Video summary

AI 데이터센터 시장의 현재와 미래

Main summary

Key takeaways

Business

Business Summary: AI Data Centers Market (Present & Future)

The webinar frames “AI data centers” as a strategic shift away from traditional data centers that primarily store data, toward facilities that must support model training, inference, and continuous learning—an “AI Factory” concept (referencing NVIDIA CEO Jensen Wang).

Key Growth Drivers

Market growth is driven by:

  • Explosive data traffic (cloud migration, digital services)
  • Rapid AI adoption by enterprises and consumers
  • Large-scale investment by global hyperscalers (e.g., AWS, Microsoft, OpenAI)
  • Entry of domestic conglomerates and policy reforms

Execution Complexity (Why Delivery Is Hard)

Execution is complex due to:

  • Large capex
  • Permitting/licensing requirements
  • Need for strategic power infrastructure
  • Multi-stage risk across:
    • design → construction → commissioning → operation

Key Frameworks / Decision Playbooks (Investor & Operator Execution)

Business Feasibility Review (“3 Pages”) — Decision Prerequisites

The “Business feasibility review” must answer:

  • What is the growth consensus of the domestic data center industry?
  • Why is development slow despite rising demand?
  • What conditions are necessary for successful development?
  • How can missing capabilities be supplemented?

Structure: Four Detailed Review Areas

  1. Analyze the current status of the domestic IDC industry plus cloud acceleration, interconnection, and metropolization effects (including server redundancy and concentration).
  2. Analyze CSP/global/domestic trends to identify why some projects fail to progress.
  3. Define core competitive aspects, including:
    • operation strategy
    • operational capability definition
    • how to build operational strength/capacity
  4. Optimize CARFAX/OPEX for viability and perform contractual risk mitigation.

IDC Development Lifecycle Stages (With Risk Hotspots)

Stage 1: Single-site Development (Early Economics)

  • Market trends analysis
  • P&L analysis
  • Site evaluation

Stage 2: Legal & Licensing

  • Legal consultation
  • Licensing/permits
  • Water feasibility checks

Stage 3: Leasing & Pre-Construction Planning

  • Conclude lease agreement
  • Validate investment P&L feasibility
  • Prepare facility operation plan

Stage 4: Construction

  • Finalize technical design concept
  • Cost/quantity calculations
  • Civil/MEP engineering execution
  • Incorporation & installation

Stage 5: Commissioning

  • Integrated reliability testing
  • Put into actual operation

“Five Strategic Areas” in Development Strategy

  • Establish early consensus with tenants/operators starting from the planning stage
  • Secure the optimal site (best demand position + best ability to procure power infrastructure)
  • Terminate/mitigate contractual risks using accurate MS contract risk analysis
  • Secure business viability
  • Ensure project finance (PF) funding readiness, accounting for hyperscale power infrastructure needs (e.g., substation voltage like ≥154kV)

Risk Control Principle (Design → Execution → Verification)

Treat the project as mission-critical infrastructure, not “typical real estate.”

Core idea: Small early design mistakes amplify exponentially over time via a risk amplification chain such as:

  • design changes
  • equipment changes
  • schedule delays
  • COD (commercial operation date) disruptions
  • revenue loss

Success comes from proactive early risk design, not reactive fixes.

Market Metrics & KPIs / Targets (With Concrete Numbers)

Global & Regional Market Size Forecasts

  • Global data center market: ~$380B (2023) → ~$630B (2029), ~+9% CAGR (approx.)
  • Global colocation market: ~$77B (2023) → $220B by the forecast period, ~+12% CAGR

Regional growth:

  • Americas: ~6x growth
  • Europe: ~2.3x growth
  • Asia-Pacific: ~1–2x growth, but still structural growth

Operating Capacity & Vacancy (Demand/Supply Signals)

  • Americas operating capacity: ~43GW, vacancy ~4.2%
  • Asia operating vacancy: ~10.9% (also cited ~9%); interpreted as growth/maturity stage → expansion opportunity

Domestic (Korea) Market Growth & Demand Drivers

  • Domestic data center scale estimate: ~2.42T KRW (2018) → ~10T KRW (future; the “2018” reference is likely a transcription error)
  • Drivers:
    • domestic internet rate expected to reach ~95% by 2025
    • increased cloud usage + AI + hybrid cloud
    • AI demand requires high-power, high-density GPU computation

Power Demand and Supply Gap

  • Electricity demand for domestic data centers expected to reach ~8x by 2027 vs. 2023
  • The supply-demand gap widens by 2027 vs. 2023
  • Many sites require >100MW per site, making electrical capacity acquisition a “core business”

Supply Pipeline (Korea)

  • Planned new data centers: ~60–64
  • Pipeline timing: 26–28 years mentioned (likely a misread; intent appears to be “concentrated within a near planning window”)
  • Supply location trend:
    • short term: still Seoul metro
    • mid-to-long term: decentralization to non-metropolitan areas
  • Count trend:
    • data center count increased about 3x to 165 (2024)
    • collocation DCs increased about 5x from 10 → 50, ~7% annual average growth rate

Investor KPIs Mentioned (Investor / Business Owner Perspective)

Most critical KPI:

  • Commercial operations data / “promised volume at promised time”

Other KPIs:

  • CAM rate (cash flow / NOI capitalization inputs)
  • Development cost per watt tied to NOI
  • IRR (IR) as dominant profitability indicator
  • DSCR for lender capital limit checks
  • DSCR & cash waterfall by investor, including capex/opex and contract terms
  • Additional valuation indicators:
    • NOI / cap rate
    • EBITDA multiples (for operationally heavy segments such as colocations/retail)
    • Value per MW as a supplementary metric

Financial Structure & Example Numbers (Finance Model)

  • Example total investment: 755.4B KRW
    • Equity 20% / Debt 80%
    • Debt split “5:3 ratio depending on tranche” (specific tranche labels not fully defined in the subtitle text)
  • Leverage variability depends on:
    • sponsor credit
    • business stability
    • market conditions

Offtake / Lease Economics Concepts

  • Example “rent-free period”: ~1 to 1.5 years
    • revenue starts later, but cost outflow occurs earlier → needs a refinancing plan

Policy & Operational Constraints Shaping Strategy (Execution Implications)

Distributed Energy / Grid Impact Assessment Changes

  • 2024 Distributed Energy Act + Power Grid Impact Assessment introduced/implemented
  • By June 2026, metropolitan passing threshold expected to be raised to 75 points
    • implication: harder to build large-scale DCs in metro areas
  • Aug 2024–Jun 2025 metrics:
    • submissions: 195
    • reviews: 33
    • metropolitan pass rate: 4/19 → ~20%
    • similar category elsewhere: 10/14 → ~70%

IDC Location Policy Direction (Decentralization)

  • Government inducing AIDC in non-Seodo areas
  • Some regions exempt from power system impact assessments
  • Relaxing power location rules and building regulations (e.g., time-out system, easing building rules)

Concrete Examples & Case Studies (Risk and Execution)

Power-System / UPS Fire Incidents Used to Demonstrate Risk Amplification

National Information Resources Management Agency Daejeon Center

  • Prolonged critical government digital service impact due to a UPS battery fire
  • Implication: not only equipment—also power system design verification gaps (e.g., failure-spread prevention, structure, backup/recovery)

S4 Pangyo Data Center

  • Tenant’s major service (K) suspended for about 127 hours due to UPS battery thermal runaway
  • Highlighted causes: design defects (battery compartment, cable routing)

Construction Cost Escalation / Contract Outcome Examples

Gimae Data Center Project

  • CARFAX increased cost: 80B KRW → 180B KRW
  • Project scrapped
  • Illustrates the need for:
    • price fluctuation/contract adjustment clauses
    • FX hedge and dispute resolution mechanisms

Yongin Site

  • Licensing/social acceptance issue led to cancellation; resolved by relocating
  • Takeaway: “social license” should be contractually treated as a prerequisite (public hearing, civil-complaint-related agreements)

Commissioning / Operations Failure Examples

S4 Gwacheon Data Center

  • Fire occurred while the emergency generator operated for a long time
  • Signs detected during commissioning phase
  • Compensation ruling >200B KRW
  • Takeaway: clarify liability scope between business operator and operator; enforce commissioning standards

Pangyo Data Center (Company N)

  • Air-conditioning error → server room temperature rise → storage failure → suspension of public cloud services
  • Takeaway: monitoring/response/reporting systems + SOP/EOP + integrated acceptance must be contractually and operationally secured

Strategic Recommendations (Actionable Takeaways)

Site Selection (When Power / Permits / Social Acceptance Matter)

Treat final power availability as the primary site condition. Require:

  • Redundant backup power
  • Power ≥ 20MW as a prerequisite (as stated)
  • Placement within/near data center clusters (around ~20km radius)
  • Adequate telecom interconnection availability

Planning rule of thumb mentioned:

  • ~2,000 pyeong per 10MW for feasibility

Licensing:

  • follow the Distributed Energy Special Zone Act (2024) / direction evaluation requirements for 10MW+

Tenant Demand Stability & Vacancy Risk Assessment

  • Metro vs. provinces:
    • Metro: new supply decreases due to stricter assessments → vacancy risk “almost non-existent” (per panel)
    • Regional: require anchor tenants (large corps + government); retail-only development has high vacancy risk
  • Positioning in provinces:
    • treat as an energy infrastructure project, not just real estate

Data Center Procurement (CAPEX, Power, GPU) and Business KPIs

For AI data centers, because equipment share increases:

  • Include GPU procurement schedules
  • Include AI capability requirements
  • Include power/GPU planning in cashflow models

Optimize:

  • total business viability, not only “EPC cost minimization”

Operational KPI focus:

  • minimize IT load mismatch vs contracted capacity
  • align power cost optimization with actual demand

Contract Structuring & Risk Transfer

Contract risk areas emphasized:

  • Unilateral lease termination (e.g., 90 days prior termination-for-convenience clauses)
  • Late delivery compensation relative to revenue
  • Excess power usage “PU” beyond contractual standard → cost overrun & damages

Contract philosophy:

  • integrated commissioning standards and acceptance procedures
  • SOP/EOP + real-time monitoring obligations
  • clearly defined responsibility boundaries during commissioning/handover

Notes: Traditional vs AI vs Edge Data Centers (Business Operating Differences)

Traditional vs AI Data Centers (Customer & Technology)

  • Customer use case
    • AI: training + inference + AI service operations
  • Core equipment
    • AI: GPU-centric
    • Traditional: CPU-centric
  • Site selection criteria
    • AI: large power supply + scalability + high-density requirements
  • Cooling & power architecture
    • AI: high-power, different cooling approach; layered/execution-oriented power/cooling integration
  • Operations
    • AI: integrated management across power + cooling facilities

Edge/H Data Center Trend (Urban Demand Mitigation)

Benefits:

  • Can bypass power system impact assessment if <10MW
  • Shorter construction timeline (repurpose office/commercial assets)
  • Serves urban demand centers; reduces latency

Tradeoffs:

  • Retrofit requires server/UPS + structural reinforcement → potentially high costs
  • Limited density/single-site scale (e.g., ~2.5MW–5MW)
  • Example urban conversion targets mentioned (e.g., Magok, Hill Dilton, etc.), tied to office-district demand clusters

Investing / Finance Angle (High-Level; Execution Emphasized)

The final finance segment stresses building a cash-flow-linked financial model:

  • Verify historical financial information (DD, P&L, balance sheet, development costs, invested funds execution)
  • Model a “two-month financing method” (as phrased) and project-by-project funding procurement plan
  • Project future estimated cash flows:
    • revenue module (rental/hosting/services/power revenue)
    • OPEX structure (electricity is largest; pass-through depends on electricity rates)
    • CAPEX timing and GPU purchase schedule
  • Validate profitability and performance using:
    • investor cash waterfall
    • profitability indicators (IRR, DSCR)
    • scenario/sensitivity analysis (key-variable swings)

Presenters / Sources (As Stated)

  • Jang Eun-young (Director, PWC) — host
  • Yu Won-seok (Vice President, PWC Consulting) — opening/Q&A lead
  • Park Sung-jin (Partner, real estate advisory solutions)
  • Seo Yong-tae (Partner; leader of AI Data Center Platform; head of large-scale infrastructure investment business division, PWC)
  • Eunhee Cho / Jo Eun-hee (Partner; Co-leader, AI Data Center Platform, PWC Consulting) — design & construction risk
  • Choi Seong-eun / Choi Seong-won (subtitle inconsistency) (Partner, Deal Division; Il Accounting Firm / PwC-related content—finance/economic feasibility session)
  • Choi Seong-hoon (subtitle mentions “Pato”) (Partner; profitability variables response in Q&A)
  • (Implied panelists include the above; some Q&A answers attributed to Park Seong-hee regarding pass rate/policy gap)
  • Park Seong-hee (team leader mentioned during the pass-rate/policy gap question)

Original video