Building-to-IDC Transformation: Challenges and Opportunities

Introduction

With the rapid development of the AI industry, data center demand has surged. Building new IDCs takes time and is costly, making the transformation of existing buildings into AI-ready data centers a new trend. This article shares our experience and insights from actual transformation projects.

Project Background

Transformation Target: A commercial office building in Taipei (B2-4F)
Original Use: Traditional office space
Transformation Goal: Support 1,000 GPU AI training center
Project Timeline: 6 months (2024/04-2024/09)
Total Investment: NT$350 million

I. Preliminary Assessment

1.1 Building Structure

Load-bearing Capacity Testing

Original design load: 500 kg/m²
AI server requirements: 800-1,200 kg/m²
Solution: Reinforce floor structure, distribute load

Ceiling Height

Original floor height: 3.6m
Required clear height: 2.8m (including cable trays)
Conclusion: Meets requirements, no additional treatment needed

1.2 Power System

Existing Capacity

Building total capacity: 2,000 kVA
Available for IDC: 1,500 kVA (excluding public facilities)

Demand Analysis

GPU power consumption: 700W × 1,000 = 700 kW
CPU/Network/Storage: 200 kW
Cooling system: 300 kW
Total Demand: 1,200 kW + 20% margin = 1,440 kW

Transformation Plan

New dedicated transformers (2 × 1,000 kVA)
Dual feed A+B power supply
N+1 UPS configuration (500kVA × 3 units)

1.3 Cooling System

Original System

Traditional central air conditioning
Cooling capacity: 500 RT (refrigeration tons)

AI Workload Requirements

Heat dissipation: 1,200 kW ≈ 340 RT
Problem: Existing system insufficient capacity and low efficiency

Transformation Plan

Cooling Tower (Roof)
    ↓
Chiller Units (400 RT × 2 units, N+1)
    ↓  
Chilled Water Circulation System
    ↓
In-row Air Conditioners (CRAC) in Data Hall

Key Design

Hot/Cold Aisle Containment
Raised floor air supply
Reserved liquid cooling upgrade interface

II. Construction Phase

2.1 Month 1: Demolition and Foundation Work

Remove existing partitions and ceilings
Floor reinforcement work
Power wiring pre-installation

Challenge: Building still operational on other floors, requiring phased construction

2.2 Months 2-3: MEP System Installation

M (Mechanical)

Chiller installation
Pipeline configuration
Air conditioning units positioning

E (Electrical)

Transformer installation
Distribution panel layout
UPS system configuration
Monitoring system wiring

P (Plumbing)

Fire sprinkler system
Leak detection system

2.3 Month 4: Data Hall Finishing

Raised floor installation (600mm height)
Cable tray installation
Access control system configuration
Environmental monitoring sensor deployment

2.4 Month 5: Equipment Installation

42U standard racks (120 units)
InfiniBand switches
Server rack mounting
Network cabling

2.5 Month 6: Testing and Acceptance

Power system testing
Cooling system commissioning
Network connectivity testing
GPU burn-in testing
Fire drill

III. Technical Challenges and Solutions

3.1 Floor Load-bearing

Problem: B2 floor insufficient load capacity

Solution:

Add support columns
Optimize rack layout to distribute weight
Use lightweight racks (aluminum alloy)

3.2 Heat Exhaust

Problem: Limited roof space, difficult cooling tower installation

Solution:

Use high-efficiency compact cooling towers
Phased deployment, install capacity for 500 GPU first
Reserve space for second-phase expansion

3.3 Network Cabling

Problem: Insufficient conduit space in existing building

Solution:

Use vertical cable trays
Some routes along exterior walls
Use high-density fiber (MPO/MTP)

3.4 Noise Control

Problem: Office building sensitive to noise

Solution:

Double-layer soundproof doors for data hall
Sound absorption treatment in equipment rooms
Silencers added to cooling towers

IV. Cost Analysis

| Item | Amount (NT$10K) | Percentage | |------|----------------|------------| | Power System | 8,000 | 23% | | Cooling System | 10,000 | 29% | | Building Finishing | 5,000 | 14% | | IT Equipment | 10,000 | 29% | | Other (Fire/Monitoring) | 2,000 | 6% | | Total | 35,000 | 100% |

Investment Recovery

Rack rental: NT$150K/month per rack
Total racks: 120 units
Target occupancy: 80%
Monthly revenue: NT$14.4M
Payback Period: Approximately 2.5 years

V. Operational Data (After 3 Months)

Power Efficiency

PUE: 1.35 (target <1.4)
Electricity cost: Approximately NT$6M per month

Cooling Efficiency

Supply water temperature: 12°C
Return water temperature: 18°C
Data hall temperature: 24±2°C

Availability

Uptime: 99.95%
Planned maintenance: Once per month
Unplanned outages: 0

VI. Experience Summary

Success Factors

Thorough Preliminary Investigation: Spent 2 months on detailed assessment
Flexible Design: Reserved 30% expansion capacity
Professional Team: Cross-disciplinary collaboration of IDC + Building + MEP
Phased Deployment: 50% capacity first, reducing risk

Points to Note

Regulatory Approval: Building use change took 2 months
Neighbor Coordination: Construction noise required prior communication
Supply Chain: Long lead-time equipment like UPS needed early ordering
Testing Time: Don't compress acceptance testing cycle

VII. Future Planning

Short-term (Within 6 months)

Achieve 80% occupancy
Attract more AI customers

Medium-term (Within 1 year)

Second-phase expansion (+500 GPU)
Introduce liquid cooling technology
Upgrade to GB200 systems

Long-term (2+ years)

Replicate experience in other buildings
Establish Building-to-IDC transformation service brand

Conclusion

Transforming traditional buildings into AI data centers is a viable and economical path. The key lies in:

Solid preliminary assessment
Professional design and construction
Reasonable cost control
Long-term operational planning

We are happy to share more experience and assist building owners and investors interested in entering this field.

Contact Us
If you are interested in building-to-IDC transformation projects, please contact:
Email: building@huisuanlabs.com
Tel: +886-2-XXXX-XXXX