The Complete Overview of Azure AI Services and Tools

Microsoft has renamed nearly everything in its AI portfolio over the last two years. Some things have been renamed twice.

If you’ve tried to work out what Azure actually offers in this space, you might have ended up more confused than when you started. So below, we’ll lay it all out clearly. You’ll find out which AI services are part of Azure right now, what each one does, and when you’ll want to reach for it.

But the naming chaos hides a genuinely impressive toolkit. Azure has a lot to offer businesses that want to build, automate, and get creative with AI. From prebuilt services that drop into an app with a few API calls, through to platforms for training your own models from scratch. Once you can see past the labels, there’s plenty to get excited about.

Two parallel rebrands have caused most of the trouble.

The first is the prebuilt AI services. They started life as Azure Cognitive Services, became Azure AI Services in 2023, and were rebranded again in late 2025 as Foundry Tools. Same products, mostly. New name, new positioning as building blocks for AI agents.

The second is the development platform. Azure AI Studio became Azure AI Foundry, which then became Microsoft Foundry at the end of 2025. The “Azure” prefix was dropped to position Foundry as a standalone Microsoft AI platform rather than just an Azure feature.

So if you read an article from 2024 talking about Azure Cognitive Services inside Azure AI Studio, that’s the same thing as Foundry Tools inside Microsoft Foundry today. We’ll use the current names throughout the rest of this piece, but we’ll flag the old ones where it helps.

Here’s the short version in table form:

Before we get into the details, here’s how Azure’s AI offering breaks into roughly four groups:

• Microsoft Foundry: the overarching platform for building, deploying, and managing AI workloads on Azure.
• Azure OpenAI Service: access to OpenAI’s frontier models (GPT-5, embeddings, image and audio models) delivered as an Azure service.
• Foundry Tools: prebuilt, task-specific AI services for things like vision, speech, document processing, and translation. Formerly known as Cognitive Services.
• Azure Machine Learning: the custom MLOps platform for teams building and training their own models from scratch.

Plenty of projects lean on one or two of these, rather than all four. Picking the right entry point is an important step, either way.

Microsoft Foundry is the umbrella platform; an environment for building AI applications. It pulls together model access, agent building, prebuilt AI tools, search and retrieval, observability, and governance. Inside Foundry you’ll find:

• Foundry Models: the model catalogue, including Azure OpenAI models, Anthropic models, Meta’s Llama family, Mistral, and a wide range of open-source options. Azure OpenAI is now available here as “Azure OpenAI in Foundry Models”.
• Foundry Agent Service: the managed runtime for building and deploying AI agents, with multi-agent orchestration, tool use, and enterprise controls baked in.
• Foundry Tools: the prebuilt task-specific services we mentioned above (more on these shortly).
• Foundry IQ: a unified retrieval layer for grounding agents in enterprise data, built on top of Azure AI Search.
• Foundry Local: a way to run Foundry models on local hardware (like laptops, edge devices, or on-prem servers) for offline or sovereignty-sensitive scenarios. Foundry Local hit general availability in April 2026.
• Foundry Control Plane: the governance, cost management, and observability layer.

You might use Foundry as the central place to prototype an agent, first hooking it up to your own data. You’d then evaluate it against test sets, and deploy it with proper monitoring and access controls. Foundry is where Microsoft is putting most of its AI engineering effort, so expect it to keep changing shape.

Azure OpenAI is OpenAI’s model family delivered through Azure. Same models you’d get from OpenAI directly (GPT-5 series, the GPT-4 family, embeddings, image generation, audio), but running inside your Azure subscription with Azure’s security, networking, billing, and compliance wrapped around them. You decide where to deploy your model to comply with data residency obligations.

The practical difference from calling OpenAI’s API directly is that it’s more complex, but more capable too. You provision an Azure OpenAI resource, deploy a specific model version into it, and call that deployment (rather than calling the model by name). That’s a few extra steps, but it’s what unlocks private endpoints, Entra ID authentication, regional pinning, customer-managed keys, and integration with the rest of your Azure estate. Worth it if you’re taking enterprise AI seriously.

What you might use it for:

• Building a copilot for your internal knowledge base
• Generating product descriptions at scale
• Summarising support tickets
• Powering any LLM feature inside an existing Azure-hosted application

Azure Machine Learning is the custom MLOps platform. Where Foundry Tools give you prebuilt models for common tasks, Azure ML is for teams building their own models from scratch (or fine-tuning existing ones at scale).

It covers the full lifecycle: data preparation, experiment tracking, distributed training across GPU clusters, model registry, deployment to managed endpoints, and monitoring in production.

If you’ve got data scientists who write PyTorch or use libraries like transformers or scikit-learn and need somewhere to run it properly, Azure ML is the platform for that.

What you might use it for:

• Training a demand forecasting model on years of your own sales data
• Fine-tuning a computer vision model for a specialised industrial inspection task
• Running a recommendation engine at production scale

Foundry Tools (formerly Azure AI Services, formerly Cognitive Services) are the off-the-shelf services for common AI tasks. You don’t need a data science team to use them. You make API calls and you get results. Here’s a rundown of the main ones.

Azure AI Vision

Image analysis, object detection, OCR, spatial analysis, and face detection.

What you might use it for: automatically tagging product photos, reading text from scanned receipts, or counting people in a retail space.

Azure AI Document Intelligence

Extracts structured data from documents like invoices, receipts, contracts, and forms. Comes with prebuilt models for common document types and the ability to train custom models for your own formats.

What you might use it for: automating accounts payable by pulling line items out of supplier invoices.

Azure AI Language

Natural language understanding: sentiment analysis, entity recognition, key phrase extraction, summarisation, language detection, and conversational language understanding.

What you might use it for: routing inbound emails to the right team based on their content.

Azure AI Speech

Speech-to-text, text-to-speech, real-time translation, speaker recognition, and custom voice creation. This is what people are talking about when they mention “Azure AI Voice”, as it’s a versatile tool.

What you might use it for: transcribing meeting recordings, building a voice interface for a mobile app, or creating a multilingual customer service IVR for your call centre.

Azure AI Translator

Real-time text translation across more than 100 languages, with custom translation models for industry-specific terminology.

What you might use it for: translating user-generated content on a marketplace, or localising documentation on the fly.

Azure AI Search (and Foundry IQ)

Enterprise search with vector and hybrid retrieval, the foundation for most retrieval-augmented generation (RAG) patterns on Azure. Foundry IQ sits on top of it as the unified retrieval layer for AI agents.

What you might use it for: grounding a chatbot in your company’s policies, contracts, and product docs. This way it can answer from your data rather than making things up.

Azure AI Content Safety

Detects harmful content in text and images (like hate, violence, sexual, self-harm), with prompt shields and groundedness detection for LLM scenarios.

What you might use it for: moderating user-generated content on a community platform, or wrapping safety guardrails around a customer-facing chatbot.

Azure AI Video Indexer

Pulls insights out of video: transcripts, faces, objects, scenes, sentiment, and topics.

What you might use it for: making a large video archive searchable by what’s actually said and shown on screen.

Azure AI Bot Service

The managed runtime for conversational bots. Worth a caveat here: Microsoft archived the Bot Framework SDK in December 2025, and new bot development is being steered toward Foundry Agent Service. The Bot Service itself is still active for existing workloads, but if you’re starting something new, Foundry Agent Service is where the investment is going.

Foundry Agent Service

The current home for building AI agents on Azure. Handles multi-agent orchestration, tool use, memory, and the enterprise controls you’d expect from an Azure service (like identity, networking, evaluations and observability).

Foundry MCP Servers and Tools

Foundry is the enterprise grade platform for consuming and deploying MCP servers. Foundry has a wide range of pre-built MCP servers to ensure tools used by agents are secure as well as functionality to host custom MCP servers and tools.

What you might use it for: To be effective, agents need tools. Using foundry creates a seamless experience for managing agents and their tools.

A few services in this space are on their way out, and it’s worth knowing before you build anything on them:

• Azure AI Anomaly Detector is being retired on 1 October 2026. Microsoft recommends moving to Microsoft Fabric, Azure Monitor, or the open-source anomaly detector engine instead.
• Azure AI Metrics Advisor is being retired on the same date.
• The Bot Framework SDK was archived in December 2025, as mentioned above.

If you’ve got existing workloads on any of these, now’s the time to start planning a migration. If you’re starting something new, pick a different tool.

Pricing across Azure’s AI portfolio is consumption-based. You pay for what you use, whether that’s tokens (for LLMs), transactions (for most Foundry Tools), compute hours (for Azure Machine Learning), or a combination of those.

Azure OpenAI in particular has a few different deployment models that change the cost shape. The Standard option is pay-as-you-go per token. Global Standard routes requests through a global capacity pool for better availability but data zone deployment within the EU is also available. Provisioned Throughput Units (PTUs) let you reserve dedicated capacity at a flat rate, which is a good call for high-volume production workloads that need guaranteed latency.

We’re not quoting figures here because they change often and any number we put in this article would be out of date within a few months. Microsoft publishes current rates on the Azure AI Foundry pricing page and the Azure OpenAI pricing page. The Azure pricing calculator is also useful for modelling expected spend before you commit.

A few of these services sit close enough to each other that it’s worth spelling out the difference.

Azure OpenAI vs OpenAI directly

You might be wondering why you’d use Azure OpenAI when you can call OpenAI’s API directly. Same models, different delivery.

Calling OpenAI’s API directly is the simpler path: an account, an API key, an SDK, and you’re making requests in minutes. Azure OpenAI is more involved to set up (you provision a resource, deploy a model, and call the deployment). But in return you get private networking, Entra ID authentication, regional pinning for data residency, integration with Azure billing and governance, and the rest of the Azure security model.

The short version: OpenAI’s direct API is easier and faster to ship with. Azure OpenAI takes more setup but gives you the controls enterprise environments tend to need. If you’re already on Azure and you care about governance, Azure OpenAI is usually the right answer. If you’re a small team prototyping fast and not on Azure, the direct API is fine.

Azure AI services vs AWS AI services

Both clouds have broadly equivalent portfolios at this point. AWS has Bedrock as its multi-model platform, SageMaker as its custom ML platform, and a similar set of prebuilt services. There’s Rekognition for vision, Comprehend for language, Textract for document processing, Polly and Transcribe for speech.

The picks tend to come down to which cloud you’re already in rather than feature parity arguments. Where Azure has an edge is the OpenAI relationship and the Microsoft 365 integration story. It’s a strong pull for organisations already running on Entra ID and Microsoft’s wider stack.

Copilot Studio vs Microsoft Foundry

These are different products for different audiences. Copilot Studio is a low-code environment aimed at business users and citizen developers building copilots and agents inside Microsoft 365. Microsoft Foundry is a developer platform for building custom AI applications and agents from the ground up, with full code-level control. If you want to ship a Teams bot that answers HR questions without writing much code, go for Copilot Studio. If you’re building a custom AI product, Foundry is the one.

If you’re trying to work out which corner of the portfolio your project belongs in, this might help:

Some projects end up touching more than one of these. An agent could use Foundry Agent Service for orchestration and Azure OpenAI for the underlying model. It could then call on Foundry Tools for document parsing, and Foundry IQ to ground answers in company data.

The names will probably continue to change. But the underlying offer is more stable than the labels suggest, and once you’re comfortable with the main groups, it’ll be easier to grasp. Most new announcements/tools should slot into one of those four buckets.

Trying to work out which of these services fits a project you’re scoping? Maybe you’re future-proofing your Azure estate and want a sensible plan for what to keep, retire, or consolidate. Or you might just want to know how to build cool stuff with AI.

Whichever it is, get in touch with us at Synextra today. These are the kind of conversations we’re built for.