Sovereignty beats price: the new procurement signal
Angelika Beierlein
8 min read The German federal government has commissioned SAP and Deutsche Telekom to build its central ...
Read Article
7 min read
“86% of CIOs plan cloud repatriation” has been a recurring headline across tech media for months. The figure comes from a Barclays survey and is technically accurate-but strategically misleading. Gartner forecasts public cloud spending to grow by 21.5% in 2025, reaching $723 billion. According to IDC, only 8% to 9% of companies are planning full repatriation. The real trend isn’t retreat-it’s a more precise allocation across cloud, hybrid, and on-premises environments. CIOs who miss this are fighting the wrong battle.
Key Takeaways
- The oft-cited Barclays figure of 86% refers to CIOs moving “some” workloads back-not companies fully exiting the cloud. This distinction is strategically critical.
- Gartner forecasts public cloud spending to reach $723.4 billion in 2025, up 21.5% from 2024. IaaS is growing fastest at 24.8%, contradicting any macro-level narrative of widespread repatriation.
- IDC reports that only 8% to 9% of enterprises plan to fully repatriate their workloads. The vast majority are instead pursuing selective redistribution between cloud and on-premises environments.
- Gartner’s projection that 90% of organizations will adopt hybrid infrastructure by 2027 signals the real architectural shift: away from “cloud-first” toward “workload-driven decisions.” Repatriation is a symptom, not the trend.
- Selective repatriation is primarily driven by AI workloads with high GPU demands, data gravity challenges, and compliance requirements-not broad disillusionment with public cloud.
The 86% figure-and what it really means
The headline is rhetorically powerful but analytically weak. In its CIO survey for Q4 2024, Barclays asked whether companies plan to move workloads from public cloud environments back to private cloud or on-premises infrastructure. A full 86% answered yes-the highest rate Barclays has ever recorded. Since then, this number has been cited in hundreds of articles as proof of a “cloud repatriation” trend.
What the figure doesn’t reveal is how many workloads per company are actually involved. A company moving just one workload-say, a database with especially stringent compliance requirements-answers “yes.” So does a company relocating 10% of its portfolio. And so does one exiting the cloud entirely. All three scenarios appear in the same survey response line, despite representing fundamentally different strategic decisions. Thus, the 86% figure measures movement-not direction.
A closer look reveals a more telling counter-statistic. According to IDC’s survey on server and storage workloads, only 8% to 9% of organizations plan full-scale workload repatriation. The remaining 77% to 78% of “yes” responses in the Barclays survey reflect selective redistribution-not a wholesale exit from the cloud. For CIOs basing architectural decisions on media headlines, that’s an expensive misreading.
Gartner’s Counterpoint: 21.5% Growth in 2025
The most compelling rebuttal to the repatriation narrative comes from Gartner’s latest public cloud spending forecast for 2025. According to Gartner, the global market will grow from $595.7 billion to $723.4 billion-a 21.5% increase. If 86% of CIOs were truly moving workloads back en masse, such growth would be mathematically impossible. Either the repatriation story is wrong, or Gartner’s numbers are-but both can’t be true at the same time.
Of particular interest is the segment breakdown. Infrastructure as a Service (IaaS) is growing fastest at 24.8%, followed by Platform as a Service (PaaS) at 21.6%, while Software as a Service (SaaS) reaches 19.2%. Notably, it’s precisely the IaaS layer-the one supposedly most affected by repatriation-that shows the strongest growth. Sid Nag, Vice President Analyst at Gartner, summarized it this way, as reported by The Stack: cloud use cases continue to expand, with increasing emphasis on distributed, hybrid, cloud-native, and multi-cloud environments, driven by AI acceleration. In other words: cloud spending isn’t declining-it’s shifting.
86%
of CIOs plan selective workload repatriation (Barclays Q4 2024)
8-9%
plan full-scale repatriation (IDC Workloads Survey)
+21.5%
Public cloud spending growth in 2025 (Gartner)
Sources: Barclays CIO Survey Q4 2024, IDC Server and Storage Workloads Survey, Gartner Cloud Forecast November 2024
These three figures aren’t contradictory when properly contextualized. 86% of CIOs are moving workloads-but selectively, not entirely. Only 8-9% plan a full exit-a significant minority, but still a minority. Meanwhile, the market grows by 21.5% because new workloads are being created faster than old ones are being brought back. The headline “Cloud Repatriation” describes a real trend, but it overstates its overall impact.
The Real Trend: Hybrid as Architecture, Not Retreat
Definition
Cloud repatriation refers to moving a workload previously migrated to the public cloud back into a private-cloud or on-premises environment. The term is often confused with hybrid reallocation, where a workload isn’t withdrawn but instead deliberately distributed across multiple deployment targets from the outset. Repatriation is a correction; reallocation is a strategy.
More significant than the 86 percent figure is the 90 percent forecast. Gartner predicts that by 2027, around 90 percent of all organizations will adopt multi-cloud and hybrid strategies. By the end of 2026, these hybrid architectures will support critical workloads for 40 percent of enterprises-a jump from just 8 percent in prior years. This marks the true architectural shift: away from the “cloud-first” doctrine, which automatically sent every new workload to the public cloud, toward workload-specific decision-making.
Hybrid as an architecture fundamentally differs from repatriation as a corrective measure. Repatriation implies a previous decision was wrong and must be reversed. Hybrid implies there never should have been a blanket decision in the first place-only per-workload evaluations. CIOs redesigning their cloud strategy in 2026 face a choice: Do they want to retrospectively correct past moves and frame it publicly as a learning curve, or proactively build a nuanced architecture centered on workload assessment?
The latter option is strategically stronger. It signals internally and externally that infrastructure decisions are data-driven, not dictated by trend headlines. It allows continued use of the public cloud for specific workloads without falling prey to oversimplified criticism like “the cloud let you down.” And it aligns more realistically with overall market trends, which are clearly moving in the opposite direction-with 21.5 percent growth.
Three Workloads Where Repatriation Makes Sense
To keep the argument fair: there are workloads for which repatriation is both technically and financially sensible. The repatriation movement didn’t emerge by accident. Three categories dominate real-world cases in 2025 and 2026.
The first involves AI training and inference with high GPU intensity. Organizations that continuously train models or run large-scale inference workloads can quickly face six-figure monthly bills in public clouds-GPU hours represent the most expensive pricing tier on AWS, Azure, and GCP. For these workloads, owning dedicated hardware or using colocation often pays off within just six to nine months. BizTech Magazine documented cases from the pharmaceutical industry in August 2025 where enterprises deliberately migrated their LLM training clusters from public cloud environments to private GPU farms housed in colocation data centers.
The second category comprises workloads with massive data volumes and high egress costs. Moving terabytes of data out of the cloud daily-for video processing, backup exports, or ML pipelines connected to on-premises systems-means paying egress fees three times over: once when generating the data, again for storage, and a third time when transferring it out. While the EU Data Act tightens rules around egress fees starting 12 January 2027, existing contracts remain unaffected until then. In such scenarios, repatriation is less a matter of ideology and more one of simple math.
The third category is driven by regulation. Health data in Switzerland, certain financial data governed by DORA (Digital Operational Resilience Act), and highly sensitive public-sector data subject to EU digital sovereignty requirements all come with architectural mandates that public cloud providers struggle-or fail-to meet. Here, repatriation isn’t about cost; it’s a compliance imperative. The 8-9% of full-repatriation cases reported in IDC’s survey disproportionately originate from these regulated sectors.
Three workloads where repatriation becomes expensive
On the opposite side of the equation, clear patterns also emerge. Three workload categories don’t become cheaper through repatriation-they become more expensive-and CIOs should identify them before committing to a blanket cloud exit.
| Workload type | Repatriation makes sense | Repatriation is costly |
|---|---|---|
| AI training and inference | Yes (high GPU intensity) | – |
| Spike-driven web applications | – | Yes (auto-scaling advantage) |
| Data-intensive egress pipelines | Yes (egress costs) | – |
| Global SaaS delivery | – | Yes (multi-region reach) |
| Regulatory-sensitive data | Yes (sovereignty requirements) | – |
| Serverless event processing | – | Yes (pay-per-invocation advantage) |
Sources: Gartner Cloud Forecast 2024, Puppet Cloud Repatriation 2025, Kubermatic Public Cloud Analysis
Spike-driven web applications with irregular traffic peaks are the classic example where public cloud shines. Building on-premises infrastructure to handle peak traffic means paying 24/7 for capacity that sits idle most of the time. While cloud bills may appear higher at full utilization, the total cost of ownership for variable workloads is almost always worse on-premises. The same logic applies to serverless event processing: pay-per-invocation models are structurally more cost-effective for sporadic workloads than dedicated hardware.
Global SaaS delivery is the second case. Serving users across multiple regions with latency-sensitive services requires multi-region infrastructure. Public cloud delivers this global reach at a fraction of the cost of building and maintaining your own worldwide footprint. Repatriating global SaaS workloads isn’t just expensive-it also introduces operational challenges in disaster recovery and failover.
The Peer Question: What the CIO Should Tell the CFO
CFOs have read the headlines about repatriation and are now asking in many companies, “What’s our strategy?” The answer isn’t a number or a simple yes or no. The best response is a question in return: Which workloads are we talking about? On what data basis do we evaluate them? Anyone who jumps straight onto the repatriation slide is selling savings that fail to materialize in 90 out of 100 cases. By contrast, proposing a workload-specific assessment demonstrates competence.
Concretely, this means three things for the boardroom discussion. First: The organization needs a current inventory of all workloads, including up-to-date cost and performance metrics. Without this foundation, any repatriation discussion remains speculative. Second: The organization needs a decision framework that evaluates workloads across six to eight critical dimensions-compliance, data volume, latency requirements, scaling patterns, AI intensity, and vendor risk. Third: The organization must set realistic expectations about what repatriation can actually deliver. Achieving 20% cost savings on 30% of workloads is a solid outcome-not a revolution, but a measurable improvement.
Conclusion
Cloud repatriation isn’t a myth-but it’s not a broad-based trend either. Media coverage often confuses isolated workload migrations with a fundamental architectural shift. The real trend is hybrid as a deliberate strategy-not as a correction of a flawed cloud decision, but as a more nuanced, workload-by-workload assessment. CIOs who grasp this can build a cloud strategy by 2026 that’s resilient both internally and externally. Those chasing the repatriation narrative, however, are selling an oversimplified story-and risk being left behind when the next wave of infrastructure transformation arrives.
The good news for peer discussions: the data is clear-if interpreted correctly. 21.5% cloud growth, a 90% hybrid adoption forecast, and only 8-9% full repatriation. From this, leaders can craft a clear position that withstands CFO scrutiny and provides internal guidance. Anyone asked in the coming months whether “we should migrate back, too” now has a better answer than just yes or no.
Frequently Asked Questions
Is the widely cited 86% Barclays figure wrong?
No, the number itself is accurately measured. What’s misleading is how it’s been interpreted in media coverage. The survey asks whether CIOs plan to pull workloads back from the public cloud-without distinguishing between moving a single workload versus an entire portfolio. A CIO who brings back one database while keeping 40 other workloads in the cloud still answers “yes.” This aggregation creates a statistic that appears far more dramatic than reality.
How much do AI workloads influence repatriation figures?
AI workloads are the single biggest driver behind documented repatriation cases in 2025 and 2026. GPU hours in public clouds represent the most expensive compute category-and with continuous training or high inference volumes, owning dedicated hardware often pays for itself within a year. Organizations that exclude AI workloads from their cloud strategy see significantly fewer repatriation cases in other segments.
How does the EU Data Act change cloud cost calculations?
Starting 12 January 2027, the EU Data Act bans blanket egress fees by cloud providers. Only verifiable direct costs for data migration will be permitted. This significantly weakens one of the classic drivers of repatriation, as switching costs become more transparent-and often lower. For CIOs, this means: if you’re considering repatriation due to egress fees, factor in the 2027 regulatory shift.
What role do sovereign clouds play in this debate?
Sovereign clouds offer a third option between public cloud and on-premises infrastructure, delivering cloud economics with regulatory assurance. For European companies in sensitive sectors, they’re often a better solution than full repatriation-provided the sovereign cloud provider genuinely meets specific requirements, not just markets them as a claim. The ongoing CISPE discussion around “sovereignty washing” underscores the need for caution here.
How long does a serious workload assessment for repatriation take?
For a mid-sized company with 100 to 300 workloads, six to eight weeks is realistic. The inventory phase alone can be completed in two to three weeks if a SaaS management platform or up-to-date CMDB exists. Evaluating each workload against compliance, cost, performance, and architectural fit consumes most of the time. Without external support, projects often run longer due to limited internal IT capacity.
What do hyperscalers themselves say about the repatriation trend?
AWS, Azure, and Google Cloud all reported double-digit growth in their 2025 quarterly earnings-a trajectory incompatible with a massive repatriation wave. At the same time, all three have expanded their hybrid offerings, such as AWS Outposts, Azure Arc, and Google Distributed Cloud. The strategic message is clear: hyperscalers acknowledge that not every workload belongs in the public cloud, so they’re building infrastructure to support hybrid scenarios. This nuanced reality rarely makes it into repatriation headlines.
Editor’s Reading Recommendations
More from the MBF Media Network
- → Stadtwerke Bernau Meets Smart Meter Mandate (MyBusinessFuture)
- → Valkey 9 vs. Redis 8: The Cloud Cache Showdown (cloudmagazin)
- → Secure Boot Certificates Expire in June 2026 (SecurityToday)
Source for header image: Pexels / Vladimir Nikolaevich