Bedrock vs Microsoft Foundry vs Google’s agent platform

Large organisations often build AI on the cloud they already use. Amazon Bedrock, Microsoft Foundry and Google’s Gemini Enterprise Agent Platform each offer many models, agent tooling and enterprise controls inside AWS, Azure or Google Cloud. Two of them were recently renamed: Microsoft Foundry was previously Azure AI Foundry and Azure AI Studio, and Google’s former Vertex AI address now leads to Gemini Enterprise Agent Platform.

The short answer

• Choose Amazon Bedrock if you run on AWS and want many foundation models, Guardrails and AgentCore for production agents.
• Choose Microsoft Foundry if you run on Azure or Microsoft 365 and want model routing, governance and agents that roll out to Office apps.
• Choose Gemini Enterprise Agent Platform if you run on Google Cloud and want Gemini and partner models plus classic ML training.

Side by side

Stay where your data lives

For most enterprises the deciding factor is existing infrastructure. Data, identity, networking and compliance work already done in one cloud are costly to repeat elsewhere. All three platforms offer access to many models, including some of the same third-party ones, so model availability rarely forces a move by itself. Check regional availability for the specific models you need.

Agents and governance

AWS’s AgentCore deploys agents built with any framework and adds authentication and observability. Microsoft emphasises a single control plane over agents, tools and knowledge, with Defender protection and rollout into Microsoft 365. Google focuses on agents grounded in enterprise data, alongside training and tuning of custom models.

Renames and documentation

Because Microsoft and Google have renamed their platforms, tutorials and console labels may still say Azure AI Foundry, Azure AI Studio or Vertex AI. Search under both old and new names, and confirm that guides match the current interface.

How to decide

1. Start from the cloud where your data and identity already live.
2. List the models you need and confirm availability in your regions.
3. Prototype one agent with your real data and compare governance features.
4. Estimate costs per model and service tier, including reserved capacity.
5. Plan for multi-model routing so you are not tied to a single model family.

Amazon BedrockCodingMicrosoft FoundryCodingGemini Enterprise Agent PlatformCodingAnthropic APICodingOpenAI APICodingCohereCoding

Tools to explore

Google AI StudioCodingGeminiProductivityMicrosoft 365 CopilotProductivityFactoryCodingJulesCodingMistral AICoding