If you’re leading IT at a small or mid-sized business, you’ve probably noticed that Microsoft has been rolling out a lot of new tools lately. One of the most significant—and potentially most useful—is Azure AI Foundry.

Foundry isn’t just a rebrand of Azure AI Services. It’s a new way to discover, evaluate, and deploy models to help your team automate tasks, improve workflows, and use your data. But with more options comes more complexity. And if you’re like most IT leaders we work with, you don’t have time to chase every new feature. You need to know what’s useful, what’s safe, and what’s worth your time.

That’s what this article is here to help with.

What Is Azure AI Foundry?

Azure AI Foundry is Microsoft’s new model catalog and deployment platform. It brings together over 1,900 models from Microsoft and third-party providers, all accessible through a single interface. Users can search, compare, and deploy models for a wide range of tasks, including language processing, image analysis, reasoning, and more.

Some of the key capabilities that will benefit small businesses revolve around Microsoft, reducing some of the advanced complexity of managing and curating AI models. With Azure AI Foundry, you have a transparent marketplace to support your decision-making. For example, you can:

  • Compare models side-by-side using your data
  • Deploy models using either Microsoft-managed infrastructure or your own Azure resources
  • Fine-tune models, monitor performance and apply content safety filters
  • Choose from models hosted by Microsoft or by external providers

This flexibility is great, but it also means you need to understand the differences to make the right choice.

Microsoft Hosted or Not?

There are two types of models available in the Azure AI Foundry: models sold directly by Microsoft and Models sold by third-party providers. This is similar to Azure Marketplace, where you can purchase Virtual Machines managed by Microsoft or Virtual Machines managed by other providers.

1. Models Sold Directly by Microsoft

Models sold directly by Microsoft make compliance and management eassier. They are hosted and supported by Microsoft under its product terms including being backed by Microsoft’s SLA and support. They are connected directly to Azure’s infrastructure and they meet Microsoft’s requirements for security, compliance, and performance.

Examples include:

  • Azure OpenAI models like GPT-4o
  • Microsoft’s own document processing and translation models
  • Select models from providers like Meta and Mistral that Microsoft has chosen to host and support directly

These are the safest bet for small businesses that need reliability, compliance, and support.

2. Models from Partners and the Community

These models make up the majority of the Foundry catalog. They’re developed and supported by third-party providers, such as Hugging Face, Cohere, Meta (in some cases), and others.

Key characteristics:

  • Hosted by the provider or in Microsoft-managed infrastructure
  • Support and SLAs vary by provider
  • Often cutting-edge or industry-specific
  • May not meet Microsoft’s Responsible AI standards
  • Require more due diligence from IT teams

Examples include:

  • Hugging Face models for sentiment analysis, classification, and more
  • Cohere’s RAG (retrieval-augmented generation) models for knowledge-heavy tasks
  • Meta’s LLaMA models (when not hosted directly by Microsoft)
  • Industry-specific models (e.g., legal, healthcare, and finance)

Deployment Options:

Microsoft offers both standard and managed compute options. With a standard deployment, Microsoft hosts the model. With Managed compute, you deploy the model on an Azure Virtual Machine.

 Azure AI Foundry Deployment Options: Side-by-Side Comparison

Feature Standard Deployment Managed Compute
Hosting Microsoft hosts the model You deploy the model to your own Azure virtual machine
Access Method Access via API Access via REST API on your VM
Billing Billed per use (tokens in, tokens out) Billed based on VM usage
Ease of Use Easier to get started Requires Azure VM quotas and setup
Customization Limited customization Supports fine-tuning and advanced monitoring
Control & Security Less control over infrastructure More control over performance, security, and network isolation

 

Compare Models Side-by-Side Using Your Data

A great new feature in Azure Foundry is the ability to evaluate models with your own data.

Here’s how it works:

  • In the Foundry portal, you can browse the model catalog and select multiple models supporting the same task type (e.g., summarization, classification, translation).
  • You can upload your sample data—documents, text, or other inputs—and run the same task across different models.
  • Foundry will show you how each model performs on your data, including:
    • Output quality
    • Latency
    • Cost metrics
    • Benchmark scores (where available)

You can also filter models by:

  • Industry-specific training (e.g., legal, healthcare, finance)
  • Capabilities like reasoning or tool calling
  • Deployment options (standard API vs. managed compute)
  • Licensing terms and support levels

This kind of hands-on evaluation is rare in small business tech and it’s a big reason why Foundry is worth exploring, even if you’re not ready to deploy yet.

Can You Use Foundry Models in Power Platform?

Yes—and this is where things get exciting for small business IT teams.

As of the 2025 Wave 1 release, Microsoft has introduced native integration between Azure AI Foundry and Power Platform’s Prompt Builder. This means you can securely connect to models you’ve deployed in Foundry and use them directly in:

  • Power Automate flows
  • Copilot agents
  • Power Apps (via Power Automate or API connectors)

This allows you to:

  • Use specialized models (including ones you’ve fine-tuned) for tasks like summarization, classification, or document processing
  • Incorporate domain-specific intelligence into your automations
  • Optimize costs by selecting models that are right-sized for your needs

But it’s not plug-and-play. You’ll still need to:

  • Deploy the model in Foundry (standard or managed compute)
  • Expose it via API
  • Set up authentication and governance

What Small Business IT Leaders Need to Watch For

  • Not all models are created equal. Just because a model is in the catalog doesn’t mean it’s right for your business.
  • Support varies widely. Microsoft-hosted models come with enterprise-grade support. Partner models may not.
  • Compliance is your responsibility. If you’re using a third-party model, you’re responsible for ensuring it meets your data handling and privacy requirements.
  • Copilot and automation tools don’t replace expertise. These tools can help you move faster—but they don’t design solutions for you.

How TechHouse Can Help

Azure AI Foundry is new and like many of Microsoft’s most powerful tools, it’s evolving quickly. At TechHouse, we haven’t built production applications with Foundry yet, but we’re actively learning, testing, and staying on top of what’s possible.

What we do bring to the table is deep experience with Microsoft’s platform, a strong understanding of how small businesses work, and a track record of helping IT leaders make smart, strategic decisions.

If you’re curious about Foundry—what it can do, how it fits into your environment, and whether it’s worth exploring—we’re here to help you figure that out. We’ll help you:

  • Understand what’s available and how it works
  • Evaluate models and deployment options
  • Connect Foundry to your Power Platform solutions
  • Plan for governance, security, and support

You don’t need to be a data scientist to use Foundry—but you do need a partner who’s paying attention, asking the right questions, and ready to help you move forward.

Let’s explore what’s possible—together.

TechHouse

Innovative Solutions. Practical Software. Friendly Support.

Schedule a call  |  View Upcoming Events  |  Access our TechHouse Library of Articles