Data Center Cost Optimization
Optimize costs for on-premises and colocation infrastructure. Covers power usage effectiveness, hardware lifecycle management, capacity planning, hybrid cloud economics, and the total cost of ownership model that compares cloud vs data center.
Not everything belongs in the cloud. For predictable, high-utilization workloads, data centers can be 30-50% cheaper than public cloud over a 3-5 year horizon. Understanding data center economics is essential for hybrid strategies and for evaluating whether cloud migration truly saves money.
Total Cost of Ownership (TCO)
Data Center Costs
Capital Expenditure (CapEx):
Servers: $5,000-15,000 per server (3-5 year lifecycle)
Networking: $50,000-500,000 per rack (switches, cabling)
Storage: $500-2,000 per TB
Facility build: $10-25M per MW of capacity
Operating Expenditure (OpEx):
Power: $0.08-0.15 per kWh
Cooling: 30-50% of power cost
Staff: $150,000-250,000 per engineer
Colocation rent: $500-2,000 per rack/month
Network transit: $0.50-2.00 per Mbps/month
Maintenance: 15-20% of hardware cost annuallyCloud vs Data Center Break-Even
Cloud makes sense when:
- Workloads are variable (> 3:1 peak to average)
- Time to market matters more than unit cost
- Team lacks data center expertise
- Workload lifespan < 3 years
Data center makes sense when:
- Workload is stable and predictable
- Utilization > 60% sustained
- Data sovereignty requirements
- Workload lifespan > 3 years
- Compliance requires physical controlPower Usage Effectiveness (PUE)
PUE measures how efficiently a data center uses power:
PUE = Total Facility Power / IT Equipment Power
PUE 1.0: Perfect — all power goes to IT (impossible)
PUE 1.2: Excellent — hyperscaler level
PUE 1.4: Good — modern enterprise
PUE 1.6: Average — older facilities
PUE 2.0+: Inefficient — needs modernizationReducing PUE
| Strategy | PUE Impact | Cost |
|---|---|---|
| Hot/cold aisle containment | -0.1 to -0.3 | Low |
| Free cooling (outside air) | -0.2 to -0.4 | Medium |
| Liquid cooling | -0.3 to -0.5 | High |
| Raise inlet temperature | -0.05 to -0.1 | None |
| Efficient UPS (> 96%) | -0.05 to -0.1 | Medium |
Hardware Lifecycle Management
Year 0: Purchase and deploy
Year 1: Full production, warranty active
Year 2: Peak utilization
Year 3: Warranty expires, evaluate refresh vs extend
Year 4: Higher failure rates, consider repurpose to dev/test
Year 5: Decommission, recycle, or sellRefresh Decisions
| Factor | Keep (Extend) | Replace (Refresh) |
|---|---|---|
| Performance meets needs | ✅ | |
| Power efficiency declining | ✅ | |
| Failure rate increasing | ✅ | |
| Software requires newer hardware | ✅ | |
| Maintenance cost > 40% of new | ✅ |
Capacity Planning
def forecast_capacity(current_usage, growth_rate, planning_horizon_months):
"""Simple exponential growth model"""
projected = current_usage * (1 + growth_rate) ** planning_horizon_months
return {
'current': current_usage,
'projected': projected,
'buffer': projected * 1.3, # 30% headroom
'lead_time_months': 3, # Hardware procurement
'order_trigger': projected * 0.7 # Order at 70% utilization
}Anti-Patterns
| Anti-Pattern | Consequence | Fix |
|---|---|---|
| No PUE tracking | Cannot optimize power costs | Monitor PUE monthly |
| Hardware beyond 5 years | Failure rates, power inefficiency | Lifecycle management with refresh cycles |
| Cloud for stable workloads | 2-3x higher cost than needed | Evaluate repatriation for stable workloads |
| No capacity planning | Emergency purchases at premium pricing | 12-month rolling capacity forecast |
| Ignoring decommission costs | End-of-life hardware accumulates | Budget for decommission in lifecycle |
Data center cost optimization is a discipline that combines electrical engineering, financial modeling, and IT operations. The goal is maximum compute per dollar over the infrastructure lifecycle.
