The Great Repatriation: Why Half of Enterprises Pulled Workloads Back — and What It Means for UK SMEs

A retailer we spoke with last year had a monthly cloud bill that looked like a phone number. Their finance director had stopped reading the itemised breakdown because it changed every month and never in the right direction. The kicker: most of that spend came from workloads that ran the same way, at the same volume, every single day. They were paying elastic prices for something that never flexed.

They are not alone. Survey after survey now shows the same pattern — Barclays, IDC and others have all reported that somewhere around half of enterprises have moved at least some workloads out of public cloud and back onto private or on-premises infrastructure. A few years ago this would have been treated as heresy. Cloud-first was gospel, and moving anything back was an admission of failure. That has changed, and it has changed for good reasons.

This is not the old cost argument

Repatriation used to be a story about bill shock. A company would lift-and-shift a legacy application into AWS or Azure, forget to right-size anything, and get walloped by egress fees and idle instances. The fix was usually better housekeeping, not a wholesale move back.

What is happening now is different, and it comes down to three pressures that have all intensified at once.

AI infrastructure costs. GPU compute in public cloud is eye-wateringly expensive, and demand has made it scarce. Firms that started training or running models in the cloud quickly found that steady, heavy AI workloads are exactly the kind of predictable, high-utilisation task that public cloud prices badly. If you are running GPUs at 80% utilisation around the clock, owning or leasing dedicated hardware often costs a fraction of the on-demand rate. The economics that made public cloud attractive for bursty, unpredictable work invert completely for sustained AI inference and training.

Compliance pressure. UK SMEs in healthcare, legal, finance and the public sector supply chain are being asked harder questions about where data physically lives and who can touch it. Data residency, the fallout from Schrems II, and sector-specific rules around patient and financial records mean that "it's in the cloud somewhere in Europe" is no longer a good enough answer. A private cloud in a known UK data centre gives a clear, defensible answer to an auditor.

Performance and predictability. Latency-sensitive applications, large databases, and anything that shuffles enormous amounts of data between services can perform inconsistently on shared public infrastructure. And egress charges — the fee for moving data out — punish exactly the workloads that need to move a lot of data. For some firms, dedicated hardware simply runs faster and behaves the same way every day.

Which workloads are actually candidates

The worst thing an IT partner can do is treat repatriation as an all-or-nothing decision. It almost never is. The right approach is to sort workloads by their characteristics and move only the ones where the maths and the risk profile point the same way.

Good candidates for repatriation tend to share a few traits:

• Steady, predictable load. Anything that runs at a consistent level rather than spiking. Line-of-business applications, internal systems, established databases.
• High data volume with lots of movement. Where egress fees add up fast.
• Strict data residency or compliance requirements. Where knowing the physical location matters more than global reach.
• Sustained AI inference. Models serving predictions at a steady rate, once you know your utilisation.
• Legacy systems that never needed to be elastic. The ones that got lifted-and-shifted without any real re-architecting.

Workloads that should usually stay in public cloud are the mirror image: spiky and unpredictable traffic, seasonal peaks, greenfield projects still finding their shape, anything that genuinely benefits from global distribution, and services where a managed platform saves you from running the plumbing yourself. A seasonal e-commerce site that quadruples traffic on Black Friday is exactly what public cloud was built for. Do not drag that back.

Most SMEs end up somewhere in the middle — a hybrid setup where the steady, sensitive, heavy workloads live in private cloud and the variable, experimental ones stay public. That is not a compromise. That is the point.

Managed private cloud is not a step backward

The reason repatriation still carries a stigma is that people picture it as going back to a cupboard full of ageing servers, a tangle of cables, and one member of staff who is the only person who understands how any of it works. That version of on-premises deserved to die.

Managed private cloud is a different thing entirely. It is dedicated infrastructure — sometimes in your building, more often in a professional data centre — run with the same automation, monitoring, and self-service tooling that made public cloud pleasant to use, but on hardware you control and with costs you can predict. You get virtualisation, provisioning on demand, backups, patching and security handled as a service. What you do not get is a bill that surprises you every month or a support queue that treats your SME like account number 4,000,000.

For an IT services firm, this is where the real work sits. The value is not in racking servers. It is in helping a client understand their own workloads well enough to make good decisions, and then running the result reliably.

What IT services firms should build now

If you are an IT partner — or an SME choosing one — here is what a credible repatriation capability looks like.

A proper assessment process. Before anything moves, you need real data: utilisation figures, data transfer volumes, actual monthly spend broken down by workload, and the compliance requirements attached to each system. Most repatriation regrets come from decisions made on gut feel rather than numbers. Build a repeatable way to gather and read that data.

Honest total-cost-of-ownership modelling. Public cloud has hidden costs, but so does private. You have to account for hardware refresh cycles, data centre space, the staff time to manage it, and the migration effort itself. A model that ignores those is just as misleading as an unmonitored cloud bill. Show clients the three-year picture, not the month-one saving.

Hybrid networking that actually works. Almost no one goes fully private. The skill is connecting private and public environments cleanly, with secure links, consistent identity management, and sensible data flows that do not rack up egress charges by accident.

Migration discipline. Moving a workload back is a project with real risk. Test environments, staged cutovers, rollback plans, and clear success criteria all matter. Treat it with the same seriousness as the move into cloud, because it is the same size of task.

Ongoing management. Dedicated infrastructure only stays a good deal if someone keeps it patched, secure, backed up and monitored. That recurring service is what turns repatriation from a one-off gamble into a stable arrangement.

The sensible conclusion

Repatriation is not a rejection of cloud. It is the market growing up and realising that different workloads belong in different places. The firms doing well right now are the ones who stopped treating "cloud" as a destination and started treating it as a set of trade-offs to be matched, workload by workload, to what the business actually needs.

For UK SMEs, the practical question is not "should we go back?" but "which of our systems are quietly costing us more than they should, and where would they run better?" Answer that honestly, with real numbers, and the right mix usually becomes obvious. If you would like help working through your own workloads, that is exactly the kind of conversation we have every week.