SK Telecom AI Cloud: Gigawatt-Scale NVIDIA DSX AI Factory
Status: Active / Early Site Civil Engineering & Technology Blueprint Integration (Tracking June 2026) | Location: Greater Seoul Regional Node, South Korea
Project Overview
The SK Telecom "AI Cloud" Factory is a groundbreaking, gigawatt-scale digital infrastructure initiative officially announced on June 7, 2026. Developed through a high-profile technology alliance between SK Group and NVIDIA, this mega-scale development represents South Korea's most aggressive push to establish absolute autonomy over its national sovereign artificial intelligence processing pipeline. Advancing into localized site engineering, foundation designs, and structural substation integration in mid-2026, the campus transitions away from traditional general-purpose hyperscale designs. Instead, the facility is purpose-built as an "AI Factory" optimized explicitly for accelerated GPU computing, multi-trillion parameter neural network training, and high-throughput agentic AI deployment. Built on the unified NVIDIA DSX full-stack reference architecture, the complex integrates advanced compute nodes directly with SK Group's next-generation high-bandwidth memory (HBM) and telco transport networks. This specialized blueprint maximizes token output while achieving the industry's lowest operational cost per megawatt
Quick Facts
| Field | Value |
| Project Name | SK Telecom AI Cloud: Gigawatt-Scale NVIDIA DSX AI Factory |
| Location | Greater Seoul Regional Node, South Korea |
| Status | Active / Early Site Civil Engineering & Technology Blueprint Integration (Tracking June 2026) |
| Commissioning | Phase 1 dedicated AI factory production block scheduled to come online in 2027 |
| Total IT Load | ~350.0 MW net critical operational compute capacity across early cluster blocks |
| Total Capacity | 1.00 Gigawatt (1,000 MW) gross ultimate master-planned sovereign AI cloud network |
| Tier Level | Tier IV Compliant (Fault-tolerant architecture with dual isolated distribution paths) |
| Project Type | Greenfield Gigawatt-Scale AI Cloud Data Center & Sovereign Agentic Hub |
City Profile
| Header | Details |
| City Name | Seoul |
| Population | ~9.4 million (Seoul city proper) |
| Urban Agglomeration | ~26 million (Seoul Capital Area metropolitan region) |
| City GDP | ~US$400 billion |
| Per Capita Income | ~US$42,000–45,000 |
| City Tier | Tier 1 / Alpha Global City |
| Key Strengths | Global financial center, technology and innovation hub, headquarters of major conglomerates (Samsung, LG, Hyundai), advanced infrastructure, world-class public transportation, strong education and research institutions, vibrant culture and tourism sector |
Companies Involved
| Header | Details |
| Developer / Operator | SK Telecom |
| Strategic Infrastructure Partner | NVIDIA Corporation (DSX Reference Architecture Blueprint & Full-Stack Silicon Alliance) |
| Construction Contractor | Leading South Korean Tier-1 Prime General Industrial EPC Consortia |
| MEP Engineering | Joint SK Telecom Core Engineering Group & NVIDIA Advanced Critical Systems Division |
| Network Connectivity | Ultra-high-speed, carrier-neutral private fiber optics utilizing custom NVIDIA quantum-packet switching blocks to interconnect thousands of accelerator nodes with zero network jitter |
| Power Infrastructure | Korea Electric Power Corporation (KEPCO) Regional Grid (Multi-feed high-voltage utility substations isolating heavy processing tiers from consumer load lines) |
Technical Specifications
| Header | Details |
| Power Capacity | Initial multi-megawatt rapid-deployment cluster blocks scaling into a long-term, multi-site 1.0 GW gross utility power footprint |
| UPS Redundancy | $2(N+1)$ block-redundant solid-state static UPS topologies engineered to absorb massive GPU transient power spikes and continuous cluster loads |
| Cooling System | 100% direct-to-chip closed-loop liquid cooling technology engineered to dissipate the massive thermal density of next-generation processing platforms |
| Connectivity | Minimum of 6 physically isolated, concrete-hardened network entries providing complete carrier route diversity and low-latency transport loops |
| PUE Target | 1.15 design-optimized annualized operating profile driven by severe-weather airflow optimization and software-managed power throttling |
| Energy Mix | High-efficiency mechanical designs paired with long-term clean-energy supply pacts and localized storage options to ensure stable 24/7 operations |
Milestones
| Header | Details |
| Announcement | 2026 - Q2 |
| Construction Start | 2026 - Q3 |
| Phase 1 Go-Live | 2027 - Q4 |
| Full Buildout | 2031 - Q2 |
Investment Details
| Header | Details |
| Total Investment | Forms the core piece of a multi-billion dollar long-term AI infrastructure allocation, with advanced processing silicon representing the majority of capital expenditures |
| Funding | Direct corporate capital allocations from SK Group, backed by global technology development funds and structured private infrastructure assets |