Last reviewed: August 2026
Not every workload is suited to the public cloud. There is demand for extending cloud infrastructure to on-premises/edge environments for reasons such as data sovereignty, ultra-low latency, and protecting existing investments.
| Case |
Requirement |
Suitable solution |
| Data sovereignty |
Data must not leave a specific country/facility |
Dedicated Region, Azure Stack, GDC Air-gapped |
| Ultra-low latency |
Factory automation, real-time gaming, AR/VR |
Local Zones, Wavelength, Edge Zones |
| Protecting existing investment |
On-premises equipment still has useful life + wants to use cloud services |
Arc, GKE Enterprise, EKS Anywhere |
| Regulatory (network isolation) |
Environment with no internet connectivity |
GDC Air-gapped, Dedicated Region |
| Pattern |
Description |
Example |
| Split-stack |
Different vendors for different layers |
Frontend (AWS CloudFront) + backend/DB (OCI) |
| Data gravity |
Keep large-volume data in one place; access it from multiple vendors for analytics/AI |
Data lake (GCS) + ML (Vertex AI) + serving (AWS) |
| Best-of-breed |
Choose the optimal vendor per service area |
AI (Google Cloud) + enterprise apps (Azure) + DB (OCI) |
| Cloud-bursting |
On-premises normally, burst to public cloud at peak |
On-premises K8s + EKS/AKS/GKE bursting |
| DR/Failover |
Switch to a secondary vendor if the primary vendor has an outage |
AWS (primary) + Azure (DR) |
- Underestimating network latency in hybrid architectures — Designing synchronous calls without accounting for round-trip latency (tens of ms) between on-premises and cloud sharply degrades performance.
- Overlooking the operational burden of edge equipment — Deploying to the edge means patching, incident recovery, and monitoring must be done remotely in environments with difficult physical access. Adopting this without operational automation causes management costs to spike.
- Defaulting to multi-cloud — Adopting multi-cloud without a clear business need (regulation, DR, avoiding vendor lock-in) only increases complexity and cost.