Microsoft Azure
Last reviewed
Jun 1, 2026
Sources
21 citations
Review status
Source-backed
Revision
v1 ยท 3,029 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 1, 2026
Sources
21 citations
Review status
Source-backed
Revision
v1 ยท 3,029 words
Add missing citations, update stale details, or suggest a clearer explanation.
Microsoft Azure is the public cloud computing platform operated by Microsoft. It offers compute, storage, networking, databases, identity, analytics, and a large catalog of managed services that customers rent on demand rather than running in their own facilities. Azure was first shown to developers in 2008 under the codename Project Red Dog, launched commercially as Windows Azure in February 2010, and renamed Microsoft Azure on March 25, 2014.[1][2] By 2025 it was the second-largest cloud provider in the world, behind Amazon Web Services (AWS) and ahead of Google Cloud, holding roughly a fifth of the global cloud infrastructure market.[3]
Azure runs on a network of data center regions that spans more than 70 announced regions and over 400 physical datacenters, which Microsoft says is the broadest regional footprint of any cloud provider.[4] In its 2025 fiscal year the platform passed 75 billion US dollars in annual revenue, a figure Microsoft disclosed as a standalone number for the first time, and growth in the high 30 percent range was driven in large part by demand for artificial intelligence workloads.[5]
Much of Azure's recent prominence comes from its role in AI. It hosts the Azure OpenAI Service, which provides enterprise access to models from OpenAI including the GPT-4 family, GPT-4o, the o-series reasoning models, and image generation models.[6] Azure is the primary cloud behind Microsoft's deep partnership with OpenAI and the compute layer underneath Microsoft's Copilot products. To meet AI demand, Microsoft has built out enormous datacenter capacity packed with NVIDIA GPU clusters and has designed its own silicon, the Maia 100 AI accelerator and the Cobalt 100 Arm processor.[7]
| Field | Detail |
|---|---|
| Developer | Microsoft |
| Launched | February 2010 (as Windows Azure) |
| Renamed | March 25, 2014 (to Microsoft Azure) |
| Type | Public cloud computing platform |
| Key AI services | Azure OpenAI Service, Microsoft (Azure AI) Foundry, Azure Machine Learning, Azure AI services |
| Custom chips | Maia 100 AI accelerator, Cobalt 100 Arm CPU |
| Main competitors | Amazon Web Services, Google Cloud, Oracle Cloud, IBM Cloud |
| FY2025 revenue | Over 75 billion US dollars annually |
Microsoft announced its cloud platform at the Professional Developers Conference in October 2008, where the project had been developed internally under the codename Project Red Dog.[1] The service became commercially available as Windows Azure in February 2010, initially centered on a platform-as-a-service model for running .NET applications, with later support added for languages such as Java and PHP.[1][2]
The platform's early years saw a steady move toward a more general-purpose cloud. In 2012 and 2013 Microsoft added infrastructure-as-a-service capabilities, including persistent virtual machines and support for Linux, broadening Azure beyond its original developer-focused roots. On March 25, 2014, the company renamed the service from Windows Azure to Microsoft Azure to reflect that it ran far more than Windows workloads.[2] Under chief executive Satya Nadella, who took over in 2014, Microsoft reorganized around a "cloud-first" strategy, and Azure became the centerpiece of the company's Intelligent Cloud reporting segment.
Azure expanded into AI services well before the current generative AI wave. Microsoft offered Cognitive Services for vision, speech, and language, along with Azure Machine Learning for model training and deployment. The company first opened the Azure OpenAI Service in a limited preview in November 2021 and made it generally available on January 16, 2023, giving enterprises access to OpenAI models with content filtering, role-based access control, and regional deployment.[8] That release, followed within weeks by the public launch of ChatGPT, marked the point where Azure's AI business began to grow quickly.
Azure organizes its offerings into broad categories. The platform advertises hundreds of individual services, and the main groups are summarized below.
| Category | Representative services | What it provides |
|---|---|---|
| Compute | Azure Virtual Machines, Virtual Machine Scale Sets, Azure Kubernetes Service, Azure Functions, App Service | On-demand servers, container orchestration, and serverless code execution |
| Storage | Blob Storage, Azure Files, Disk Storage, Data Lake Storage | Object, file, and block storage for applications and analytics |
| Networking | Virtual Network, Load Balancer, Azure Front Door, ExpressRoute, Azure DNS | Private networks, traffic distribution, content delivery, and dedicated connectivity |
| Databases | Azure SQL Database, Cosmos DB, Azure Database for PostgreSQL and MySQL | Managed relational and NoSQL data stores |
| Identity | Microsoft Entra ID (formerly Azure Active Directory) | Authentication, single sign-on, and access management |
| AI and machine learning | Azure OpenAI Service, Microsoft Foundry, Azure Machine Learning, Azure AI services | Model hosting, custom model training, and prebuilt AI APIs |
Compute is the foundation of the platform. Azure Virtual Machines let customers run Windows or Linux servers of many sizes, while Azure Kubernetes Service manages containerized applications and Azure Functions runs event-driven code without a managed server. Storage spans Blob Storage for unstructured objects, Azure Files for shared file systems, and Data Lake Storage for large analytics datasets.
Identity is handled by Microsoft Entra ID, the cloud directory and access-management service. Microsoft renamed Azure Active Directory to Microsoft Entra ID in 2023, with the change rolling out across its products from August 15 of that year, to reduce confusion with the older Windows Server Active Directory and to signal the service's multicloud reach.[9] Existing deployments, licensing, and service-level agreements were unaffected by the rename.[9] Entra ID underpins sign-in for Azure itself as well as for Microsoft 365 and many third-party applications.
On the data side, Azure SQL Database provides a managed relational engine based on Microsoft SQL Server, and Azure Cosmos DB is a globally distributed NoSQL database. The platform also includes analytics services such as Azure Synapse Analytics and Microsoft Fabric, along with developer tooling, monitoring, and security products.
Azure groups its physical capacity into regions, each of which is a cluster of one or more datacenters connected by a high-capacity, low-latency network.[10] Microsoft reports a global footprint of more than 70 announced regions and over 400 datacenters, linked by more than 370,000 miles of terrestrial and subsea fiber and supported by over 190 edge sites.[4] Many regions are subdivided into availability zones, which are physically separate groups of datacenters with independent power, cooling, and networking so that a failure in one zone does not take down the others.[10]
Microsoft operates specialized cloud regions to meet government and sovereignty requirements, including Azure Government for US public-sector customers and dedicated sovereign clouds in other jurisdictions. The company has continued to add regions and capacity rapidly through 2025 and 2026 to keep pace with AI demand, and it disclosed that it brought roughly 2 gigawatts of new datacenter capacity online during fiscal 2025.[5]
Azure has become one of the most important platforms for building and running AI systems, and AI is now a major driver of its revenue. In Microsoft's fiscal 2025 third quarter, Azure and other cloud services grew 33 percent year over year, with about 16 percentage points of that growth attributed to AI services.[11] For the full fiscal year, Azure surpassed 75 billion US dollars in annual revenue, up 34 percent.[5]
The platform's AI development hub went through two rebrands in as many years. Microsoft renamed Azure AI Studio to Azure AI Foundry at its Ignite conference in 2024, positioning it as a unified platform for building generative AI applications with model selection, benchmarking, prompt engineering, retrieval-augmented generation, agent building, and safety guardrails.[12] At Ignite on November 18, 2025, Microsoft renamed Azure AI Foundry again, this time to Microsoft Foundry, and shifted its emphasis toward autonomous agents rather than models alone.[12] The platform offers a model catalog that combines OpenAI's GPT models with Microsoft's own Phi family, Meta's Llama models, and others, alongside the Azure AI Agent service for building agents that automate business tasks.[12]
Beyond Foundry, Azure provides Azure Machine Learning for training and deploying custom models and a set of Azure AI services (formerly Cognitive Services) that expose prebuilt capabilities for vision, speech, language, and document processing through APIs.
The Azure OpenAI Service gives enterprises access to OpenAI's models within Microsoft's cloud, with the security, compliance, and regional controls that large organizations expect. Microsoft made the service generally available on January 16, 2023, initially offering GPT-3.5, Codex, and DALL-E 2 together with content filtering, role-based access control, and private networking.[8]
Since then the catalog has tracked OpenAI's releases closely. It includes the GPT-4 family and GPT-4o, a multimodal model that handles text, images, and audio and is faster and cheaper than the earlier GPT-4 Turbo.[6] The service also hosts OpenAI's o-series reasoning models, including o3 and o4-mini, which spend more time working through problems and perform well on math, coding, and science tasks.[6] For images, Microsoft added GPT-image-1 in 2025, a model with stronger instruction following and text rendering than the older DALL-E line; the dall-e-3 model was retired for new deployments in March 2026.[6]
Enterprises can consume these models through different deployment modes. Standard deployments bill per token and include a global option that routes requests across Azure's worldwide capacity, while provisioned deployments let customers reserve dedicated throughput measured in provisioned throughput units for predictable performance.[6] Azure OpenAI runs in multiple regions and integrates with the rest of the platform, so the same identity, networking, and monitoring controls apply to AI workloads as to other Azure services.
The service is also the infrastructure behind much of Microsoft's own AI product line. Azure OpenAI powers the Copilot family, including Microsoft 365 Copilot, which Microsoft announced in March 2023 as a generative AI assistant built into Word, Excel, PowerPoint, Outlook, and Teams.[13] By 2026, Microsoft had broadened Copilot beyond a single supplier: its orchestration layer can route requests across several models, including OpenAI models hosted on Azure and, for some workloads, third-party models such as Anthropic's Claude, reflecting a shift toward a multi-model strategy rather than reliance on one provider.[14]
Microsoft's relationship with OpenAI is the most consequential commercial partnership in the current AI industry, and it has reshaped Azure's position in the market. Microsoft first invested 1 billion US dollars in OpenAI in 2019 and became its preferred cloud provider, then deepened the commitment with further rounds that brought its total committed investment to about 13 billion US dollars by early 2023.[15] In exchange, Azure became OpenAI's exclusive cloud, OpenAI's models ran on Azure infrastructure, and Microsoft gained broad rights to commercialize OpenAI's technology in its own products.[15]
Those terms changed substantially in 2025. After OpenAI completed a corporate recapitalization that created OpenAI Group PBC, Microsoft and OpenAI signed a new definitive agreement on October 28, 2025, that loosened the exclusivity at the heart of the original deal.[16] Under the new structure, Microsoft holds an investment in OpenAI Group PBC valued at roughly 135 billion US dollars, equal to about 27 percent of the company on an as-converted diluted basis.[16] Microsoft keeps exclusive rights to OpenAI's models and products through Azure's API and an intellectual property license that runs through 2032, though the agreement now extends those rights to models developed after artificial general intelligence is reached, subject to safety conditions.[16]
The most important shift was around compute. Microsoft gave up its right of first refusal to be OpenAI's compute provider, which means OpenAI can buy capacity from other cloud providers.[16] At the same time, OpenAI committed to purchasing an incremental 250 billion US dollars of Azure services, a commitment that anchors a large share of Azure's future revenue even as the exclusivity ends.[16] The agreement also created an independent expert panel to verify any future declaration that AGI has been reached, a clause that affects when certain rights and revenue-sharing arrangements expire.[16] Microsoft's accounting for its OpenAI stake has become large enough to move its results: the company recorded a 3.1 billion US dollar reduction to net income in one quarter from its share of OpenAI's losses under equity-method accounting.[17]
Microsoft designs its own chips to control cost and performance for cloud and AI workloads, an effort it made public at Ignite in November 2023 with two parts, the Maia 100 AI accelerator and the Cobalt 100 Arm CPU.[7]
The Maia 100 is a large AI accelerator built for training and inference on generative models. Microsoft describes it as a 5-nanometer chip with about 105 billion transistors, among the largest chips made on that process at the time.[7][18] Technical disclosures put the die at roughly 820 square millimeters, manufactured on TSMC's N5 process with CoWoS-S packaging, with four HBM2E memory stacks providing 64 gigabytes of capacity and 1.8 terabytes per second of bandwidth.[18] The chip is designed for up to 700 watts but provisioned at 500 watts, supports low-precision data formats including sub-8-bit types, and ships with a software stack that includes a PyTorch backend so existing models can be ported with limited changes.[18] Because Maia runs hot, Microsoft built custom server racks with liquid-cooling "sidekick" units to deploy it.[7]
The Cobalt 100 is Microsoft's first in-house general-purpose Arm processor for the cloud, a 128-core CPU built on Arm Neoverse cores and aimed at scale-out workloads such as web servers, databases, and analytics.[7] Microsoft made Cobalt 100-based virtual machines generally available in October 2024, offering up to 50 percent better price performance than its previous Arm-based virtual machines, and expanded availability to dozens of Azure regions.[19] At Ignite in November 2025, Microsoft announced a successor, Cobalt 200, as the next generation of its cloud-native CPU.[19]
Microsoft's custom silicon complements rather than replaces its very large fleet of NVIDIA GPUs, which still carry the heaviest AI training and inference workloads on Azure.
Microsoft has spent heavily on a new class of AI-focused datacenters. In September 2025 it introduced Fairwater, a datacenter in Wisconsin that it called the largest and most sophisticated AI facility it had built, and described it as the first in a series of identical Fairwater sites planned across the United States.[20] Microsoft then connected the Wisconsin site with a second Fairwater datacenter in Atlanta, Georgia, to form what it calls an AI superfactory, a single distributed system spanning multiple locations.[21]
The Fairwater design uses a flat network that can knit hundreds of thousands of NVIDIA GB200 and GB300 GPUs into one unified supercomputer.[20][21] Within that fabric, each rack holds 72 NVIDIA Blackwell GPUs joined in a single NVLink domain that delivers 1.8 terabytes per second of GPU-to-GPU bandwidth and gives every GPU access to 14 terabytes of pooled memory.[20] Microsoft has said these facilities represent tens of billions of dollars of investment and hundreds of thousands of AI chips, and the broader buildout pushed the company's capital spending to record levels through 2025 and 2026.[20] The Wisconsin and Atlanta sites went into operation in 2025 and 2026, and the company continued construction on additional Fairwater datacenters elsewhere in the country.[20][21]
Azure competes primarily with Amazon Web Services and Google Cloud, the other two of the three large hyperscale providers that together control most of the global cloud market.[3] AWS remains the market leader with the largest share, Azure sits second with about a fifth of the market, and Google Cloud holds third place.[3] Beyond the top three, Azure faces competition from Oracle Cloud Infrastructure, IBM Cloud, and Alibaba Cloud, as well as from specialized AI-cloud providers such as CoreWeave that focus on GPU capacity.
The competitive dynamics in AI are unusual because the providers are also customers, partners, and investors in one another's ecosystems. Microsoft's OpenAI partnership gave Azure an early lead in hosting frontier models, but the loosening of exclusivity in 2025 means OpenAI workloads can now run on rival clouds, and Microsoft's own Copilot products have started to incorporate models from other vendors.[14][16] Each major provider has also invested in custom silicon, with AWS building its Trainium and Inferentia chips and Google its Tensor Processing Units, in parallel with Microsoft's Maia and Cobalt lines, all while continuing to deploy large quantities of NVIDIA GPUs.