SK Group Wants $350 Billion to Turn South Korea Into an AI Data Center Hub

Chung Suk-geun, chief executive of SK Hyper, told Bloomberg Television on October 23 that his company needs about $350 billion — roughly 463 trillion won, including servers and other equipment — to build out the first 5 gigawatts of a data center program that SK Group hopes will eventually reach 15 gigawatts of capacity in South Korea by 2035. That first phase alone is roughly five times the size of Korea's entire current data center market, which SK Hyper puts at around 1 gigawatt.

A bet that the world will rent Korean compute

The number is a target for a 'proposed buildout,' not capital that has been raised or committed, as one outlet covering the talks was careful to note. SK Hyper is in discussions with hyperscalers and financial investors and is aiming to land its first outside backers by late 2026 or early 2027, then keep raising over the following three to four years. SK Telecom, which owns SK Hyper outright, has already committed up to 750 billion won in stages through 2030, but that is a sliver of what the plan requires. Chung said the biggest obstacles are not power or land — he expects those to be manageable for the first phase — but locking in actual global demand and persuading outside capital to commit to it.

The strategy depends on exporting Korea's compute capacity rather than consuming it domestically. Chung argues the domestic market alone cannot support data centers at this scale, so SK Hyper is pitching itself to global hyperscalers as a base for AI inference and training workloads serving Japan, Taiwan, and the rest of Asia — plus domestic 'sovereign AI' demand from companies and governments wary of depending entirely on foreign-hosted models. The plan calls for a cluster exceeding 2 gigawatts in Korea's Yeongnam region, anchored by a facility already under development in Ulsan, with additional gigawatt-scale sites planned for the Chungcheong and southwestern regions. Phased operations are targeted to begin in 2029 for the first 5 gigawatts, with revenue potentially starting as early as next year from early-stage services, well before any of the flagship campuses are finished.

Splitting the balance sheet to de-risk the bet

SK Hyper is the newer and riskier of two vehicles SK Telecom has built to pursue this strategy. In August, SK Telecom split its subsidiary SK Broadband into a surviving fixed-line and media business and a new entity, SK Horizon, which took over already-operating data centers across sites including Seocho, Bundang, Gasan, Centum, Yangju, Pangyo, and Ilsan, along with submarine-cable infrastructure, with plans to expand that footprint to 318 megawatts of capacity. KKR and an IMM Investment-Stonebridge consortium immediately put about $2.2 billion of equity into SK Horizon, taking stakes of 29% and 20% respectively — private equity buying into cash-generating infrastructure that already has tenants and revenue. SK Hyper, by contrast, is the development-stage arm chasing gigawatt-scale projects that do not yet exist, which is why it is the one hunting for $350 billion rather than a few billion in growth equity.

That structural split — proven assets in one box for yield-seeking infrastructure funds, speculative build-out in another for strategic and growth capital — mirrors how other hyperscale operators have tried to separate stabilized cash flow from construction risk as AI infrastructure financing has grown more expensive and more scrutinized. Chung's own framing points to why: he told Bloomberg that competition among data center operators has shifted from simply securing GPUs to producing AI tokens at the lowest cost and highest efficiency, meaning SK Hyper has to prove it can build and run capacity more cheaply than rivals in Japan, Singapore, or the Gulf states also chasing hyperscaler tenants, not just that it can pour concrete.

The timing problem

SK Hyper's pitch is landing in an unusually skeptical market for exactly this kind of asset. Three days before Chung's interview, Elice Group — a separate South Korean AI infrastructure company — debuted on the KOSDAQ and immediately fell as much as 27% below its 90,500-won IPO price, despite having priced at the top of its marketed range and raised about $151 million. Reuters, reporting the debut, framed it as a test of investor appetite for AI and data center companies generally, and noted that planned listings tied to AirTrunk, Firmus, and SB Energy had already been pulled or paused amid an AI-led selloff and investors turning more selective about high leverage in the sector.

That backdrop makes SK Hyper's ask a useful stress test for the broader AI infrastructure financing market: Chung is asking global capital to commit tens of billions of dollars to speculative, multi-year Korean construction projects in the same week that public investors punished a comparable Korean AI infrastructure stock on sight. His answer, in effect, is to de-risk by sequencing — confirming demand from specific hyperscaler customers before each tranche of capital gets deployed, rather than committing the full $350 billion up front. Whether that is enough to attract first-close investors by his early-2027 target, with construction not finishing until 2029 and the full 15-gigawatt vision not due until 2035, is the question that will determine whether Korea becomes the regional AI compute hub SK Group is betting on, or another cautionary data point in a market that is suddenly less willing to fund gigawatts on faith.