Approach Comparison
Data Center Modernization Approaches
| Approach | Capital Cost | Timeline | Risk | Best For |
|---|---|---|---|---|
| Refresh | Medium | 6–18 months | Low–Medium | Facilities with sound structure, targeted capability gaps |
| Rebuild | High | 18–36 months | Medium–High | Facilities with fundamental structural limitations |
| Colocation | Low–Medium (OpEx shift) | 3–12 months | Medium | Organizations eliminating facility management overhead |
| Cloud Migration | Low CapEx (high OpEx) | 6–24 months | Medium | Variable workloads, development, burst capacity |
| Hybrid | Varies | Varies | Medium | Most enterprise organizations, on-prem + cloud by workload type |
Refresh: When It Works
A refresh program upgrades specific infrastructure components: compute, storage, network, or cooling: without replacing the facility structure or power distribution architecture. It is the right approach when the facility has adequate power capacity, adequate cooling capacity, and adequate floor space for the workloads it must support, and the gap between current capability and required capability is addressable through targeted component upgrades.
Refresh is appropriate when
Facility power capacity is adequate for target workloads
Cooling infrastructure can support target power densities
Floor space is sufficient for planned deployments
Network architecture can be upgraded without structural changes
Compliance requirements can be met with targeted controls
Refresh is not appropriate when
Power capacity is at or near the facility ceiling
Cooling cannot support the power densities of target workloads
Floor space constraints prevent required deployments
Network architecture requires fundamental redesign
Structural limitations prevent required redundancy levels
Rebuild: When It Is Required
A rebuild program replaces the facility infrastructure at a fundamental level: power distribution, cooling architecture, or both: because the existing infrastructure has structural limitations that cannot be addressed through component upgrades. Rebuild programs are more expensive and more disruptive than refresh programs, but they are the only option when the facility cannot support the workloads the organization requires.
The rebuild trigger most organizations miss
Colocation: Trade-offs
Colocation moves the organization's compute, storage, and network equipment into a third-party facility. The colocation provider is responsible for power, cooling, physical security, and facility operations. The organization retains responsibility for the equipment it installs.
Colocation: What You Gain and What You Give Up
| Dimension | What You Gain | What You Give Up |
|---|---|---|
| Facility management | Eliminate facility operations overhead | Control over facility decisions and upgrades |
| Capital cost | Convert CapEx to OpEx (monthly fees) | Long-term cost may exceed owned facility |
| Power density | Access to high-density cages (10–30 kW/rack) | Premium pricing for high-density deployments |
| Compliance | Provider certifications (SOC 2, ISO 27001) | Shared responsibility model requires verification |
| Latency | Carrier-neutral facilities with low-latency interconnects | Higher latency than on-premises for internal applications |
Cloud Migration: What It Cannot Replace
Cloud migration is the right answer for specific workload types: and the wrong answer for others. The workloads that cloud handles well are variable, burst, and development workloads where the elasticity of cloud infrastructure provides cost and operational advantages. The workloads that cloud handles poorly are regulated, latency-sensitive, high-bandwidth, and AI training workloads.
Cloud is well-suited for
Variable workloads with unpredictable demand
Development and test environments
Burst capacity for seasonal peaks
SaaS applications with cloud-native architecture
Disaster recovery and backup targets
Cloud is poorly suited for
Regulated data with strict data residency requirements
Latency-sensitive applications (sub-5ms requirements)
High-bandwidth workloads (100+ Gb/s sustained)
AI training on proprietary data at scale
Applications with predictable, steady-state demand (higher TCO)
The Hybrid Reality
Most enterprise organizations end up with a hybrid infrastructure model: on-premises or colocation for regulated, latency-sensitive, and high-bandwidth workloads; cloud for variable, burst, and development workloads. The modernization question is not "on-premises vs. cloud", it is "which workloads belong where, and what infrastructure is required to support each."
Decision Criteria
Does the workload have data residency or sovereignty requirements?
If yes, on-premises or colocation in the required jurisdiction. Cloud may be available in the required region but requires verification of data residency controls.
What is the latency requirement?
Sub-5ms latency requirements typically require on-premises infrastructure. 5–50ms can be served from colocation. Higher latency tolerances can be served from cloud.
What is the sustained bandwidth requirement?
Workloads requiring 10+ Gb/s sustained bandwidth are typically more cost-effective on-premises or in colocation than in cloud, where egress costs are significant.
Is the workload variable or steady-state?
Variable workloads with unpredictable demand benefit from cloud elasticity. Steady-state workloads with predictable demand are typically more cost-effective on-premises over a 3–5 year horizon.
What is the compliance framework?
Identify all applicable frameworks and verify that the chosen infrastructure model can satisfy them before committing to an approach.