The Repatriation Signal
Cloud repatriation — moving workloads from public cloud back to private cloud or on-premises infrastructure — is now mainstream. The Barclays CIO Survey found 86% of CIOs plan to repatriate some public cloud workloads, and IDC research shows 70–80% of companies repatriate at least some data each year. Around 67% of enterprises have already moved some workloads back, with the large majority planning further moves within 12–24 months.
This is not a cloud exodus. Only about 8% of organisations are moving entirely off public cloud; the rest are making selective, workload-by-workload decisions. Repatriation in 2026 is a sign of cloud maturity, not abandonment — enterprises are placing predictable, steady-state workloads where they are cheapest and keeping elastic, variable workloads in the cloud. That nuance matters for negotiation, and it sits alongside the competitive shifts we track in cloud market share and the broader market intelligence pillar.
The catalyst this cycle is AI. Training and steady-state inference are exactly the kind of high-volume, predictable compute that public cloud prices least competitively, and as AI workloads scale, the monthly bill crosses the threshold where owned hardware pays back within two to three years. Deloitte’s finding that on-premises AI can save 50% or more over three years, once token volume passes a certain point, is pushing AI infrastructure to the front of the repatriation queue — and giving CIOs a fresh, board-visible reason to revisit cloud commitments they signed when their workloads were smaller and more variable.
The headline figure to keep in view is the 86% of CIOs now planning some repatriation — a proportion unthinkable five years ago. It signals that questioning a cloud-first default is no longer contrarian but mainstream board-level practice, which makes the conversation with your provider far easier to open from a position of strength.
Which workloads come home
The repatriation decision is rarely all-or-nothing; it is a sorting exercise. Steady-state production databases, large-scale storage, predictable batch processing and high-volume AI inference are the usual candidates to move, because their demand curve is flat and the cloud premium on flat demand is highest. Spiky, seasonal or genuinely elastic workloads — and anything early in its lifecycle — stay in the cloud, where on-demand scaling earns its premium. A buyer who can show the provider exactly which 20–30% of the estate is a repatriation candidate is negotiating from evidence, not threat.
The Economics Behind the Move
The driver is cost. 84% of organisations cite managing cloud spend as their biggest cloud challenge, and roughly 27% of cloud infrastructure spending is wasted on under-used resources, per Flexera’s 2025 data. For workloads with consistent, predictable demand, the public cloud premium runs 30–50% above equivalent private infrastructure, and modern private cloud delivers 40–50% lower total cost of ownership for steady-state workloads.
| Repatriation data point | Figure | Negotiation implication |
|---|---|---|
| CIOs planning repatriation | 86% | A credible, common alternative — not a bluff |
| Steady-state TCO saving | 40–50% | Quantify the gap before renewal |
| Public cloud premium | 30–60% | Target committed-use discounts to close it |
| Cloud spend wasted | ~27% | Right-size before you commit or renew |
| Moving off entirely | ~8% | Selective exit is the realistic lever |
The case studies are concrete. After leaving AWS, 37signals reported around $2 million in annual savings — roughly $10 million over five years. For AI workloads specifically, Deloitte’s analysis found on-premises delivers 50% or more savings over three years once token volume crosses a threshold, a calculation increasingly relevant given the AI infrastructure surge in our IT spending forecasts.
Contract Clauses That Enable Exit
Repatriation only works as a strategy — and as leverage — if your contract permits a clean exit. The single biggest barrier is data egress cost: moving petabytes out of a hyperscaler can erase the savings if egress fees are not capped. Negotiate egress waivers or caps for documented migration, ideally before you sign, not when you are trying to leave. Equally important are committed-use terms that do not penalise volume reductions, defined assistance for data and workload extraction, and clear timelines for the return of your data in a usable format.
Portability is the other half of the equation. Workloads built on proprietary managed services are far harder to repatriate than those built on open, portable foundations — one of the strongest commercial arguments for the open source enterprise adoption trend. Designing for portability from the outset keeps repatriation credible and therefore keeps your provider honest at renewal.
The hidden cost most enterprises underestimate is the operating model. Public cloud bundles infrastructure, resilience and a great deal of operational labour into the bill; bringing workloads home means rebuilding some of that capability, whether through colocation, managed private cloud or in-house staff. A credible repatriation business case has to count those costs honestly — roughly 27% of cloud spend may be wasted on under-used resources, per Flexera’s 2025 data, but capturing that saving requires the discipline to right-size and operate the returned workloads well. The cases that disappoint are the ones that modelled hardware savings and ignored the people and resilience costs.
Repatriation as Renewal Leverage
A documented repatriation analysis is the cloud equivalent of a competitive bid. When your provider knows you have quantified the on-premises alternative — with real TCO numbers, a migration plan and egress costs modelled — committed-use discount authority expands quickly. The threat does not need to be a full exit; a credible plan to move 20–30% of steady-state workloads is usually enough to reset the conversation, the same dynamic we see across enterprise software market trends.
FinOps discipline determines whether the savings are real. Before modelling repatriation, strip out the roughly 27% of cloud spend that is simply waste — idle instances, over-provisioned reservations, unattached storage — because moving waste on-premises just relocates it. A clean baseline of genuinely needed, steady-state capacity is the only fair comparison against owned infrastructure, and it often reveals that right-sizing alone captures much of the saving without any migration at all.
The negotiation value holds even when you do not move. Providers know that a customer who has done the repatriation maths is a customer who can leave, and that knowledge alone expands committed-use discount authority. The strongest position is to present the analysis as a fact, not a threat: here is our steady-state workload, here is its on-premises cost, here is the gap your committed-use pricing needs to close. Framed that way, repatriation becomes a benchmark the provider must beat rather than an ultimatum it can dismiss.
Timing the analysis to the contract clock matters as well. A committed-use agreement signed for three years locks your cost base for that period, so the moment to model repatriation is in the six months before that commitment expires — not after you have renewed. Building the on-premises comparison into the standard pre-renewal checklist, the way benchmarking is built into a software renewal, ensures the option is live every cycle rather than rediscovered too late to act on.
The disciplined play is to model repatriation before every major cloud renewal, whether or not you intend to execute it. The analysis pays for itself either way: it either lowers your cloud bill through better committed-use terms, or it identifies workloads genuinely cheaper to bring home. Our cloud contract framework sets out the egress, portability and commitment clauses to negotiate, and you can request a confidential briefing on building a credible repatriation position for your next renewal.