Ten years ago the advice was simple: move everything to public cloud and never look back. In 2026 that advice has quietly reversed for a large slice of the workloads NZ organisations run. Cloud repatriation — moving workloads off public cloud back to private or hybrid infrastructure — is now a mainstream planning exercise, not a fringe reaction. The intent is striking: 89% of ANZ organisations are planning some degree of repatriation, according to research reported by IT Brief.
But “repatriate everything” is the wrong lesson to take from that number. The right move is workload-by-workload triage. Some things genuinely belong in hyperscale public cloud. Others have been quietly bleeding budget and control for years. This guide gives NZ IT managers and CIOs a framework for telling the two apart.
What cloud repatriation actually means
Cloud repatriation is the deliberate relocation of a workload from a public cloud provider (AWS, Azure, Google Cloud) to private cloud, on-premises, or colocated infrastructure. It is not anti-cloud. Most NZ organisations doing it end up multi-hybrid — keeping public cloud for the workloads it suits, and pulling back the ones where the economics or compliance no longer add up.
The driver is rarely a single bill shock. It is the slow realisation that some workloads were never a fit for consumption pricing. A steady-state database running at 70% utilisation around the clock pays the same elasticity premium as a workload that scales from zero to thousands of cores overnight — except only one of them uses that elasticity.
Why the shift is happening now in NZ
Three things changed at once.
Cost predictability eroded. Egress charges, inter-region transfer, and per-API pricing make public cloud bills hard to forecast — and almost always harder than the original business case assumed. For data-heavy workloads, transfer alone can run 15–30% of the total, which is the kind of line item that triggers a finance review.
AI changed the maths. Continuous GPU workloads in public cloud carry a heavy data-gravity problem: moving terabytes of training data in and out of a cloud region becomes both a bottleneck and a cost centre. NZ organisations are responding — 57% of ANZ firms now plan to invest in on-premises infrastructure for AI, per the same IT Brief research, keeping at least part of their AI compute inside their own estate.
Data sovereignty got teeth. The Privacy Act 2020 and NZISM (the New Zealand Information Security Manual) push public-sector and regulated organisations toward knowing exactly where their data lives. Hyperscalers now run local regions — Microsoft opened its Azure New Zealand North region in Auckland, and AWS has its Asia Pacific (Auckland) region — which helps. But a local region is not the same as a contractual, auditable sovereignty guarantee with a NZ entity, and for some agencies that distinction is the whole point.
The workloads worth moving back
Here is where repatriation tends to pay off for NZ organisations, ranked by how often the case holds up under scrutiny.
Steady-state compute is the clearest win. A regional council running a finance system or a property database at constant load is paying elasticity premiums for elasticity it never uses. Predictable, always-on workloads run more cheaply on owned or private-cloud capacity sized to their actual demand.
Sovereignty-bound data comes next. Health records, student information, and government-classified data carry residency and auditability obligations that are simpler to satisfy on infrastructure with a documented NZ footprint. When a CISO has to evidence where data sits for a NZISM audit, “in a private NZ cloud with a named operator” is an easier sentence to write than untangling a global provider’s shared-responsibility model.
Continuous AI workloads are the newest entry. If you are running inference around the clock or training on large local datasets, repatriating to NZ-resident GPU capacity sidesteps the egress trap and keeps training data in-country. ASI’s
GPU as a Service
runs NVIDIA hardware (H200, L40S, RTX Pro 6000) inside New Zealand, which removes the data-gravity penalty entirely for locally-held datasets.
The workloads to leave in public cloud
Repatriation has a failure mode: dragging back a workload that was genuinely well-suited to hyperscale, then discovering you now own a scaling problem you used to rent away.
Keep public cloud for genuinely spiky demand — a ticketing site for a one-off event, a tax-season portal that idles for ten months. Keep it for global edge delivery, where the hyperscaler’s points of presence beat anything you would build. And keep it for workloads leaning on managed services (serverless, managed Kubernetes, proprietary AI APIs) you would otherwise have to rebuild and run yourself. The migration effort there rarely clears the savings.
The honest test: if a workload uses real elasticity, leans on a managed service you would struggle to replicate, or serves a global audience, it stays.
A practical way to decide
Score each candidate workload on four axes before moving anything:
- Utilisation pattern — flat and predictable points toward repatriation; spiky points toward staying.
- Data gravity — large datasets that mostly stay put (and rack up egress when they move) favour bringing compute to the data.
- Compliance load — Privacy Act 2020 or NZISM obligations that are easier to evidence locally tilt the decision.
- Managed-service dependency — heavy reliance on a provider’s proprietary services raises the cost of leaving.
Run a small pilot first. Pick one steady-state workload, move it, and measure the real total cost — including the engineering hours, not just the infrastructure line. The pilot tells you more about your own environment than any vendor case study will.
Getting started
Most NZ organisations land on hybrid, not a wholesale exit. The goal is to put each workload where it runs best on cost, control, and compliance. A private platform with documented NZ data residency, like
ASI Cloud
, gives you somewhere to land repatriated workloads without rebuilding sovereignty guarantees from scratch.
If your public cloud bill no longer matches the value you get from it, start with the triage, not the migration.
Book a meeting
with a NZ-based engineer to work through which of your workloads belong back home — and which are better left where they are.