South Korea’s largest mobile operator, SK Telecom, says its AI-focused GPU cluster platform “Haein” is emerging as a central pillar of the country’s sovereign AI push, as governments and enterprises worldwide race to secure domestic compute capacity for large model training.
The company this week marked six months since the launch of Haein, a large-scale GPU cluster built to deliver GPU-as-a-Service (GPUaaS) for foundation model development and high-intensity AI workloads. Introduced in August 2025, the platform is now being used as part of Korea’s Sovereign AI Foundation Model Project, a national initiative aimed at building globally competitive domestic AI models and reducing reliance on foreign hyperscaler infrastructure.
Haein reflects a broader global shift: telecom operators are increasingly positioning themselves not just as connectivity providers, but as AI infrastructure players offering compute, orchestration and platform services.
AI infrastructure moves from cloud to carrier
AI model training has created unprecedented demand for high-performance compute clusters, low-latency interconnects and tightly managed GPU resources. Until now, most of that capacity has been concentrated inside hyperscale cloud providers. SK Telecom’s strategy with Haein is to create a carrier-grade alternative optimized for national and enterprise AI workloads.
The Haein cluster is designed specifically for large-scale AI training and experimentation, supporting multi-tenant usage and distributed model development. SKT says the platform has moved from initial rollout to stable operations, enabling faster model training cycles and broader collaboration between research, enterprise and government teams.
The naming of Haein — drawn from a historic Korean temple associated with long-term knowledge preservation — signals SKT’s positioning of the platform as foundational digital infrastructure rather than a short-term compute offering.
Full-stack differentiation versus raw GPU capacity
While many AI infrastructure projects focus primarily on GPU scale, SKT is emphasizing software integration as its main differentiator.
Haein combines three integrated software layers: Petasus AI Cloud for AI-optimized data center virtualization, AI Cloud Manager for large-scale training job scheduling and multi-tenant management, and a GPUaaS Service Orchestrator that provides real-time visibility and control across GPU, network and storage resources.
Petasus AI Cloud supports heterogeneous compute environments and virtualizes high-speed interconnect technologies such as NVLink, InfiniBand and RoCEv2. According to the company, this allows logical isolation of GPU and network resources for each tenant in under an hour — significantly faster than physically segmenting clusters, which can take days or weeks.
The orchestration and observability layers are designed to address a growing operational challenge in AI clusters: utilization efficiency. As GPU infrastructure becomes more expensive and supply-constrained, software-driven scheduling, monitoring and anomaly detection are becoming as important as raw hardware scale.
Sovereign AI and national compute capacity
Haein’s selection for Korea’s Sovereign AI Foundation Model Project highlights how national governments are increasingly treating AI compute as strategic infrastructure.
Countries across Asia, Europe and the Middle East are now funding sovereign AI stacks — combining local data, domestic models and in-country compute — to reduce exposure to geopolitical risk, export controls and platform dependency. GPUaaS models operated by telecom and national cloud providers are emerging as one way to deliver shared, policy-aligned compute capacity.
By anchoring Haein inside a national AI program, SK Telecom gains both anchor demand and policy alignment, while positioning itself as a long-term AI platform provider rather than only a network operator.
Telecom operators expand up the AI stack
SK Telecom’s move mirrors a wider operator trend: telecom groups are extending into AI clouds, edge AI platforms and GPU clusters to capture value from AI-driven traffic growth and enterprise transformation.
As AI workloads drive heavier east–west data flows between clouds, data centers and edge locations, operators with dense fiber, data center and interconnect assets see an opportunity to bundle connectivity with compute and orchestration.
The competitive question will be whether telco-led AI infrastructure platforms can match hyperscaler ecosystems in tooling, developer adoption and pace of innovation — or whether they will primarily serve regulated, sovereign and specialized enterprise segments.
With Haein now tied to Korea’s flagship foundation model effort, SK Telecom is making an early bet that national AI infrastructure will be a durable and strategic market — not just a temporary capacity gap.
