Skip to content
CSA Loom — the Microsoft Fabric experience for Azure tenants where Fabric isn't yet available: lakehouses, warehouses, notebooks, semantic models, Activator rules, Data Agents, across Commercial, GCC, GCC-High, and DoD IL5

For Azure Government (GCC-High / IL4)

Comparative positioning note

This document is written from the perspective of Microsoft Azure, Cloud Scale Analytics, and CSA Loom. Any description of third-party or competing products, services, pricing, or capabilities is derived from publicly available documentation and sources believed accurate at the time of writing, and is provided for general comparison only. We do not claim expertise in, or authority over, any non-Microsoft product or service; the respective vendor's official documentation is the authoritative source for their offerings, which may change over time. Nothing here is intended to disparage any vendor — where a competing product has genuine advantages, we aim to note them honestly. Verify all third-party details against the vendor's current official documentation before making decisions.

The full Gov boundary: Azure Government cloud + M365 GCC-High tenant. FedRAMP High + DoD IL4 + ITAR-eligible.

What "GCC-High" means here

Layer Where it lives
Microsoft 365 tenant M365 GCC-High
Azure subscriptions Azure Government (usgovvirginia, usgovtexas, usgovarizona)
Compliance FedRAMP High + DoD IL4 + ITAR-eligible

Prerequisites

Item Notes
Azure Government subscription Deployed to via GitHub Actions only — see Deploy
Region usgovvirginia recommended (most services + AOAI chat models) + usgovarizona for OpenAI embeddings
Microsoft 365 GCC-High tenant Identity provider
AZURE_GOV_* repository secrets The deploy identity. Missing secrets do not fail the workflow — they make it skip and report success
AOAI quota in usgovvirginia gpt-4o + gpt-4.1 + o3-mini + gpt-5.1
AOAI quota in usgovarizona text-embedding-3-large (Standard mode is usgovarizona-only)

Optional — only with LOOM_SEMANTIC_MODEL_BACKEND=powerbi. Azure Analysis Services is not offered in Azure Government, so the default GCC-High semantic-model path is the Loom-native tabular layer (definition persisted with the item, emitted as TMSL, queried via Synapse Serverless OPENROWSET(... FORMAT='DELTA')) — no Power BI capacity is required. If you opt into the Power BI backend, GCC-High supports F-SKU (F8 minimum).

Critical GCC-High dispatch deltas (vs Commercial)

Per Per-boundary dispatch matrix:

Component GCC-High
Container host AKS (Container Apps not at IL4+)
Functions host Premium EP1 (Flex not in Gov)
APIM Classic Premium (v2 not confirmed in Gov)
Catalog primary Microsoft Purview (UC managed not yet in Gov)
Databricks Classic clusters + Hive metastore (no UC, no SQL Warehouse)
SQL Warehouse Synapse Serverless (no Databricks SQL Warehouse in Gov)
Agent orchestration Microsoft Agent Framework + AOAI direct (Foundry Agent Service Gov-GA unconfirmed)
Foundry portal Not available (use classic Azure ML Hub)
Defender for Cloud AI Threat Protection Commercial-only (use Sentinel pipeline)
OpenAI Batch API Not in Gov (use synchronous calls or provisioned throughput)
OpenAI Content Safety Not in Gov (use self-hosted Presidio)

Deploy — Actions only

There is no local-CLI deployment path for GCC-High. All Azure Government deployment and verification runs through GitHub Actions; running Azure Government az commands from a workstation is prohibited. The workflows carry the Gov login endpoints and the AzureUSGovernment cloud switch internally.

An earlier version of this page published az cloud set --name AzureUSGovernment + az login + a single az deployment sub create as the GCC-High procedure. That contradicted the Actions-only rule stated in Greenfield, and it also omitted deployAppsEnabled=false, the image phase and the bootstrap phase — so even setting the boundary rule aside, it could not have produced a working Console on a fresh subscription. It is corrected here rather than quietly removed.

The full walkthrough, including the three ways a green Gov run can mean nothing, is in Greenfield → Azure Government. Brownfield adoption on this boundary is in Brownfield → Azure Government.

# 1. Build the app images into the sovereign ACR.
gh workflow run gov-build-images.yml -f boundary=GCC-High

# 2. Provision + bootstrap. All three inputs differ from the defaults, and all
#    three matter: whatif-only provisions nothing, and keep_resources=false
#    tears the estate down and skips the post-deploy bootstrap.
gh workflow run deploy-fiab-gcch.yml \
  -f run_mode=full \
  -f topology=tenant \
  -f keep_resources=true

The deploy requires manual approval on the gcc-high-deploy environment protection rule before it touches the Gov subscription.

Verification is also Actions-only:

gh workflow run gov-verify-facts.yml
gh workflow run gov-gates.yml

Prerequisite: the AZURE_GOV_CLIENT_ID / _SECRET / _TENANT_ID / _SUBSCRIPTION_ID repository secrets. If any is missing, the workflow warns, skips every downstream job, and still concludes success — see when a green Gov run means nothing.

gcc-high.bicepparam sets: - environment = 'AzureUSGovernment' - location = 'usgovvirginia' - All endpoint suffixes use .usgovcloudapi.net / .us / .databricks.azure.us - containerPlatform = 'aks' - apimSku = 'Premium' (classic) - functionsHostSku = 'EP1' - databricksUnityCatalogEnabled = false - databricksSqlWarehouseEnabled = false - databricksMetastore = 'hive' - catalogPrimary = 'purview' - agentOrchestrator = 'maf' - foundryPortalEnabled = false - defenderForAIEnabled = false - contentSafetyEnabled = false - openaiBatchEnabled = false - openaiLocation = 'usgovvirginia' - openaiEmbeddingsLocation = 'usgovarizona' - openaiChatModel = 'gpt-4o' - powerBiSku = 'F64'

Plus compliance tags:

param complianceTags = {
  Environment: 'GCC-High'
  FedRAMP_Level: 'High'
  DISA_IL: 'IL4'
  Data_Classification: 'CUI'
  M365_Boundary: 'GCC-High'
}

Bootstrap with an existing console app registration

The post-deploy bootstrap (csa-loom-post-deploy-bootstrap.yml) normally discovers the console app registration by listing app registrations. Listing is tenant-wide, so it needs the Graph permission Application.Read.All or a directory role; a deploy service principal with neither fails there with Insufficient privileges to complete the operation (GCC-High run 39319044221), which skips the Estate.Power role step and the repo-variables step after it.

If the console app registration already exists (operator-owned), supply it instead. Brownfield rule R5: an existing value is a first-class input.

Input (dispatch / workflow_call) Repo variable fallback Meaning
console_app_client_id LOOM_CONSOLE_APP_CLIENT_ID Client (app) id of the existing console app. Not a secret.
console_app_object_id (optional) LOOM_CONSOLE_APP_OBJECT_ID Object id; read by object id and cross-checked against the client id.
none LOOM_CONSOLE_APP_NAME (optional) Display name the app must carry.

When the client id is supplied:

  • The MSAL step reads that one app by id, validates it (app id, object id, signInAudience = AzureADMyOrg, optional display name) and changes nothing if it is not the expected app. It never lists app registrations and never creates one. A refusal is reported as could not read, existence unknown (naming Application.Read.All), a 404 as not found; the two are not mixed up. Not covered: the later az ad app ... --id calls in that script resolve the id with the Azure CLI's own filtered lookup, so the deploy identity still needs to be able to read that one app.
  • The Estate.Power step reads the app and the two service principals by id (no listing). If the role and its assignment already exist it makes zero writes. It runs even when the MSAL step failed, and the repo-variables step runs whenever the Estate.Power step succeeded.
  • The audience written to CSA_LOOM_GCCH_CONSOLE_AUDIENCE is derived from the app's live identifier URIs: api://<appId> when that URI is set, the bare app id when the app exposes none.

Without the input or variable nothing changes: discover-or-create, as before.

What needs an admin and what does not

Action Needs
List app registrations tenant-wide (discovery) Application.Read.All or a directory role
Read one named app by id Application.Read.All, or Application.ReadWrite.OwnedBy with the deploy SP as an owner of the app, or a directory role
Edit the app's appRoles (create Estate.Power) Application.ReadWrite.OwnedBy + ownership of the app, or Application.ReadWrite.All, or Application Administrator
Assign the role to the deploy SP (appRoleAssignment) AppRoleAssignment.ReadWrite.All or a directory role; app ownership does not cover it
Everything above, once the role and assignment exist Reads only; no admin action per run

The permission names in this table come from the Microsoft Graph permission reference and were not exercised against a live Gov tenant; whether adding the deploy SP as an owner of the app, with no Graph permission, is enough for an app-only token has not been verified, and the step's error text says so. If the Estate.Power step is refused (exit 3) the log names the exact change that was missing and the fix. Once the operator has created the role and the assignment, re-runs need no admin.

Deploy time

  • Admin Plane: 45-65 min (AKS cluster spin-up is slower than Container Apps)
  • First DLZ: 20-40 min
  • Total to working Console: 70-110 min

Validation

Gov verification is Actions-only (gov-verify-facts.yml / gov-gates.yml, above). A health probe from a machine that can reach the Gov console:

curl -i https://<your-gov-console-hostname>/api/health

Then sign in via browser using your GCC-High M365 identity. A curl 200 is not a verification receipt — a deploy is verified by a live in-browser walk.

Compare the result against the recorded Gov readiness ceiling, not against Commercial: the Gov ceiling is infrastructure-bound and legitimately lower, so a Gov estate reporting fewer green capabilities than Commercial is not necessarily broken.

ITAR considerations

GCC-High supports ITAR-eligible workloads. Customer responsibility: - Mark ITAR-restricted data with sensitivity labels - Apply Purview ITAR classification rules - Verify cross-cloud B2B is disabled or scoped per ITAR policy - Configure Sentinel rules to detect ITAR-data egress

See ITAR compliance page.

Cost (Azure-native GCC-High baseline)

GCC-High pricing is typically 10-25% above Azure Commercial:

Component Approximate $/month
Databricks Premium classic $600-3,000
Synapse Serverless $5-50
ADX cluster D14_v2 $600
ADLS Gen2 (10 TB) $250
AOAI Gov (50K TPM gpt-4o) $250-600
AI Search S1 $300
Purview $350
AKS cluster + workloads $200-500
Misc (KV Premium HSM, LA, Sentinel) $200
Total ~$2,800-5,800/mo

Optional add-on — only with LOOM_SEMANTIC_MODEL_BACKEND=powerbi: a Power BI Premium F8 capacity for the opt-in Power BI backend adds ~$1,200/mo. It is not part of the Azure-native baseline above — the default Loom-native / Synapse Serverless semantic-model path carries no separate capacity charge.

Forward migration

When Fabric Gov-H reaches GA: - F-SKU + Direct Lake natively available - Loom and Fabric Gov run side-by-side - Per-workload migration via Forward to Fabric runbook

Runbooks