The "move everything to the cloud" narrative peaked a few years ago. Since then, a more measured counter-narrative has emerged: cloud repatriation, outage postmortems, and finance teams questioning whether the cost projections anyone believed in 2020 were ever real.
Neither side is right. The organizations making the most defensible infrastructure decisions in 2026 have slowly abandoned the ideological framing altogether. They're asking a more useful question: which workloads belong where, and who governs the answer?
When a major cloud provider experiences a regional outage, the incident reports get shared widely. Boards ask questions. CIOs who staked everything on a single public cloud have uncomfortable conversations.
That reaction is reasonable, but the follow-on logic, "therefore we should run more on premises," deserves scrutiny before it shapes capital expenditure decisions.
On-premises infrastructure fails too, frequently without the SLA documentation, automated failover, or transparent incident reporting that hyperscalers provide. The difference is that a private data center outage rarely makes headlines. Availability risk in hybrid environments is bidirectional, and any serious infrastructure review should model failure scenarios in both directions.
The more legitimate driver for cloud skepticism is cost. Forrester's cloud computing research has noted increased momentum toward private cloud and on-premises deployments as organizations work through data sovereignty concerns, cost overruns, and data ownership challenges that emerged from aggressive public cloud migrations. That pattern is notable but reflects execution and governance failures rather than a structural argument against cloud infrastructure.
The organizations that made poor cloud decisions typically made them at the portfolio level rather than the workload level. "We're cloud-first" is a procurement posture. Deciding where a specific application should run requires a different kind of analysis.
A few dimensions that consistently matter:
Microsoft's hybrid and multicloud strategy guidance in the Cloud Adoption Framework offers a useful taxonomy here, distinguishing hybrid-first, Azure-first, and multicloud-first organizational profiles. The point is not that one is better; it's that the right model depends on where your workloads, data, and operational capabilities are.
A technically sound workload placement decision that lacks clear ownership tends to drift. Workloads that were supposed to be transitioned to cloud get left running on premises indefinitely. Instances that were provisioned for a specific project stay active after the project closes. Licensing entitlements go underutilized or double-purchased across environments.
These aren't edge cases. They're the standard outcome in organizations that treat hybrid as an architecture pattern rather than an operating model. The governance layer has to define:
Our cloud infrastructure services work frequently surfaces these gaps in organizations that have been in hybrid mode for years. The technical debt is usually manageable while the governance debt tends to be more significant.
One development worth specific attention for Microsoft-aligned environments: Azure Arc substantially reduces the operational penalty of hybrid infrastructure. Organizations can manage on-premises servers, Kubernetes clusters, and SQL instances through Azure Resource Manager, applying consistent governance, security policy, and monitoring tooling regardless of where resources physically run.
That doesn't make the placement decision for you. A workload that shouldn't be in Azure is still in the wrong place even if Azure Arc gives you a unified control plane view of it. But it does mean that "we can't govern our on-premises environment consistently" is no longer a valid reason to move workloads to the cloud prematurely. The management surface now extends to where the workloads are located.
Before any hybrid strategy conversation, the most useful input is a current-state inventory: which workloads are running where, what they cost in total (compute, licensing, operational overhead), and what the documented reasons are for their placement.
Most organizations don't have this. They have billing dashboards for Azure spend and a general sense of what's on-premises, but not a reconciled view that surfaces the actual total cost of ownership per workload and the last time placement was deliberately evaluated.
That audit is the starting point. Everything else, cloud-first, hybrid-first, repatriation, is a conclusion that should follow from the data, not a strategy that the data is then recruited to justify.
The IT leaders getting this right in 2026 aren't making ideology-driven infrastructure decisions. They're running disciplined workload reviews and accepting that the answer varies, sometimes by application, sometimes by region, sometimes by team maturity. If you haven't done that analysis recently, our free cloud infrastructure assessment is a structured way to start.