Microsoft Azure in 2025: Year in Review

As 2025 got underway, Microsoft hinted at the directions Azure would take throughout the year with AI plans, partnerships, and hyper-cloud competition in focus. 

Early in January, executives touted the business case for Responsible AI and ways to boost performance by combining models. 

At the 2025 Retail BIG Show from NRF, they profiled the power of AI agents and squared off with Salesforce, offering a different pitch on AI and retail. Soon after, Microsoft and SAP announced a partnership, offering RISE with SAP on the Microsoft Azure Global Acceleration Program.

Microsoft and its competitors began to contend with how to comply with the European Union’s AI Act. New obligations came into effect in 2025, prompting Microsoft to publish AI Act documentation on its Trust Center. In what it described as its inaugural Transparency Report, the company took a risk management approach across the AI development lifecycle doing impact assessments and red-teaming to spot potential risks. Additionally, it offered classifiers are a part of AI Content Safety. 

Near the end of January, Microsoft announced Q2 earnings. Cloud and AI revenue had risen 175 percent year-over-year. At the time, the company experienced a six percent sell off after reporting the results. Azure revenue growth showed a bifurcation, driven by AI services but with lagging channel growth. 

In early February, CVP Asha Sharma highlighted the availability of DeepSeek R1 for AI Foundry and GitHub. The company recapped a win for Azure DDoS during the 2024 Christmas shopping season, when the service fended off highly volumetric DDoS attacks of over 10M pps.

At the start of March, Microsoft implemented the EU Data Boundary for Microsoft Cloud. Customers in Europe gained the ability to process their customer data and pseudonymized personal data for Microsoft core cloud services — including Microsoft 365, Dynamics 365, Power Platform, and most Azure services — within the EU and EFTA regions.

Microsoft was named a leader in the Forrester Wave for Cloud Platforms the same month. Microsoft placed in the lead category alongside AWS, Alibaba Cloud, and Google Cloud Platform. Forrester took note of Microsoft’s focus on AI-powered app development, the launch of AI Foundry in 2024, integration between hybrid and multicloud, as well as Microsoft Fabric.

NVIDIA and Microsoft integrated NVIDIA’s Blackwell platform with Azure AI infrastructure services, announcing the change in mid-March. Aksha Sharma, Microsoft’s CVP for AI Platform noted the integration of AI Foundry and NVIDIA NIM. Through this work, Microsoft endeavored to offer zero-configuration deployments, support seamless Azure integration, and provide “enterprise-grade” reliability. NVIDIA’s AgentIQ is also part of the picture, providing an open-source toolkit to connect and optimize groupings of AI agents. It is intended to support profiling, optimization, dynamic inference enhancement, and be integrated with Semantic Kernel.

AI Foundry continued to add features to support agent apps into early April. Microsoft added some of the first models from the Llama 4 herd to AI Foundry. he models, which originate with Meta, are meant to seamlessly integrate text and vision tokens in a unified model backbone. AI Foundry added Llama 4 Scout and Maverick models as managed compute offerings. 

At KubeCon Europe 2025, the company announced Azure Storage updates. Numerous customers deploy stateful workloads to Kubernetes with open-source databases like MariaDB, PostgreSQL, and MySQL. When customers need low latency and high input/output operations per second, Container Storage allows them to use ephemeral non-volative memory express (NVMe) drives in a node pool. By doing so, customers can achieve latency of less than a millisecond, up to half a million IOPS, and a five-fold increase in transactions per second. During the event, Microsoft demonstrated how customers can use local NVMe and Premium SSD v2 disks to achieve high availability PostgreSQL deployments.

In mid-April, Steve Sweetman, OpenAI Service product lead at Microsoft, shared the release of the GPT-4.1 model series for AI Foundry and GitHub developers. Models including GPT-4.1, 4.1-mini, and 4.1-nano launched for both GitHub and OpenAI Service. With GPT-4.1, for instance, customers can boost coding, leverage a long context model, and enhance instruction following. Customers increasingly have the option to do supervised fine-tuning. Sweetman also touted an array of models like o3 and 04-mini to support multimodality and output reasoning.

By the end of April, Microsoft could count 1,917 models available for AI Foundry. CVP Jessica Hawk described the product as “like the GPS for your AI ambitions.”

Azure expert Jeff Christman offered insights into the differences between Azure Government and Commercial Cloud. Security, access limitations, technical differences, and tooling adjustments can all impact the nature of the work, he wrote.

Working in Azure Government Cloud means embracing a different operational mindset. You must prioritize compliance, documentation, and communication. Projects typically take longer due to approval processes and access constraints. However, the opportunity to support critical infrastructure and national security missions makes the extra effort worthwhile.

VP of Storage Aung Oo shared the general availability of Storage Actions in mid-May. The managed platform is aimed at transforming the way customers manage tasks with Azure Blobs and Data Lake Storage to reduce data sprawl and lack of uniform standards for data management. 

In June, Databricks and Microsoft announced deeper integrations for Azure Databricks with an extended partnership. Together, the two companies planned to release SAP Databricks on Azure.

IDC released a study, commissioned by Microsoft, which found a 306% three-year return on investment with an 11-month payback on investment for companies that migrated their Ubuntu workloads from on-prem to Azure. At nearly the same time, CVP Douglas Phillips shared details about Microsoft Planetary Computer Pro. The need to connect hyperscale clouds with the growing number of satellite constellations to manage geospatial data prompted Microsoft to work to overcome silos between geospatial data stores. Planetary Computer Pro--in public preview at the time--is intended to help ingest, catalog, store, process, and disseminate for real-world scenarios.

After Microsoft headquarters returned to its regular pace of activity in the new fiscal year in July, Azure thought leaders shared insights on using Serverless Compute and scaling AI transformation with AI Center of Excellence guidance. 

Microsoft has deployed multiple iterations of the Phi model family. In mid-July, the company premiered Phi-4-mini-flash reasoning. Although the new model shares a name with Phi-4-mini, it is built on a different hybrid architecture with 10 times higher throughput with an average of two to three times average reduction in latency that allows significantly faster inferencing without reducing reasoning performance. Around the same time, Microsoft introduced Deep Research in AI Foundry Agent Service. 

At the close of July, Microsoft announced that its Q4 2025 earnings outperformed expectations, marking a strong year of Azure growth. In announcing the results, CEO Satya Nadella highlighted cloud and AI performance for the year, stating, “We’re innovating across the tech stack to help customers adapt and grow in this new era, and this year, Azure surpassed $75 billion in revenue, up 34 percent, driven by growth across all workloads.”

Throughout August, Microsoft slowed its pace of announcements amid summer vacations. Subject-matter experts shared insights on protecting Azure infrastructure, cloud migration best practices, and opportunities with Databricks. Microsoft achieved “leader” status in the 2025 Gartner Magic Quadrant for Container Management. 

In September, AKS Automatic entered general availability. AKS Automatic is intended to reduce the learning curve for new Kubernetes users, free up resources, mitigate security, and cut down on reliability problems associated with misconfigurations. 

Microsoft announced an update to its marketplace strategy with Microsoft Marketplace, a single destination that unifies their existing AppSource and Azure Marketplace sites. It is rolling out today in the US and “coming soon” to the rest of the world.

Combined, AppSource and Azure Marketplace contain tens of thousands of solutions, apps, and service offers for Azure, Microsoft 365, Dynamics 365, Power Platform, Microsoft Security and more. 

AppSource launched in mid-2016 along with the Dynamics 365 cloud brand. 3,000 AI agents and apps are newly available on Microsoft Marketplace as of late September. Microsoft launched these offerings with the help of at least 96 partners.

Jeff Christman contributed an excellent new article on securing Azure tenants for government agencies. He wrote:

Even small configuration changes can have cascading effects, especially in government environments where dependencies are tightly coupled.

Microsoft maintains distinct Azure environments, including the Commercial Cloud, US Government Cloud, and China Cloud, each with unique operational characteristics. For instance, PowerShell commands and service availability differ between the commercial and government versions. Understanding these differences—and the associated FedRAMP impact levels—is critical for maintaining compliance and operational integrity.”

In mid-October, Microsoft announced the availability of Sora 2 in AI Foundry. Other announcements followed like OpenAI GPT-image-1-mini, GPT-realtime-mini, and GPT-audio-mini, as well as safety upgrades to GPT-5, plus Grok 4 as part of the AI Foundry toolkit. Storage Discovery entered general availability as well, offering ways to consolidate data sprawl.

Microsoft’s strong financials continued into it’s announcement of 2026 Q1 financial performance. The company noted growing investments related to AI that are impacting factors like capital expenditure, which rose 74%. Cloud margin due to AI infrastructure decreased 68%, with the company witnessing growing cost of revenue in various areas.

At GitHub Universe in November, GitHub and Microsoft emphasized a new era of collaboration with agents becoming a key part of software development. GitHub can run in the cloud or locally, supporting expanded security, scalability, and trust with Agent HQ. During the event, GitHub touted achievements like 180 million developers, 80 percent of developers using Copilot in the first week, and 4.3 million AI-related repositories. 

Ignite 2025 delivered a wave of announcements across cloud, biz apps, productivity, and the Microsoft channel. Some mid-size and large enterprises are already executing sophisticated AI plans, but Microsoft believes SMBs will look to their partners for the bulk of their guidance on challenges like agent governance, user adoption, and cost management.

In early December, Microsoft introduced Mistral Large 3 in Microsoft Foundry. The new offering provides customers with an open-weight Apache-licensed frontier model set, adding to options like DeepSeek and GPT OSS. It is intended to meet enterprise application needs like long-context comprehension, multimodal reasoning, predictable performance, and reliable instruction following. The company also introduced Claude Opus 4.5. Designed by Anthropic, Claude Opus 4.5--in public preview for Foundry, GitHub Copilot, and Copilot Studio—can support complex multi-tool workflows.

…

Photo by Towfiqu barbhuiya on Unsplash