GPU service turns hardware capacity into a utilization business.

Reported evidence.

SK Telecom's February 2025 results confirm that it opened the Gasan AI data center with Lambda in December 2024 and introduced SKT GPUaaS. Its July account again identifies December as the GPUaaS launch period. This case concerns that operating service milestone, not another quarterly earnings observation. Neither source isolates Gasan GPUaaS revenue, installed utilization or contribution.

Investment interpretation.

Rented compute can give customers access without owning equipment while creating a service business for the operator. The operator bears utilization, hardware renewal and reliability obligations. A completed opening is meaningful operating evidence, but its economic return depends on paid workload over the equipment's useful commercial life.

Economic assessment.

Model compute receipts against equipment funding, power, cooling, maintenance and idle time. Capacity sold through subscription or usage has a different cash profile from a long-term facility lease. The releases' broader data-center revenue should not be assigned to this specific service without a segment bridge.

Utilization Is the Main Denominator

Compute equipment can be available while generating little paid work. The economic denominator should therefore be sold capacity under actual service conditions, not the number of machines installed. Utilization includes workload compatibility, scheduling and customer demand, not simply physical uptime. A customer may value immediate access while reserving little predictable volume. Pricing must compensate for that flexibility and the operator's idle capacity. Longer commitments can improve visibility but may require discounts or specific hardware allocation. The operator should distinguish training, inference and intermittent experimentation because they create different occupancy and support patterns. The opening establishes a service route, but neither cited page provides the utilization needed to calculate its standalone return. A useful funding model would connect customer commitments with the equipment configuration and test lower occupancy together with price pressure. Strong overall demand for AI does not guarantee that a particular hardware pool remains fully sold at an attractive contribution.

Commercial Life of the Equipment

GPU hardware can remain functional after newer systems change the market price of equivalent work. The recovery period should match the equipment's useful commercial life rather than its physical lifespan. A service operator can improve return through efficient scheduling and appropriate workloads, while customers can shift to alternatives as prices and capability change. Financing should allow for renewal and replacement, not assume initial capacity produces the same cash indefinitely. The supplier and partner arrangements also determine who owns the equipment, funds upgrades and bears residual value. The source names cooperation with Lambda but does not disclose the full asset and payment allocation. A capital appraisal should obtain that structure before assigning debt coverage. A completed data-center opening can create useful infrastructure while the compute layer experiences a faster depreciation cycle. Those assets should not automatically share one financing horizon or one recovery assumption.

Electricity expenditure should follow the actual contract and workload profile. A customer price fixed for a long period can leave the operator exposed to changed input costs, while a pass-through may weaken competitiveness. The useful forecast tests those terms alongside utilization, rather than applying a constant power-cost assumption.

Reliability and Customer Support

A compute service must provide more than powered hardware. Customers need access control, provisioning, network performance, monitoring and help when workloads do not behave as expected. Service-level promises can create credits, replacement or support expense. The operator should identify which obligations sit with SK Telecom and which depend on Lambda or another supplier. A recognizable partner can strengthen technical capability without eliminating interface risk. Customers also may need assistance moving models and data into the service. That cost can be reusable or account-specific, affecting the contribution from different customer types. The financial model should measure service costs alongside paid capacity. A high nominal occupancy can still produce weak contribution if workloads require extensive support or if performance problems create concessions. The sources establish an actual operating milestone, not a complete service margin. Follow-through should show repeat usage, dependable delivery and a cost to serve that remains manageable as the customer base expands.

Cash Commitments Before Usage

Equipment and facility expenditure may be committed before customer demand becomes predictable. The operator needs liquidity for that interval, including power, maintenance and support. Customer deposits or prepaid capacity can improve conversion while transferring obligations to supply service later. Receipts should therefore be reconciled with the associated commitments, not treated as unrestricted surplus. Usage-based customers can generate recurring revenue but may leave a less secure borrowing base than non-cancellable contracts. The operator should also distinguish domestic group demand from external accounts, since internal usage can have a different transfer-price and payment structure. Broader AI-infrastructure plans provide strategic context but do not establish this site's cash. A disciplined appraisal would allocate capital to the Gasan service's own contracts and hardware cycle, with a downside case that survives lower usage and a faster decline in rental prices. Opening the route is progress; financing its repeated use requires more specific evidence.

Underwriting questions.

1. Who owns and finances the GPU equipment and its upgrades?
2. What paid utilization and non-cancellable commitments support the service?
3. How do partner fees, service obligations and hardware renewal affect available cash?

Primary sources.

1. SK Telecom / dated confirmation of Gasan opening and GPUaaS2025-02-12
2. SK Telecom / later confirmation of December launch period2025-07-04