Azure AI Foundry vs AWS Bedrock vs Google Vertex AI: Where to Build Enterprise AI
When you build your own AI applications and agents, rather than buying an assistant, you choose a platform to run them on. All three major clouds now offer models from many providers, including several of the same ones. So the model menu rarely decides it any more. Your cloud commitments, where your data must stay and how you buy do. Here is the honest comparison.
Build where your data and your cloud commitments already are, and keep the option to switch. Microsoft Foundry suits Azure and Microsoft 365 estates, Amazon Bedrock suits AWS estates, and Google's platform, formerly Vertex AI and now the Gemini Enterprise Agent Platform, suits Google Cloud estates and Gemini first strategies. All three offer models from several providers, including Anthropic's Claude. The differences that matter are where each model can run, how third party models are bought and contracted, and how much of your build is tied to one cloud.
C4C is an independent technology consultancy that helps organisations build, buy and govern enterprise AI. We advise on which platform fits, design the architecture, and negotiate the commercials, drawing on years spent on the vendor side building enterprise software and cloud deals. Send us your situation and we will give you a straight view.
First, the names
Two of the three have been renamed. Azure AI Foundry is now Microsoft Foundry, and Google's Vertex AI now sits under the name Gemini Enterprise Agent Platform. Most people still search for the old names, so we use both here. Microsoft, Foundry models from partners; Google Cloud, Gemini Enterprise Agent Platform.
How they compare
| Microsoft Foundry | Amazon Bedrock | Google (Vertex AI) | |
|---|---|---|---|
| Natural home for | Azure and Microsoft 365 estates | AWS estates | Google Cloud estates and Gemini first strategies |
| Models | OpenAI models plus partners including Anthropic, Meta, Mistral, Cohere and Microsoft's own | Hundreds of models including Anthropic, Meta, Mistral and Amazon's Nova, plus OpenAI through Bedrock Managed Agents | Gemini plus a model garden including Anthropic's Claude, Meta's Llama and Google's open Gemma |
| Building blocks | Model catalogue, Agent Service, AI gateway in Azure API Management | Guardrails, Knowledge Bases and AgentCore for agents | Agent building and runtime services with deep Google Cloud integration |
| Watch for | Partner models bought through Azure Marketplace on the provider's terms | Spreading across many models without a reason | Strategy that assumes Gemini for everything |
Sources: Microsoft, Foundry models from partners; AWS, Amazon Bedrock; Google Cloud, Gemini Enterprise Agent Platform.
The details that actually decide it
Where the model runs
The same model can be available on all three clouds but not in the same regions. In Microsoft Foundry, for example, Microsoft's own list shows Claude models deployable in Europe from Sweden Central rather than the UK regions, and its EU data zone option does not cover them. For a UK organisation with data location commitments, check the region for each model you plan to use, not just the platform. Microsoft, Foundry models from partners.
How third party models are bought
On Foundry, models from partners such as Anthropic are bought through Azure Marketplace, priced and licensed by the model provider, and treated as non Microsoft products under Microsoft's terms. Microsoft also states that Cloud Solution Provider subscriptions cannot purchase these third party models. Many UK organisations buy Azure through a Cloud Solution Provider, so that single line can decide which models are available to you. Check the equivalent terms on any platform before you design around a model. Microsoft, Foundry models from partners.
Your existing cloud commitments
If you have a committed spend agreement with one cloud, ask whether model and AI platform spend counts towards it. It can change the economics substantially, and it is a point worth negotiating rather than assuming. Our guide to cloud commitment deals covers how those agreements work.
Models change faster than platform contracts. The best model for a task this year may come from a different provider next year. Put an AI gateway in front of your applications, so they call one endpoint you control and you can route to a different model or platform without rewriting them.
How they are bought, and what to weigh
All three charge mainly for model usage, measured in tokens, with options to reserve capacity for predictable workloads. Before committing, model the cost of your actual workloads at realistic volumes, compare reserved and on demand pricing, check how partner models are billed and supported, and agree what happens to price and capacity as usage grows. Our AI inference cost calculator helps with the first pass.
How we help
We help you choose the platform on evidence, design an architecture that keeps you free to change models and clouds, and negotiate the terms. For the gateway layer that keeps that flexibility, Kong is usually where we start, because it runs in front of all three clouds, while Azure API Management is a strong choice if you are committed to Azure. And if the right answer is to run models in your own data centre instead, our private AI work covers that.
Choosing where to build your AI?
Tell us which clouds you run, what you want to build and any data location requirements. We will tell you which platform fits, what to check before committing, and how to keep your options open. We design and deliver the architecture too.
Prefer email? Reach us directly at hello@c4cgroup.co.uk.
Frequently asked questions
What is the difference between Azure AI Foundry, AWS Bedrock and Vertex AI?
They are the AI platforms of Microsoft, Amazon and Google, used to build your own AI applications and agents. Each offers models from several providers, including Anthropic's Claude, along with tools for retrieval, safety and agents. The main differences are which cloud and data they integrate with, which models run in which regions, and how third party models are bought and contracted.
Has Azure AI Foundry been renamed?
Yes. Azure AI Foundry is now called Microsoft Foundry, and Google's Vertex AI now sits under the name Gemini Enterprise Agent Platform. The services carry on under the new names, so existing work does not need to move.
Can we use Claude on Azure, AWS and Google Cloud?
Yes. Anthropic's Claude models are available on Microsoft Foundry, Amazon Bedrock and Google's model garden. Regional availability and commercial terms differ by platform, so check where each model can run and how it is billed before designing around it.
Which AI platform should a UK organisation choose?
Usually the one aligned with your existing cloud estate, data and commitments, because moving data and duplicating governance is a real cost. Then check that the specific models you need are available in regions that meet your data location requirements, and that your way of buying the cloud allows you to purchase them.
How do we avoid lock in to one AI platform?
Put an AI gateway in front of your applications so they call one endpoint you control, keep prompts and business logic out of platform specific features where you can, and avoid committing spend you cannot redeploy. That keeps you free to switch models or platforms as the market moves.