The repatriation question in 2026.

Here is a framework for which workloads actually pay to leave, the costs buyers underestimate, and why optimisation beats repatriation more often than the headlines suggest.

Why optimisation usually beats leaving

The cloud is expensive when it is used as if it were a datacenter: instances sized for peak and run at idle, no commitment coverage, storage left on hot tiers, and architectures that pay premium rates for data movement. An estate run this way can look two or three times more expensive than owned hardware, and repatriation looks compelling. But the comparison is unfair, because it pits an untuned cloud estate against a hypothetical efficient datacenter. Tune the cloud estate first and the gap usually narrows or closes. Rightsizing removes the idle capacity you are paying for, commitment coverage through Savings Plans, Reservations, CUDs, or Universal Credits cuts the rate by twenty to seventy percent on steady workloads, storage tiering moves cold data off expensive tiers, and architecture changes remove the quiet data transfer and gateway charges. After that work, the workloads that still look expensive on the cloud are the real repatriation candidates, and there are far fewer of them.

Which workloads actually pay to leave?

Repatriation economics favour a specific profile, and it is worth being precise about it:

• Steady, high utilisation workloads that run near full all the time, because owned hardware is only cheap when it is busy and the cloud advantage is elasticity you are not using.
• Data heavy workloads with strong gravity, where egress and storage at scale dominate the bill and owning the storage avoids per gigabyte cloud rates.
• Predictable workloads with a long horizon, because hardware is a multi year capital commitment that only pays back if demand is stable.
• Workloads with control or residency requirements that owned infrastructure satisfies more cheaply than the equivalent cloud controls.

Workloads that are variable, bursty, seasonal, or early in their life almost never pay to repatriate, because the cloud lets you stop paying for idle capacity and owned hardware does not. The same logic explains why most AI inference, with its uneven and fast changing demand, belongs on cloud or committed cloud capacity rather than on a fixed fleet.

What are the hidden costs buyers underestimate?

The headline comparison of cloud rate against hardware amortisation is the easy part. The costs that erase the saving are the ones buyers forget. Egress charges to move data out of the cloud can be substantial for a large estate, and they are a one time tax on leaving that must be weighed against the recurring saving. Hardware carries not just purchase cost but refresh cycles, spares, and the risk of buying for a demand forecast that proves wrong. Then there is the operational burden the cloud was quietly absorbing: patching, power, cooling, physical security, capacity planning, and the staff to do it. Colocation shifts some of this but not all. An honest repatriation case includes every one of these lines, and when it does, the set of workloads that still pay to leave shrinks further.

How do you decide without guessing?

Treat repatriation as a placement decision per workload, not an estate wide verdict. For each candidate, build the fully optimised cloud cost as the baseline, then the all in repatriated cost including egress, hardware refresh, colocation, and operations, over the same multi year horizon. Compare like for like, and require the repatriation case to win clearly rather than marginally, because the move carries execution risk the spreadsheet does not show. The independence point matters here. A provider has no incentive to tell you a workload should leave its cloud, and a hardware vendor has every incentive to tell you it should. A buyer side advisor sized the saving on either side of the table is the only party whose recommendation follows the economics rather than a commission, which is the position we occupy.

Answer the repatriation question with us

We model repatriation as a per workload placement decision across AWS, Azure, GCP, OCI, and owned infrastructure, with the fully optimised cloud cost as the honest baseline. As an independent buyer side advisor we take zero provider commissions, so the recommendation follows the economics. Our guarantee: we reduce your cloud spend or we reimburse our service fee, on a Fixed Fee or a no risk Gainshare basis.

Our cross cloud cost guide sets placement in the context of every other lever, and we share new analysis through The Cloud Spend Navigator.