Overview
The cloud was sold as obviously cheaper and simpler. For many workloads it is β but at scale, cloud bills can become one of a company's largest expenses, and a notable few have moved workloads back to their own infrastructure. This report explains the real economics.
What the cloud is great at
The cloud's core value is elasticity: you pay for what you use and can scale up or down instantly. For startups, spiky traffic, experimentation, and unpredictable demand, that flexibility is worth a premium β you avoid buying hardware you might not need and you ship faster. Early on, cloud is usually the right call.
Where bills explode
The same elasticity that helps early can hurt at steady-state scale. When you run large, predictable, constant workloads 24/7, you pay a continuous premium for flexibility you no longer need. Data egress fees, managed-service markups, and idle-but-provisioned resources compound. At scale, the convenience premium can become enormous β sometimes multiples of owning the equivalent hardware.
The repatriation trend
A handful of high-profile companies (famously, some at large scale) publicized big savings by repatriating steady workloads from cloud to their own or leased servers. It's not a stampede, and it's not right for everyone β but it punctured the assumption that cloud is always cheapest. For the right workload, owning hardware can cut costs dramatically.
The honest answer: it depends on the workload
This isn't "cloud bad" or "on-prem good." It's workload-by-workload:
- Variable, spiky, unpredictable, early-stage β cloud usually wins.
- Large, steady, predictable, mature β owned/leased hardware can be far cheaper. Most mature companies end up hybrid, matching each workload to the cheapest viable home.
What this means for you
Treat cloud spend as a managed cost (FinOps), not a fixed fact. Monitor it, kill idle resources, watch egress, and periodically ask whether your steady, heavy workloads belong in the cloud at all. The default of "everything in the cloud forever" is not automatically optimal.
Honest limits
Repatriation has real hidden costs too β hardware, ops staff, data-center risk, and lost agility. The cloud premium often buys speed and focus that are worth it. The point is to decide, with the actual numbers, rather than assume.
