$10 Million Saved From Leaving The Cloud
Based on The PrimeTime's video on YouTube. If you like this content, support the original creators by watching, liking and subscribing to their content.
37signals moved seven applications off AWS onto its own hardware, with contract roll-off timing making 2024 the first fully measurable savings year.
Briefing
Seven cloud applications were moved off AWS and onto 37signals’ own hardware, and 2024 became the first full year where the savings were clean enough to measure. The shift began with long-term contract commitments that only fully rolled off by the end of 2023, but the payoff showed up in 2024: the annual cloud bill dropped from an original estimate of $3.2 million to $1.3 million—about $2 million per year saved.
The company credits the stronger-than-expected results to practical deployment choices. New equipment was placed into existing data center racks and worked within power limits, avoiding major infrastructure expansion costs. It also points to a key financial nuance: “on-prem” isn’t free, because hiring and managing engineers, plus the overhead around HR, insurance, legal, office space, and benefits, can multiply the true cost of running infrastructure. Even so, the company argues that the savings can still fund a capable team—especially when planning ahead and scaling responsibly.
Hardware spending was about $700,000 for the new gear, and that cost was fully recouped during 2023 as the longer cloud commitments gradually ended. The company frames the remaining cloud spend as increasingly narrow and contract-driven: after the exit, the $1.3 million still spent on AWS is largely for S3 file storage. At the time of the update, 10 petabytes of data sat in S3, including critical customer files stored in duplicate across separate regions. Storage costs were optimized using a mix of storage classes to balance reliability, access, and price, but the bill still exceeded $1 million annually even after long-term commitment discounts.
The next phase is a planned move away from S3. When the current 4-year file storage contract expires next summer, 37signals expects to deploy a dual data-center, peer storage setup with 18 petabytes of capacity. The initial hardware for that setup is expected to cost about the same as one year of AWS S3, but ongoing costs should be limited to service contracts because the company will use high-density, power-efficient Pure Storage Flash arrays. From there, it projects an additional $4 million in savings over five years.
Taken together, the company projects total savings from the cloud exit to be more than $10 million over five years—while also gaining faster compute and more storage. It also pushes back on a common fear: that leaving the cloud would force a bigger team. The same group that previously operated the apps in AWS is now running them on the company’s own hardware, and the company says the exit didn’t require changing team composition.
Finally, the company acknowledges that cloud can still make sense in early-stage uncertainty or for highly variable workloads, but argues that once cloud bills become substantial, basic cost math and staged migration—starting with one service or moving to a VPS before going fully on-prem—can reveal whether the “all cloud” approach has become more expensive than it needs to be. The company says it plans to delete its AWS account after the remaining migration completes.
Cornell Notes
37signals moved seven applications off AWS onto its own hardware and says 2024 was the first year where savings were fully measurable. Cloud spending fell from an original $3.2M/year estimate to $1.3M in 2024, saving about $2M annually, helped by fitting new hardware into existing data center racks and power limits. The remaining cloud cost is mostly AWS S3 for about 10 petabytes of duplicated, critical customer files. After a 4-year S3 contract expires next summer, the plan is to deploy a dual data-center peer storage setup using Pure Storage Flash arrays, with 18 petabytes capacity and modest ongoing service-contract costs. The company projects total savings of more than $10M over five years while keeping the same operations team.
Why did the savings show up clearly in 2024 rather than immediately after the migration began?
What specific factor made the on-prem cost outcome better than the original savings estimate?
How does the company justify that on-prem costs won’t be swallowed by staffing overhead?
What portion of remaining AWS spending is still in use, and what drives it?
What changes are planned once the S3 contract expires, and how does that affect cost?
What migration approach does the company suggest for others considering leaving the cloud?
Review Questions
- What were the two main reasons 2024 became the first “clean” year for measuring savings?
- How does the company’s S3 usage (data volume, duplication, and storage classes) connect to its remaining AWS bill?
- What assumptions underlie the projected additional $4M savings after the S3 contract expires next summer?
Key Points
- 1
37signals moved seven applications off AWS onto its own hardware, with contract roll-off timing making 2024 the first fully measurable savings year.
- 2
Cloud costs dropped from an original $3.2M/year estimate to $1.3M in 2024, saving roughly $2M annually.
- 3
On-prem transitions still require real overhead—management, HR, insurance, office space, and benefits—but the company argues savings can fund a capable team.
- 4
The remaining AWS spend is largely AWS S3 for about 10 petabytes of duplicated, critical customer files optimized with multiple storage classes.
- 5
After the current 4-year S3 contract expires next summer, the plan is a dual data-center peer storage setup with 18 petabytes using Pure Storage Flash arrays.
- 6
Hardware spending (~$700,000) was recouped during 2023 as longer cloud commitments rolled off, supporting the longer-term savings thesis.
- 7
The company says it didn’t expand team composition after the cloud exit; the same operations group runs the systems on-prem.