Edge, Centralised, or Hybrid? The Infrastructure Question You're Probably Asking Wrong
In 2025, 86% of CIOs planned to repatriate at least some public cloud workloads, the highest figure ever recorded, according to the Barclays CIO Survey. Yet only around 8% planned a full exit. Most organisations are not abandoning cloud. They are realising that a single architectural default was never the right answer.
Industry signal
Gartner projects that by 2027, 85% of workload placements made before 2022 will no longer be optimal. The decisions made during the cloud-first era may already be wrong not because cloud failed, but because the question has changed.
The organisations navigating this well are not asking which architecture to adopt. They are asking what each workload actually needs and letting that drive the placement decision. That shift is what this article is about.
Why 'Which Architecture?' Is the Wrong Place to Start
Cloud-first was a common (and common-sense) default in the 2010s. It reduced upfront capital, unlocked speed, and removed the burden of managing physical infrastructure. The problem is not that cloud was wrong. The problem is that treating any single architecture as the answer to every situation was wrong.
The evidence for this is accumulating. GEICO saw cloud costs increase 2.5 times after migrating over 600 applications. 37signals saved $2 million in 2024 by reducing cloud reliance. Approximately 21% of cloud infrastructure spending is typically wasted on underused resources. We should be clear though, these are not arguments against cloud, they are arguments against defaulting to the same solution regardless of the problem.
Latency-sensitive workloads do not behave like bursty analytics workloads. Regulated data does not have the same placement requirements as public-facing web assets. The workload-first reframe is straightforward: start with what each workload needs, then choose the environment.
What Each Architecture Actually Offers
Centralised / Public Cloud
Elastic, scalable, and low on upfront capital. The right environment for variable workloads, burst demand, and rapid prototyping. Cost becomes unpredictable at high volume, and data sovereignty constraints are real — hyperscalers cannot guarantee UK data residency in all configurations. Skills and tooling are mature; the trade-offs are well understood.
Edge Computing
Processes data close to its source, delivering low latency for manufacturing automation, real-time AI inference, and high-throughput IoT environments. Reduces bandwidth costs significantly. The operational complexity is genuine — distributed hardware requires rigorous management, and the skills gap is real. Notably, 84% of IT leaders want solutions that consolidate edge and cloud operations (Forrester/Microsoft, 2025), which signals that governance, not capability, remains the hard problem.
Hybrid Architecture
Not a compromise, but a deliberate choice to match workload requirements to the environment that serves them best. Gartner predicts over 40% of leading enterprises will have formally adopted hybrid computing paradigm architectures by 2028, up from 8% today. Deloitte describes the mature model as three-tier: cloud for elasticity, private infrastructure for consistent high-volume workloads, edge for immediacy. The discipline hybrid demands is governance — workload classification, placement criteria, and regular review cycles.
Four Questions That Should Drive the Decision
1. How latency-sensitive is this workload?
If a delay of more than a few milliseconds creates a business problem (i.e. real-time inference, financial transactions, operational technology), then centralised cloud is structurally disadvantaged. Edge or private infrastructure is preferential.
2. How predictable is the demand pattern?
Predictable, high-volume workloads consistently cost less on private or colocated infrastructure over time. Variable workloads benefit from cloud elasticity. Most organisations have both, which means most organisations need more than one answer.
3. Where does your data need to live?
GDPR, and sector-specific compliance obligations are tightening, and rightly so. The question of where data must reside should come before the question of which architecture to use. Map your data residency requirements first and let them inform architecture decisions.
4. What does your team have the capacity to operate?
Edge computing requires discipline to manage at scale. Hybrid requires governance frameworks. Cloud repatriation requires refactoring skills that cloud-era teams may not have built. Infrastructure decisions that ignore operational capacity tend to fail in implementation, not design.
The Right Strategy Is the One That Matches Your Workloads
The right infrastructure strategy in 2026 is not necessarily edge, cloud, or hybrid. It is the one that emerges from an honest classification of what each workload actually needs. For most organisations, that process produces a hybrid result, and that is not a convenient middle ground, it is what the evidence points to.
The organisations doing this well are not chasing a new orthodoxy. They are building the discipline to place workloads deliberately, and to revisit those placements as requirements change. That may be a more demanding way to run infrastructure, but it is also a more honest one.