capacity tightens. SK’s apparent decision to distribute sites across Ulsan and reportedly Busan, Sejong, and coastal Jeonnam and Jeonbuk locations is not simply geographic diversification — it is a direct response to grid topology.
Why It Matters for Global Heads of Data Center Energy
The SK buildout crystallizes a procurement model that will affect supply-side competition in several markets simultaneously. The group is explicitly planning to combine LNG generation, PPAs, battery storage, and SMRs into a hybrid power stack. That multi-source approach, now being validated at potential GW scale by a non-hyperscaler, will compete for the same generation assets, interconnection slots, and long-duration storage capacity that hyperscalers have been locking up. If SK executes across five sites, it becomes a material buyer in Korean power markets and — through its Japan ambitions — a new entrant in one of Asia’s most constrained grid environments.
The organizational implication is more immediate than the procurement one. SK’s appointment of a former KEPCO and Korea Gas CEO to a dedicated energy strategy role signals that the industry has moved past the point where energy procurement can be managed as a procurement function. At GW scale, it requires former regulators with utility-level relationships and grid authority access. That talent shift is already visible in hyperscaler energy teams globally. Operators who have not built equivalent institutional capacity face a widening gap in negotiating leverage with grid operators and government counterparts.
A secondary pressure point is SMR pipeline access. SK’s investment in TerraPower places it in a queue alongside Microsoft, Amazon, and other hyperscalers. As that queue lengthens, early SMR capacity may depend less on capital than on relationship priority established years in advance of commercial operations.
Forward View
If SK confirms a full five-site program, the first market consequence will be Korean grid interconnection queues. Grid operators managing constrained transmission infrastructure will face simultaneous GW-scale requests from a single conglomerate, likely prompting regulatory engagement and potentially accelerated grid investment or selective site approval. How Korea’s grid authority manages that pressure will reveal how governments handle conglomerate-scale AI power demand — a policy test that other jurisdictions will face in sequence.
In Japan, where SK has signaled GW-scale ambitions and where grid capacity is already strained by industrial demand recovery, the competitive signal for existing data center operators is direct: a vertically integrated rival with domestic LNG supply and SMR optionality is entering the market with a fundamentally different cost structure than a conventional colocation developer.
What Is Still Uncertain
The five-site plan is reported but not officially confirmed as of publication. Investment figures circulating in industry reports — approximately 70 trillion Korean won per data center — have not been verified against a formal SK announcement. The interconnection strategy for each site remains unknown: whether SK intends behind-the-meter generation, direct grid connection, or hybrid arrangements will determine how quickly capacity can be commissioned and whether grid congestion becomes an execution constraint rather than a planning assumption.
The customer pipeline beyond Ulsan is not publicly confirmed. At GW-scale economics, one anchor tenant is insufficient; the financial model requires multiple large-volume commitments before groundbreaking. How quickly SK can secure those agreements will govern the actual pace of buildout and the degree to which the five-site ambition translates into real interconnection demand within the next two to three years.
One Question for Your Team
If a vertically integrated conglomerate with its own LNG supply, SMR equity, and former grid regulators running energy strategy enters your markets — Korea, Japan, or adjacent Asian power corridors — which elements of your current PPA and interconnection strategy assume a competitive landscape that no longer exists?
Sources
- Co — SK’s Big Picture: Uniting an AI Ecosystem from Semiconductors to Energy (Link)
