AI, Data and IoT Services Overview

This content was generated with the assistance of AI and should be verified against the official documentation before being relied on in production.

See the section bibliography for the reference material consulted while preparing these pages.

This page targets Azure CLI 2.90.0 — re-verify exact command names and current GA/preview status against the Azure CLI release notes before relying on them, since almost every service below has been renamed, re-scoped, or had a sibling product launched around it in the last two years. This is a landing/overview page, not a hands-on walkthrough: each section below gets one paragraph, a portal or CLI first step, and a link to the service’s own official documentation for the depth this page intentionally skips. Where this site already has a dedicated reference for a topic, that page is linked instead of repeated.

Where the old names went

Before mapping individual services, it helps to see the renames as a group, because several of them touch the same underlying resource:

Old name Current name What actually changed

Old name	Current name	What actually changed
Azure Cognitive Services	Azure AI services, now Foundry Tools under Microsoft Foundry	Same `Microsoft.CognitiveServices/accounts` resource type and `az cognitiveservices account` CLI group throughout — only the marketing name changed, twice: "Cognitive Services" became "Azure AI services" in 2023, and Microsoft’s current documentation now brands the same catalog of prebuilt APIs (Vision, Speech, Language, Content Safety, Azure AI Search, and more) as Foundry Tools, reachable from the Microsoft Foundry portal. New resources should use the Foundry resource kind (`AIServices`) rather than the legacy multi-service `CognitiveServices` kind; see Resources, Endpoints and Authentication.
Azure AI Studio / Azure AI Foundry	Microsoft Foundry	Renamed at Microsoft Ignite in November 2025, effective in Microsoft’s product terms from 1 January 2026. The platform, portal experience (still at `https://ai.azure.com`), and agent/model catalog are functionally the same service under a new umbrella name; the underlying Azure resource kind is unaffected.
Azure Machine Learning Studio (classic)	Retired 31 August 2024	Not the same product as today’s "Azure Machine Learning studio" (`https://ml.azure.com`) — classic was a separate drag-and-drop ML Studio product, unable to create new instances since December 2021 and fully retired in 2024. Anything from that era should be rebuilt on current Azure Machine Learning, not migrated in place.
Azure Data Lake Storage Gen1	Retired 29 February 2024; superseded by Data Lake Storage Gen2	Gen1 was a standalone service; Gen2 is a capability set (hierarchical namespace) layered onto ordinary Azure Storage accounts, covered in depth in Storage Accounts and Blob Storage.
Azure Synapse Analytics as "the" analytics platform	Microsoft Fabric as the newer, actively developed platform; Synapse still supported, not deprecated	Microsoft has not announced an end-of-life date for Synapse, and both are generally available today — see Microsoft Fabric vs. Azure Synapse Analytics, Databricks, Data Factory and friends below for how the two actually relate.

Azure Cognitive Services

Azure AI services, now Foundry Tools under Microsoft Foundry

Same Microsoft.CognitiveServices/accounts resource type and az cognitiveservices account CLI group throughout — only the marketing name changed, twice: "Cognitive Services" became "Azure AI services" in 2023, and Microsoft’s current documentation now brands the same catalog of prebuilt APIs (Vision, Speech, Language, Content Safety, Azure AI Search, and more) as Foundry Tools, reachable from the Microsoft Foundry portal. New resources should use the Foundry resource kind (AIServices) rather than the legacy multi-service CognitiveServices kind; see Resources, Endpoints and Authentication.

Azure AI Studio / Azure AI Foundry

Microsoft Foundry

Renamed at Microsoft Ignite in November 2025, effective in Microsoft’s product terms from 1 January 2026. The platform, portal experience (still at https://ai.azure.com), and agent/model catalog are functionally the same service under a new umbrella name; the underlying Azure resource kind is unaffected.

Azure Machine Learning Studio (classic)

Retired 31 August 2024

Not the same product as today’s "Azure Machine Learning studio" (https://ml.azure.com) — classic was a separate drag-and-drop ML Studio product, unable to create new instances since December 2021 and fully retired in 2024. Anything from that era should be rebuilt on current Azure Machine Learning, not migrated in place.

Azure Data Lake Storage Gen1

Retired 29 February 2024; superseded by Data Lake Storage Gen2

Gen1 was a standalone service; Gen2 is a capability set (hierarchical namespace) layered onto ordinary Azure Storage accounts, covered in depth in Storage Accounts and Blob Storage.

Azure Synapse Analytics as "the" analytics platform

Microsoft Fabric as the newer, actively developed platform; Synapse still supported, not deprecated

Microsoft has not announced an end-of-life date for Synapse, and both are generally available today — see Microsoft Fabric vs. Azure Synapse Analytics, Databricks, Data Factory and friends below for how the two actually relate.

Microsoft Fabric vs. Azure Synapse Analytics, Databricks, Data Factory and friends

Microsoft Fabric is a software-as-a-service analytics platform: Data Factory, Data Engineering (Spark), Data Science, Data Warehouse, Real-Time Intelligence and Power BI all run as workloads inside one tenant, sharing a single logical data lake called OneLake (itself built on Data Lake Storage Gen2) instead of each service owning its own storage. Because everything reads and writes OneLake directly, a Fabric Data Engineering notebook, a Fabric Data Warehouse table and a Power BI report can all reference the same underlying data with no copy step. Microsoft’s own guidance positions Fabric as the platform most new engineering investment (Copilot, the newest Spark runtime, Fabric IQ) lands on first, and describes it as a natural successor to Synapse for new projects — but it explicitly has not announced an end-of-life date for Azure Synapse Analytics, which remains a supported, independently provisioned PaaS service (dedicated SQL pools, Spark pools, pipelines) for teams that need per-resource ARM control, existing large-scale deployments, or Synapse-specific capabilities such as GPU Spark pools that Fabric does not (yet) offer. Azure Databricks is a separate, third-party-run big-data/ML platform (not a Fabric workload) that many teams keep using stand-alone or side by side with either Synapse or Fabric for its own notebook/MLflow ecosystem; Fabric Data Science can attach to it, but Databricks is not being retired or absorbed by anything above. Data governance and cataloguing across whichever of these platforms you pick — sensitivity labels, data lineage, access policies — is Microsoft Purview's job, not a feature of any one analytics engine; it is only mentioned here in passing because it is the dedicated subject of Governance and Compliance, not repeated on this page.

In short: Fabric is the SaaS umbrella for new analytics work with the least infrastructure to manage; Synapse is the older PaaS platform, still fully supported, for teams already invested in it or needing lower-level control; Databricks is its own third-party engine that both can integrate with.

First step — create a Fabric capacity from the portal (https://app.fabric.microsoft.com, admin portal → Capacity settings → Fabric capacities → New) or, for an Azure-billed capacity, from the CLI:

export RG=rg-bookshelf
export LOCATION=eastus2

az resource create \
  --resource-group "$RG" \
  --resource-type "Microsoft.Fabric/capacities" \
  --name fc-bookshelf \
  --location "$LOCATION" \
  --properties '{"administration":{"members":["<admin-upn>"]},"sku":{"name":"F2"}}'

See the Microsoft Fabric overview and the Azure Synapse Analytics overview.

Data Factory

Data ingestion/orchestration ("pipelines": copy activities, mapping data flows, triggers, 200+ connectors) exists in two forms today: standalone Azure Data Factory (its own ARM resource, az datafactory CLI group) and Data Factory in Microsoft Fabric, the same pipeline authoring experience running as a Fabric workload against OneLake instead of a separate resource. New Fabric-first projects use the latter; existing pipelines on standalone ADF keep working unchanged. First step (standalone): az datafactory create --resource-group "$RG" --factory-name adf-bookshelf --location "$LOCATION". See Microsoft Learn — Data Factory.

Stream Analytics

Azure Stream Analytics is a fully managed, SQL-like (temporal windowing) engine for real-time processing of streams from Event Hubs, IoT Hub or Blob Storage into a sink such as Power BI, a database, or Blob/Data Lake Storage; Fabric’s Real-Time Intelligence workload covers overlapping ground (event streams, KQL databases) for teams already on Fabric, without retiring standalone Stream Analytics. First step: az stream-analytics job create --resource-group "$RG" --name asa-bookshelf --location "$LOCATION" --sku-name Standard. See Microsoft Learn — Stream Analytics introduction. Ingestion sources are covered in Messaging and Integration Services (Event Hubs, Event Grid) and below (IoT Hub).

Data Lake Storage Gen2

Not a separate service to provision: enabling the hierarchical namespace flag on an ordinary Azure Storage account turns it into a Data Lake Storage Gen2 account, giving POSIX-like directory semantics, fine-grained ACLs, and the storage layer every engine above (Synapse, Fabric’s OneLake, Databricks, HDInsight) ultimately reads and writes. This site covers it in depth in Storage Accounts and Blob Storage rather than repeating it here; first step: az storage account create --name stbookshelf --resource-group "$RG" --location "$LOCATION" --sku Standard_LRS --kind StorageV2 --hierarchical-namespace true.

HDInsight

Azure HDInsight is a managed cluster platform for open-source big-data frameworks (Apache Spark, Hive, Kafka, HBase, Hadoop/YARN) that predates both Synapse and Fabric; it is still generally available and actively supported, but Microsoft’s current investment for new analytics work goes to Fabric and Synapse Spark pools, so HDInsight is now mainly the right choice for teams that need a specific open-source component (e.g. a self-managed Kafka or HBase cluster) rather than a managed lakehouse experience. First step: az hdinsight create --name hdi-bookshelf --resource-group "$RG" --cluster-type spark --component-version Spark=3.3 --http-user admin --http-password '<password>' --headnode-size Standard_D13_v2 --workernode-size Standard_D13_v2 --workernode-count 2. See Microsoft Learn — What is Azure HDInsight.

Power BI Embedded

Power BI Embedded lets an application embed Power BI reports, dashboards and tiles for end users without each viewer needing a Power BI license, billed through a dedicated capacity (Azure A-series SKUs for the classic Embedded offer, or a Microsoft Fabric F-series capacity) rather than per user. Two integration patterns exist: embed for your customers ("app owns data" — your app authenticates as a service principal, viewers need no Power BI licence, the common ISV pattern) and embed for your organization ("user owns data" — each viewer signs in with their own Entra ID and Power BI licence). First step: az powerbiembedded workspace-collection create --resource-group "$RG" --name pbi-bookshelf --location "$LOCATION", then create a capacity from the portal’s Power BI Embedded blade. See Microsoft Learn — Power BI embedded analytics overview.

Analysis Services

Azure Analysis Services is a managed tabular semantic-model engine (the same modeling engine underneath Power BI and SQL Server Analysis Services) used to publish a single, governed set of measures and relationships that multiple BI tools query. Microsoft’s current guidance for new work is a Power BI Premium/Fabric-hosted semantic model instead of a standalone Analysis Services server, so treat Analysis Services as a fit only for an existing investment in that exact engine; new Bookshelf-style reporting should model directly in Power BI/Fabric. See Microsoft Learn — Azure Analysis Services overview.

Microsoft Foundry, Azure OpenAI and Azure AI services

Microsoft Foundry (the January-2026 rebrand of Azure AI Foundry, itself the 2024 rebrand of Azure AI Studio) is the unified portal (https://ai.azure.com) and platform for building, evaluating and deploying generative-AI and agentic applications: a model catalog spanning Azure OpenAI, Meta, Mistral, Cohere and other providers, prompt/ agent authoring, evaluation, and observability, all against the same underlying Microsoft.CognitiveServices resource type used by the rest of this section. Azure OpenAI — "Azure OpenAI in Foundry Models" — is part of the Foundry Models catalog, not a Foundry Tool: it is the model family for OpenAI’s GPT, embedding and reasoning models, deployed on a Foundry resource (or, for older setups, a dedicated --kind OpenAI resource) and sharing that resource’s endpoints and keys. Foundry Tools — what used to be called Azure AI services and, before that, Cognitive Services — is the catalog of prebuilt APIs (Speech, Translator, Language, Vision with Face, Document Intelligence, Content Understanding, Content Safety, Immersive Reader, and Azure AI Search, covered next) that Foundry composes into agentic applications. Several older members of that catalog are already retired and should not anchor new work: QnA Maker (2025-03-31), Language Understanding/LUIS (2026-03-31), and Anomaly Detector, Metrics Advisor and Personalizer (2026-10-01). Content Moderator is deprecated and retires on 2027-03-15, replaced by Content Safety.

The Foundry Tools sub-section covers each tool in depth — resources and authentication, networking, pricing and containers, one or more pages per tool with use cases and code, agent integration and end-to-end Bookshelf scenarios — along with the full retirement table.

First step — create a Foundry resource (--kind AIServices, with a custom subdomain so Microsoft Entra ID authentication works) from the CLI:

export RG=rg-bookshelf
export LOCATION=eastus2

az cognitiveservices account create \
  --name foundry-bookshelf \
  --resource-group "$RG" \
  --kind AIServices \
  --sku S0 \
  --location "$LOCATION" \
  --custom-domain foundry-bookshelf \
  --yes

Application code should read the resource’s endpoint/key from Key Vault or, better, authenticate with DefaultAzureCredential/a managed identity rather than an API key — see Identity and Access and Key Vault and Secrets. See Microsoft Learn — What are Foundry Tools? and Microsoft Learn — What is Microsoft Foundry?.

Azure AI Search

Azure AI Search is a fully managed search-and-retrieval service: it indexes content for classic full-text, vector, hybrid and multimodal queries, and underpins agentic retrieval (a knowledge-base abstraction an agent queries instead of a raw index) inside Microsoft Foundry. It is the standard retrieval layer for retrieval-augmented generation (RAG) against Azure OpenAI — when the application side is Spring Boot / Spring AI rather than the Foundry SDKs directly, follow Integrating with Spring AI for the VectorStore/EmbeddingModel integration pattern used throughout this site’s vector-database section, and Comparing Vector Databases for how it compares to a self-hosted vector database. How AI Search calls Foundry Tools during indexing (skillsets, integrated vectorization) and serves agents (agentic retrieval, Foundry IQ) is covered in Foundry Tools, on Azure AI Search with Foundry Tools. First step: az search service create --name search-bookshelf --resource-group "$RG" --sku basic --location "$LOCATION". See Microsoft Learn — Introduction to Azure AI Search.

Azure Machine Learning

Azure Machine Learning is the platform for the classical MLOps lifecycle — experiment tracking, pipelines, managed training/inference compute, model registry, CI/CD-driven redeploys — accessible from Azure Machine Learning studio (https://ml.azure.com, distinct from the retired classic ML Studio above), the Python/CLI v2 SDK, or ARM. It also has its own model catalog and Automated Machine Learning (AutoML) feature for automating featurization and algorithm selection without hand-tuning a training loop, and it interoperates with Microsoft Foundry for LLM-centric work — both surfaces can serve the same generative-AI/LLM scenarios, and Azure ML’s Prompt flow authoring tool is itself scheduled to retire on 20 April 2027 in favour of the Microsoft Agent Framework, so new prompt-orchestration work should start there rather than in Prompt flow. For classical ML training and MLOps (not a generative agent), Azure Machine Learning is the default choice on this site. First step: az ml workspace create --name mlw-bookshelf --resource-group "$RG" (installs the ml CLI extension on first use). Responsible-AI tooling (model fairness/explainability dashboards, model cards) ships as part of the same studio rather than a separate product. See Microsoft Learn — What is Azure Machine Learning? and Microsoft Learn — What is automated machine learning (AutoML)?.

AI Builder, the low-code model-building surface inside Power Platform (Power Apps/Power Automate) for tasks such as form processing or prediction without writing training code, is a Power Platform capability rather than an Azure resource in its own right — it is out of scope for this Azure-focused guide beyond this pointer; see Microsoft Learn — What is AI Builder?.

IoT: Hub, Central, Edge and Digital Twins

A typical Azure IoT solution is built from a small set of composable building blocks rather than one monolithic product: a device identity/connection layer (IoT Hub), an optional no-code application layer on top of it (IoT Central), an optional gateway runtime that runs modules on the device itself (IoT Edge), and an optional spatial/graph model of the environment those devices report into (Digital Twins). None of the four retires or replaces another — they compose.

Azure IoT Hub is the managed, bidirectional message hub between devices and the cloud: device provisioning and identity, device-to-cloud telemetry ingestion, cloud-to-device commands, and device twins (a per-device JSON document of reported/desired properties) that other services — including Digital Twins below — read from. First step: az iot hub create --name iothub-bookshelf --resource-group "$RG" --sku S1. See Microsoft Learn — What is IoT Hub?.

Azure IoT Central is an application platform-as-a-service built on top of IoT Hub: a hosted web UI for device templates, dashboards, rules and bulk device management, aimed at teams that want a working IoT application quickly instead of building the UI/rules layer themselves. It remains a current, actively documented product, not a legacy or retiring one. First step: az iot central app create --name bookshelf-iot --resource-group "$RG" --subdomain bookshelf-iot. See Microsoft Learn — What is Azure IoT Central?.

Azure IoT Edge runs containerized modules directly on a device or gateway, so telemetry can be filtered, aggregated or acted on locally (lower latency, reduced bandwidth, offline operation) before anything reaches the cloud; an IoT Edge device still shows up as a regular device identity in IoT Hub. See Microsoft Learn — Azure IoT Edge overview.

Azure Digital Twins is a platform-as-a-service for modeling an entire environment — a building, factory, or supply chain — as a graph of typed entities (defined in the Digital Twins Definition Language, DTDL) with live properties and relationships, connecting to IoT Hub’s device twins to stay current and routing events onward to Event Hubs/Event Grid/Service Bus or into Azure Data Explorer for history. It is explicitly distinct from a plain IoT Hub device twin and from Fabric’s own (preview) digital-twin builder — Azure Digital Twins is the general-purpose graph/spatial modeling product. First step: az dt create --dt-name dt-bookshelf --resource-group "$RG" --location "$LOCATION" (installs the azure-iot CLI extension on first use). See Microsoft Learn — What is Azure Digital Twins?.

Azure Maps

Azure Maps is a location-platform service — geocoding/reverse geocoding, routing, traffic, weather, and rendered map tiles — consumed via REST APIs and web/mobile SDKs, commonly paired with IoT Hub/Digital Twins data (for example, plotting Bookshelf delivery-fleet positions on a live map) or with Azure AI Search’s geo-spatial filters. First step:

az maps account create \
  --name maps-bookshelf \
  --resource-group "$RG" \
  --sku G2 \
  --accept-tos

See Microsoft Learn — What is Azure Maps?.

Azure Quantum

Azure Quantum is Microsoft’s cloud quantum-computing service: it gives access to quantum hardware from several industry partners (IonQ, Pasqal, Quantinuum, Rigetti), the open-source Q# language and Quantum Development Kit (QDK) for writing hardware-agnostic quantum programs, and a resource estimator for predicting what a program would cost to run on a future large-scale quantum machine before any hardware access is purchased. It is a specialized research/experimentation service, unrelated to the classical-compute services covered elsewhere in this section, and is out of scope for a typical line-of-business application such as Bookshelf. First step: create an Azure Quantum workspace from the portal (search Azure Quantum → Create) or install the QDK extension for VS Code to start writing Q# without an Azure account at all. See Microsoft Learn — What is Azure Quantum?.

Clean up

Every first-step command above is shown for illustration and does not need to persist beyond trying it out. If you did create any of these resources against a real subscription, remove them the same way as anywhere else in this section:

az group delete --name "$RG" --yes --no-wait

Azure Machine Learning workspaces are soft-deleted by default — pass --permanently-delete to az ml workspace delete if you need the name back immediately rather than waiting out the soft-delete window.