Cloudera published a survey this month that quietly contradicts a default a lot of enterprises have been running on for years: that AI workloads belong in the public cloud. In a study conducted by Wakefield Research of 1,500 enterprise architects, cloud infrastructure leads, and data architects across nine markets, 66% said their organization moved AI workloads from public cloud back to private cloud or on-premises infrastructure in the past year. That's not a fringe result. That's two out of every three organizations surveyed.
The obvious explanation isn't the real one
It would be easy to read this as a story about the public cloud failing to keep up with AI workloads technically. The data doesn't support that. Two numbers matter more here than the headline stat. First, 84% of respondents reported higher infrastructure costs from AI workloads than they'd planned for, a straightforward budget problem. Second, and larger, 95% said they had delayed or canceled an AI project specifically because of governance, compliance, or regulatory issues. Governance beat cost as a blocker. The infrastructure works. The control and predictability around it often don't.
Repatriation doesn't fix the underlying problem
There's a third number worth sitting with: 72% said their current data architecture needs a significant overhaul to actually meet AI requirements. Moving a workload from public cloud to on-premises doesn't resolve that on its own. A data architecture that isn't ready for AI is still not ready for AI once it's running on hardware you own instead of hardware you rent. Repatriation changes where the problem lives. It doesn't make the problem go away.
This is a rebalancing, not a retreat
The forward-looking numbers keep this from reading as a wholesale rejection of cloud. Only 24% of respondents expect to increase on-premises spending over the next two years. Meanwhile 29% still expect cloud spending to grow, and 25% are planning a hybrid-first architecture going forward. Almost nobody in this survey is walking away from the public cloud entirely. What's changing is the default assumption that every AI workload automatically belongs there. Enterprises are starting to make that call workload by workload instead of applying one policy to everything.
What to do with this before your next infrastructure review
- Map which AI workloads actually require the compliance posture and cost predictability that private cloud or on-premises infrastructure provides, instead of defaulting every workload to public cloud by habit.
- Budget for AI infrastructure cost growth explicitly, as its own line item with its own assumptions. Eighty-four percent getting surprised by AI infrastructure costs means most plans didn't account for it going in.
- Fix the data architecture gap before moving workloads, not after. Relocating a workload without fixing the architecture underneath it just moves the same problem to a new address.
Where TrueHorizon fits
We help enterprises make workload placement decisions based on their actual cost structure and compliance requirements, not on whichever infrastructure pattern was the default when a project started. That's not a checklist we're learning on your project. It's the expertise we bring to it. The right answer is rarely all public cloud or all on-premises. It's knowing which workload belongs where, and why.
If you're rethinking where your AI workloads should actually run, take our AI readiness assessment before your next infrastructure commitment.
Written by
Deepankar Bhadrasen
Founding Engineer
Deepankar is an AI automation specialist and Founding Engineer at TrueHorizon AI, where he builds practical AI systems that help businesses streamline operations, reduce costs, and scale efficiently. He focuses on integrating custom AI agents and workflows with existing tools so teams can grow without expanding headcount.
