18 Percent Pay Gap, an EU Deadline, and Little Preparation: Salary Transparency from June 2026
Benedikt Langer
8 min. reading time Starting June 2026, salary ranges must appear in job postings. Inquiring about current ...
Read Article
9 min Reading Time
86 percent of CIOs plan to repatriate workloads from public cloud environments back into private cloud or on-premises infrastructure. After exiting AWS, 37signals saves $2 million annually. GEICO is repatriating at least 50 percent of its workloads after a decade on Azure. The cloud hasn’t failed. But the illusion that cloud is always cheaper? That has.
TL;DR
- 86 percent of CIOs plan repatriation of workloads from the public cloud. 80 percent aim to execute this within 12 months (IDC 2024).
- 30 to 60 percent infrastructure cost savings are realistic. 37signals saves $2 million annually; Dropbox saved $75 million over two years.
- 27 percent of cloud spending is wasted: underutilized resources, oversized instances, and forgotten development environments drive up costs (Flexera 2025).
- GEICO as a cautionary tale: a decade-long Azure migration across 600+ applications resulted in costs 2.5× over budget – and declining reliability. Now, 50 percent of workloads are being moved back onto OpenStack.
- Hybrid architectures as the target model: not cloud vs. on-premises, but each workload placed where it delivers maximum value. Gartner forecasts $723 billion in global cloud spending for 2025 – even amid the repatriation trend.
The Cloud Cost Trap: Why the Math Doesn’t Add Up
The cloud promised no upfront investment, elastic scaling, and operational expenditure instead of capital expenditure. For many organizations, that promise held true – for the first few years. Then came the hidden costs: egress fees for data transfer, disproportionately rising storage expenses, vendor lock-in via proprietary services, and the complexity of managing multi-cloud environments.
According to Flexera, 27 percent of all cloud spending is wasted. For a company with €10 million in annual cloud costs, that’s €2.7 million vanishing into unused or oversized resources. And 84 percent of organizations cite cost management as their top cloud challenge.
For leadership, this isn’t an IT issue – it’s a return-on-investment issue. If cloud bills grow faster than the business value they generate, the strategy is flawed.
Source: Flexera State of the Cloud Report, 2025
Three Companies That Came Back
37signals (Basecamp/HEY): David Heinemeier Hansson’s company exited AWS entirely and invested in its own servers. Its projected five-year savings of $7 million were exceeded: it now saves $2 million annually – roughly $10 million over five years. The key? Predictable workloads without sharp scaling spikes.
Dropbox: The cloud storage provider built its own infrastructure and saved $75 million in two years. Dropbox faced a specific problem: its workloads were so large and steady that public cloud pricing models were structurally more expensive than owning hardware outright.
GEICO: The U.S. insurer migrated more than 600 applications to Azure over a decade. The result? Costs 2.5× higher than expected – and declining reliability. GEICO is now repatriating at least 50 percent of its workloads onto an OpenStack-based private cloud. Expected savings: 50 percent per compute core and 60 percent per gigabyte of storage.
“Cloud repatriation isn’t a step backward. It’s the maturation of cloud strategy. Organizations are increasingly making informed decisions about where their workloads truly belong.”
Puppet, Cloud Repatriation Report 2025
When Repatriation Makes Sense – and When It Doesn’t
Not every workload belongs back on your own servers. The decision hinges on three factors:
Predictability. Workloads with stable, predictable demand – databases, internal applications, archiving – are strong repatriation candidates. Workloads with sharp scaling spikes – campaign landing pages, seasonal e-commerce peaks – are better left in the cloud.
Data Volume. Large volumes of data moved regularly trigger high egress costs in public cloud. Dropbox and 37signals shared exactly this profile: massive, low-variability data flows.
Regulation. For DACH-region companies, data sovereignty, GDPR, and industry-specific regulations – including DORA (Digital Operational Resilience Act) for financial services and the KRITIS umbrella law – play a central role. Hosting infrastructure in Germany grants full control over data location and access. Data sovereignty has become a strategic argument.
Not suitable for repatriation: AI workloads requiring GPU clusters (too costly to run in-house), SaaS applications (the cloud is the product), globally distributed applications with strict latency requirements, and startups in growth mode (flexibility outweighs cost optimization).
The Hidden Costs of Bringing Workloads Home
What success stories omit: repatriation isn’t free. Leadership must understand total cost – not just infrastructure savings.
Hardware investment: Servers, storage, networking, and data center capacity. For a mid-sized workload, an upfront investment of €500,000 to €2 million is realistic – depreciated over 3-5 years.
Personnel: Cloud-managed services replace in-house operations staff. Returning to owned hardware means hiring or training system administrators, network engineers, and security specialists. In today’s tight talent market, this is the biggest hurdle.
Migration: The repatriation process itself typically takes 6-18 months and ties up developer resources. Applications built around cloud-native services (Lambda, DynamoDB, Cosmos DB) require rewriting.
Opportunity cost: Every month teams spend on migration is a month lost to product development. That must factor into ROI calculations.
Five Questions for Leadership
1. What are our actual cloud costs? Not list prices – but total cost, including hidden egress fees, premium support, and oversized instances. Most companies underestimate their cloud spend by 20-40 percent.
2. Which workloads have predictable demand? These are the repatriation candidates. Anything highly variable stays in the cloud.
3. Do we have the personnel? Without an operations team, repatriation is risky. Managed hosting or colocation with a DACH-based provider can be a bridge: your hardware, their operated infrastructure.
4. How deeply are we locked into cloud-native services? The more proprietary services you use (Lambda, Step Functions, Cosmos DB), the costlier the migration. Platform dependencies cut both ways.
5. What’s our target architecture in five years? Full repatriation is rarely the answer. Hybrid models – stable workloads on-premises, variable ones in the cloud – are the most pragmatic path for most organizations.
Conclusion
Cloud repatriation isn’t an anti-cloud trend. It’s a strategic correction. The cloud remains the right choice for variable workloads, AI training, and globally distributed applications. But for stable, data-intensive workloads, owned infrastructure is often cheaper, more controllable, and more regulatorily secure. The numbers speak clearly: 30-60 percent infrastructure savings for the right workloads. Leadership must decide – not cloud or on-premises, but workload-by-workload. Hybrid architecture isn’t a compromise. It’s the mature answer to a complex question.
Frequently Asked Questions
What does cloud repatriation mean?
Moving workloads, applications, or data from public cloud providers (AWS, Azure, GCP) back into private cloud environments, colocation data centers, or your own on-premises infrastructure. Not to be confused with cloud exit – the complete abandonment of cloud services.
How much can you save through repatriation?
Typically 30 to 60 percent of infrastructure costs for suitable workloads. 37signals saves $2 million annually; Dropbox saved $75 million over two years. Actual savings depend on workload profile, data volume, and existing cloud dependencies. Crucially: factor in upfront hardware investment (€500,000-€2 million) and personnel costs.
Does repatriation contradict the broader cloud trend?
No. Gartner forecasts $723 billion in global cloud spending for 2025 – up 21 percent. Cloud adoption continues to grow. Repatriation targets selective workloads – not entire infrastructures. Most repatriating organizations retain 40 to 70 percent of their workloads in the cloud.
Is repatriation relevant for mid-market companies?
Yes – especially those with stable workloads and high data volumes. In the DACH region, regulatory drivers add weight: GDPR, DORA, and the KRITIS umbrella law make data sovereignty a strategic imperative. German managed hosting providers offer hybrid solutions: your hardware, hosted and operated in a German data center.
How long does cloud repatriation take?
Typically 6 to 18 months for core workloads. Biggest time sinks: applications built on cloud-native services (serverless, managed databases, proprietary APIs) require rewriting or replacement. Lift-and-shift workloads (VMs without cloud-native dependencies) can be migrated in 2 to 4 months.
Further Reading
Platform Ecosystems: Build, Buy, or Join
The Digital Operating Model: Restructuring the CIO’s IT Organization
Data Culture at the C-Level: 78 Percent Fail on People
More from the MBF Media Network
cloudmagazin: The VMware Cost Trap 2026 – Alternatives for IT Teams
MyBusinessFuture: Exiting U.S. Cloud Providers – What Mid-Market Companies Must Assess
SecurityToday: Cloud Security as a German Export Product
Header Image Source: Pexels / Brett Sayles (px:5203849)