Cloud Repatriation Case Studies

Posted: March 27, 2026 to Technology.

Cloud Repatriation Is No Longer Controversial

For years, the technology industry narrative was unidirectional: on-premises is legacy, cloud is the future. But a counter-trend has emerged as organizations that migrated to the cloud discover that certain workloads are more cost-effective to run on owned infrastructure. This process, known as cloud repatriation, involves moving workloads back from public cloud to on-premises or colocation facilities.

A 2024 survey by Andreessen Horowitz found that among companies spending more than $2 million annually on cloud infrastructure, 72% had repatriated at least one workload. The primary motivation in 83% of cases was cost reduction. This is not a rejection of cloud computing. It is a maturation of cloud strategy that places each workload in its optimal location based on economics, performance, and operational requirements.

Case Study 1: SaaS Company Cuts Infrastructure Costs by 60%

Basecamp (now 37signals) publicly documented one of the most high-profile cloud repatriation projects. The company was spending approximately $3.2 million per year on AWS for their suite of project management and communication tools. After analysis, they determined that their workloads were steady-state (predictable, consistent demand) and did not leverage elastic scaling, the primary value proposition of cloud computing.

They invested approximately $600,000 in server hardware and built out their infrastructure in two leased colocation facilities. The result: annual infrastructure costs dropped to approximately $1.3 million, a savings of roughly 60%. The hardware investment paid for itself in under 6 months. Over 5 years, the projected savings exceed $8 million.

Key factors that made this successful:

• Workloads were steady-state with predictable resource requirements
• The team had deep operations expertise (they had run their own infrastructure before migrating to cloud)
• They did not need multi-region geographic distribution
• Their scale justified dedicated hardware rather than multi-tenant shared resources

Case Study 2: Video Streaming Platform Reduces Bandwidth Costs by 50%

A mid-size video streaming company spending $1.8 million annually on AWS, with approximately 65% of that cost attributed to data transfer (egress) fees, repatriated their video delivery infrastructure to bare-metal servers in a Tier III colocation facility.

Video streaming is one of the worst workloads for public cloud economics because of the combination of high bandwidth requirements, predictable demand patterns, and cloud egress pricing. AWS charges $0.09 per GB for the first 10 TB of data transfer, dropping to $0.05 per GB at higher volumes. Colocation bandwidth costs a fraction of this through committed bandwidth agreements.

After repatriation, their bandwidth costs dropped from approximately $1.17 million to under $400,000 annually. Total infrastructure costs (including hardware depreciation, colocation fees, and staff time) decreased from $1.8 million to approximately $900,000, a 50% reduction.

Key factors:

• Data-intensive workload where egress costs dominated the bill
• Predictable traffic patterns allowed accurate capacity planning
• Content delivery requirements were regional rather than global (they used a CDN for edge delivery)
• The team had existing data center operations experience

Case Study 3: Healthcare Analytics Firm Achieves 45% Savings with Compliance Benefits

A healthcare analytics company processing PHI for hospital systems was spending $2.4 million annually on a HIPAA-compliant AWS deployment. Beyond cost, they faced ongoing compliance complexity with the AWS shared responsibility model, BAA management, and the constant audit burden of proving their cloud configuration met HIPAA technical safeguards.

They repatriated their core analytics platform to dedicated servers in a HIPAA-compliant colocation facility. By owning the infrastructure, they simplified their compliance posture: physical security was provided by the colocation facility (with SOC 2 Type 2 and HIPAA certifications), and they had complete control over the technical environment without needing to navigate cloud provider configuration complexity.

Annual costs dropped to $1.32 million (45% savings), and their HIPAA audit preparation time decreased by approximately 40% because they no longer needed to document and verify cloud-specific controls across hundreds of AWS services and configurations.

Key factors:

• Compliance overhead in cloud exceeded the compliance burden of owned infrastructure
• Workloads processed sensitive data (PHI) requiring stringent controls
• Steady-state compute and storage requirements
• The team preferred the operational simplicity of a smaller, fully-controlled environment

Case Study 4: Fintech Startup Repatriates Database Layer for 55% Cost Reduction

Rather than full repatriation, a fintech company took a hybrid approach. Their application tier remained in AWS (leveraging auto-scaling for variable web traffic), but they moved their PostgreSQL database cluster to bare-metal servers in a colocation facility connected to AWS via Direct Connect.

Their database workload was the most expensive component: multiple RDS instances with provisioned IOPS, cross-region replication, and automated backups totaled approximately $45,000 per month. The same database cluster on owned hardware with NVMe storage cost approximately $20,000 per month (including hardware amortization, colocation, and Direct Connect fees), a 55% reduction.

Database performance actually improved because bare-metal servers eliminated the noisy neighbor problem inherent in multi-tenant cloud environments. Query latency decreased by 30%, and p99 latency became significantly more predictable.

Key factors:

• Hybrid approach kept elastic workloads in cloud while repatriating steady-state workloads
• Database workloads benefited from dedicated I/O rather than shared cloud storage
• Direct Connect provided reliable, low-latency connectivity between environments
• The team had strong database administration skills

When Cloud Repatriation Makes Sense

Based on these case studies and broader industry data, cloud repatriation is most beneficial when:

• Workloads are steady-state: Consistent, predictable resource requirements that do not benefit from elastic scaling
• Cloud spend exceeds $500K annually: Below this threshold, the operational overhead of managing your own infrastructure may outweigh the savings
• Egress costs are significant: Data-intensive workloads (video, large dataset distribution, backup replication) where cloud data transfer fees dominate the bill
• Performance requirements are stringent: Workloads needing consistent I/O performance, low and predictable latency, or dedicated hardware resources
• Your team has operations expertise: Running your own infrastructure requires skills in hardware procurement, networking, storage, monitoring, and capacity planning
• Compliance is simpler on owned infrastructure: When cloud configuration complexity makes compliance harder rather than easier

When Cloud Repatriation Does NOT Make Sense

• Variable workloads: Applications with significant traffic spikes or seasonal patterns benefit enormously from cloud elasticity
• Rapid growth: If your compute needs are growing 50%+ per year, cloud allows you to scale without hardware procurement delays
• Global distribution: Applications serving users across multiple continents benefit from cloud providers' global infrastructure
• Small scale: Organizations spending under $200K annually on cloud typically cannot achieve meaningful savings through repatriation after accounting for operational overhead
• Limited ops team: If you do not have or cannot hire infrastructure operations staff, the cloud's managed services are worth the premium
• Innovation velocity: If you need rapid access to managed services (ML platforms, managed databases, serverless functions), cloud provides capabilities that take months to build on-premises

The Repatriation Process

1. Cost analysis: Detailed breakdown of current cloud spend by workload, with separation of compute, storage, data transfer, and managed service costs
2. Workload assessment: Categorize each workload as repatriation candidate, cloud-native, or hybrid based on the criteria above
3. Infrastructure planning: Specify hardware, colocation facility, network connectivity, and monitoring infrastructure for repatriated workloads
4. Migration execution: Move workloads in waves, starting with lowest-risk, highest-savings candidates
5. Optimization: After repatriation, optimize the hybrid architecture for cost, performance, and operational efficiency

The Hybrid Cloud Architecture: Best of Both Worlds

The most sophisticated organizations do not think in terms of cloud versus on-premises. They design hybrid architectures that place each workload in its optimal location and connect the environments through dedicated network links.

Hybrid Architecture Design Principles

• Data gravity: Place compute near the data it processes. If your primary database is on-premises, keep compute-intensive analytics workloads on-premises too. Moving compute to the cloud when data remains on-premises creates egress costs and latency.
• Burst capability: Use cloud for capacity that exceeds your on-premises baseline. Run steady-state workloads on owned hardware and burst to cloud during demand spikes, seasonal peaks, or special projects.
• Unified management: Use tools that span both environments (Terraform for infrastructure, Kubernetes for orchestration, Prometheus for monitoring) rather than maintaining separate toolchains that create operational silos.
• Consistent security: Apply the same security policies, monitoring, and access controls across both environments. A compromised on-premises system connected to your cloud via Direct Connect provides a bridge for attackers to reach cloud resources.

Connectivity Options

Reliable, high-bandwidth connectivity between on-premises and cloud environments is critical for hybrid architectures. AWS Direct Connect provides dedicated 1 Gbps or 10 Gbps connections with consistent latency and no data transfer over the public internet. Azure ExpressRoute offers similar dedicated connectivity to Azure with options for different bandwidth tiers. Google Cloud Interconnect provides dedicated or partner connections to Google Cloud. For smaller deployments, site-to-site VPN provides encrypted connectivity over the internet at lower cost but with higher latency and less consistent performance.

Lessons Learned from Failed Repatriations

Not every repatriation succeeds. Understanding common failures helps avoid costly mistakes.

The Ops Team That Was Not Ready

A technology company repatriated their primary application stack from AWS to colocation without adequately staffing for infrastructure operations. In the cloud, they relied on RDS for database management, ELB for load balancing, and CloudWatch for monitoring, all managed services that required minimal operational overhead. On-premises, these became their responsibility. Within three months, they experienced two significant outages caused by database maintenance procedures they did not have expertise to perform, and monitoring gaps that masked performance degradation until users noticed. After six months, they migrated back to AWS at significant additional cost. The lesson: repatriation requires operations skills that cloud-native teams may not possess.

The Capacity Planning Failure

A SaaS company repatriated based on current resource utilization without adequately planning for growth. They purchased hardware sized for current demand plus 20% headroom. Within 8 months, their user base grew 40% and they hit capacity limits. Procuring additional hardware took 8 weeks (supply chain delays on specific server models), during which they experienced performance degradation that affected customer satisfaction. Cloud's elastic scaling had masked their rapid growth pattern. The lesson: build significant headroom into capacity planning, and maintain a cloud burst capability for demand spikes that exceed on-premises capacity.

Total Cost of Ownership: Cloud vs. On-Premises Detailed Breakdown

Understanding the full cost picture requires looking beyond the monthly cloud bill. Here is a detailed comparison framework that organizations use to evaluate repatriation economics.

Cloud Costs (Often Underestimated)

• Compute: EC2, Azure VM, or GCE instance charges including reserved and on-demand costs
• Storage: Block storage (EBS, Azure Disk), object storage (S3, Blob), and snapshot charges. Storage costs compound over time as data grows.
• Data transfer: Egress fees are the hidden tax that catches many organizations. AWS charges $0.09/GB for the first 10 TB, with volume discounts at scale. A service transferring 50 TB per month pays approximately $4,500 in egress alone.
• Managed services premium: Managed databases (RDS, Azure SQL), managed Kubernetes (EKS, AKS), and other PaaS services include a significant markup over self-managed equivalents. RDS PostgreSQL costs 3 to 5 times more than self-managed PostgreSQL on equivalent compute.
• Support plans: Enterprise support plans run $15,000+/month on AWS, 100% of monthly bill at minimum on Azure Enterprise.
• Reserved capacity commitment: Reserved instances and savings plans lock you into 1 to 3-year commitments. If your needs change, you are paying for unused capacity.

On-Premises Costs (Often Underestimated)

• Hardware: Server purchase (amortized over 4 to 5 years), networking equipment, storage arrays, and backup infrastructure
• Colocation: Monthly rack fees ($500 to $2,000 per rack), power costs ($0.08 to $0.15/kWh), bandwidth commitments ($500 to $5,000/month for dedicated bandwidth), cross-connects ($200 to $500/month per connection)
• Staff: System administrators, network engineers, and storage engineers required to manage the infrastructure. Factor in fully loaded costs including benefits, training, and tooling.
• Redundancy: On-premises redundancy requires purchasing spare hardware, maintaining DR sites, and implementing data replication that cloud provides natively.
• Opportunity cost: Staff time spent managing infrastructure is time not spent on business-differentiating projects.

The Breakeven Calculation

For a typical workload spending $10,000/month on cloud compute and storage, the on-premises equivalent might cost $3,000/month in colocation and power plus $150,000 in hardware (amortized to $2,500/month over 5 years) plus an incremental $2,000/month in staff time, totaling $7,500/month. The monthly savings of $2,500 means the $150,000 hardware investment pays back in 5 years, but only if the workload remains stable. If the workload is growing 30% per year, you will need additional hardware investment in 18 months that may change the calculation.

This is why repatriation works best for stable, predictable workloads and why variable workloads should remain in the cloud regardless of the current cost comparison.

Building a Repatriation Business Case

To get organizational buy-in for cloud repatriation, structure your business case around these elements:

1. Current cloud spend breakdown: Detailed analysis of what you spend by service, by workload, and by cost category (compute, storage, transfer, managed services)
2. Workload classification: Categorize every cloud workload as repatriation candidate, cloud-optimal, or hybrid, with justification for each classification
3. Projected on-premises costs: Detailed TCO for repatriated workloads including hardware, colocation, connectivity, staff, and DR
4. Migration costs: One-time costs for hardware procurement, colocation setup, migration labor, and testing
5. Risk assessment: Identify and quantify risks including migration downtime, operational complexity, and the risk of needing to re-migrate to cloud if requirements change
6. 3-year financial projection: Year-by-year comparison showing the investment payback period and cumulative savings
7. Non-financial benefits: Performance improvements, compliance simplification, data sovereignty, and reduced vendor dependency

Frequently Asked Questions