Skip to main content

Microsoft Foundry & AI Platform — Consolidated Enterprise Guidance

Prepared by: Microsoft Cloud Solution Architecture
Date: April 2026
Audience: Enterprise AI Platform & Engineering Teams
Context: Microsoft Foundry Adoption, Multi-Region AI Resilience, EU Capacity Planning


Table of Contents

  1. Executive Summary
  2. Question 1 — Microsoft Foundry Roadmap & Key Features
  3. Question 2 — Private Endpoint Support for Foundry
  4. Question 3 — Cloud Model Capacity in EU Regions
  5. Question 4 — Responses API Regional Availability
  6. Architecture — Multi-Region AI Platform Resilience
  7. Gaps & Limitations — Transparent Assessment
  8. Recommended Actions
  9. Microsoft Learn Reference Links

1. Executive Summary

Microsoft Foundry (formerly Azure AI Studio / Azure AI Foundry) is Microsoft's unified platform for building, evaluating, and deploying AI applications and agents. It brings together Azure OpenAI models, the Foundry model catalog, agent services, evaluation tools, and enterprise networking under a single platform.

Key Facts for Enterprise AI Platform Teams

AreaCurrent StatusImpact
Foundry platformGA — with features continuing to roll out by regionProduction-ready for core scenarios
Private endpointsGA for Foundry Tools; Managed VNet is in previewEnterprise networking supported, but managed VNet has limitations
Agent ServiceGA — depends on Responses API region availabilityAvailable in 25+ regions including EU
Responses APIGA in 25 regionsNot available in West Europe — see Section 5
Model capacity (EU)Varies by model and deployment typeCapacity can sell out; use Data Zone or Global deployments for flexibility
EvaluationsGA in 27 regions including West EuropeBroadly available

Key Limitation: The Responses API — which powers the Foundry Agent Service — is not available in West Europe as of April 2026. This has direct implications for multi-region AI resilience strategies that include West Europe as a failover region.


2. Question 1 — Microsoft Foundry Roadmap & Key Features

Customer Question

"What are the upcoming features and roadmap for Microsoft Foundry, including key capabilities available now and coming soon?"

Current GA Features

FeatureStatusDescription
Azure OpenAI Model DeploymentsGADeploy GPT-4o, GPT-4.1, o3, o4-mini, and newer models
Foundry Model CatalogGAAccess 1,800+ models from Microsoft, Meta, Mistral, Cohere, and others
Foundry Agent ServiceGABuild AI agents with tools (code interpreter, file search, function calling, Bing grounding, MCP)
EvaluationsGABatch evaluation, agent response evaluation, safety evaluations
Prompt FlowGABuild and orchestrate LLM workflows
Fine-TuningGAFine-tune GPT-4o, GPT-4o-mini, and other supported models
Content SafetyGAConfigurable content filters, guardrails, jailbreak detection
Private LinkGAPrivate endpoint support for Foundry resources
Custom VNet (BYO)GADeploy Foundry agents into customer-managed virtual networks
Managed VNetPreviewMicrosoft-managed network isolation for agents

Key Capabilities by Category

Models & Deployment Types

Deployment TypeDescriptionData Residency
StandardPay-per-token, shared capacityRegional
Global StandardOptimized routing across regionsData may cross regions
Data Zone StandardRegional data processing within a geographyWithin geography (EU/US)
Provisioned Throughput (PTU)Reserved capacity with guaranteed TPMRegional
Data Zone ProvisionedReserved capacity within geographyWithin geography
Global ProvisionedReserved capacity with global routingData may cross regions

Recommendation for EU enterprises: Use Data Zone Standard or Data Zone Provisioned to keep data within EU geography while accessing broader capacity pools.

Agent Service Tools

ToolDescriptionRegion Restrictions
Code InterpreterExecute Python code in sandboxAvailable in most regions
File SearchRAG over uploaded filesNot available in Italy North, Brazil South
Function CallingCall external APIsBroadly available
Bing GroundingGround responses in web searchBroadly available
Azure AI SearchRAG over enterprise dataBroadly available
MCP (Model Context Protocol)Connect to external tool serversBroadly available
Computer UseAutomated UI interactionLimited — only East US 2
Image GenerationDALL-E powered image creationBroadly available

Reference: Tool support by region and model


3. Question 2 — Private Endpoint Support for Foundry

Customer Question

"Does Microsoft Foundry support private endpoints, and how can we ensure network isolation for AI workloads?"

Answer: Yes — Two Options Available

Option 1: Custom VNet (BYO) — GA

Deploy Foundry agents into a customer-managed virtual network with full control:

  • You provide a VNet with a dedicated subnet delegated to Microsoft.App/environments
  • Full control over firewall, UDR, network peering
  • Private endpoints to all dependent services (Storage, Cosmos DB, AI Search, Key Vault)
  • Supports production workloads

Reference: Configure custom virtual networks for Agents

Option 2: Managed VNet — Preview

Microsoft-managed network isolation:

FeatureManaged VNetCustom VNet (BYO)
Setup complexitySimple — Microsoft managesMore complex — customer manages
Network controlLimited — Microsoft-managed firewallFull control — BYO firewall, UDR, peering
Private endpointsManaged PEs to Azure servicesStandard PEs in your VNet
Isolation modesAllow Internet Outbound / Allow Only Approved OutboundFull customer control
On-premises accessVia Application GatewayVia ExpressRoute / VPN Gateway
LoggingNo outbound traffic loggingFull NSG flow logs
StatusPreviewGA

Managed VNet supports private endpoints to:

  • Azure Storage, Cosmos DB, AI Search, Key Vault, ACR
  • Azure SQL, PostgreSQL, MySQL
  • Event Hubs, Redis, Application Insights
  • Azure Data Factory, Databricks
  • Other Foundry resources

Managed VNet supported regions (as of April 2026): East US, East US 2, West Europe, Sweden Central, Germany West Central, France Central, UK South, Japan East, Australia East, Brazil South, UAE North, Spain Central, Italy North, South Central US, West Central US, Canada East, South Africa North, West US, West US 3, South India

Gap: Managed VNet is preview only — not recommended for production if your enterprise policy prohibits preview features. Use the GA Custom VNet option instead.

Reference: Configure managed virtual network for Foundry
Reference: Configure Private Link for Foundry


4. Question 3 — Cloud Model Capacity in EU Regions

Customer Question

"We experience persistent capacity issues with cloud models in EU regions. What options exist and when will capacity improve?"

Current EU Model Availability (April 2026)

Latest Models — Standard Deployment

Regiono3o4-minigpt-4.1gpt-4.1-minigpt-4.1-nanogpt-4ogpt-4o-mini
Sweden Central
Germany West Central
France Central
West Europe
Norway East
Poland Central
Switzerland North
UK South

Provisioned Throughput — EU Availability

RegionPTU AvailableData Zone PTU
Sweden Central
Germany West Central
France Central
West Europe
Norway East
Poland Central

Understanding Capacity Constraints

ConceptExplanation
QuotaMaximum PTU/TPM your subscription can use in a region — does not guarantee capacity
CapacityActual GPU availability at deployment time — can sell out
Quota ≠ CapacityYou can have quota but no capacity if the region is full

Strategies to Mitigate Capacity Issues

StrategyDescriptionTrade-off
Data Zone deploymentsPool capacity across EU regionsData stays in EU; broader GPU access
Global StandardRoute to any available regionData may leave EU — check compliance
Multi-region deploymentDeploy same model in 2-3 EU regionsHigher cost; application routing needed
Provisioned Throughput (PTU)Reserve dedicated capacityMust commit to minimum PTUs; higher cost
Quota increase requestRequest more TPM/PTU via support formSubject to availability
Monitor capacity APICheck real-time availability before deployingProgrammatic approach

How to Check Capacity

# Query capacity API for a specific model
az rest --method GET \
--url "https://management.azure.com/subscriptions/{sub-id}/providers/Microsoft.CognitiveServices/modelCapacities?api-version=2024-10-01&modelName=gpt-4o&modelVersion=2024-08-06"

Or use the Azure AI Foundry portalQuota page to view real-time capacity by region.

Reference: Quota and capacity management
Reference: Provisioned throughput concepts

Gap — Transparent Communication: Microsoft does not currently provide proactive notifications when EU capacity is constrained. Customers must check the capacity API or portal manually. This is a known friction point. Consider engaging your Microsoft account team for priority capacity updates.


5. Question 4 — Responses API Regional Availability

Customer Question

"The Responses API is not supported in West Europe, which affects our multi-region failover. When will it be available?"

Current Responses API Region Availability

The Responses API is available in 25 regions (as of April 2026):

EU RegionsAvailable
Sweden Central
Germany West Central
France Central
Norway East
Poland Central
Switzerland North
UK South
Italy North
Spain Central
West EuropeNot available

Impact on Agent Service

The Foundry Agent Service requires the Responses API to function. This means:

  • Agent Service is NOT available in West Europe
  • Any multi-region AI resilience strategy using West Europe as a failover target for Agent Service will not work
  • Standard model deployments (Chat Completions API) do work in West Europe — only the Responses API is missing

Gap Assessment

CapabilityWest EuropeSweden CentralGermany West Central
Standard model deployments
Provisioned throughput
Evaluations
Responses API
Agent Service
Managed VNet (preview)

Workaround for Multi-Region Resilience

Recommended resilience pattern:

  1. Primary: Sweden Central (full Agent Service + Responses API)
  2. Secondary: Germany West Central (full Agent Service + Responses API)
  3. West Europe: Use for standard Chat Completions API workloads only — NOT for Agent Service

Gap — No ETA: Microsoft has not published a timeline for Responses API availability in West Europe. Engage your Microsoft account team to register demand and receive updates.


6. Architecture — Multi-Region AI Platform Resilience

Key Design Decisions

DecisionRecommendationReasoning
Primary regionSweden CentralBroadest EU feature support + Responses API
Secondary regionGermany West CentralFull Agent Service support + data sovereignty
Data Zone deploymentsUse for flexibilityEU data residency with pooled capacity
West Europe roleChat Completions + Evaluations onlyNo Responses API / Agent Service
Cosmos DBMulti-region writesLow-latency data access for agents
AI SearchIndex replicas in both primary/secondaryRequired for RAG-based agents

7. Gaps & Limitations — Transparent Assessment

GapImpactMitigationStatus
Responses API not in West EuropeAgent Service unusable in West Europe; breaks failover assumptionsUse Sweden Central + Germany West Central for agent workloadsNo ETA for resolution
Managed VNet is previewCannot use in production if preview features are prohibitedUse Custom VNet (GA) for production agent deploymentsPreview — GA timeline unclear
Capacity constraintsModels may not be deployable in desired region even with quotaUse Data Zone / multi-region deployment strategyOngoing — Microsoft adding GPU capacity
No proactive capacity alertsCustomers discover constraints only at deployment timeIntegrate capacity API checks into deployment pipelinesFeature gap
Agent tool region gapsFile Search not in Italy North / Brazil South; Computer Use only in East US 2Deploy in regions with full tool supportRegion expansion ongoing
Model availability lagLatest models may not be available in all EU regions immediatelyUse Global Standard for early access (if data residency allows)Normal — models roll out gradually
Foundry roadmap not publicCustomers cannot plan ahead with confidenceEngage Microsoft account team for roadmap briefingsKnown friction

#ActionOwnerPriority
1Deploy Foundry resources in Sweden Central (primary) and Germany West Central (secondary)AI Platform TeamP0
2Configure Private Link using Custom VNet (GA) for production agent workloadsPlatform / Network TeamP0
3Evaluate Data Zone deployments to maximize EU capacity accessAI Platform TeamP1
4Integrate capacity API into deployment automation to pre-check availabilityDevOps TeamP1
5Request Provisioned Throughput (PTU) for mission-critical AI workloadsAI Platform Team + MicrosoftP1
6Engage Microsoft account team for Responses API West Europe timelineMicrosoft CSAP1
7Request regular capacity and roadmap updates from MicrosoftMicrosoft CSAP2
8Test Managed VNet preview in non-production to prepare for GAAI Platform TeamP2

Microsoft Foundry — Core

TopicURL
Microsoft Foundry Architecturehttps://learn.microsoft.com/azure/foundry/concepts/architecture
Foundry Region Supporthttps://learn.microsoft.com/azure/foundry/reference/region-support
Foundry Security Baselinehttps://learn.microsoft.com/security/benchmark/azure/baselines/azure-ai-foundry-security-baseline

Networking & Private Endpoints

TopicURL
Configure Private Link for Foundryhttps://learn.microsoft.com/azure/foundry/how-to/configure-private-link
Managed VNet for Foundry (preview)https://learn.microsoft.com/azure/foundry/how-to/managed-virtual-network
Custom VNet for Agentshttps://learn.microsoft.com/azure/foundry/agents/how-to/virtual-networks
AI Services Private Endpointshttps://learn.microsoft.com/azure/ai-services/cognitive-services-virtual-networks

Models & Capacity

TopicURL
Model Availability by Regionhttps://learn.microsoft.com/azure/ai-foundry/openai/concepts/models
Provisioned Throughput Conceptshttps://learn.microsoft.com/azure/ai-foundry/openai/concepts/provisioned-throughput
Quotas and Limitshttps://learn.microsoft.com/azure/ai-foundry/openai/quotas-limits
Quota Increase Request Formhttps://aka.ms/oai/stuquotarequest

Agent Service

TopicURL
Agent Service Limits, Quotas, Regionshttps://learn.microsoft.com/azure/foundry/agents/concepts/limits-quotas-regions
Tool Best Practices & Region Matrixhttps://learn.microsoft.com/azure/foundry/agents/concepts/tool-best-practice

Responses API & Evaluations

TopicURL
Responses API (Region Availability)https://learn.microsoft.com/azure/foundry/openai/how-to/responses
Evaluation Region Supporthttps://learn.microsoft.com/azure/foundry/concepts/evaluation-regions-limits-virtual-network

Document prepared based on Microsoft Learn documentation as of April 2026. AI service capabilities and regional availability evolve rapidly — always verify against the latest Foundry documentation and the model availability page.

📖Learn