The Numbers

What Moves With the Data

Unforeseen Migration Gaps

The Path Ahead Cloud providers have always sold convenience. Compute on demand, storage that scales, and somewhere in the fine print, the implied promise that someone else is watching the infrastructure. For a lot of teams, that last item was the most valuable thing they were paying for, whether they knew it or not. That arrangement is starting to come apart. Cloudian's 2026 research report surveyed 212 senior IT decision makers and found that 75% had moved workloads from the cloud back to on-premises infrastructure in the prior 24 months. That is not a rounding error or a niche trend. Three out of four senior IT professionals at organizations large enough to have senior IT professionals made a deliberate choice to bring their data and compute closer to home. The reasons are not surprising. Security and compliance pressure is one driver, and the growth of AI workloads is another. Michael Gale, CMO at EDB, put it plainly in a recent IT Brew piece, “If you want to use AI and data, you’ve got to be secure and compliant, they’ve got to be next to each other.” Sending proprietary data to a third-party cloud provider to feed a general-purpose model is increasingly hard to justify when purpose-built, containerized, on-premises alternatives exist. Egress fees are the third driver, and arguably the most compelling one. Cloud providers charge you to store data, and then they charge you to process it. And when you eventually decide you want it back, they charge you for that as well. Andy Stone, CTO for the Americas at Everpure, described it clearly: “They’re saying, as long as your data lives here, we’re cool; you want to take your data out, we’re going to charge you on the back end. In your data center, you don’t have that, you’re not going to pay an egress charge. It’s a benefit you derive, but the move itself takes time, a lot of planning and effort, and it’s certainly not easy in most cases.” In addition to charging for usage, companies are now paying not only to get their data back, but now the onus of monitoring and the associated costs are transferred back to the company as well. The part of this conversation that does not get enough attention is what teams lose when they leave the cloud, beyond the convenience of managed services. AWS CloudWatch, Azure Monitor, Google Cloud Operations. These tools exist because cloud providers understand that customers need to be able to see their infrastructure to troubleshoot it, and customers who cannot troubleshoot it generate support tickets. Visibility was bundled into the cost of cloud compute because the cloud needed it to function at scale. Informed customers generate fewer support tickets, so monitoring in a cloud environment became an amenity, when in reality it lowers their support costs. When a company repatriates its workloads, that visibility disappears. Now that the servers and the data are in house, so is the burden of monitoring the system. In the IT Brew Stone notes that repatriation requires a lot of architecting and planning, including managing the applications consuming and producing data. That’s accurate, and monitoring sits at the center of it. It’s hard to manage what you can’t see, and managing infrastructure on-premises creates a monitoring gap that needs to be filled, either internally or externally. The teams making this move are not all large enterprises with dedicated platform engineering staff. It is reasonable to assume that some portion of that 75% are organizations with lean technical teams making a deliberate architectural choice to prioritize control. They have the skills to manage their own hardware, they’ve made the cost calculation and decided it made sense. What they frequently do not have is the time or the desire to build and maintain a production-grade observability stack on top of everything else that’s migrating from the cloud. This is where the repatriation trend creates a genuinely new problem rather than just a different version of an old one. The cloud abstracted away the operational burden of monitoring. On-premises infrastructure exposes it directly. Companies need to be made aware that a disk is filling up before it causes an outage, alert routing needs to reach someone when a service goes down in the middle of the night, and log retention should go back far enough to reconstruct the events that occurred during an incident. Building a monitoring stack is not the hard part, most teams can easily deploy the tooling. The open source tooling available for collecting telemetry is genuinely excellent. The real problem created by building an in-house monitoring system is the burden of ongoing operational overhead and figuring out which team members will own the maintenance. It’s an ongoing process that requires dedicated personnel to configure the tools, tune them, keep them running, and revisit the alert thresholds as the infrastructure changes. After dealing with planning and executing data repatriation for several months, they’re now faced with creating and maintaining monitoring for their infrastructure, and allocating resources they may not have to that endeavor. The repatriation trend is not likely to lose momentum in any meaningful way. The AI data sovereignty argument is too strong, the cost of cloud computing is too high, and security is becoming a bigger issue. If anything, the next wave of AI agent deployments will accelerate it. Gale's estimate of up to 300 million agents operating in US enterprises is speculative but directionally correct. Agents need data, that data needs to be governed, and governance is substantially easier when you control the physical location of the data. As companies continue to pull their data in-house, a large and growing number of technical teams will find themselves responsible for infrastructure that requires monitoring, with limited time and resources to build and maintain it. Cloud-provided tools and infrastructure demonstrated the need for good visibility, and altering the deployment model should not mean changing how teams monitor their systems. Companies that navigate this well will be the ones who treat observability as a priority from the start of the repatriation process, not something to revisit once the migration is complete. Templates let you quickly answer FAQs or store snippets for re-use. Are you sure you want to hide this comment? It will become hidden in your post, but will still be visible via the comment's permalink. Hide child comments as well For further actions, you may consider blocking this person and/or reporting abuse