Quick answer: Cloud repatriation (moving workloads from public cloud back to on-premises or colocation infrastructure) has gone from a fringe cost-cutting move to mainstream strategy. Cloudian’s 2026 survey of 212 senior IT decision-makers found 89% plan to expand their on-premises footprint over the next two years, and 75% have already moved workloads back in the past 24 months. Cloudian: Nearly Nine in Ten Enterprises Plan Cloud Data Repatriation The two biggest drivers: data sovereignty requirements and the true, all-in cost of running steady-state workloads in the public cloud.
This Isn’t an All-or-Nothing Decision
The shift underway isn’t a wholesale abandonment of the cloud. Organisations are surgically pulling back specific data-heavy, regulated, or AI-training workloads while leaving burst-capacity and less-sensitive applications in the cloud where elasticity still makes sense. Cloudian describes the resulting model plainly: “hybrid isn’t a compromise anymore, it’s a deliberate strategy.” Cloudian: Data Repatriation Why Now?
Separately, MRC Productivity’s 2026 analysis found 83% of CIOs planned to move at least some public cloud workloads back to private or on-premises infrastructure — the highest rate on record. mrc’s Cup of Joe Blog: What’s Driving Cloud Repatriation in 2026?
Driver #1: Regulation Is Forcing the Question
Under frameworks like the EU’s Digital Operational Resilience Act (DORA) and the NIS2 Directive, financial services, government, and critical-infrastructure workloads face data-handling rules that are far simpler to enforce on infrastructure an organisation owns outright. Combined with the EU AI Act reaching full application in August 2026, these three regulations have effectively made sovereign, in-jurisdiction infrastructure mandatory for a growing set of workloads. Not optional. SoftwareSeni: DORA, NIS2, and the EU AI Act Are Making Sovereign Cloud Mandatory
Driver #2: The Sovereignty Premium Is Real and It’s Expensive
This is where the cost math gets interesting. Public cloud platforms already charge significant egress fees to move data across borders. Now layer on the cost of “sovereign” cloud regions built specifically to address jurisdiction concerns.
Take AWS’s European Sovereign Cloud (ESC), launched in January 2026 with a €7.8 billion investment centreed on a new region in Brandenburg, Germany, and independent identity, billing, and access-management planes with no operational dependency on the U.S. AWS: AWS Launches AWS European Sovereign Cloud It’s a serious architectural commitment and it carries a price tag to match: industry analysis puts the ESC premium at roughly 15% over standard AWS European regions. tecRacer: AWS European Sovereign Cloud — Launch, Pricing, and What’s Next
There’s a deeper catch, too: ESC is still 100% owned by Amazon, a U.S. corporation. Legal experts note that this means it offers risk mitigation, not absolute immunity from the U.S. CLOUD Act. Organisations pay a premium for a structure that reduces exposure but doesn’t eliminate it. At launch, ESC also shipped with roughly 90 services versus 240+ in standard EU regions, and only two availability zones meaning a narrower platform for a higher price. tecRacer: AWS European Sovereign Cloud — Launch, Pricing, and What’s Next
Comparing the Options
| Standard Public Cloud | Hyperscaler Sovereign Cloud (e.g., AWS ESC) | On-Premises / Owned Infrastructure | |
| Data residency | In-region | In-region | In-region |
| CLOUD Act exposure | High | Residual (parent company risk) | None |
| Pricing | Baseline | ~15% premium | Fixed, known upfront |
| Cost predictability | Variable, usage-based billing | Variable + premium | Predictable, capital-planned |
| Exit cost | Egress fees apply | Egress fees apply | N/A, you own it |
Driver #3: AI Has Changed the Physics of Data Movement
Generative AI training and inference workloads are often measured in petabytes. Moving that much data into and across public clouds is expensive, slow, and per the regulatory pressures above, increasingly legally risky. For steady, high-volume AI workloads, keeping data and compute on-premises can win on both total cost of ownership and jurisdictional insulation, since proprietary datasets and models never leave infrastructure the organisation directly controls. Data Centre POST: Getting Control Over AI Cloud Data with On-Premises Object Storage
Driver #4: The Hardware Gap Has Closed
The old argument for staying in the cloud of “we can’t match hyperscaler infrastructure ourselves” doesn’t hold up the way it used to. Modern server platforms, hyperconverged private cloud stacks, and enterprise object storage now let organisations build highly capable private infrastructure that replicates public cloud agility without the ongoing egress fees, premiums, or foreign jurisdiction exposure. mrc’s Cup of Joe Blog: What’s Driving Cloud Repatriation in 2026?
The TCO Case for Nexsan
This is where the repatriation conversation gets concrete. Cloud economics are built around recurring, usage-based billing of compute, storage, egress, and now sovereignty premiums, all metered monthly. On-premises infrastructure inverts that model: the cost is known upfront, and it stays known.
Nexsan’s approach is built directly around that predictability. Rather than pushing organisations into short refresh cycles, Nexsan systems are engineered for a 7+ year lifecycle that’s validated for reliability before they ever ship, and supported well past the point where many storage platforms are pushed into forced upgrades. Nexsan: Extend the Life of Your Storage Investment — A 7+ Year Strategy for Predictable TCO That reliability is engineered in, not assumed: every system undergoes rigorous stress-testing and QA validation before it leaves the factory. Nexsan: The Nexsan Advantage Proven Reliability, Validated Before It Ships
For an IT budget owner, that translates into a straightforward comparison:
- Public cloud: variable monthly billing, egress fees on every migration, and a sovereignty premium of roughly 15-20% if you need in-jurisdiction guarantees.
- Nexsan on-premises: a known capital cost, a multi-year support runway, and zero foreign-jurisdiction exposure because there’s no foreign cloud parent in the chain at all.
For predictable, high-volume, always-on workloads exactly like the profile organisations are now required to protect under DORA, NIS2, and the EU AI Act, that’s not just a compliance advantage. It’s a better financial model.
The Bottom Line
Cloud repatriation in 2026 isn’t a retreat from modern infrastructure. It’s organisations getting more deliberate about which workloads belong where. When the workload is sensitive, regulated, or running 24/7 at scale, the math increasingly favours infrastructure you own outright: no egress fees, no sovereignty premium, no surprise on next month’s bill, and no foreign CLOUD Act exposure to explain to your board.
Read the rest of the series: The Residency Illusion — why local storage alone isn’t sovereignty and what the €890M Google fine reveals about tech sovereignty geopolitics.
Sources: Cloudian: Nearly Nine in Ten Enterprises Plan Cloud Data Repatriation · mrc’s Cup of Joe Blog: What’s Driving Cloud Repatriation in 2026? · AWS: AWS Launches AWS European Sovereign Cloud · tecRacer: AWS European Sovereign Cloud — Launch, Pricing, and What’s Next · SoftwareSeni: DORA, NIS2, and the EU AI Act · Data Centre POST: On-Premises Object Storage for AI · Nexsan: 7+ Year Strategy for Predictable TCO