For most of the last decade, the direction was clear: get off on-premise infrastructure, move everything to the cloud, and never look back. That narrative has not collapsed, but it has been significantly complicated. In 2026, the question is no longer whether to use the cloud. It is which workloads belong there, and which ones are costing you more than they should. And it's driven by multiple factors of the post-digital era. 

The Scale of the Shift

The numbers are hard to ignore. 86% of CIOs plan to move some workloads from public cloud back to private cloud or on-premises infrastructure, the highest rate ever recorded. A Barclays survey found that 83% of enterprises reported plans to repatriate at least some workloads from public cloud to private infrastructure. And IDC found that 59% of organizations spent more than budgeted on cloud in 2024, with 84% citing managing cloud spend as their single biggest challenge.

This is not a fringe movement. It is a correction at scale.

Importantly, Gartner predicts that by 2027, approximately 90% of organizations will adopt a hybrid cloud model, not a return to purely on-premise infrastructure, but a deliberate split where workloads are placed based on their actual requirements rather than a blanket policy.

Reason 1: The Cloud Bill Stopped Making Sense

The promise of the cloud was lower costs through pay-as-you-go flexibility. For elastic, variable workloads like seasonal spikes, startup growth, and rapid experimentation, that promise still holds. For stable, always-on workloads, it often does not.

In reality, hidden costs distorted cloud economics. Egress fees, cross-zone traffic, premium managed services, and layered security tooling added up quietly. Individually, these charges looked manageable. Together, they eroded margins.

On AWS, bandwidth costs can run upwards of $10,000 per month in egress fees alone. For data-intensive companies, this difference is existential. Repatriation converts variable, unpredictable operating expenses into fixed, predictable costs.

The real-world examples are well documented. 37signals, the company behind Basecamp and HEY, was spending roughly $3.2 million annually on AWS. After investing approximately $600,000 in Dell servers, they cut their annual compute bill by $1.5 to $2 million. GEICO, facing over $300 million in annual cloud spend, cut compute costs by 50% per core through repatriation to Open Compute Project hardware.

Broadcom's internal analysis found that modern private cloud delivers 40 to 50% lower total cost of ownership for steady-state workloads compared to public cloud.

Reason 2: AI Changed the Economics Entirely

Artificial intelligence has become the accelerant nobody fully anticipated when cloud pricing was designed. Training and inference at any serious scale in the public cloud is expensive, and the data volumes involved make egress costs punishing.

A February 2026 enterprise survey found three converging forces driving on-premises AI adoption: data sovereignty concerns, cloud cost unpredictability, and real-time performance requirements. When deploying AI involving sensitive company data, 91% of respondents would choose on-premises, private cloud, or hybrid infrastructure over public cloud.

Deloitte's analysis found that on-premises AI delivers 50% or more cost savings over three years compared to cloud API alternatives, once token volume crosses a certain threshold. Open-source models like Llama and Mistral now handle 85 to 90% of enterprise AI use cases at quality indistinguishable from cloud APIs.

Gartner predicts that 40% of enterprises will adopt hybrid compute architectures for mission-critical workflows by end of 2026, up from just 8% in prior years, driven largely by AI workload economics.

Reason 3: Data Sovereignty Is No Longer Optional

Regulatory pressure has tightened across every major jurisdiction. What was once a compliance consideration has become a compliance requirement, and public cloud, by its nature, makes it harder to guarantee where data physically lives.

Under the US CLOUD Act, data hosted with US-owned hyperscalers, even in their European data centers, can theoretically be accessed by US law enforcement. For European organizations in particular, that exposure is increasingly untenable.

In Europe, GDPR sets a high bar for personal data processing, and the Digital Operational Resilience Act (DORA) for financial entities entered into application in January 2025. Several national schemes further influence cloud service choice, including Germany's C5 catalogue, France's SecNumCloud, and the GAIA-X initiative, all of which push toward data portability and sovereignty.

According to the Nutanix Enterprise Cloud Index 2026, 57% of IT leaders feel the need to run infrastructure within a single country. Regulatory pressure in financial services, healthcare, and government is intensifying.

By 2026, over 70% of large organizations are expected to evaluate or adopt sovereign cloud solutions for regulated data.

Reason 4: Vendor Lock-In Has Become Visible

The early years of cloud adoption were spent building on managed services, proprietary APIs, and platform-native tools. That was often the right call at the time. The cost of that decision is now showing up in exit estimates.

Moving away from vendor-specific services in the public cloud often necessitates significant architectural changes and refactoring. Applications built on serverless or highly cloud-centric designs may require substantial re-engineering to function optimally on on-premise or bare metal infrastructure.

Organizations that are moving now are doing so before that lock-in becomes complete, and before a future renegotiation leaves them with no credible alternative.

Reason 5: Centralization Risk Is No Longer Theoretical

High-profile cloud outages in 2024 and 2025 reminded organizations of a structural risk that had been largely overlooked during the cloud-first era: when a major hyperscaler region goes down, a significant portion of the internet goes with it.

These outages are changing how enterprises architect for resilience, making selective repatriation a risk management strategy rather than just a cost optimization tactic. Distributing workloads across owned and hosted infrastructure creates a redundancy that a purely cloud-dependent architecture cannot provide.

What Companies Are Actually Doing

The destination is not a return to the data center of 2015. It is a deliberate hybrid model where placement follows the workload's actual characteristics.

The Flexera State of the Cloud Report indicates that 42% of workloads are being moved from public clouds to private clouds or on-premise solutions to optimize costs, while the remainder stay in public cloud where the economics and flexibility still make sense.

Some workloads run better close to the user. Others require external scaling or integration with SaaS tools already hosted in the cloud. Hybrid infrastructure makes space for both.

The organizations doing this well are not making an ideological decision. They are running the numbers workload by workload and placing each one where the cost, performance, compliance, and control profile makes the most sense.

This Is Cloud Maturity, Not Cloud Rejection

The companies moving workloads off purely hosted cloud are not abandoning cloud infrastructure. They are refining their relationship with it. "Cloud-first" has quietly evolved into "cloud-appropriate."

That distinction matters. The goal is not to minimize cloud usage (you really can't), but to stop paying for cloud where it does not deliver value, and to stop accepting the compliance exposure, performance limitations, and vendor dependency that come with a purely hosted model.

At Smartt, we work with organizations that are doing exactly this assessment, figuring out what belongs in the cloud, what belongs on-premise or in a private environment, and how to build the hybrid infrastructure that connects both. If that conversation is on your roadmap, our FlexHours program can give you the structured support to work through it with expert advisors, hardware discounts, and even included cloud hosting!