SK Telecom announced plans to construct a 15-gigawatt AI data center across South Korea, marking a strategic shift in how nations compete for AI infrastructure dominance. The project represents a fundamental recognition that as artificial intelligence (AI) demand explodes globally, the real constraint is not computing chips but the power needed to run them. This move signals a broader industry awakening: energy control is becoming the new moat in AI infrastructure competition.

Why Is South Korea Suddenly Building Massive AI Data Centers?

SK Telecom's 15GW buildout is not happening in isolation. Global consulting firm McKinsey forecasts that worldwide data center demand will grow 19 to 22 percent annually, but supply will fail to keep pace, resulting in an estimated shortfall of about 15 gigawatts in the United States alone by 2030. This supply crunch is pushing major technology companies to expand data center investments beyond the U.S. to new regions.

South Korea has emerged as an attractive destination for this expansion. The country holds strong competitiveness in core AI components such as high-bandwidth memory (HBM), which is essential for training large language models (LLMs). Additionally, South Korea benefits from stable power supply conditions based on nuclear power and liquefied natural gas (LNG), along with gigawatt-class infrastructure operating capabilities accumulated through operating semiconductor fabrication plants.

SK Telecom plans to activate 5 gigawatts in stages starting in 2029, eventually reaching the full 15-gigawatt capacity. The first phase will focus on building a cluster of over 2 gigawatts across the southeastern region (Gyeongsang), using it as a base to attract AI infrastructure demand from global technology companies. An additional 1 gigawatt will be built in the southwestern region (Jeolla).

What Makes This Different From Previous Data Center Buildouts?

SK Telecom frames this project as Korea's third national infrastructure revolution, following the Gyeongbu Expressway (1968) and high-speed internet (1998). This framing reflects the strategic importance the company and government place on AI infrastructure as a national asset. SK Group is bringing together full-stack AI infrastructure capabilities across its affiliates, with SK Telecom taking the central role of leading design, construction, and operation.

The project is explicitly linked to the government's "AI G3" strategy, which aims to position South Korea as one of the world's three leading AI powers alongside the United States and China. Constructing a typical 1-gigawatt AI data center may require substantial project costs reaching approximately 70 trillion Korean won (roughly $50 billion USD). These costs are expected to be financed through the company's own investment, strategic partner investment, long-term customer contracts, and project financing.

How Is Energy Becoming the Real Competitive Advantage?

While SK Telecom focuses on South Korea, a parallel story is unfolding in the United States. SoftBank has launched SB Neo, a new cloud infrastructure subsidiary targeting 10 gigawatts of deployed AI data center capacity by around 2030. The timing and strategy reveal a critical insight: the competitive moat in AI infrastructure is shifting from GPU procurement to energy control.

SoftBank CEO Junichi Miyakawa explicitly framed the launch decision around power visibility rather than GPU pricing. He stated that the company's edge comes from "the ability to secure sources of power, mainly from gas-fired plants." SB Neo will draw on two major infrastructure projects: the Portsmouth AI Technology Campus in Pike County, Ohio, backed by $10 billion from SB Energy and expected to come online around 2028, and a 1.2-gigawatt data center in Milam County, Texas, built in partnership with OpenAI.

This energy-first strategy reflects a sobering reality in the neocloud market. McKinsey's analysis found that bare-metal-as-a-service (BMaaS), the model most cloud infrastructure companies use, carries gross margins of roughly 55 to 65 percent before depreciation but can fall to as low as 14 to 16 percent after labor, power, and hardware depreciation. The critical threshold is utilization: if it slips below 80 percent, returns flatline. With GPU rental prices for high-performance hardware falling roughly 50 percent from their 2023 peak, the pure procurement-and-resale model becomes progressively harder to defend.

What Are the Key Factors Driving This Infrastructure Race?

• Global Supply Shortage: McKinsey forecasts a 15-gigawatt shortfall in U.S. data center capacity by 2030, forcing companies to build infrastructure internationally.
• Power Stability and Cost: South Korea's nuclear and LNG-based power infrastructure, combined with SoftBank's gas-fired generation strategy, provides the stable, large-scale energy supply that AI data centers demand.
• Hyperscaler Competition: Major technology companies including Meta, OpenAI, and others are developing or expanding their own AI infrastructure, creating pressure for alternative providers to differentiate on energy control rather than GPU access alone.
• Regional Development Goals: Both SK Telecom and SoftBank are linking infrastructure projects to broader national and regional development objectives, securing government support and long-term financing.

How Are Companies Differentiating in the AI Infrastructure Market?

The market for AI cloud infrastructure generated over $25 billion in revenue in 2025, with revenues growing more than 200 percent year-over-year in the fourth quarter. However, this explosive growth masks structural fragility. On July 1, 2026, Bloomberg reported that Meta Platforms was developing "Meta Compute," an initiative to commercialize its surplus AI infrastructure. That single report sent CoreWeave (NASDAQ: CRWV) down 13.92 percent and Nebius Group (NASDAQ: NBIS) down 17.01 percent in a single session.

CoreWeave's financial results illustrate the tension. Revenue more than doubled year-over-year to $2.08 billion in the first quarter of 2026, and the revenue backlog reached $99.4 billion. However, net losses widened to $740 million, and total debt reached approximately $24.9 billion. Interest expense alone consumed roughly 46 percent of adjusted earnings before interest, taxes, depreciation, and amortization (EBITDA).

SoftBank's strategy to differentiate through energy control and proprietary software reflects this market reality. SB Neo will deploy Infrinia AI Cloud OS, a proprietary software stack that automates tasks from BIOS and RAID configuration through operating system, GPU drivers, networking, Kubernetes controllers, and storage. The stack runs on NVIDIA GB200 NVL72 hardware, which integrates 72 Blackwell GPUs across 18 hosts connected by NVLink, a high-speed interconnect that allows all-to-all GPU communication at 1.8 terabytes per second within the rack.

"This AI data center project is aimed at preemptively preparing the computing infrastructure that the global AI ecosystem needs," said Jung Jai-hun, President and CEO of SK Telecom. "We will work closely with the government, industry, and local communities to help Korea grow into Asia's core AI infrastructure hub."

Jung Jai-hun, President and CEO of SK Telecom

What Does This Mean for the Broader AI Infrastructure Landscape?

The convergence of SK Telecom's 15-gigawatt commitment and SoftBank's 10-gigawatt target signals a fundamental shift in how AI infrastructure competition will unfold. The early 2000s saw specialized compute startups fill gaps that Amazon, Microsoft, and Google could not address quickly enough, but nearly all were eventually acquired, sidelined, or forced into niche roles as hyperscalers caught up. The risk for today's pure-play GPU rental companies is similar, but the defense is now energy control and software differentiation rather than GPU access alone.

For enterprises and governments, this shift has practical implications. Companies seeking AI infrastructure capacity should evaluate not just GPU availability and pricing but the stability and cost of the underlying power supply. Nations without secure, large-scale power infrastructure will struggle to attract AI data center investment, regardless of other advantages. The infrastructure race is no longer about who can procure the most chips; it is about who can secure the most gigawatts.