South Korea AI data-center build-out Report Interpretation

Overview

The report examines the rapid emergence of AI data-center capacity plans in South Korea, the economics of competing delivery models and the implications for covered telecom, internet and IT-services companies. Morgan Stanley argues that localized enterprise inference and government support underpin the opportunity, but capacity alone is not a reliable guide to value because risk, ownership and returns vary sharply by model.

Core views

Morgan Stanley argues that the Korean AIDC build-out has begun in earnest: five operators under coverage have announced plans that together reach about 5GW over the next five years. The report links this acceleration to three forces: enterprise adoption of AI applications requiring inference near customers; data-security requirements that keep sensitive Korean data, particularly financial-services data, within national borders; and hyperscalers' need to secure capacity outside the US. The first two drivers are especially interconnected because large Korean enterprises handle sensitive information in training and inference workloads. Morgan Stanley also cites NVIDIA's planned US$1bn investment in Naver as evidence of interest from global AI-service providers as well as local customers. The announced plans are sizable but varied. Samsung SDS targets about 834MW and W3.6trn of annual revenue by 2031, including Gumi, the AI Computing Center and 500MW of DBO capacity. Naver plans 200MW by 2028 and a longer-term 1GW global target, with a proposed US$9bn Brookfield financing arrangement for much of the physical data-center and GPU ownership. SK Telecom disclosed up to 2GW of AIDC capacity using NVIDIA Vera Rubin systems and SK hynix HBM, while SK Hyper has ambitions of 5GW by 2029 and 15GW by 2035. KT targets 1GW over five years and W4.4trn in data-center revenue by 2031; LG Uplus plans to expand from 165MW to 400MW by 2030. The report emphasizes that these are not like-for-like capacity figures because each plan combines different degrees of asset ownership and operating exposure. Morgan Stanley distinguishes three core models. In neocloud, the provider owns GPUs and the supporting networking, storage and software stack, allowing it to retain a large share of token-sale margins when demand is strong. It also bears the full downside from multibillion-dollar GPU and infrastructure commitments for a 100MW facility, including low utilization, depreciation, GPU obsolescence, financing costs and fixed expenses. Colocation providers own the building, power, cooling and connectivity while customers bring servers and GPUs. This remains the expected near-term core model for Korean telcos' GPU AIDCs: it requires significant investment in land, power and facilities but avoids GPU-obsolescence exposure and can produce more stable contracted revenue. DBO providers design, build and operate facilities for asset owners; the model is asset-light and lower risk, but its upside is generally capped by fees or revenue-sharing arrangements. The report sees Naver's proposed hybrid AI Factory structure as a potentially important extension of neocloud. Naver would retain the project and customer relationship while financial partners own the data-center and GPU assets. This could allow less capital-rich operators to build large neocloud platforms and could become mainstream, potentially including SK Hyper. However, Morgan Stanley considers it a "show me" model: Naver guides for high-teens margins, yet it must maintain high utilization and reliably operate tens of thousands of GPUs while sharing economics with asset owners. If the model proves highly profitable and is easily replicated, it could lower barriers to entry, expand token supply and pressure prices and industry margins. Government backing is a meaningful enabler. Korea's strategy treats AI as a strategic national industry and focuses on compute sovereignty across compute, semiconductors, cloud, foundation models, applications and public-sector adoption. The report cites a KRW150tn five-year National Growth Fund, of which KRW50tn is targeted to AI and semiconductors; a national AI computing-center program; and an approximately KRW10tn 2026 AI budget intended to catalyze up to KRW30tn of public and private investment. At the same time, the report identifies policy continuity as a risk because projects extend beyond 2030 and require approval, electricity and public support. Power is the major physical constraint. Morgan Stanley notes that Korea's current electricity reserve exceeds 20%, providing a near-term buffer, but the 11th Basic Electricity Plan falls well short of the roughly 5GW of maximum aggregate AIDC capacity announced for the next five years. Beyond that horizon, generation plans would need significant upward revisions, with planning, approvals and funding taking time. Aggregate generation is not sufficient by itself: transmission and distribution capacity at project sites, plus connection and approval timing, can delay projects. The supply-demand balance is another central uncertainty. If announced projects arrive on schedule, supply could cluster in 2029-30 and sharply lower token prices while capacity is absorbed. A more scalable hybrid model could force further upward revisions to supply forecasts. Demand depends on enterprise AI adoption, operational efficiency at very large GPU clusters and contracting terms such as spot versus take-or-pay arrangements. Slower adoption could create oversupply concerns and result in project deferrals or cancellations. For stock selection, Morgan Stanley names Samsung SDS as its preferred AIDC play and reiterates Overweight with a W270,000 target. It argues that SDS's early neocloud commitment at Gumi and AICC, direct control of GPU computing assets and operational learning curve at scale place it ahead of peers. Its Samsung Group customer base, led by Samsung Electronics, could support utilization, recurring cash flows and pricing power. The report also highlights 500MW of potential DBO upside by 2031, AI-driven IT-services demand and robotics as a longer-term option. Its Samsung SDS base valuation uses a sum-of-the-parts approach: core operations are valued through DCF using an 8.5% WACC, 9.6% cost of equity, 80% equity/20% debt structure and 2% terminal growth; the National AI Computing Center stake has a W402bn NPV using a separate DCF with a 10% WACC; net cash receives a 30% discount. Morgan Stanley also prefers KT among telecoms, reiterating Overweight and a W62,000 target. KT plans W6trn of AI-infrastructure investment through 2031, including W5trn for 1GW of AIDC capacity and W1trn for subsea cables. The report views its legacy colocation, network and real-estate assets as supportive, while management says capacity investment will scale with confirmed demand. Morgan Stanley says the shares trade below 10x 2026 estimated P/E and around 4x EV/EBITDA, with more than 6% TSR, which it believes does not reflect AIDC upside. Its base sum-of-the-parts valuation applies DCF to parent telecom operations at a 7.2% WACC and 0% terminal growth, alongside discounted market values for listed subsidiaries and target P/B multiples for KT Estate and BC Card. Naver remains Equal-weight with a W240,000 target because the report awaits evidence that AI Factory can generate sound returns when operations begin in 2027. Morgan Stanley sees possible value creation but uses conservative assumptions about the economic split with asset owners and notes continuing uncertainty in Search and Commerce. Naver's base sum-of-the-parts valuation applies 16x 2026 core NOPLAT, 3.5x P/B to Naver Financial, and holding-company discounts to Webtoons and A-Holdings. SK Telecom remains Equal-weight with a W90,000 target: Morgan Stanley is constructive on its infrastructure initiative but requires better visibility on capacity timing, capex, revenue and margins at SK Hyper. LG Uplus remains Equal-weight with a W17,000 target; its expansion to 400MW by 2030 is viewed as credible but less differentiated than peers' plans and lacking an immediate re-rating catalyst. LG CNS is discussed as a DBO and Cloud & AI beneficiary, but Morgan Stanley seeks clearer project wins, scalable DBO economics and re-acceleration in IT spending by key LG affiliates.

Analysis framework

Morgan Stanley first identifies demand, sovereignty and policy drivers for localized Korean AIDCs, then compares announced capacity plans across operators. It assesses each plan through the ownership, capital intensity, service offering and risk profile of neocloud, colocation, DBO and hybrid models, before testing the build-out against power supply, timing, demand and policy constraints. It then applies company-specific operating and valuation frameworks, including DCF, sum-of-the-parts and peer-multiple cross-checks.

Methodology notes

• Industry AnalysisSupply-demand framework

  AI data-center supply-demand analysis

  The report compares announced capacity with power availability and expected enterprise AI demand, highlighting the risk of concentrated 2029-30 supply and lower token prices if adoption lags.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Data-center business-model comparison

  The analysis traces how ownership of GPUs, powered shells and customer contracts determines capital needs, margins and risk across neocloud, colocation and DBO models.

• Valuation methodsDCF (Discounted Cash Flow)

  Discounted cash flow valuation

  Morgan Stanley values several core operations using forecast cash flows, discount rates and terminal-growth assumptions, including Samsung SDS, KT, LG Uplus and LG CNS.

• Valuation methodsSOTP (Sum-of-the-Parts) Valuation

  Sum-of-the-parts valuation

  The report separately values operating businesses, stakes, financial businesses, real estate and other assets for companies such as Samsung SDS, Naver and KT.

• Valuation methodsP/E and PEG Valuation

  Earnings-multiple valuation

  The report uses P/E or NOPLAT multiples in scenario analysis and peer comparisons, including Naver's core operations and telecom valuation cases.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Samsung SDS (018260.KS)

  Preferred AIDC play, anchored by neocloud build-outs and a captive customer base.

  Strengths

  Early Gumi and AICC projects, direct control of GPU assets, Samsung Group demand, IT-services upside and longer-term robotics optionality.

  Weaknesses

  High capital intensity and exposure to utilization, GPU economics and execution at scale.

  Comparison

  Morgan Stanley views SDS as better positioned than peers because it combines neocloud capability with captive demand.

  Risks

  GPU-compute demand below expectations, value-dilutive M&A and lower IT-services budgets.

• KT Corp (030200.KS)

  Preferred telecom AIDC play.

  Strengths

  Legacy colocation and network infrastructure, planned 1GW capacity, subsea-cable investment and value-up initiatives.

  Weaknesses

  Future business-model mix and the returns on planned AIDC capex remain to be demonstrated.

  Comparison

  Morgan Stanley considers KT's valuation to reflect no AIDC upside, unlike the potential embedded in its planned build-out.

  Risks

  Telecom regulatory pressure, tariff cuts, higher capex, renewed competition and slower AI progress.

• Naver Corp (035420.KS)

  Covered AI Factory and hybrid-model operator.

  Strengths

  Potential to scale neocloud capacity with partner financing and leverage AI across ads and services.

  Weaknesses

  Economics with asset-owning partners, utilization and margin sustainability remain unproven; Search and Commerce uncertainty persists.

  Comparison

  Its hybrid structure is more asset-light than direct neocloud ownership but may have lower barriers to entry.

  Risks

  AI disruption to core advertising, slower Search and Commerce recovery, higher AI costs and failure to demonstrate AI Factory returns.

• SK Telecom Co Ltd (017670.KS)

  Covered AIDC developer through SK Hyper.

  Strengths

  Ambitious infrastructure plans and potential access to NVIDIA systems and SK hynix HBM.

  Weaknesses

  Limited visibility on capacity timing, capex, revenue, margins and the SK Hyper operating model.

  Comparison

  Morgan Stanley is constructive on the strategic initiative but sees less investable clarity than for Samsung SDS and KT.

  Risks

  Lower-than-expected AIDC returns, competitive pressure and volatility tied to the valuation of AI investments.

• LG Uplus Corp (032640.KS)

  Covered colocation-led AIDC operator.

  Strengths

  Paju cluster expansion and a stated path to 400MW by 2030.

  Weaknesses

  The AIDC growth story is viewed as less differentiated than peers and lacks an immediate catalyst.

  Comparison

  Its expected colocation focus carries less GPU risk than neocloud but offers a more modest risk-return profile.

  Risks

  Telecom pricing or capex regulation and renewed wireless competition.

• LG CNS Co Ltd (064400.KS)

  Covered Cloud & AI and DBO-related beneficiary.

  Strengths

  Potential Cloud & AI, data-center-management and robotics opportunities.

  Weaknesses

  Near-term growth could be constrained by tighter IT budgets at LG affiliates; DBO competition is intensifying.

  Comparison

  Its asset-light DBO exposure has lower risk but more limited upside than a direct neocloud model.

  Risks

  Tighter affiliate IT budgets, increased DBO competition and weaker-than-expected recovery in Digital Business Services.

Key data

• Aggregate covered-operator AIDC plans5GW over the next five yearsMorgan Stanley's total for five covered operators' recently announced build-out plans.
• Samsung SDS capacity and revenue target834MW and W3.6trn annual revenue by 2031Includes Gumi, AICC and incremental DBO capacity.
• Naver capacity plan200MW by 2028; 1GW longer term including global capacityNaver is negotiating a US$9bn Brookfield financing arrangement for the build-out.
• KT AIDC plan1GW over five years; W4.4trn data-center revenue target by 2031KT also disclosed W6trn of AI-infrastructure spending, including W5trn for AIDC and W1trn for subsea cables.
• LG Uplus capacity plan400MW by 2030 from 165MW currentlyThe planned portfolio includes 270MW in-house and 130MW leased capacity.
• Korean electricity reserveOver 20%Provides a near-term buffer, but Morgan Stanley says longer-term generation and grid investment must rise materially.
• Samsung SDS DCF assumptions8.5% WACC and 2% terminal growthCore-operations DCF; the National AI Computing Center stake is valued using a 10% WACC and 2% terminal growth.

Impact & implications

The report says the AIDC opportunity is real but valuation should reflect the quality of capacity rather than headline megawatts. Companies with direct GPU control, scalable operating capability and anchored demand may capture stronger economics, while asset-light or hybrid approaches reduce capital needs but leave questions around profit sharing, differentiation and utilization. Near-term progress depends on demonstrated demand, financing, grid connections, approvals and credible operating returns.

Risks

• Build-out beyond 5GW would require substantial new power generation, while site-level grid connections and approvals can delay projects.
• A cluster of capacity additions in 2029-30 could depress token prices and industry returns if enterprise AI adoption is slower than expected.
• Hybrid financing could lower barriers to entry, expand supply and pressure margins if operators cannot demonstrate sustainable differentiation.
• A negative policy or public-perception shift could reduce the approvals, power allocation and public support needed for projects extending beyond 2030.
• Neocloud operators face GPU obsolescence, financing, depreciation and utilization risk because they own the computing stack.

What to watch

• Evidence of enterprise AI demand, contract structures and utilization at Korean AI data centers.
• Power-generation revisions, transmission capacity and project-level grid connection approvals.
• Naver's AI Factory launch in 2027 and the economics agreed with financing partners.
• Samsung SDS's execution at Gumi and AICC, GPU-compute pricing and AI-driven IT-services growth.
• KT's confirmed customer demand, AIDC business-model mix and progress toward its 1GW plan.
• Greater disclosure from SK Hyper on capacity timing, capex, revenue and margin expectations.