For the best part of a decade, “put it in the cloud” was close to a universal answer. It was rarely wrong, often cheaper than the alternative in the short term, and it let a lot of infrastructure decisions get made quickly without much debate.
That era of easy defaults appears to be ending, and the data behind it is more interesting than the “cloud repatriation” headlines usually suggest.
A quiet tip in the balance
Uptime Institute runs the largest recurring survey of data centre operators in the world, now in its sixteenth year. Its 2026 edition, published at the end of July, found something that hasn’t happened before across all sixteen years of data: third-party infrastructure, cloud, colocation, managed hosting and SaaS, now holds 46% of enterprise IT workloads, against 44% sitting in organisations’ own data centres. Off-premises has, for the first time, taken the larger share.
Read quickly, that sounds like cloud finally won outright. Read alongside older data, it tells a more precise story. IDC’s research from 2024 found that roughly 80% of organisations expected to repatriate at least some workloads back from the cloud within a year. Yet the same research found only around 8% intended to move everything back. Barclays’ CIO survey around the same period found 86% of CIOs planning to shift at least some workloads out of public cloud, the highest figure it had ever recorded, up from 43% just a few years earlier.
Put those numbers together and a genuinely different picture appears. Cloud isn’t losing ground overall, as Uptime’s own headline number shows. But the era of treating “cloud” as a single, universal answer is over. Organisations are increasingly making the decision workload by workload, not once for the whole estate.
Why this is the more expensive habit to break
I’ve sat in enough of these conversations to know that the costly mistake is rarely choosing cloud, or choosing on-premises. It’s choosing one answer and applying it everywhere without asking whether it fits the workload in front of you.
A handful of questions tend to separate the decisions that hold up from the ones that quietly bleed money for years. How predictable is the demand: a steady, always-on workload at consistent scale has a different economic answer to something spiky and unpredictable, which is exactly the case cloud elasticity was built to solve. How much data does it move, and to where: egress costs and data residency requirements shape the real economics far more than most cost models account for. Who else touches it: a workload used only by your own team behaves very differently to one shared across a global business. And critically, what does it cost you if the decision is wrong for a year before anyone notices: cloud lets you reverse a bad call cheaply, owned infrastructure usually doesn’t.
What good now looks like
The organisations getting this right in 2026 aren’t the ones that picked a side in the cloud debate years ago and have stuck to it since. They’re the ones treating placement as an ongoing decision rather than a one-off migration project, revisiting workloads periodically rather than assuming last year’s answer still holds.
That’s a genuinely different discipline to run. It means someone owning the question properly, not as an annual audit exercise, but as a live part of how infrastructure decisions get made. It also means being honest that the answer for a given workload might change again in eighteen months, and building in the flexibility to act on that rather than treating today’s placement as permanent.
If you asked your own technology function to justify, workload by workload, why each one runs where it currently does, I’d be curious how many could give you a confident, current answer rather than “that’s just where we put it originally.”
That question, more than any repatriation statistic, is probably the more useful one to be asking right now.
Sources:
Uptime Institute, 16th Annual Global Data Center Survey (published 30 July 2026).
IDC, “Assessing the Scale of Workload Repatriation” (Natalya Yezhkova, June 2024).
Barclays CIO Survey, Q4 2024.