The Pendulum Swings Back
For a decade, the industry mantra was "Cloud First." The assumption was that running your own servers was archaic — like a restaurant growing its own wheat. Every workload, regardless of its characteristics, was destined for AWS, Azure, or GCP.
But as cloud bills for mature companies started hitting $10M, $50M, or $100M per year, the math changed. Engineers who had celebrated "no more hardware" were now staring at invoices that dwarfed what their old data centers ever cost.
Cloud Repatriation (sometimes called "Unclouding" or "Cloud Exit") is the strategic move of migrating workloads away from public cloud providers and back to on-premise data centers or colocation facilities. It is not a retreat — it is a recalibration.
It's Not About Anti-Cloud
Repatriation isn't about saying "The Cloud is bad." It's about recognizing that the cloud is fundamentally a rental model, and like all rental models, it is priced to include a premium for flexibility. That premium is worth paying when you need the flexibility. It is waste when you don't.
- Cloud is great for: Startups, variable workloads, rapid experimentation, geographic expansion, and burst capacity.
- Cloud is expensive for: Steady-state, predictable, high-utilization workloads where you pay a flexibility premium you never exercise.
The modern approach is "Workload Optimal": run variable and unpredictable workloads in the cloud, and run stable, heavy, predictable workloads on your own hardware. The question is never "cloud or not cloud" — it is always "which workload belongs where?"