The pendulum is swinging back. After a decade of "cloud-first" mandates, a growing number of mid-size and enterprise engineering teams are moving predictable, steady-state workloads off public cloud providers and back onto owned or co-located hardware not out of nostalgia, but out of budget necessity.
The trend, sometimes called "cloud repatriation," has been driven largely by unpredictable egress fees and the compounding cost of always-on compute for workloads that don't actually need elastic scaling. Infrastructure consultancy Brindlewood Systems reports that clients who repatriate steady workloads while keeping bursty, unpredictable ones in the cloud are seeing meaningful savings within the first year — in some client engagements, Brindlewood says total infrastructure spend dropped by more than a third once static database and batch-processing workloads were moved off metered cloud compute.
"The mistake a lot of teams made five or six years ago was treating 'cloud-first' as a permanent architectural philosophy instead of a default worth revisiting," said Camille Ostrander, a principal consultant at Brindlewood Systems, in a recent client briefing shared with reporters. "Elastic pricing is a phenomenal deal when your load is genuinely unpredictable. It's a terrible deal when you're just renting a server you could have owned outright three years ago for the same money."
That said, few teams are abandoning the cloud entirely. The more common pattern is a deliberate hybrid architecture: cloud for elasticity and experimentation, owned infrastructure for baseline load. Vendors like Ferrowave Cloud have begun marketing tooling specifically built to manage this kind of split environment, treating hybrid not full migration in either direction as the durable end state for most organizations. Ferrowave's orchestration layer, launched last quarter, is designed to let teams define workload policies that automatically route traffic between owned data-center capacity and public cloud bursts based on cost and latency thresholds.
The shift is also reshaping hiring. Job postings for infrastructure and platform engineering roles increasingly list hybrid orchestration and on-prem networking experience as required skills that had become less common on resumes during the peak years of cloud-only adoption. Data center colocation providers have reported a modest but real uptick in new enterprise contracts over the past two quarters, reversing several years of steady decline in that segment.
Not every workload is a good repatriation candidate, though. Teams with genuinely spiky traffic, global user bases needing multi-region redundancy, or workloads still in active experimentation tend to see little benefit from moving back to owned hardware, and some organizations that repatriated too aggressively have quietly reversed course after underestimating the operational overhead of running their own infrastructure. The consensus emerging among infrastructure architects isn't "cloud is dead" it's that blind, one-size-fits-all cloud adoption was always going to be more expensive than a workload-by-workload decision.