4 min Reading Time

More and more companies are moving workloads back from the public cloud into their own data centers or private-cloud environments. This trend – known as cloud repatriation – runs counter to the long-standing “cloud-first” mantra. For CIOs, however, this isn’t a step backward; it’s a sign of strategic maturity. Organizations that understand precisely where each workload belongs optimize costs, compliance, and control simultaneously.

TL;DR

Why Companies Are Leaving the Cloud

“Cloud-first” defined the past decade. Enterprises migrated workloads en masse to the public cloud, lured by promises of flexibility, scalability, and freedom from hardware management. Yet after years of operation, downsides have emerged: runaway costs, vendor lock-in that impedes migration, and compliance requirements increasingly at odds with U.S.-controlled hyperscalers.

IDC found in a recent study that 71 percent of enterprises have already repatriated workloads from the public cloud. Motivations vary: cost optimization ranks first, followed by data sovereignty and performance demands. In the DACH region (Germany, Austria, Switzerland), regulatory developments play an ever-larger role: NIS2, DORA, and the EU AI Act impose strict data control requirements that certain cloud configurations simply cannot meet.

The most prominent example is 37signals – the company behind Basecamp and HEY. Founder David Heinemeier Hansson documented its move from AWS to owned hardware, projecting $7 million in savings over five years. While this case isn’t directly transferable to German mid-sized businesses, it illustrates a fundamental principle: beyond a certain scale and workload predictability, owning infrastructure becomes more economical than renting cloud capacity.

“In most cases, the decision to repatriate workloads from the public cloud is a cost decision. Enterprises with predictable workloads can save 30 to 50 percent by running them on owned infrastructure.”
IDG/Supermicro Cloud Survey (2024)

The Three Drivers of Repatriation

Driver 1: The cost truth after years of cloud usage. Many organizations only calculated their true total cost of ownership (TCO) after several years of cloud operations. Egress fees, storage charges, premium support contracts, and the labor cost of cloud architects all add up. Flexera estimates average cloud waste at 29 percent. For predictable, steady-state workloads, owned infrastructure is often 30-50 percent cheaper than an equivalent cloud configuration. The break-even point depends on scale – but typically falls between 100 and 200 virtual machines or their container equivalent.

Driver 2: Regulatory mandates for data sovereignty. The EU Data Act, NIS2, and the EU AI Act tighten requirements around data control. Meanwhile, the U.S. CLOUD Act permits U.S. authorities to access data held by U.S. companies – even if servers reside physically in the EU. An AWS or Microsoft data center located in the EU does not automatically shield data from such access. For sensitive data in regulated industries, the choice narrows to: sovereign cloud or owned infrastructure, tertium non datur.

Driver 3: Performance and latency. Certain workloads benefit significantly from physical proximity to data sources. Industrial IoT applications, real-time manufacturing analytics, and specific AI inference tasks demand sub-10-millisecond latencies – something cloud architectures don’t always guarantee. Edge computing and on-premises infrastructure deliver advantages here that cloud solutions simply cannot match.

The Workload Placement Matrix

Cloud repatriation doesn’t mean bringing all workloads back. CIOs need a nuanced framework to evaluate each workload individually. The decision hinges on four factors: cost, compliance, performance, and scalability.

Sources: IDC 2025, Flexera State of Cloud 2026, 37signals Blog

Cloud remains the right choice for: GPU-burst-intensive AI training, globally distributed applications with variable traffic, disaster recovery and backup, development and test environments, and SaaS applications with mature ecosystems. Here, scalability and flexibility outweigh the cost premium.

Repatriation pays off for: Stable production workloads with predictable load, I/O-intensive databases, regulatorily sensitive data in the DACH region, legacy applications that would require costly cloud-native rewrites, and HPC workloads with sustained GPU utilization above 70 percent. These benefit from lower unit costs and full data control.

The hybrid approach as the optimal path: Most enterprises will settle on a hybrid model – regulatorily critical and cost-intensive workloads on owned or sovereign infrastructure; variable and global workloads in the public cloud. The challenge lies not in making the decision – but in executing it: network interconnectivity, data synchronization, and unified management across both environments demand investment in platform engineering and multi-cloud expertise.

How CIOs Should Present Repatriation to the Board

The biggest hurdle is communication. Saying “We’re leaving the cloud” sounds like regression. CIOs must frame the message as strategic maturity: “We’re optimizing our IT architecture based on four years of cloud experience. Workloads that cost more in the cloud than on owned infrastructure – and that don’t leverage any cloud-specific advantages – we’re bringing back. That saves X euros annually, while maintaining or even improving compliance.”

The business case must be concrete: TCO comparisons per workload over three to five years, compliance gains through data sovereignty, performance improvements for latency-sensitive applications, and risk reduction via decreased vendor lock-in. With these numbers, cloud repatriation becomes the logical outcome of data-driven decision-making – not an admission of strategic failure.

Frequently Asked Questions

Header Image Source: Brett Sayles / Pexels