The category confusion that produces bad federal cloud builds
Most cloud architects arrive at their first federal engagement assuming the work is a stricter version of the commercial landing zone they already know. The story they tell themselves is: same Cloud Adoption Framework, tighter policies, more audit logging, slower change control. With those three adjustments, they reason, a federal landing zone is just a hardened commercial one.
It is not. Federal-grade landing zones are a different architecture. The identity model is different. The network topology is different. The data classification handling is different. The operating-procedure overlay is different. The stakeholder relationships are different. Trying to ship a federal landing zone by tightening a commercial one is one of the most common reasons year-one federal cloud programmes run months over plan.
This post walks through what actually changes: the dimensions where federal-grade design differs from commercial CAF. Most of the specifics assume Microsoft Azure UAE North, which is a standard public Azure region located in the UAE rather than a government-only cloud. What makes a deployment there federal-grade is the landing-zone design on top of it, not the region itself. The architectural principles transfer to AWS Middle East UAE for AWS-anchored federal portfolios with the obvious vendor translations.
Identity federation: the most consequential difference
Commercial Azure landing zones federate Microsoft Entra ID with whatever the customer's primary identity source is, usually their on-premise Active Directory, sometimes a parent-company Entra ID tenant, occasionally a partner federation. The federation flow is well-understood and the configuration is largely template-driven.
Federal Entra ID federations integrate with federal directories that have specific trust, audit and operating-procedure requirements not present in commercial federation. Three things change. First, the trust relationship is governed by federal policy and requires formal approval through the federal-stakeholder process. Second, the audit logging for federated authentication events has to be retained and shareable in the form the entity's oversight bodies require, not only kept in the tenant's own log workspace. Third, the operating procedures for managing the federation, credential rotation, federation metadata refresh, trust certificate replacement, follow federal standards rather than the commercial best-practice the engineer might be used to.
For cleared work, identity becomes more involved. The entity's personnel security vetting determines which engineers can administer which parts of the tenant, and PIM enforces that decision at the role level. Cleared administrators have access. Uncleared ones do not, not just at the role level but at the network and physical-access level. The audit trail for any clearance-gated action is part of the standard federal evidence base.
Classification-driven residency enforcement
Commercial residency is binary. Either the workload is in the residency-compliant region or it is not. Configuration is a per-resource decision.
Federal classification is multi-tier. Different classification levels have different residency requirements, different access controls, different audit-logging characteristics and different egress controls. The landing zone has to encode classification as a structural property of where workloads land, not as a per-resource property managed by humans. Each classification level gets its own management group branch, and Azure Policy assigned at that branch enforces it: allowed regions are restricted to the UAE, and Deny policies reject any resource whose classification tag does not match the branch it is being deployed into.
Public cloud is not the answer for every tier. The highest classification levels may not be permitted in a public Azure region at all, whatever the landing-zone design, and belong on-premises or in a sovereign cloud offering approved for that level. The landing zone should make that boundary explicit rather than leave it to individual project decisions.
Classification labelling has to flow consistently from data to workload to resource to log. The same classification tag a data steward applies to a document needs to map cleanly to the resource group that hosts the application processing that document and to the log workspace that stores the audit trail. Drift between these layers is one of the most common federal audit findings.
Federal-stakeholder monitoring integration
Commercial landing zones stream logs to a SIEM the customer operates (or a managed SOC the customer engages). The audit boundary stops at the customer's edge.
Federal entities answer to more than their own SOC. Incident reporting and information-sharing obligations towards the UAE Cyber Security Council and, where relevant, sector regulators mean the landing zone has to be able to produce and share selected log categories and incident data on request and within the required timelines. Where an entity's obligations call for automated sharing, the technical side is straightforward: export through Event Hub or an API from the SIEM. The operating model is what changes. The customer's SOC has to work on the assumption that incidents will be reported outward and reviewed by someone else.
This is not a bad thing once the operating model adjusts to it. National-level threat intelligence brings visibility that a single customer's SOC cannot match, and the relationship with oversight bodies becomes part of the operational defence. The first few months of any federal engagement are mostly the operating team learning to work with this rhythm rather than against it.
Network topology and the role of ExpressRoute
Federal landing zones rely heavily on private connectivity, and it helps to be precise about what ExpressRoute does. An ExpressRoute circuit connects the entity's own network to Azure: it is ordered from a connectivity provider at a UAE peering location, often added as another site on the entity's existing MPLS/IPVPN WAN, and terminates on the ExpressRoute gateway in the hub. It does not connect Azure directly to other government entities.
Inter-entity traffic to federal directories, shared government platforms and partner entities travels over the government networks and circuits the entity is already approved to use, landing at its own data centre or network edge. Azure workloads reach those services through the entity's network over ExpressRoute rather than over the internet, and public internet egress is restricted by policy for classification-bearing workloads.
The hub-and-spoke design therefore has a different shape. The hub is denser: redundant ExpressRoute circuits across diverse providers or peering locations, additional firewall capacity for inter-entity traffic, additional security inspection points. The spokes are more segmented, classification levels have their own spokes with the egress rules enforced at the spoke boundary, not the resource boundary.
Operating-procedure overlay
Commercial change management runs on the customer's own cadence. Federal change management runs on the federal-entity's cadence, which is usually slower for production-affecting changes and has additional approval steps for changes that touch federated identity, audit logging or classification-handling configuration.
This is the dimension that surprises commercial-experienced cloud engineers the most. A configuration change that takes hours in a commercial engagement may take weeks in a federal engagement. Not because the federal change-management is inefficient, but because the change has stakeholder reach that the commercial change does not. The operating model has to accept this and plan against it.
What does not change
Three things are common to commercial and federal landing zones. The underlying Azure resource model. The Microsoft technical documentation. The hyperscaler's shared-responsibility model. These three are the reason cloud engineers think federal is "just a stricter commercial". They recognise the building blocks. The architecture that uses those building blocks is what changes.
Bottom line
Federal-grade landing zones are a different architecture, not a stricter commercial one. The identity federation is different. The classification handling is structural rather than per-resource. The network topology assumes private connectivity through the entity's own network. The operating model accepts federal-stakeholder reach as part of standard operations.
For federal entities planning a cloud journey, the practical lesson is that year one is design-heavy and stakeholder-heavy. By year two, with the landing zone signed off and the operating relationships established, the engagement looks much more like a normal managed-cloud rhythm. Continuity matters here more than in almost any other category of work. The relationships with federal stakeholders compound, and so does the trust.