SK Telecom's Trillion-Won AI Bet: From Telco to AI-Native Operator
At MWC 2026 in Barcelona, SK Telecom President Jeong Jae-heon drew a hard line: "If we miss the golden time to adapt to AI, we die." He announced a company-wide AI transformation and an infrastructure investment plan measured in the trillions of won. The goal is blunt-become an AI-native enterprise, monetize infrastructure, and export sovereign AI.
What SKT Is Changing Right Now
- Rebuilding the integrated computing backbone for AI-first operations across sales, line management, and billing.
- Deploying hyper-personalized rate plans and memberships powered by real-time data.
- Applying zero-trust security to every system, with network segmentation and AI-driven, integrated security control.
Jeong's diagnosis is clear: legacy IT and networks weren't built for AI data flows or service delivery. The fix isn't incremental. It's a ground-up re-architecture.
Scale: Trillions in Capex and a 1T-Parameter Model
While SKT didn't disclose an exact figure, Jeong said the investment "will exceed trillions of won." The company plans to scale its sovereign AI foundation model, A.X K1, from 519 billion parameters to the 1 trillion range.
He stressed this isn't a head-to-head with Big Tech. It's a sovereign AI play-meeting demand from countries that want models built and operated on their terms, with export potential for SKT.
Manufacturing AI with SK Hynix and Nvidia
SKT is co-developing manufacturing AI solutions using Nvidia's models while SK Hynix maps the most valuable use cases. The package targets real-time process analytics to reduce defects and maximize equipment efficiency. The intent is straightforward: compress cycle times, lift yield, and tighten cost per unit.
Infrastructure Build: 1+ GW of AI Data Centers
To support training and inference at scale, SKT will build ultra-large AI data centers totaling more than 1 gigawatt nationwide. The ambition: grow into Asia's largest AI hub. This raises familiar constraints-power, cooling, land, and grid interconnects-and forces a long view on PUE, supply chain, and siting.
Why This Matters to Executives
- Telcos are repositioning as AI infrastructure providers, not just connectivity utilities.
- New monetization vectors: sovereign AI exports, managed AI services, and usage-based AI plans.
- Data gravity shifts to operators with privileged access to network signals and customer context.
- Security posture moves to zero-trust by default-table stakes for AI-era services.
- Partnerships with chipmakers and manufacturers compress time-to-impact in real industries.
Risks and Constraints to Watch
- Capex intensity vs. cash flow and interest rates; avoiding stranded assets.
- Power availability, grid timelines, and sustainability targets.
- Scarce AI talent for platform engineering, MLOps, and security.
- Model governance, evaluation, and IP/licensing clarity for sovereign deployments.
- Unit economics of inference (GPU utilization, scheduling, and QoS).
- Vendor lock-in risk across model stacks, accelerators, and tooling.
Moves to Put on Your Roadmap
- Audit legacy systems blocking data access; set a staged migration to AI-first architectures.
- Stand up a zero-trust program with clear milestones and red-team validation.
- Modernize your data pipeline: event streaming, feature stores, and governance by design.
- Co-develop with OEMs on high-value use cases (yield, OEE, predictive maintenance).
- Define a sovereign AI offering: residency, compliance, language, and sector tuning.
- Model TCO scenarios for training and inference; secure long-dated power and capacity.
- Plan for edge inference near networks and factories to cut latency and cost.
For strategic depth and case-driven playbooks, see AI for Executives & Strategy. If you own infrastructure and security outcomes, the AI Learning Path for CIOs maps directly to re-architecting at SKT's scale.
Signals from the Show Floor
Jeong met Samsung's Roh Tae-moon at MWC. The Galaxy S26 Ultra's privacy display-"hide specific apps or areas"-drew a sharp comment from Jeong: "Screen protector companies must be ruined." After stops at Meta and Xiaomi, his read: device integration is accelerating, and Xiaomi showed tight links across cars and telecom.
Context and References
- Event: MWC Barcelona
- Security approach: NIST Zero Trust Architecture (SP 800-207)
Jeong's message to operators and industrial CEOs is hard to misread: miss the AI transition window, and the market moves without you. The counter move is equally clear-commit capital, refactor for AI, and build products countries will buy.
Your membership also unlocks:
