Cloud Repatriation: Moving Off AWS and Azure

For over a decade, the business world operated under a singular directive: move everything to the cloud. The promise was infinite scalability and lower costs. However, a significant shift is occurring in 2024 and 2025. Chief Information Officers (CIOs) are discovering that the rent bill for the public cloud has become unsustainable. By moving workloads back to on-premise private servers—a process known as cloud repatriation—companies are realizing millions of dollars in annual savings.

The Economic Reality: Renting vs. Owning

The initial allure of Amazon Web Services (AWS), Microsoft Azure, and Google Cloud was the ability to convert Capital Expenditures (CapEx) into Operating Expenditures (OpEx). You didn’t need to buy $50,000 servers; you just rented them by the hour.

This model works perfectly for startups with unpredictable traffic. However, for established companies with stable, predictable workloads, the math has flipped. Renting a server instance in the cloud 24⁄7 often costs three to five times more over a three-year period than buying equivalent hardware.

The High Cost of Compute

When you pay for an AWS EC2 instance, you are paying a premium for flexibility you might not be using. If your application runs at a steady rate, you are effectively paying hotel prices for a long-term apartment lease.

The Egress Fee Trap

The hidden killer in cloud pricing is data egress fees. Cloud providers charge minimal fees to upload data, but they charge heavily to take it out or move it between regions. For data-intensive businesses, these transfer fees can account for 20% to 30% of the total monthly bill. Moving back to a colocation facility (a data center where you rent space and power but own the servers) usually eliminates per-gigabyte egress fees in favor of flat-rate bandwidth connections.

Real-World Case Study: 37signals

The most prominent example of successful cloud repatriation comes from 37signals, the company behind Basecamp and the email service HEY. In 2023, CTO David Heinemeier Hansson (DHH) announced a strategic exit from the cloud.

The company was spending roughly $3.2 million annually on AWS. By moving their applications to their own hardware hosted in two data centers, they projected savings of $7 million over five years.

The Hardware Numbers: To replace the cloud capacity, 37signals purchased Dell R7625 servers. These powerful machines, equipped with AMD EPYC processors, cost significantly less upfront than the cumulative rental fees of equivalent cloud instances.

  • Cloud Cost: Continued rental of AWS services would remain high indefinitely.
  • Repatriation Cost: A one-time hardware investment of roughly $600,000 completely replaced a massive chunk of their monthly operational spend.

The result was an immediate reduction in operational complexity and a dramatic increase in profit margins.

The Ahrefs Example: Value Through Hardware

Ahrefs, a popular SEO toolset that crawls the entire web, has long championed on-premise infrastructure. They maintain a massive server farm in Singapore to handle petabytes of data.

According to Ahrefs’ internal calculations, running their infrastructure on AWS would cost roughly $400 million per year. By owning their hardware and colocation space, their actual costs are a tiny fraction of that figure. This efficiency allows them to price their product competitively while maintaining high margins, something that would be mathematically impossible if they relied on public cloud infrastructure.

Identifying Workloads for Repatriation

Not every application should leave the cloud. The “Cloud Smart” approach suggests keeping certain elements in AWS or Azure while moving the heavy lifting on-prem.

Keep in the Cloud:

  • Variable Traffic: If you run a ticket sales site that stays quiet for months but explodes during a Taylor Swift concert sale, the cloud’s elasticity is essential.
  • Global Content Delivery: CDNs (Content Delivery Networks) like Cloudflare or AWS CloudFront are unmatched for serving static assets globally.
  • Experimental Projects: If you don’t know if a product will survive six months, do not buy hardware for it.

Move to On-Prem:

  • Core Databases: Large, steady databases often run faster and cheaper on bare metal NVMe storage you control.
  • Continuous Integration (CI) Runners: Automated testing environments often run 24⁄7. Paying per-minute cloud rates for these is inefficient.
  • Large Media Archives: Storing petabytes of video on AWS S3 is convenient but notoriously expensive compared to buying high-density storage servers.

The Operations Gap

The primary argument against repatriation is the need for a team to manage hardware. Critics argue that the money saved on AWS bills is simply lost to hiring engineers.

However, modern data center management has evolved. Tools like Kubernetes allow teams to manage bare metal servers with the same ease as cloud instances. The gap between “cloud” and “on-prem” operations has narrowed. 37signals, for example, did not increase the size of their operations team despite moving completely off the cloud. They simply shifted the team’s focus from managing AWS complexity (IAM roles, VPC peering, cost optimization tools) to managing Linux servers.

Making the Switch: A Strategic Approach

If your organization spends over $500,000 annually on cloud services, an audit is necessary. Look for line items labeled “Compute” and “Data Transfer.” If these are high and your usage is flat (meaning it doesn’t spike wildly up and down), you are a prime candidate for repatriation.

You do not need to build a data center. You can rent a rack in a Tier 3 colocation facility (like those operated by Equinix or Digital Realty). They provide the power, cooling, and security. You simply slide your Dell or HPE servers into the rack and connect them to the internet.

Frequently Asked Questions

Is cloud repatriation the same as a hybrid cloud? Not exactly. Hybrid cloud implies using both simultaneously. Repatriation is the specific act of moving a workload permanently away from the public cloud to private infrastructure to reduce costs. Many companies end up with a hybrid model after repatriating their core workloads.

Does moving off the cloud hurt reliability? It can actually improve it. When you rely on AWS, you are subject to their outages (like the famous us-east-1 failures). When you control your hardware, you control your redundancy. However, you are responsible for setting up that redundancy, such as having servers in two geographically distinct locations.

What is the “Cloud Exit” strategy? This is a regulatory or strategic plan required by some industries (like banking) to ensure they can continue operations if a cloud provider bans them or fails. Repatriation is the execution of a cloud exit strategy.

Are reserved instances a solution to high costs? AWS and Azure offer “Reserved Instances” where you commit to 1 or 3 years of usage for a discount. While this lowers the bill, it is usually still more expensive than owning the hardware outright. Furthermore, it locks you into a specific instance type, reducing your flexibility.