Cloud repatriation headlines tend to imply a stampede away from AWS, Azure, and Google Cloud, but the underlying data describes something narrower. IDC's research on enterprise server and storage workloads found that roughly 80% of organizations expect to repatriate some share of compute or storage within the next twelve months yet only 8% to 9% of enterprises are planning a full exit from public cloud. The two figures are not in conflict: most companies are moving specific, predictable workloads home while leaving elastic and experimental ones in the cloud where they still make sense. Repatriation, at the scale it's actually happening, is a rebalancing exercise, not a retreat.
The clearest documented case of large-scale repatriation comes directly from a regulatory filing rather than a survey. Dropbox's S-1, filed with the SEC ahead of its 2018 IPO, disclosed that its multi-year “Infrastructure Optimization” initiative moving the majority of user data off Amazon Web Services and onto company-owned, colocated hardware saved the company $74.6 million in operating costs over two years. The company's gross margin rose from 33% to 67% over the same period, a shift Dropbox attributed directly to the migration combined with revenue growth. It remains one of the few instances where a public company has quantified repatriation savings in an audited financial disclosure rather than a marketing case study.
Companies rarely repatriate workloads until they can actually see what those workloads cost which is why the rise of cloud financial-operations (FinOps) practices tracks so closely with the repatriation conversation. The FinOps Foundation's State of FinOps 2026 survey, drawing on 1,192 practitioners representing more than $83 billion in combined annual cloud spend, found that 98% of respondents now actively manage AI-related cloud costs, up from just 31% two years earlier. That kind of granular cost tracking is what turns a vague sense that “cloud is expensive” into a specific, defensible business case for moving an individual workload back on-premises.
The cost-control pressure isn't confined to the private sector. A U.S. Government Accountability Office audit of cloud procurement across 24 major federal agencies found that federal spending on cloud service contracts has grown from $2.3 billion to more than $10 billion a year over the past decade, while most agencies still rely on outdated procurement definitions and imprecise historical data to track what they're actually spending. Seventeen of the 24 agencies GAO reviewed reported that managing cloud costs requires changing day-to-day IT management practices they haven't yet implemented. If a government with subpoena power over its own contractors is struggling to track cloud spend, the private-sector cost-visibility problem driving repatriation is not a fringe complaint.
The destination for most repatriated workloads isn't a return to traditional data centers it's hybrid infrastructure that blends public cloud with private or colocated capacity. Gartner's cloud infrastructure forecast projects that roughly 90% of organizations will operate a hybrid infrastructure model by 2027, treating cloud-versus-on-premises as a per-workload decision rather than a company-wide architecture commitment. That framing matches what IDC's repatriation data shows: enterprises are not choosing sides so much as sorting workloads by which environment actually fits their cost and performance profile.
Beyond individual company case studies, the financial stakes of ignoring cloud cost discipline are large enough to show up in equity valuations. Reporting on venture firm Andreessen Horowitz's “Cost of Cloud” analysis, CIO Dive noted that cloud infrastructure costs and their drag on margins were estimated to be eating into roughly $100 billion in combined market value across 50 major public software companies, including names like Adobe and Zoom a figure the firm's analysts extended to more than $500 billion when applied to the broader universe of scaled public companies running similar workloads. That framing is a large part of why repatriation, once viewed as an engineering decision, is increasingly a boardroom one.
Cloud repatriation in 2026 is real, measurable, and almost entirely selective. IDC's data shows most enterprises repatriating something, but hardly any repatriating everything. Dropbox's SEC filing proves the savings can be substantial when a company has the scale and predictability to justify the engineering lift. And the common thread across the FinOps Foundation's numbers, GAO's federal audit, and Gartner's hybrid forecast is the same: organizations are not fleeing the cloud, they are finally pricing it correctly, and moving the workloads that lose that pricing argument back home.