Industry Example: This case study showcases 37signals' public cloud exit story to demonstrate the potential savings achievable through strategic infrastructure decisions. This represents the type of results we help our clients achieve.
Background
37signals, the company behind Basecamp and HEY, had been running on AWS since 2006. By 2022, their cloud bills had grown to over $3.2 million annually, despite their relatively stable and predictable workloads.[1]
David Heinemeier Hansson (DHH), the company's CTO and creator of Ruby on Rails, began questioning whether the cloud premium was worth it for their use case. What followed was a carefully planned migration that would become a landmark case study in cloud repatriation.
The Challenge
- Annual AWS spend exceeding $3.2 million
- Mostly predictable workloads that didn't need elastic scaling
- Paying premium prices for commodity hardware
- Growing concerns about vendor lock-in
- Desire for more control over infrastructure
The Solution
Hardware Investment
37signals invested approximately $600,000 in hardware:[2]
- • Dell servers for compute workloads
- • High-performance NVMe storage arrays
- • Redundant networking equipment
- • Colocation in two geographically distributed data centers
Migration Strategy
The migration was executed in phases over 6 months:
- 1. Started with non-critical applications for testing
- 2. Built parallel infrastructure while maintaining AWS
- 3. Migrated databases with real-time replication
- 4. Gradually shifted traffic using DNS and load balancers
- 5. Maintained AWS as backup during transition
Technology Choices
- • KVM for virtualization instead of EC2
- • ZFS for storage instead of EBS
- • Open-source alternatives for managed services
- • Kamal (their open-source deployment tool) for orchestration
Results
Cost Savings
Reduced infrastructure costs from ~$300K/month to under $30K/month - a 90%+ reduction. The $600K hardware investment was recouped in about 6 months.[3]
Performance Gains
30% improvement in application response times due to dedicated hardware and elimination of virtualization overhead.
Operational Simplicity
Fewer moving parts, no complex AWS services to manage, and complete control over the stack.
Team Empowerment
The existing ops team managed the transition without additional hires, gaining valuable skills in the process.
Key Lessons
"The cloud is great for startups and variable workloads, but once you have predictable patterns, you're just paying a massive premium for flexibility you don't need."
- David Heinemeier Hansson, CTO[4]
What They Learned
- Predictable workloads don't need elastic infrastructure - Their traffic patterns were stable enough to capacity plan effectively
- Modern hardware is incredibly powerful - A single modern server can replace dozens of cloud instances
- Complexity has a cost - Simpler infrastructure is easier to debug and maintain
- Skills are transferable - Their team adapted quickly to managing physical infrastructure
Migration Timeline
Initial analysis and business case development
Hardware procurement and data center contracts
Infrastructure setup and initial application migrations
Database migrations and traffic shifting
Full cutover and AWS shutdown
Updated projection: $10M+ savings over 5 years
Industry Impact
37signals' transparent sharing of their cloud exit journey has inspired thousands of companies to reconsider their cloud strategies. Their blog posts, conference talks, and open-source tools have made cloud repatriation accessible to companies of all sizes.
The success has led to:
- Development of Kamal, their open-source deployment tool
- Widespread media coverage questioning cloud economics
- A movement of companies sharing their own repatriation stories
- Renewed interest in bare metal and hybrid strategies
Key Takeaways for Your Business
Analyze Your Workloads
If your traffic is predictable and doesn't need instant elastic scaling, you're likely overpaying for cloud.
Do the Math
Calculate the TCO including hardware, colocation, and staff time. The savings often exceed 50%.
Start Small
Begin with non-critical workloads to build confidence and expertise before tackling core systems.
Plan for Success
With proper planning and gradual migration, the risks are minimal and the rewards substantial.