Cloud Bills Rising? How to Decide Whether to Move Back On-Premises

The finance team reviews cloud bills and projects costs several years ahead. As workloads grow, so do charges. If those costs are rising much faster than the business, an approaching contract renewal can prompt a question: should we keep using the cloud?

Some companies are moving from the cloud back to infrastructure they operate themselves. But a high bill does not mean moving every system will save money. The deciding factors are how predictable usage is and how much waste can be cut from the bill.

How 37signals Reassessed Millions of Dollars in Cloud Spending#

37signals, the company behind HEY and Basecamp, moved to the public cloud following the launch of Amazon S3 in 2006. At its peak, the company was spending millions of dollars a year on cloud services. It began reducing its AWS workloads in 2022 and, by 2023, had moved from AWS and Google Cloud to hardware it owned.

The company puts its migration cost at $500,000 for hardware, while saying it eliminated more than $3.5 million in annual cloud spending. Another key detail: it placed its servers in colocation facilities instead of building its own data center. Its infrastructure uses tools including KVM, Docker, and Kamal.

Of course, these figures cannot be applied directly to another company. Server requirements and migration labor costs vary from business to business. What matters is comparing more than the monthly cloud bill, including hardware purchases, maintenance, and the staffing needed to operate the infrastructure.

Other companies have also raised concerns about rising cloud costs. Rebecca Weekly, vice president of infrastructure at U.S. auto insurer GEICO, says its cloud spending rose to 2.5 times its original level over a decade of migration to public clouds, including Azure.

SEO software company Ahrefs, meanwhile, says it has never used a public cloud such as AWS and estimates that it has saved more than $400 million since 2020. That is the company’s own estimate; other businesses would not necessarily achieve the same results.

The Bill Alone Does Not Explain the Cost#

A single large bill is not always what triggers a reassessment of cloud use. According to Spencer Kimball, CEO and co-founder of Cockroach Labs, contract renewal is a common turning point. When finance teams forecast spending over several years, they sometimes discover that costs are growing faster than workloads.

In a 2025 study of 2,100 organizations, Cast AI found that Kubernetes clusters running on AWS, Google Cloud, and Azure used an average of 10% of their allocated CPU resources and 23% of their allocated memory. The findings show substantial unused capacity in the configurations surveyed. They do not mean every company has the same proportion of idle resources.

Sometimes, simply deallocating unused servers can lower costs. Kunal Agarwal, CEO of Unravel Data, says investigating the gap between usage and billing can make a discussion about moving off the cloud unnecessary. Migration requires preparation and carries risks. Clearing out wasted resources first may deliver comparable savings without a move.

Server charges are not the only costs that are easy to overlook. It is also worth checking egress charges for moving data out of the cloud, transfer fees between regions, and platform licensing fees that do not match actual usage. The UK’s Competition and Markets Authority (CMA) is investigating Microsoft over cost barriers faced by companies seeking to switch to another cloud provider.

Predictable Workloads Are Stronger Candidates for In-House Infrastructure#

One factor in deciding whether a return to on-premises infrastructure makes sense is how predictable workload demand is. Kimball notes that owning infrastructure can be cheaper when a company can forecast demand a year ahead and maintain high utilization.

Conversely, the flexibility of the public cloud helps with workloads whose usage fluctuates sharply or systems that need to scale quickly as the business grows. It is therefore more practical to assess suitability workload by workload than to switch the entire company’s environment at once.

Platform9 says it moved roughly 40% of its workloads to an on-premises environment after cloud resource costs surged in December 2025. According to the company, running its development, testing, and lab environments on-premises reduced their costs by 70%. That result is specific to Platform9; other companies would not necessarily achieve the same savings. Environments where service-level agreements (SLAs) are less critical could be a useful starting point for identifying migration candidates.

Gather the Numbers Before You Move#

To assess whether to move back on-premises, work through the following steps:

• Break down the past 12 months of bills into compute, storage, data transfer, and licensing.
• Check actual CPU and memory usage, and remove resources that remain allocated but unused.
• Document demand fluctuations, SLAs, and data location requirements for each workload.
• Compare the total cost of staying in the cloud with operating infrastructure in-house over three to five years. Include maintenance, electricity, networking, and staff costs as well as hardware.
• If you decide to migrate, start with a limited scope, such as development and testing environments. Measure the operational effort and actual costs before expanding.

When comparing costs, avoid simply projecting today’s cloud bill into the future. Using the cost after unused resources have been removed as your baseline makes it easier to judge whether optimization is enough or a migration is needed.

Avoid Treating Cloud and On-Premises as an Either-Or Choice#

A report published by Andreessen Horowitz in 2021 analyzed cloud spending at 50 leading publicly traded software companies and estimated that cloud costs had reduced their combined market value by roughly $100 billion. Extrapolating to the entire publicly traded software market, it estimated at least $500 billion in wasted market value compared with running the same workloads on-premises. These are estimates from the report, not figures showing how much any individual company could save.

The cloud is a strong option for companies beginning a digital transformation. For workloads with stable usage, or systems that can maintain high utilization, operating infrastructure in-house can offer a cost advantage. The key is to assess each workload using actual bills and usage data, rather than making a single cloud-or-on-premises decision for the whole business.

Before your next contract renewal, start by reviewing the past 12 months of bills alongside CPU and memory utilization. Remove any unused resources you find. Then, for the workloads whose costs still do not make sense, compare the alternatives—including the cost of migration.

Source: https://japan.zdnet.com/article/35253356/