Microsoft Foundry, Amazon Bedrock, and Gemini Enterprise Agent Platform (formerly Google Vertex AI) are the foundation layer beneath the rest of the landscape. They package model access, agent services, safety controls, SDKs, and deployment infrastructure differently.
For regulated enterprises, this is the control-plane decision before any individual tool choice. The platform sets the baseline for identity, network boundaries, model access patterns, logging, regional deployment options, and how much governance is available natively versus assembled later.
Read this page top-down: compare the platforms first, use the category mapping section to see where each cloud lands in agents, orchestration, governance, and assistants, then use the detailed comparison table for deeper vendor-to-vendor evaluation.
Microsoft
Unified enterprise AI platform spanning models, agents, tools, evaluations, and governance.
Why teams pick it: Broadest model catalog plus agents, governance, and sovereign deployment options.
AWS
Managed foundation model platform with Bedrock Agents, AgentCore, and Bedrock Guardrails.
Why teams pick it: Framework-agnostic model access with strong AWS-native agent and safety integration.
Enterprise GenAI platform combining Model Garden, agent tooling, MLOps, and safety controls.
Why teams pick it: Model Garden plus Gemini, Agent Development Kit, and enterprise MLOps in one stack.
Protocol support is starting to shape enterprise platform choices. This snapshot tracks where the major cloud foundations already expose protocol alignment and surfaces recent dataset mentions for MCP, A2A, and OpenAPI.
Model Context Protocol is becoming the default interoperability layer for tool and service connections.
AWS introduced AgentCore payments in preview, adding managed agent micropayments for APIs, web content, MCP servers, and other agents via x402-backed transaction flows.
Mastra 1.32.0 introduced fine-grained authorization checks across agent runs, tool and workflow execution, memory thread access, server adapters, and MCP paths.
Agent2Agent support matters where teams expect cross-agent handoffs instead of single-vendor agent silos.
Microsoft Agent Framework 1.7.0 added harness agents, A2A session support, and new Foundry integration hooks, extending the framework's orchestration and inter-agent workflow surface.
Microsoft Agent Framework 1.4.0 introduced experimental skills API breaking changes, migrated its A2A layer to a2a-sdk v1.0, and tightened default DevUI access controls and CORS posture.
OpenAPI still matters as the stable enterprise contract for API exposure, governance review, and gateway alignment.
This bridges the platform layer with the category hubs, so visitors can quickly see the native cloud product most relevant to each tracked category.
| Category | Microsoft Foundry | Amazon Bedrock | Gemini Enterprise Agent Platform |
|---|---|---|---|
| Agent Frameworks | Foundry Agent Service | Bedrock Agents + AgentCore | Agent Builder + ADK |
| Orchestration | Logic Apps (AI-enhanced) | Step Functions | Cloud Workflows |
| Governance | AI Content Safety | Bedrock Guardrails | Model Armor |
| Assistants (Coding) | GitHub Copilot | Amazon Q Developer | Gemini Code Assist |
| Assistants (Productivity) | M365 Copilot | Amazon Q Business | Gemini for Workspace |
| Assistants (Build Your Own) | Copilot Studio | Amazon Q Apps | Gemini Enterprise |
Use the full table below after the platform summaries above. This is the detail layer for direct vendor-to-vendor comparison rather than the best entry point for first-time visitors.
| Attribute | Microsoft Foundry | Amazon Bedrock | Gemini Enterprise Agent Platform |
|---|---|---|---|
| Formerly known as | Azure AI Foundry / Azure AI Studio / Azure OpenAI Service | AWS Bedrock (common informal name) | Vertex AI; AI Studio is a separate free playground |
| What it is | Unified PaaS for enterprise AI: models, agents, tools, evaluations, governance | Managed service for foundation models with agent building and guardrails | Full ML + GenAI platform: model garden, agent engine, MLOps, evaluation |
| Model catalog | 11,000+ models | 100+ foundation models | 200+ via Model Garden |
| Unique models | GPT-5.4, GPT-5.4 Pro, Phi-4, MAI-Image-2, Sora 2 | Amazon Nova, Amazon Titan | Gemini 3.1 Pro, Nano Banana Pro, Veo 3.1, Imagen 4 |
| Agent service | Foundry Agent Service (GA) | Bedrock Agents + AgentCore (GA) | Agent Builder + ADK + Agent Engine |
| Guardrails | AI Content Safety | Bedrock Guardrails | Model Armor + Vertex AI Safety |
| Protocol support | MCP, A2A, OpenAPI | MCP, A2A | MCP, A2A, OpenAPI |
| SDK languages | Python, C#, JS/TS, Java | Python, JS, Java, Go, .NET | Python, Java, Go, TypeScript |
| Unified SDK | azure-ai-projects v2 | boto3 + AWS SDK | google-cloud-aiplatform |
| Connectors | 1,400+ via Logic Apps + Tool Catalog | 220+ AWS services | GCP services + Workspace |
| Free tier | $200 Azure credit | $200 AWS credit | $300 GCP credit + AI Studio free playground |
| Pricing model | Pay-per-token + per-tool | Pay-per-token + per-feature | Pay-per-token + per-compute |
| On-premises | Yes (Foundry Local) | No | No |
| MCP server | Yes (mcp.ai.azure.com) | Preview | Preview |
| EU data residency | Yes | Yes | Yes |
| SOC 2 / ISO 27001 | Yes / Yes | Yes / Yes | Yes / Yes |
| HIPAA | Yes | Yes | Yes |
| Key updates | SDK 2.0 GA, Agent Service GA, GPT-5.4 GA, PromptFlow deprecated | AgentCore GA, Automated Reasoning GA, 85% guardrails price cut | ADK open-sourced, Gemini 3.1 Pro, Gemini Enterprise rename, Wiz acquisition |
Use the category hubs to drill from the cloud foundation layer into tracked agent, orchestration, governance, assistant, and update pages.
Compare managed cloud agent stacks with open source agent frameworks used in enterprise deployments.
Review workflow engines, pipeline builders, and automation layers connected to the platform layer.
Check guardrails, safety controls, and policy tooling mapped across the major cloud vendors.
Explore coding assistants, productivity copilots, and build-your-own assistant platforms.
Follow releases, deprecations, acquisitions, and other market changes across the tracked landscape.
Review sourcing standards, contribution rules, and project scope before editing tracked platform data.