Maybe some of your cloud workloads should come home

For a long time, moving to cloud was treated almost like a measure of progress. I think that made some sense when companies were fighting slow hardware procurement and wanted to ship faster. It becomes less obvious when a workload has been running at roughly the same capacity for five years.

Flexera’s 2026 cloud report shows that companies are still moving plenty of work into public cloud, while also reevaluating what they keep there. That is not a contradiction. Different workloads have different economics.

Start with the boring servers

Imagine a database and application service that run all day, every day, with relatively predictable traffic. Compare the complete monthly cloud cost to the real cost of owning hardware, including replacement, power, storage, support and someone to operate it. Sometimes owning the infrastructure wins. Sometimes the cloud still wins decisively, especially when you use managed services the company could not reasonably operate itself.

I also would not move an application solely because the compute is cheaper. If it talks constantly to cloud storage or managed databases, you might create new network costs and failure modes. You have to look at the full system.

My preference is to avoid ideological arguments about cloud. Keep what benefits from cloud in cloud. Move the predictable pieces when the numbers and operational requirements justify it. A hybrid environment is not a failure to commit.

The hidden cost isn't always the monthly bill

There's a reason people like the cloud. It gives you fast provisioning, managed databases, backups and a way to add capacity without planning a hardware order. When someone proposes leaving, I want to know which of those capabilities they'll have to replace and who will own the replacement.

I'd start by separating workloads that change constantly from ones with predictable demand. A stable internal service running all year might be a good candidate for dedicated infrastructure. A seasonal workload, a new product or something that needs global distribution may be a poor one. The same company can have both, and that's not a failure of planning.

Moving back also doesn't have to mean building a private data center from scratch. Colocation, managed bare metal and hybrid options can be worth comparing. The question is not whether owning servers is inherently more serious than renting them. It's where the operational burden and economics fit the business.

A spreadsheet showing a smaller hosting bill is a starting point. I'd still insist on calculating backup, recovery, network, labor and the next hardware refresh before calling it a win.