o3-mini’s advanced capabilities, combined with its efficiency gains, make it a powerful tool for developers and enterprises looking to optimize their AI applications.

With faster performance and lower latency, o3-mini is designed to handle complex reasoning workloads while maintaining efficiency.

New features of o3-mini

As the evolution of OpenAI o1-mini, o3-mini introduces several key features that enhance AI reasoning and customization:

• Reasoning effort parameter: Allows users to adjust the model’s cognitive load with low, medium, and high reasoning levels, providing greater control over the response and latency. 
• Structured outputs: The model now supports JSON Schema constraints, making it easier to generate well-defined, structured outputs for automated workflows.
• Functions and Tools support: Like previous models, o3-mini seamlessly integrates with functions and external tools, making it ideal for AI-powered automation. 
• Developer messages: The “role”: “developer” attribute replaces the system message in previous models, offering more flexible and structured instruction handling.
• System message compatibility: Azure OpenAI Service maps the legacy system message to developer message to ensure seamless backward compatibility.
• Continued strength on coding, math, and scientific reasoning: o3-mini further enhances its capabilities in coding, mathematics, and scientific reasoning, ensuring high performance in these critical areas. 

With these improvements in speed, control, and cost-efficiency, o3-mini is optimized for enterprise AI solutions, enabling businesses to scale their AI applications efficiently while maintaining precision and reliability. 

From o1-mini to o3-mini: What’s changed? 

o3-mini is the latest reasoning model released, with notable differences compared with the o1 model released last September. While both models share strengths in reasoning, o3-mini adds new capabilities like structured outputs and functions and tools, resulting in a production-ready model with significant improvements in cost efficiencies. 

Feature comparison: o3-mini versus o1-mini

Watch o3-mini in action, helping with banking fraud, in the demo below:

And watch this o3-mini demo for financial analysis:

Join us on this journey

We invite you to explore the capabilities of o3-mini and see how it can transform your AI applications. With Azure OpenAI Service, you get access to the latest AI innovations, enterprise-grade security, and global compliance, and data remains private and secure.

Learn more about OpenAI o3-mini in GitHub Copilot and GitHub Models here.

Get started today! Sign up in Azure AI Foundry to access o3-mini and other advanced AI models.

The post Announcing the availability of the o3-mini reasoning model in Microsoft Azure OpenAI Service appeared first on Microsoft AI Blogs.

Accelerating AI reasoning for developers on Azure AI Foundry

AI reasoning is becoming more accessible at a rapid pace transforming how developers and enterprises leverage cutting-edge intelligence. As DeepSeek mentions, R1 offers a powerful, cost-efficient model that allows more users to harness state-of-the-art AI capabilities with minimal infrastructure investment. 

One of the key advantages of using DeepSeek R1 or any other model on Azure AI Foundry is the speed at which developers can experiment, iterate, and integrate AI into their workflows. With built-in model evaluation tools, they can quickly compare outputs, benchmark performance, and scale AI-powered applications. This rapid accessibility—once unimaginable just months ago—is central to our vision for Azure AI Foundry: bringing the best AI models together in one place to accelerate innovation and unlock new possibilities for enterprises worldwide. 

Develop with trustworthy AI

We are committed to enabling customers to build production-ready AI applications quickly while maintaining the highest levels of safety and security. DeepSeek R1 has undergone rigorous red teaming and safety evaluations, including automated assessments of model behavior and extensive security reviews to mitigate potential risks. With Azure AI Content Safety, built-in content filtering is available by default, with opt-out options for flexibility. Additionally, the Safety Evaluation System allows customers to efficiently test their applications before deployment. These safeguards help Azure AI Foundry provide a secure, compliant, and responsible environment for enterprises to confidently deploy AI solutions. 

How to use DeepSeek in model catalog

• If you don’t have an Azure subscription, you can sign up for an Azure account here.
• Search for DeepSeek R1 in the model catalog.
• Open the model card in the model catalog on Azure AI Foundry.
• Click on deploy to obtain the inference API and key and also to access the playground. 
• You should land on the deployment page that shows you the API and key in less than a minute. You can try out your prompts in the playground.
• You can use the API and key with various clients.

Get started today

DeepSeek R1 is now available via a serverless endpoint through the model catalog in Azure AI Foundry. Get started on Azure AI Foundry here and select the DeepSeek model.

On GitHub, you can explore additional resources and step-by-step guides to integrate DeepSeek R1 seamlessly into your applications. Read the GitHub Models blog post.

Customers will be able to use distilled flavors of the DeepSeek R1 model to run locally on their Copilot+ PCs. Read the Windows Developer blog post.

As we continue expanding the model catalog in Azure AI Foundry, we’re excited to see how developers and enterprises leverage DeepSeek R1 to tackle real-world challenges and deliver transformative experiences. We are committed to offering the most comprehensive portfolio of AI models, ensuring that businesses of all sizes have access to cutting-edge tools to drive innovation and success. 

The post DeepSeek R1 is now available on Azure AI Foundry and GitHub appeared first on Microsoft AI Blogs.

1. Copilot is your UI for AI, with Copilot Studio enabling low-code creation of agents and extensibility to your data.
2. Azure AI Foundry is the only AI app server for building real-world, world-class, AI-native applications.
3. Microsoft Fabric is the AI data platform that provides one common way to reason over your data —no matter where it lives.

All three of these platforms are open and work synchronously to enable the development of modern AI solutions; and each is surrounded by our world-class security offerings so leaders can move their AI-first strategies forward with confidence.

As we look ahead to what we can achieve together, I remain inspired by the work we are doing today. Below are a handful of the many stories from the past quarter highlighting the differentiated AI solutions our customers and partners are driving to move business forward across industries and realize pragmatic value. Their success clearly illustrates that real results can be harnessed from AI today, and it is changing the way organizations do business.

To power its industrial IoT and AI platform, ABB Group leveraged Microsoft Azure OpenAI Service to create Genix Copilot: a generative AI-powered analytics suite aimed at solving some of the most complex industrial problems. The solution helps customers analyze key functions in their operations —such as asset and process performance, energy optimization and emission monitoring — with real-time operational insights. As a result, customers are seeing up to 35% savings in operations and maintenance, and up to 20% improvement in energy and emission optimization. ABB also saw an 80% decrease in service calls with the self-service capabilities of Genix Copilot.

Serving government healthcare agencies across the US, Acentra Health turned to Microsoft to help introduce the latest AI capabilities that maximize talent and cut costs in a secure, HIPAA-compliant manner. Using Azure OpenAI Service, the company developed MedScribe — an AI-powered tool reducing the time specially trained nursing staff spend on appeal determination letters. This innovation saved 11,000 nursing hours and nearly $800,000, reducing time spent on each appeal determination letter by about 50%. MedScribe also significantly enhanced operational efficiency, enabling nurses to process 20 to 30 letters daily with a 99% approval rate.

To ease challenges for small farmers, Romanian agribusiness group Agricover revolutionized access to credit by developing MyAgricover. Built with help from partner Avaelgo, the scalable digital platform utilizes Microsoft Azure, Azure API Management and Microsoft Fabric to automate the loan process and enable faster approvals and disbursements. This has empowered small farmers to grow their businesses and receive faster access to financing by reducing loan approval time by 90 percent — from 10 working days to a maximum of 24 hours.

Building on its status as a world-class airline with a strong Indian identity, Air India sought ways to enhance customer support while managing costs. By developing AI.g, one of the industry’s first generative AI virtual assistants built on Azure OpenAI Service, the airline upgraded the customer experience. Today, 97% of customer queries are handled with full automation, resulting in millions of dollars of support costs saved and improved customer satisfaction — further positioning the airline for continued growth.

BMW Group aimed to enhance data delivery efficiency and improve vehicle development and prototyping cycles by implementing a Mobile Data Recorder (MDR) solution with Azure App Service, Azure AI and Azure Kubernetes Service (AKS). The solution achieved 10 times more efficient data delivery, significantly improved data accessibility and elevated overall development quality. The MDR monitors and records more than 10,000 signals twice per second in every vehicle of BMW’s fleet of 3,500 development cars and transmits data within seconds to a centralized cloud back end. Using Azure AI Foundry and Azure OpenAI Service, BMW Group created an MDR copilot fueled by GPT-4o. Engineers can now chat with the interface using natural language, and the MDR copilot converts the conversations into KQL queries, simplifying access to technical insights. Moving from on-premises tools to a cloud-based system with faster data management also helps engineers troubleshoot in real time. The vehicle data covered by the system has doubled, and data delivery and analysis happen 10 times faster.

Coles Group modernized its logistics and administrative applications using Microsoft Azure Stack HCI to scale its edge AI capabilities and improve efficiency and customer experience across its 1,800 stores. By expanding its Azure Stack HCI footprint from two stores to over 500, Coles achieved a six-fold increase in the pace of application deployment, significantly enhancing operational efficiency and enabling rapid innovation without disrupting workloads. The retailer is also using Azure Machine Learning to train and develop edge AI models, speeding up data annotation time for training models by 50%.

Multinational advertising and media company Dentsu wanted to speed time to insights for its team of data scientists and media analysts to support its media planning and budget optimization. Using Microsoft Azure AI Foundry and Azure OpenAI Service, Dentsu developers built a predictive analytics copilot that uses conversational chat and draws on deep expertise in media forecasting, budgeting and optimization. This AI-driven tool has reduced time to media insights for employees and clients by 90% and cut analysis costs.

To overcome the limitations of its current systems, scale operations and automate processes across millions of workflows, Docusign created the Intelligent Agreement Management (IAM) platform on Azure. Using Azure AI, Azure Cosmos DB, Azure Logic Apps and AKS, the platform transforms agreement data into actionable insights to enhance productivity and accelerate contract review cycles. IAM also ensures better collaboration and unification across business systems to provide secure solutions tailored to diverse customer needs. For example, its customer KPC Private funds reported a 70% reduction in time and resources dedicated to agreement processes.

Emirates Global Aluminium (EGA) transformed its manufacturing operations by leveraging a hybrid environment with Azure Arc, Azure Stack HCI and Azure Kubernetes Service. This digital manufacturing platform resulted in 86% cost savings for AI image and video analytics and a 13-fold improvement in AI response times. The seamless hybrid cloud architecture has enhanced EGA’s operational efficiency and agility, supporting its Industry 4.0 transformation strategy.

EY collaborated with Microsoft to enhance the inclusivity of AI development using Azure AI Foundry. By involving neurodivergent technologists from EY’s Neuro-Diverse Centers of Excellence, they improved the accessibility and productivity of AI tools, resulting in more inclusive AI solutions, fostering innovation and ensuring that AI tools unlock the potential of all users. With an estimated 20% of the global workforce identifying as neurodivergent, inclusive AI solutions are crucial for maximizing creativity and productivity. Neurodivergent EY technologists also collaborated with Microsoft developers to make Azure AI Foundry more inclusive and help all users work productively to create innovative AI solutions.

Colombian household appliance manufacturer Haceb integrated AI to optimize processes, reduce costs and improve service quality. Using Microsoft Copilot Studio and Azure OpenAI Service, the company created a virtual technical support assistant, saving its 245 technicians 5 minutes per visit — a total of 5,000 minutes saved daily. This AI solution has enhanced efficiency and boosted customer satisfaction by allowing for faster issue resolution. Haceb’s AI adoption has also empowered employees, boosted productivity and positioned the company as a leader in AI innovation in Colombia.

To better serve its global patients, Operation Smile — in collaboration with partner Squadra — leveraged Azure AI, Machine Learning and Microsoft Fabric to develop an AI-powered solution to predict surgical outcomes and optimize resource allocation. This innovation resulted in a 30% increase in surgical efficiency, a 90% reduction in translation errors and improved patient outcomes. Additionally, report generation is now up to 95% quicker, and repeated medical events have decreased by 15%, enabling Operation Smile to provide better care to more children worldwide.

Ontada — a McKesson business dedicated to oncology data and evidence, clinical education and point-of-care technologies — needed a way to generate key insights across 150 million unstructured oncology documents. Using Microsoft Azure AI and Azure OpenAI Service, Ontada developed a data platform solution called ON.Genuity to provide AI-driven insights into the patient journey, enhance patient trial matching and identify care gaps. The company also implemented large language models to target nearly 100 critical oncology data elements across 39 cancer types, enabling the company to analyze an estimated 70% of previously inaccessible data, reduce processing time by 75% and accelerate product time-to-market from months to just one week.

As the UK’s largest pet care company, Pets at Home sought a way to combat fraud across its retail operations — particularly as its online business continued to grow. Working closely with its fraud team, it adopted Copilot Studio to develop an AI agent that quickly identifies suspicious transactions. The agent autonomously gathers relevant information, performs analysis and shares it with a fraud agent to enable a manual, data-intensive investigative process while ensuring a human remains in the loop. With this low-code agent extending and seamlessly integrating into existing systems, the company’s fraud department can act more quickly; what used to take 20 to 30 minutes is now handled by the AI agent within seconds. The company is identifying fraud 10 times faster and is processing 20 times more cases a day. Now, the company can operate at scale with speed, efficiency and accuracy — with savings expected to be in the seven figures as it continues to build more agents.

Revenue Grid, a technology company specializing in sales engagement and revenue optimization solutions, partnered with Cloud Services to modernize its data infrastructure and develop a unified data warehouse capable of handling unstructured, semi-structured and structured data. By migrating to Microsoft Fabric, Revenue Grid can now deliver data-powered revenue intelligence, driven by a unified platform, elastic scalability, enhanced analytics capabilities and streamlined operations. Revenue Grid has reduced infrastructure costs by 60% while enhancing its analytical capabilities to improve real-time data processing, empowering sales teams with accurate and diverse data. 

To better manage and integrate employee data across diverse regions and systems, UST built a comprehensive Employee Data platform on Microsoft Fabric. In under a year, UST migrated 20 years of employee data with all security measures to enhance data accessibility and employee productivity. The Meta Data Driven Integration (MDDI) framework in Fabric also helped the company cut data ingestion time by 50% so employees can focus more on analysis than preparation. As a result of this implementation, the company has seen an increase in collaboration and innovation from employees, helping put its values into action.

The Microsoft Commercial Marketplace offers millions of customers worldwide a convenient place to find, try and buy software and services across 140 countries. As a Marketplace partner, WeTransact is helping independent software vendors (ISVs) list and transact their software solutions — and find opportunities for co-selling and extending their reach to enterprise customers through development of the WeTransact platform. Powered by Azure OpenAI Service, the platform is changing the way partnerships are being built by using AI pairing to facilitate a “plug and play” reseller network. More than 300 ISVs worldwide have joined the Microsoft Commercial Marketplace using the WeTransact platform, cutting their time to publish by 75%.

The opportunity for AI to create value is no longer an ambition for the future — it is happening now, and organizational leaders across industries are investing in AI-first strategies to change the way they do business. We believe AI should empower human achievement and enrich the lives of employees; and we are uniquely differentiated to help you accelerate your AI Transformation responsibly and securely. Choosing the right technology provider comes down to trust, and I look forward to what we will achieve together as we partner with you on your AI journey.

The post The value of AI: How Microsoft’s customers and partners are creating differentiated AI solutions to reinvent how they do business today appeared first on Microsoft AI Blogs.

In each episode, we’ll explore how our guests are using data, AI, and cloud technologies to deliver meaningful and trustworthy innovation to their organizations, customers, and communities. You’ll hear from executives, developers, data scientists, and visionaries across industries in both the public and private sectors. We’ll discuss how the platform shift is affecting everything from data strategy, customer relationships, organizational structure, and culture, to their own lives and career journeys. And, most importantly, they’ll share what they’ve learned along the way, as well as their advice on how to get started and how to navigate bumps in the road.

The first four episodes of Leading the Shift cover a lot of ground, from how data and AI are being used for social impact, to the use of AI to create a fan remix experience for a world-famous band, to the opportunity to use data, AI, and cloud technologies to map the customer journey and truly personalize experiences, to the potential for proprietary data to transform an organization’s competitive advantage.

Throughout these conversations we aim to get at the heart of the approaches, tools, techniques, opportunities, threats, and emerging best practices that leaders are adopting as they learn by doing. We’ll see common themes emerge: the transformative power of data and AI, the importance of trust, the innovation potential of co-creation, and the value of getting hands-on during the process.

Here’s what you can expect from our first four episodes:

Creating the new field of data for social impact, with Perry Hewitt, data.org

You’ve heard about data-driven products and services, and even data-driven business models. But in our premiere episode, Perry Hewitt, Chief Marketing and Product Officer at data.org, discusses how the organization is using data and AI to create an entirely new field—data for social impact.

Perry shares how data.org is working with organizations around the world to co-create solutions to some of the world’s greatest challenges, such as dispelling misconceptions about women’s health in India, upskilling migrant Venezuelan women in Chile, and creating intensive AI curricula for teachers, students and businesses in rural Mississippi.

It’s a rigorous and innovative approach built around trust, respect, and co-creation, and is as applicable to business as it is to the public sector.

Remixing the Coldplay fan experience with AI, with Robby Ingebretsen, Pixel Lab

In this episode, Robby Ingebretsen, Founder and Creative Director of Pixel Lab, an award-winning creative agency, talks about how he and his team built a fan remix experience that builds on the release of Coldplay’s new album, MOON MUSiC, and its film compendium, A Film For The Future.

The fan remix experience offers people the opportunity to join a community of Coldplay fans to create their own remix of the film using inputs that match their mood and attribute choices and contribute to the film as their piece is added to the future playback.  

It’s a wide-ranging conversation that explores the impact of digitalization and platform shifts in the music and technology industries, the impact and opportunities of AI, and how co-creation can unlock new opportunities for creativity, engagement, and innovation for customers and consumers across a range of industries. Read more about how this experience was built with Azure AI Foundry.

Data doesn’t just fuel generative AI—it goes both ways, with Shirli Zelcer, dentsu

Shirli Zelcer began her career as a statistician and is now Chief Data and Technology Officer at dentsu, an integrated growth and transformation partner to the world’s leading organizations. It’s a fascinating role, and her roots in data, combined with a deep understanding of AI, have prepared her to bring data and AI together to help dentsu and its clients unlock new sources of value.

Shirli shares a bit about her journey, including how she’s seen data and analytics evolve from a back-office practice to a C-suite priority. We explore a range of topics, including new approaches to generating audiences, the opportunity to combine structured and unstructured data to better understand customer needs and behavior, the value of synthetic data, and—underlying all of this innovation—the imperative to use intelligent data and technologies in a trustworthy, responsible, and privacy-compliant way.

Your proprietary data is your competitive advantage, with Teresa Tung, Accenture

In this episode, we talk with Teresa Tung, Global Lead of Data Capability at Accenture, about the evolving role of data in AI strategy. We explore the strategic value of proprietary data, how generative AI can help organizations realize value from unstructured data, the changing landscape of data governance, and how synthetic data can model sensitive scenarios in a safer way. We also discuss how generative AI can help organizations jumpstart data capability.

Teresa is a prolific inventor, holding over 225 patents and applications, and leads the vision and strategy that ensures that Accenture is prepared for ever-changing data advancements. It’s a real treat to hear her share what she’s seeing in the industry and what she believes will help organizations deliver value to their customers and clients.

Looking ahead with Leading the Shift

Leading the Shift will release new episodes in the last week of each month. Listen, like, and subscribe wherever you get your podcasts, including Spotify, Apple Podcasts, and YouTube.

The post Introducing Leading the Shift, a new Microsoft Azure podcast appeared first on Microsoft AI Blogs.

My first concert with my son was a Coldplay show in Vancouver a couple of years ago. I was completely captivated by the show, the music, the immersive experience, and the focus on sustainability—but mostly by watching my youngest experience a concert for the first time. It was truly magical. The band has always been at the forefront of innovation and generating magic with their fans. 

To coincide with their latest project, A Film For The Future, the band collaborated with Microsoft to produce an AI-powered experience that lets fans interact with their new album MOON MUSiC in a unique and personal way.

After some initial skepticism when generative AI emerged, we’re now seeing creators and artists begin to experiment with AI as a creative booster—bringing fresh ways to interact with and enhance their work, come up with new ideas, and even get things done faster.

A Film For The Future is a highly collaborative project, bringing together a diverse group of filmmakers and animators to create unique segments for the 44-minute film, which provides a visual accompaniment to the new album. Each creator was given a broad set of themes and creative license to come up with visuals each felt best represented Coldplay’s music. That collaborative theme now extends to the fan experience, which transcends traditional passive viewing and enables you to create your own 15-second clip of the film using Microsoft Copilot and Azure AI. Your clip, or remix, is then added to the community playback here.

By going to the Community Remix section on the film’s website, you can see clips created by others accompanied by the “iAAM” track (“I Am A Mountain”) from “MOON MUSiC” and can create your own clip.

Not surprisingly, it’s a very emotive experience as you create your own personal remix. Mine is trust in summoning lightning, which is obviously awesome. You can fine-tune your own remix by adjusting the intensity of seven different attributes, such as passion, growth, and peace, that are represented as moons against a rainbow (trust me, you have to see it). Once you’re done, your remix is added to the community and you can download a video and image of your remix if you’d like. You can also create another remix.

The memories of that Vancouver concert came flooding back while I was in the app, giving me a wonderful moment to plunge into a cherished memory. It was unexpected and kind of amazing. Not every app can evoke such a response. Coldplay certainly brings some emotional chemistry to the mix, but this creative use of AI opens new ground for fans. 

Microsoft collaborated with Seattle-based Pixel Lab to build this fan remix experience. The platform AI analyzes the emotional context of each video clip and dynamically assembles them to create a unique and immersive experience for every fan. This means that no two remixes are the same, and each fan gets a personalized journey through Coldplay’s music, making the experience deeply engaging and memorable.

From the Community Remix experience, you can chat with Microsoft Copilot about Coldplay, the new album, A Film For The Future, or whatever you want to chat about using simple prompts.

The fusion of generative AI and human creativity is opening new vistas for artists and businesses alike. The fan remix experience is more than a showcase of cutting-edge technology. Fans can now become co-creators, using Microsoft AI to craft their own unique interpretations of Coldplay’s music. It highlights one of the many strategic values of integrating AI into creative processes to unlock new opportunities for innovation and differentiation. 

Made with Azure AI Foundry

This fan remix experience was built with a collection of Azure AI services available in Azure AI Foundry, Microsoft’s unified AI platform announced a few months ago at Microsoft Ignite. It integrates advanced AI services like natural language processing, computer vision, and machine learning to help organizations across industries create AI-powered solutions that accelerate innovation and differentiate them in the market. Azure AI Foundry enables everyone—from developers to data scientists, to business and IT leaders—to collaborate seamlessly design, customize, and manage innovative solutions that transform ideas into reality. 

Enhancing human creativity with AI 

At the heart of the Coldplay project is the belief that AI can enhance human creativity rather than replace it. Technology has been an important part of creative expression for a long, long time. Artists are often among the first to test the creative potential of technical innovation. Generative AI expands the role of technology in artistic expression, bringing audiences into the creative process, and in this case, elevating fans to co-creators with the artist. It’s a glimpse into how AI is changing expectations for how we engage with our favorite artists.

About Jessica

Jessica leads Azure data, AI, and digital application product marketing at Microsoft. Find Jessica’s blog posts here and be sure to follow Jessica on LinkedIn. 

The post Coldplay evolves the fan experience with Microsoft AI appeared first on Microsoft AI Blogs.

• Azure Storage, the filesystem for AI on Azure, enabled model training with exabytes of data and supported inferencing/AI applications for startups and enterprises. 
• Premium SSD v2 disks, our general-purpose block storage, continued to grow double digit MoM, with expanded integration for new workloads like Epic and Oracle. 
• Azure Elastic SAN, the industry’s only cloud native SAN offering, grew by more than 10 times within few months of general availability, highlighting strong customer interests to transition from on-prem SAN to cloud.
• Azure Container Storage, our new Kubernetes storage management solution, simplifies cloud native app development with a unified experience for all backing storage options from ephemeral disks to Azure Disks.

As we enter 2025, our focus remains on our customers, with a vision to provide an intelligent data platform that maximizes the value of their data. We measure our success by our customers’ success and will continue to develop strong fundamentals and innovations to benefit our customers and partners. We extend a heartfelt thank you to our customers and partners for their trust and collaboration, driving us to achieve more together.

Fundamentals for Innovation

Under Microsoft’s Secured Future (SFI) initiatives, we’ve strengthened security across multiple categories—security by default, design, and operations—resulting in significant platform improvements in identity and secret protection, network security, engineering systems security, monitoring, and secure design review. We also introduced customer-facing capabilities for security and data protection, including Microsoft Defender support for Storage, Purview protection policies support (in preview), Network Security Perimeter (in preview), and TLS 1.3 Blob support. We have plans in 2025 to expand our secure by default posture across various offerings, with policies leveraging identity-based authentication and discouraging anonymous access.

Our investments in platform resiliency focus on proactive mitigation of impactful customer issues via predictive analysis. We achieved 52% reduction in high impact incidents in the second half of 2024 compared to 2023. In parallel, we continue to empower customers in building highly available (HA) solutions and enhancing our business continuity and disaster recovery (BCDR) support via capabilities like multi-volume snapshots for Premium SSD v2/Ultra Disks, and planned failover for Blobs and Files (in preview). In 2025, we will leverage AI and build automation to further enhance failure prediction and detection, reducing both the occurrence and recovery time of high impact incidents; and empower customers with richer insights to track workload health signals for proactive mitigation.

Innovations for AI and emerging workloads

From powering cutting-edge AI solutions to driving seamless cloud experiences, storage innovation has become the backbone of technological progress. But how do organizations stay ahead in this rapidly evolving landscape? The answer lies in focused innovations—targeted advancements that transform how we manage, access, and utilize data.

Accelerating AI innovation with differentiated capabilities

In 2024, we advanced large language model development in collaboration with OpenAI, with Azure Storage serving as the foundation for the AI lifecycle. At Ignite, we showcased how OpenAI leveraged Azure Blob Storage throughout the lifecycle, from ingesting vast datasets for model training to enabling seamless data management and expansion without increasing operational complexity. Such achievement is made possible with Blob Storage scaled accounts providing highly scalable throughput to prevent performance bottlenecks and simplify infrastructure. More of these capabilities that we have developed in collaboration with AI-scale customers such as OpenAI will come in 2025. We will also bring these advancements into Azure Data Lake Storage (ADLS) and offer more intelligent capabilities on ADLS targeting AI workloads for more efficient data processing.

We’re working on an exciting innovation that will seamlessly connect unstructured data with Azure AI services and other companies’ AI solutions, whether that data resides on-premises, in other clouds, or within Azure. Building multi-modal enterprise AI applications requires access to vast amounts of unstructured data but enabling secure and well-governed access to critical enterprise data across cloud and on-premises environments is a major challenge for customers. This emerging solution will tackle the challenge by securely connecting your data with Azure AI services like Azure AI Foundry, and Azure Machine Learning, through their deep integration with Azure Blob—without the need for data migration or copying. By accelerating access to AI, reducing time to insight, and enabling the logical unification of data from multiple sources, this solution will pave the way for transformative advancements in enterprise AI. With robust governance powered by tools like Azure RBAC, it will give you complete control over your data estate. The limited preview of this solution will be available soon, with general availability planned for 2025.

Focused innovations for emerging workloads

As the cloud-first approach becomes the default among startups and enterprises, we are witnessing the emergence of new data-centric workloads related to AI training, security scanning built on stateful containers. In response to increasing demand of storage catered for containers, we introduced Azure Container Storage, which offers a range of storage options via the native Kubernetes interface, with support for new block storage options like ephemeral disks and Elastic SAN. This is especially beneficial for workloads hosted on virtual machines (VMs) with graphics processing units (GPUs), where terabytes of local NVMe can accelerate model training and inferencing. Moreover, developers no longer need to choose between resiliency and speed for ephemeral disks. Azure Container Storage offers replication with flexible options, ensuring high resiliency within or across zones and best performance on the market. In 2025, we will enhance integration with Azure Kubernetes Service (AKS) by streamlining the installation experience and improve cost-efficiency with general availability support for Elastic SAN. 

Optimizations for mission critical workloads

We continue to optimize the experience for running mission-critical workloads across our product portfolio. We brought to the market—Azure Elastic SAN, optimized for price-performance at scale and seamless on-premises SAN migration to Azure. With the general availability of Azure VMware Solution integration, customers can leverage it not only for VMs and containers but also as a native VMware datastore. We recently released 99.99% uptime SLA support, instant snapshot (in preview) with Azure Backup (in preview), and capacity auto-scaling (in preview). These capabilities solidify Elastic SAN as a leading choice for enterprise storage at scale. Looking ahead, we are focusing on optimization for database and bare-metal use cases, reducing access latency and improving IOPS scalability, and advancing BCDR support.

In parallel, we are advancing our Disk offerings to meet the growing needs of enterprises. To help customers scale up database performance during peak traffic, we introduced Mbv3 VM series, supporting up to 650,000 disk IOPS—more than a 50% increase. We also enabled general availability for live resizing on our Premium SSD v2/Ultra Disk offerings, simplifying on-demand storage expansion. With the added support for disk type conversion, customers can upgrade to these latest SKUs in place. In the coming year, we are landing more improvements on Premium SSD v2/Ultra Disks including automated disaster recovery via Azure Site Recovery (ASR), and Azure Migrate integration.

Azure NetApp Files (ANF) brings the strength of NetApp ONTAP to the cloud as a native Azure service, offering broad protocol compatibility, and sub-millisecond latency. The general availability release of large volumes, raising the scale from 50 TiB to 2 PiB with over 700,000 IOPS and 12.5 GiBps throughput, positions ANF to handle mission critical workloads in sectors like Electronic Design Automation and high-performance computing (HPC). We will continue optimizing cost-efficiency and enabling integration with Azure AI services.

Azure Files offers fully managed file shares for hybrid, lift-and-shift and cloud native applications. To bring you the benefits of pay-as-you-go pricing while providing predictable costs, we introduced a new provisioning model—Provisioned v2 on Standard Files. You can independently scale out the size, IOPS, and throughput per workload requirements, starting from 32 GiB. In 2025, our focus is to ease lift and shift of workloads with simplified management and seamless data migration while enabling cloud native use cases with integrated identity, and enhanced performance.

Leading with our partner ecosystem

We continue to strengthen our partner ecosystem to provide customers with choice and deliver robust hybrid management capabilities adhering to high standards for quality and security. In 2024, we collaborated with partners in delivering Komprise intelligent tiering for Azure, Qumulo Azure Native v2 and Cloud Native (introduced innovative price/performance tiering between Premium v2 Disks and Blob), and Veeam Data Cloud Vault built on Blob. We pride ourselves on joint innovation and are excited to announce upcoming releases in 2025:

• Nasuni Data Service will allow our joint customers to access data stored in Nasuni UniFS via a Blob API compatible gateway.
• Pure Storage will release a fully managed scalable block storage service on Azure.
• Dell will bring Dell-Managed as-a-Service offerings of OneFS and Data Domain to our joint customers.

Looking ahead with Azure Storage

As we step into 2025, the horizon for storage innovation is more exciting than ever. With advancements like AI-driven optimization, sustainable infrastructure, and lightning-fast access speeds, we are redefining what’s possible in Azure Storage. The year ahead isn’t just about storing more—it’s about empowering businesses to achieve more with smarter, scalable, and future-ready storage solutions. Here’s to a year of turning data into possibilities!

The post Azure Storage—A look back and a look forward appeared first on Microsoft AI Blogs.

In the Microsoft model catalog, there are more than 1,800 AI models available. Even more models and services are available via Azure OpenAI Service and Azure AI Foundry, so you can find the right models to build your optimal AI solution. 

Let’s look at how a multiple model approach works and explore some scenarios where companies successfully implemented this approach to increase performance and reduce costs.

How the multiple model approach works

The multiple model approach involves combining different AI models to solve complex tasks more effectively. Models are trained for different tasks or aspects of a problem, such as language understanding, image recognition, or data analysis. Models can work in parallel and process different parts of the input data simultaneously, route to relevant models, or be used in different ways in an application.

Let’s suppose you want to pair a fine-tuned vision model with a large language model to perform several complex imaging classification tasks in conjunction with natural language queries. Or maybe you have a small model fine-tuned to generate SQL queries on your database schema, and you’d like to pair it with a larger model for more general-purpose tasks such as information retrieval and research assistance. In both of these cases, the multiple model approach could offer you the adaptability to build a comprehensive AI solution that fits your organization’s particular requirements.

Before implementing a multiple model strategy

First, identify and understand the outcome you want to achieve, as this is key to selecting and deploying the right AI models. In addition, each model has its own set of merits and challenges to consider in order to ensure you choose the right ones for your goals. There are several items to consider before implementing a multiple model strategy, including:

• The intended purpose of the models.
• The application’s requirements around model size.
• Training and management of specialized models.
• The varying degrees of accuracy needed.
• Governance of the application and models.
• Security and bias of potential models.
• Cost of models and expected cost at scale.
• The right programming language (check DevQualityEval for current information on the best languages to use with specific models).

The weight you give to each criterion will depend on factors such as your objectives, tech stack, resources, and other variables specific to your organization.

Let’s look at some scenarios as well as a few customers who have implemented multiple models into their workflows.

Scenario 1: Routing

Routing is when AI and machine learning technologies optimize the most efficient paths for use cases such as call centers, logistics, and more. Here are a few examples:

Multimodal routing for diverse data processing

One innovative application of multiple model processing is to route tasks simultaneously through different multimodal models that specialize in processing specific data types such as text, images, sound, and video. For example, you can use a combination of a smaller model like GPT-3.5 turbo, with a multimodal large language model like GPT-4o, depending on the modality. This routing allows an application to process multiple modalities by directing each type of data to the model best suited for it, thus enhancing the system’s overall performance and versatility.

Expert routing for specialized domains

Another example is expert routing, where prompts are directed to specialized models, or “experts,” based on the specific area or field referenced in the task. By implementing expert routing, companies ensure that different types of user queries are handled by the most suitable AI model or service. For instance, technical support questions might be directed to a model trained on technical documentation and support tickets, while general information requests might be handled by a more general-purpose language model.

 Expert routing can be particularly useful in fields such as medicine, where different models can be fine-tuned to handle particular topics or images. Instead of relying on a single large model, multiple smaller models such as Phi-3.5-mini-instruct and Phi-3.5-vision-instruct might be used—each optimized for a defined area like chat or vision, so that each query is handled by the most appropriate expert model, thereby enhancing the precision and relevance of the model’s output. This approach can improve response accuracy and reduce costs associated with fine-tuning large models.

Auto manufacturer

One example of this type of routing comes from a large auto manufacturer. They implemented a Phi model to process most basic tasks quickly while simultaneously routing more complicated tasks to a large language model like GPT-4o. The Phi-3 offline model quickly handles most of the data processing locally, while the GPT online model provides the processing power for larger, more complex queries. This combination helps take advantage of the cost-effective capabilities of Phi-3, while ensuring that more complex, business-critical queries are processed effectively.

Sage

Another example demonstrates how industry-specific use cases can benefit from expert routing. Sage, a leader in accounting, finance, human resources, and payroll technology for small and medium-sized businesses (SMBs), wanted to help their customers discover efficiencies in accounting processes and boost productivity through AI-powered services that could automate routine tasks and provide real-time insights.

Recently, Sage deployed Mistral, a commercially available large language model, and fine-tuned it with accounting-specific data to address gaps in the GPT-4 model used for their Sage Copilot. This fine-tuning allowed Mistral to better understand and respond to accounting-related queries so it could categorize user questions more effectively and then route them to the appropriate agents or deterministic systems. For instance, while the out-of-the-box Mistral large language model might struggle with a cash-flow forecasting question, the fine-tuned version could accurately direct the query through both Sage-specific and domain-specific data, ensuring a precise and relevant response for the user.

Scenario 2: Online and offline use

Online and offline scenarios allow for the dual benefits of storing and processing information locally with an offline AI model, as well as using an online AI model to access globally available data. In this setup, an organization could run a local model for specific tasks on devices (such as a customer service chatbot), while still having access to an online model that could provide data within a broader context.

Hybrid model deployment for healthcare diagnostics

In the healthcare sector, AI models could be deployed in a hybrid manner to provide both online and offline capabilities. In one example, a hospital could use an offline AI model to handle initial diagnostics and data processing locally in IoT devices. Simultaneously, an online AI model could be employed to access the latest medical research from cloud-based databases and medical journals. While the offline model processes patient information locally, the online model provides globally available medical data. This online and offline combination helps ensure that staff can effectively conduct their patient assessments while still benefiting from access to the latest advancements in medical research.

Smart-home systems with local and cloud AI

In smart-home systems, multiple AI models can be used to manage both online and offline tasks. An offline AI model can be embedded within the home network to control basic functions such as lighting, temperature, and security systems, enabling a quicker response and allowing essential services to operate even during internet outages. Meanwhile, an online AI model can be used for tasks that require access to cloud-based services for updates and advanced processing, such as voice recognition and smart-device integration. This dual approach allows smart home systems to maintain basic operations independently while leveraging cloud capabilities for enhanced features and updates.

Scenario 3: Combining task-specific and larger models

Companies looking to optimize cost savings could consider combining a small but powerful task-specific SLM like Phi-3 with a robust large language model. One way this could work is by deploying Phi-3—one of Microsoft’s family of powerful, small language models with groundbreaking performance at low cost and low latency—in edge computing scenarios or applications with stricter latency requirements, together with the processing power of a larger model like GPT.

Additionally, Phi-3 could serve as an initial filter or triage system, handling straightforward queries and only escalating more nuanced or challenging requests to GPT models. This tiered approach helps to optimize workflow efficiency and reduce unnecessary use of more expensive models.

By thoughtfully building a setup of complementary small and large models, businesses can potentially achieve cost-effective performance tailored to their specific use cases.

Capacity

Capacity’s AI-powered Answer Engine® retrieves exact answers for users in seconds. By leveraging cutting-edge AI technologies, Capacity gives organizations a personalized AI research assistant that can seamlessly scale across all teams and departments. They needed a way to help unify diverse datasets and make information more easily accessible and understandable for their customers. By leveraging Phi, Capacity was able to provide enterprises with an effective AI knowledge-management solution that enhances information accessibility, security, and operational efficiency, saving customers time and hassle. Following the successful implementation of Phi-3-Medium, Capacity is now eagerly testing the Phi-3.5-MOE model for use in production.

Our commitment to Trustworthy AI

Organizations across industries are leveraging Azure AI and Copilot capabilities to drive growth, increase productivity, and create value-added experiences.

We’re committed to helping organizations use and build AI that is trustworthy, meaning it is secure, private, and safe. We bring best practices and learnings from decades of researching and building AI products at scale to provide industry-leading commitments and capabilities that span our three pillars of security, privacy, and safety. Trustworthy AI is only possible when you combine our commitments, such as our Secure Future Initiative and our Responsible AI principles, with our product capabilities to unlock AI transformation with confidence. 

Get started with Azure AI Foundry

To learn more about enhancing the reliability, security, and performance of your cloud and AI investments, explore the additional resources below.

• Find the ideal AI model at Azure AI Foundry. 

• Learn more about Azure OpenAI Service models. 

• Read about Phi-3-mini, which performs better than some models twice its size. 

The post Boost processing performance by combining AI models appeared first on Microsoft AI Blogs.

This is where Azure’s adaptive cloud approach enabled by Azure Arc comes into play. Azure is designed to help unify operations, streamline application development, and consolidate data management across distributed infrastructures to meet the demands of an AI-powered world. 

At Microsoft’s Ignite conference, we announced new capabilities that make it easier to use the power of Azure wherever you need it. These announcements include the preview of Azure Arc gateway, the introduction of Windows Server Management enabled by Azure Arc, the preview of Azure Container Storage enabled by Azure Arc, Azure Monitor pipeline, Azure Key Vault Secret Store, general availability (GA) of Azure IoT Operations, Fabric Real-Time Intelligence, and the introduction of Azure Local, among others. 

Azure Arc, the core component of this approach, now supports over 39,000 customers worldwide, connecting servers, databases, and Kubernetes clusters directly to Azure. The rapid growth of Azure Arc signals that the cloud is more than just a place; it is a paradigm shift spread throughout every organization. We are excited to share how customers such as LALIGA, Coles, Husqvarna, and Emirates Global Aluminium are leveraging the adaptive cloud approach to help achieve their business goals. 

Operate anywhere with AI-enhanced management and security

For one of our customers, LALIGA, one of the world’s largest football leagues, Azure’s adaptive cloud approach has been critical for managing an infrastructure capable of helping support real-time data processing across its stadiums and digital platforms to deliver an engaging fan experience. By adopting Azure Arc, LALIGA has achieved seamless management of both cloud and on-premises environments, processing over 3 million data points per match and enabling rapid, AI-powered insights. This unified platform provides LALIGA with the flexibility and scalability to react to dynamic fan interactions and optimize operations, to help ensure they can continue evolving alongside new technologies and market demands. 

“The challenges that we have can be solved with this adaptive cloud approach. I think that we have the right tool, which is Azure Arc. With this, we’re able to manage the infra no matter if it’s located in the cloud, on-premises, in the edge.”—Miguel Angel Leal Góngora, Chief Technology & Innovation Officer, LALIGA 

Enabling operational resilience and security across distributed systems is a foundational requirement to help maintain service and protect sensitive data.

Azure provides comprehensive tools that help streamline operations and management of both infrastructure and applications, including configuration management and governance, resiliency and observability, built-in security and control, and universal AI assistants like Copilot in Azure. An essential part of simplifying cloud connectivity from datacenter and edge sites is the new Azure Arc gateway in preview. Built in response to customer feedback, this capability provides a streamlined approach that simplifies cloud connectivity and empowers teams to manage cloud connections more easily, enhancing control over the network infrastructure. 

In addition, we are simplifying access to Windows Server Management enabled by Azure Arc. At no extra cost, customers with Software Assurance (SA) or active subscription licenses can access certain Azure Arc-enabled management capabilities. By connecting their servers to Azure, customers can use over 20 of Azure’s services, including Azure Update Manager, Machine Configuration, and Azure Monitor, as well as Windows Server features like Azure Site Recovery and Best Practices Assessment. These tools help centralize and modernize management across hybrid, multi-cloud, and edge environments.

Simplifying app development and accelerating innovation with Arc enabled Kubernetes

Leveraging cloud native development to drive innovation is a key focus for modern infrastructure and Microsoft’s adaptive cloud approach provides solutions that help make this easier for developers.

One of the primary challenges that developers face is ensuring applications remain reliable, even in disconnected or intermittent network scenarios. A significant part of this solution is the new Azure Container Storage enabled by Azure Arc, which is currently available in preview. Azure Container Storage allows developers to build containerized applications that operate seamlessly across environments, regardless of where data is stored and despite intermittent connectivity. With this capability, data is automatically synchronized between local storage and cloud environments when connectivity is restored, ensuring that developers can confidently build edge solutions that are scalable and robust. 

The new Azure Monitor pipeline allows teams to ingest and process large volumes of data efficiently, allowing for quick identification and resolution of potential issues. This streamlined data pipeline is key for maintaining operational efficiency and scaling modern cloud-native applications across distributed environments.

Security becomes increasingly complex as systems span clouds, datacenters, and edge. Azure Key Vault Secret Store provides a robust solution for managing secrets within Kubernetes clusters, offering features such as auto-rotation of secrets for enhanced security. This modern cloud security approach helps to ensure that sensitive information remains secure across Linux and Windows environments and offers a reliable and scalable way to secure applications and workloads. 

Coles operates more than 1,800 stores across Australia, including 850 Coles Supermarkets locations. Committed to seamless experiences for millions of customers in person and online, the retailer is constantly investing in innovative technologies, such as AI and computer vision. Coles uses Azure Machine Learning to train and develop edge AI models and run these models locally in their stores. Leveraging Azure’s adaptive cloud approach, the company reports it has met its target of a six-times increase in the pace that it can deploy applications to stores. 

“We’re going to be working more with Microsoft over the next 12 months to build out Azure Machine Learning operations for the edge to be able to seamlessly test and deploy new models and ensure auditability of our models and the different versions over time. The Azure automated machine learning tool was really useful for us, and it speeds up our data annotation time for training models by 50%.”—Roslyn Mackay, Head of Technology Innovation, Coles Group 

Unifying data from edge to cloud for AI-powered insights

In a recent Forrester survey commissioned by Microsoft, respondents shared that on average 46% of the data captured by their organization is currently sent to the cloud and that they expect that number to grow to 68% within just two years. With Azure Arc, organizations can more consistently manage and secure connected devices across distributed environments, to help ensure data integrity at every level. 

The GA of Azure IoT Operations further enhances this capability by simplifying the process of collecting, managing, and processing data from IoT devices. Additionally, the GA of Fabric Real-Time Intelligence enables ingestion of signals from the edge for transformation, visualization, tracking, AI, and real-time actions in the cloud. This complements Fabric’s existing suite of insights and analytics capabilities including OneLake and Power BI. Together, these services help provide businesses with the ability to leverage AI fully across their distributed estate. 

Husqvarna is a world-leader in outdoor products for forest, park, lawn, and garden care, as well as equipment and diamond tools for the light construction industry. Husqvarna’s business goals include doubling the number of connected devices it currently has in the market, doubling the sales of robotic lawn mowers, and increasing its market share of electrified solutions, all of which will require a highly effective global supply chain. Husqvarna envisions Azure IoT Operations as an important component of the platform they are defining by providing new capabilities that will allow them to build a data-powered, global supply chain and improve processes in ways that were previously difficult. For Husqvarna, the ability to harness data on a global scale will allow them to develop a stronger and more efficient supply chain, reduce costs, and help enhance their efficiency in delivering goods to their customers. 

Innovating on a blended infrastructure, together 

Microsoft’s vision for the future of distributed infrastructure centers on creating a seamless blend of cloud and customer environments, fundamentally redefining what cloud infrastructure means. Instead of treating cloud as a separate entity, Microsoft’s approach integrates customer environments with cloud services, allowing businesses to extend their technology operations across datacenter, edge, and public cloud environments.

A cornerstone of this vision is the introduction of Azure Local. Azure Local is an Arc-enabled infrastructure designed specifically for local data processing and critical operations, bringing cloud-like capabilities to on-premises environments. This solution enables organizations to manage workloads that require low-latency and robust performance while benefiting from the scalability and resilience of the cloud. 

Azure Local’s architecture is built to support near real-time processing needs and can help our customers with more options for their regulatory compliance requirements, including decentralized or disconnected scenarios.

Emirates Global Aluminium (EGA), based in the United Arab Emirates has rapidly grown from a small regional smelter into a vertically integrated aluminium provider serving more than 400 customers from aerospace to automotive to consumer goods, and is now the world’s largest premium aluminium producer. To support both its on-site operations and broader cloud-based solutions, including its digital manufacturing platform, EGA is now focused on a plan to move one-third of its server base to the cloud with Azure and another third to run hybrid and at the edge with Azure Local, bringing together the best of the public and private cloud from one provider.

After bringing the power of the cloud into its operation areas, EGA experienced 10 to 13 times faster AI response time, lower latency, and 86% cost savings associated with AI video and image recognition in comparison to building an AI model at the edge independently.

“Microsoft Azure hybrid cloud brings us not just infrastructure as a service capability, but also many software and platform capabilities that open up new possibilities we didn’t have in our former on-premises environment. One of the key capabilities is running real-time AI data analysis in our operations. For example, we used to manually inspect only 2% of anodes, which are large carbon blocks used in the aluminium smelting process. Now, we inspect 100% of all anodes using vision AI based on a neural network machine learning model running on the edge (with Azure Local). This model allows us to standardize the inspection process by automatically recognizing defects in real time going beyond what the human eye can see.”—Carlo Khalil Nizam, Chief Digital Officer, EGA 

Looking ahead: join us in shaping the future of adaptive cloud

At Microsoft, our commitment to the adaptive cloud approach is not just about addressing today’s challenges. It is about equipping organizations to thrive in the AI-powered world of tomorrow. As we continue to innovate, we are excited to partner with you to redefine what is possible across distributed environments.

Want to learn more about our roadmap and how Azure is powering transformation across industries? Check out our Microsoft Ignite sessions and blogs to dive deeper into the latest announcements, hear from our experts, and explore how the adaptive cloud approach can work for you. Let’s build the future together. 

The post Laying a secure cloud foundation for AI-powered enterprises appeared first on Microsoft AI Blogs.

From agile startups to multinational corporations, Microsoft’s infrastructure offers more choice in performance, power, and cost efficiency so that our customers can continue to innovate. At Microsoft Ignite, we’re introducing significant updates across our entire cloud and AI infrastructure, from advancements in chips and liquid cooling, to new data integrations, and more flexible cloud deployments.

Unveiling the latest silicon updates across Azure infrastructure

As part of our systems approach in optimizing every layer in our infrastructure, we continue to combine the best of industry and innovate from our own unique perspectives. In addition to Azure Maia AI accelerators and Azure Cobalt central processing units (CPUs), Microsoft is expanding our custom silicon portfolio to further enhance our infrastructure to deliver more efficiency and security. Azure Integrated HSM (hardware security module) is our newest in-house security chip, which is a dedicated hardware security module that hardens key management to allow encryption and signing keys to remain within the bounds of the HSM, without compromising performance or increasing latency. Azure Integrated HSM will be installed in every new server in Microsoft’s datacenters starting next year to increase protection across Azure’s hardware fleet for both confidential and general-purpose workloads.

We are also introducing Azure Boost DPU, our first in-house DPU designed for data-centric workloads with high efficiency and low power, capable of absorbing multiple components of a traditional server into a single dedicated silicon. We expect future DPU equipped servers to run cloud storage workloads at three times less power and four times the performance compared to existing servers.

We also continue to advance our cooling technology for GPUs and AI accelerators with our next generation liquid cooling “sidekick” rack (heat exchanger unit) supporting AI systems comprised of silicon from industry leaders as well as our own. The unit can be retrofitted into Azure datacenters to support cooling of large-scale AI systems, such as ones from NVIDIA including GB200 in our AI Infrastructure.

In addition to cooling, we are optimizing how we deliver power more efficiently to meet the evolving demands of AI and hyperscale systems. We have collaborated with Meta on a new disaggregated power rack design, aimed at enhancing flexibility and scalability as we bring in AI infrastructure into our existing datacenter footprint. Each disaggregated power rack will feature 400-volt DC power that enables up to 35% more AI accelerators in each server rack, enabling dynamic power adjustments to meet the different demands of AI workloads. We are open sourcing these cooling and power rack specifications through the Open Compute Project so that the industry can benefit. 

Azure’s AI infrastructure builds on this innovation at the hardware and silicon layer to power some of the most groundbreaking AI advancements in the world, from revolutionary frontier models to large scale generative inferencing. In October, we announced the launch of the ND H200 V5 Virtual Machine (VM) series, which utilizes NVIDIA’s H200 GPUs with enhanced memory bandwidth. Our continuous software optimization efforts across these VMs means Azure delivers performance improvements generation over generation. Between NVIDIA H100 and H200 GPUs that performance improvement rate was twice that of the industry, demonstrated across industry benchmarking.

We are also excited to announce that Microsoft is bringing the NVIDIA Blackwell platform to the cloud. We are beginning to bring these systems online in preview, co-validating and co-optimizing with NIVIDIA and other AI leaders. Azure ND GB200 v6 will be a new AI optimized Virtual Machines series and combines the NVIDIA GB200 NVL 72 rack-scale design with state-of-the-art Quantum InfiniBand networking to connect tens of thousands of Blackwell GPUs to deliver AI supercomputing performance at scale. 

We are also sharing today our latest advancements in CPU-based supercomputing, the Azure HBv5 virtual machine. Powered by custom AMD EPYC^TM 9V64H processors only available on Azure, these VMs will be up to eight times faster than the latest bare-metal and cloud alternatives on a variety of HPC workloads, and up to 35 times faster than on-premises servers at the end of their lifecycle. These performance improvements are made possible by 7 TB/s of memory bandwidth from high bandwidth memory (HBM) and the most scalable AMD EPYC server platform to date. Customers can now sign up for the preview of HBv5 virtual machines, which will begin in 2025. 

Accelerating AI innovation through cloud migration and modernization 

To get the most from AI, organizations need to integrate data residing in their critical business applications. Migrating and modernizing these applications to the cloud helps enable that integration and paves the path to faster innovation while delivering improved performance and scalability. Choosing Azure means selecting a platform that natively supports all the mission-critical enterprise applications and data you need to fully leverage advanced technologies like AI. This includes your workloads on SAP, VMware, and Oracle, as well as open-source software and Linux.

For example, thousands of customers run their SAP ERP applications on Azure and we are bringing unique innovation to these organizations such as the integration between Microsoft Copilot and SAP’s AI assistant Joule. Companies like L’Oreal, Hilti, Unilever, and Zeiss have migrated their mission-critical SAP workloads to Azure so they can innovate faster. And since the launch of Azure VMware Solution, we’ve been working to support customers globally with geographic expansion. Azure VMware Solution is now available in 33 regions, with support for VMware VCF portable subscriptions. 

We are also continually improving Oracle Database@Azure to better support the mission-critical Oracle workloads of our enterprise customers. Customers like The Craneware Group and Vodafone have adopted Oracle Database@Azure to benefit from its high performance and low latency, which allows them to focus on streamlining their operations and to get access to advanced security, data governance, and AI capabilities in the Microsoft Cloud. We’re announcing today Microsoft Purview supports Oracle Database@Azure for comprehensive data governance and compliance capabilities that organizations can use to manage, secure, and track data across Oracle workloads.  

Additionally, Oracle and Microsoft plan to provide Oracle Exadata Database Service on Exascale Infrastructure in Oracle Database@Azure for hyper-elastic scaling and pay-per-use economics. Additionally, we’ve expanded the availability of Oracle Database@Azure to a total of nine regions and enhanced Microsoft Fabric integration with Open Mirroring capabilities. 

To make it easier to migrate and modernize your applications to the cloud, starting today, you can assess your application’s readiness for Azure using Azure Migrate. The new application aware method provides technical and business insights to help you migrate entire application with all dependencies as one.

Optimizing your operations with an adaptive cloud for business growth

Azure’s multicloud and hybrid approach, or adaptive cloud, integrates separate teams, distributed locations, and diverse systems into a single model for operations, security, applications, and data. This allows organizations to utilize cloud-native and AI technologies to operate across hybrid, multicloud, edge, and IoT environments. Azure Arc plays an important role in this approach by extending Azure services to any infrastructure and supporting organizations with managing their workloads and operating across different environments. Azure Arc now has over 39,000 customers across every industry, including La Liga, Coles, and The World Bank.

We’re excited to introduce Azure Local, a new, cloud-connected, hybrid infrastructure offering provisioned and managed in Azure. Azure Local brings together Azure Stack capabilities into one unified platform. Powered by Azure Arc, Azure Local can run containers, servers and Azure Virtual Desktop on Microsoft-validated hardware from Dell, HPE, Lenovo, and more. This unlocks new possibilities to meet custom latency, near real-time data processing, and compliance requirements. Azure Local comes with enhanced default security settings to protect your data and flexible configuration options, like GPU-enabled servers for AI inferencing.

We recently announced the general availability of Windows Server 2025, with new features that include easier upgrades, advanced security, and capabilities that enable AI and machine learning. Additionally, Windows Server 2025 is previewing a hotpatching subscription option enabled by Azure Arc that will allow organizations to install updates with fewer restarts—a major time saver.

We’re also announcing the preview of SQL Server 2025, an enterprise AI-ready database platform that leverages Azure Arc to deliver cloud agility anywhere. This new version continues its industry-leading security and performance and has AI built-in, simplifying AI application development and retrieval augmented generation (RAG) patterns with secure, performant, and easy-to-use vector support. With Azure Arc, SQL Server 2025 offers cloud capabilities to help customers better manage, secure, and govern SQL estate at scale across on-premises and cloud.

Transform with Azure infrastructure to achieve cloud and AI success

Successful transformation with AI starts with a powerful, secure, and adaptive infrastructure strategy. And as you evolve, you need a cloud platform that adapts and scales with your needs. Azure is that platform, providing the optimal environment for integrating your applications and data so that you can start innovating with AI. As you design, deploy, and manage your environment and workloads on Azure, you have access to best practices and industry-leading technical guidance to help you accelerate your AI adoption and achieve your business goals. 

Jumpstart your AI journey at Microsoft Ignite

Key sessions at Microsoft Ignite

• BRK214: Operate infrastructure across distributed locations with Azure Arc.
• BRK 224: Specialized silicon and hardware optimized for cloud infrastructure.
• BRK239: Application centric assessment and guided migration with Azure Migrate.

Discover more announcements at Microsoft Ignite

• Discover the next wave of Azure innovation: Azure AI Foundry, intelligent data, and more.
• Accelerate your app innovation with an AI-powered platform.

Resources for AI transformation

• Build your Azure skills with Microsoft Learn.
• Learn how to improve resiliency on Azure.
• Adopt AI best practices with Azure Essentials.

The post Scale your AI transformation with a powerful, secure, and adaptive cloud infrastructure appeared first on Microsoft AI Blogs.

This week we’re announcing news and advancements to showcase our commitment to your success in this dynamic era. Let’s get started.

Introducing Microsoft Azure AI Foundry: A unified platform to design, customize, and manage AI solutions 

Every new generation of applications brings with it a changing set of needs, and just as web, mobile, and cloud technologies have driven the rise of new application platforms, AI is changing how we build, run, govern, and optimize applications. According to a Deloitte report, nearly 70% of organizations have moved 30% or fewer of their Generative AI experiments into production—so there is a lot of innovation and results ready to be unlocked. Business leaders are looking to reduce the time and cost of bringing their AI solutions to market while continuing to monitor, measure, and evaluate their performance and ROI.

This is why we’re excited to unveil Azure AI Foundry today as a unified application platform for your entire organization in the age of AI. Azure AI Foundry helps bridge the gap between cutting-edge AI technologies and practical business applications, empowering organizations to harness the full potential of AI efficiently and effectively.

We’re unifying the AI toolchain in a new Azure AI Foundry SDK that makes Azure AI capabilities accessible from familiar tools, like GitHub, Visual Studio, and Copilot Studio. We’ll also evolve Azure AI Studio into an enterprise-grade management console and portal for Azure AI Foundry.

Azure AI Foundry is designed to empower your entire organization—developers, AI engineers, and IT professionals—to customize, host, run, and manage AI solutions with greater ease and confidence. This unified approach simplifies the development and management process, helping all stakeholders focus on driving innovation and achieving strategic goals.

For developers, Azure AI Foundry delivers a streamlined process to swiftly adapt the latest AI advancements and focus on delivering impactful applications. Developers will also find an enhanced experience, with access to all existing Azure AI Services, and tooling along with new capabilities we’re announcing today. 

For IT professionals and business leaders, adopting AI technologies raises important questions about measurability, ROI, and ongoing optimization. There’s a pressing need for tools that provide clear insights into AI initiatives and their impact on the business. Azure AI Foundry enables leaders to measure their effectiveness, align them with organizational goals, and more confidently invest in AI technologies.

To help you scale AI adoption in your organization, we’re introducing comprehensive guidance for AI adoption and architecture within Azure Essentials so you are equipped to successfully navigate the pace of AI innovation. Azure Essentials gives you access to Microsoft’s best practices, product experiences, reference architectures, skilling, and resources into a single destination. It’s a great way to benefit from all we’ve learned and the approach you’ll find aligns directly with how to make the most of Azure AI Foundry.

In a market flooded with disparate technologies and choices, we created Azure AI Foundry to thoughtfully address diverse needs across an organization in the pursuit of AI transformation. It’s not just about providing advanced tools, though we have those, too. It’s about fostering collaboration and alignment between technical teams and business strategy.

Now, let’s dive into additional updates designed to enhance the overall experience and efficiency throughout the AI development process, no matter your role.

Introducing Azure AI Agent Service to automate business processes and help you focus on your most strategic work  

AI agents have huge potential to autonomously perform routine tasks, boosting productivity and efficiency, all while keeping you at the center. We’re introducing Azure AI Agent Service to help developers orchestrate, deploy, and scale enterprise AI-powered apps to automate business processes. These intelligent agents handle tasks independently, involving human users for final review or action, ensuring your team can focus on your most strategic initiatives. 

A standout feature of Agent Service is the ability to easily connect enterprise data for grounding, including Microsoft SharePoint and Microsoft Fabric, and tools integration to automate actions. With features like bring your own storage (BYOS) and private networking, it ensures data privacy and compliance, helping organizations protect their sensitive data. This allows your business to leverage existing data and systems to create powerful and secure agentic workflows.

Enhanced observability and collaboration with a new management center experience

To support the development and governance of generative AI apps and fine-tuned models, today we’re unveiling a new management center experience right in Azure AI Foundry portal. This feature brings essential subscription information, such as connected resources, access privileges, and quota usage, into one pane of glass. This can save development teams valuable time and facilitate easier security and compliance workflows throughout the entire AI lifecycle. 

Expanding our AI model catalog with more specialized solutions and customization options 

From generating realistic images to crafting human-like text, AI models have immense potential, but to truly harness their power, you need customized solutions. Our AI model catalog is designed to provide choice and flexibility and ensure your organization and developers have what they need to explore what AI models can do to advance your business priorities. Along with the latest from OpenAI and Microsoft’s Phi family of small language models, our model catalog includes open and frontier models. We offer more than 1,800 options and we’re expanding to offer even more tailored and specialized task and industry-specific models.  

We’re announcing additions that include models from Bria, now in preview, and NTT DATA, now generally available. Industry-specific models from Bayer, Sight Machine, Rockwell Automation, Saifr/Fidelity Labs, and Paige.ai are also available today in preview for specialized solutions in healthcare, manufacturing, finance, and more.

We’ve seen Azure OpenAI Service consumption more than double over the past six months, making it clear customers are excited about this partnership and what it offers_². We look forward to bringing more innovation to you with our partners at OpenAI, starting with new fine-tuning capabilities like vision fine-tuning and distillation workflows which allow a smaller model like GPT-4o mini to replicate the behavior of a larger model such as GPT-4o with fine-tuning, capturing its essential knowledge and bringing new efficiencies.

Along with unparalleled model choice, we equip you with essential tools like benchmarking, evaluation, and a unified model inference API so you can explore, compare, and select the best model for your needs without changing a line of code. This means you can easily swap out models without the need to recode as new advancements emerge, ensuring you’re never locked into a single model.

New collaborations to streamline model customization process for more tailored AI solutions

We’re announcing collaborations with Weights & Biases, Gretel, Scale AI, and Statsig to accelerate end-to-end AI model customization. These collaborations cover everything from data preparation and generation to training, evaluation, and experimentation with fine-tuned models. 

The integration of Weights & Biases with Azure will provide a comprehensive suite of tools for tracking, evaluating, and optimizing a wide range of models in Azure OpenAI Service, including GPT-4, GPT-4o, and GPT-4o-mini. This ensures organizations can build AI applications that are not only powerful, but also specifically tailored to their business needs.  

The collaborations with Gretel and Scale AI aim to help developers remove data bottlenecks and make data AI-ready for training. With Gretel Azure OpenAI Service integration, you can upload Gretel generated data to Azure OpenAI Service to fine-tune AI models and achieve better performance in domain-specific use cases. Our Scale AI partnership will also help developers with expert feedback, data preparation, and support for fine-tuning and training models. 

The Statsig collaboration enables you to dynamically configure AI applications and run powerful experiments to optimize your models and applications in production. 

More RAG performance with Azure AI Search 

Retrieval-augmented generation, or RAG, is important for ensuring accurate, contextual responses and reliable information. Azure AI Search now features a generative query engine built for high performance (for select regions). Query rewriting, available in preview, transforms and creates multiple variations of a query using an SLM-trained (Small Language Model) on data typically seen in generative AI applications. In addition, semantic ranker has a new reranking model, trained with insights gathered from customer feedback and industry market trends from over a year.  

With these improvements, we’ve shattered our own performance records—our new query engine delivers up to 12.5% better relevance, and is up to 2.3 times faster than last year’s stack. Customers can already take advantage of better RAG performance today, without having to configure or customize any settings. That means improved RAG performance is delivered out of the box, with all the hard work done for you.

Effortless RAG with GitHub models and Azure AI Search—just add data 

Azure AI Search will soon power RAG in GitHub Models, offering you the same easy access glide path to bring RAG to your developer environment in GitHub Codespaces. In just a few clicks, you can experiment with RAG and your data. Directly from the playground, simply upload your data (just drag and drop), and a free Azure AI Search index will automatically be provisioned. 

Once you’re ready to build, copy/paste a code snippet into your dev environment to add more data or try out more advanced retrieval methods offered by Azure AI Search. 

This means you can unlock a full-featured knowledge retrieval system for free, without ever leaving your code. Just add data.

Advanced vector search and RAG capabilities now integrated into Azure Databases  

Vector search and RAG are transforming AI application development by enabling more intelligent, context-aware systems. Azure Databases now integrates innovations from Microsoft Research—DiskANN and GraphRAG—to provide cost-effective, scalable solutions for these technologies.

GraphRAG, available in preview in Azure Database for PostgreSQL, offers advanced RAG capabilities, enhancing large language models (LLMs) with your private PostgreSQL datasets. These integrations help empower developers, IT pros, and AI engineers alike, to build the next generation of AI applications efficiently and at cloud scale. 

DiskANN, a state-of-the-art suite of algorithms for low-latency, highly scalable vector search, is now generally available in Azure Cosmos DB and in preview for Azure Database for PostgreSQL. It’s also combined with full-text search to power Azure Cosmos DB hybrid search, currently in preview.  

Equipping you with responsible AI tooling to help ensure safety and compliance  

We continue to back up our Trustworthy AI commitments with tools you can use, and today we’re announcing two more: AI reports and risk and safety evaluations for images. These updates help ensure your AI applications are not only innovative, but safe and compliant. AI reports enable developers to document and share the use case, model card, and evaluation results for fine-tuned models and generative AI applications. Compliance teams can easily review, export, approve, and audit these reports across their organization, streamlining AI asset tracking, and governance. 

We are also excited to announce new collaborations with Credo AI and Saidot to support customers’ end-to-end AI governance. Credo AI pioneered a responsible AI platform enabling comprehensive AI governance, oversight, and accountability. Saidot’s AI Governance Platform helps enterprises and governments manage risk and compliance of their AI-powered systems with efficiency and high quality. By integrating the best of Azure AI with innovative AI governance solutions, we hope to provide our customers with choice and foster greater cross-functional collaboration to align AI solutions with their own principles and regulatory requirements.   

Transform unstructured data into multimodal app experiences with Azure AI Content Understanding  

AI capabilities are quickly advancing and expanding beyond traditional text to better reflect content and input that matches our real world. We’re introducing Azure AI Content Understanding to make it faster, easier, and more cost-effective to build multimodal applications with text, audio, images, and video. Now in preview, this service uses generative AI to extract information into customizable structured outputs.  

Pre-built templates offer a streamlined workflow and opportunities to customize outputs for a wide range of use-cases—call center analytics, marketing automation, content search, and more. And, by processing data from multiple modalities at the same time, this service can help developers reduce the complexities of building AI applications while keeping security and accuracy at the center.

Advancing the developer experience with new AI capabilities and a personal guide to Azure 

As a company of developers, we always keep the developer community top of mind with every advancement we bring to Azure. We strive to offer you the latest tech and best practices that boost impact, fit the way you work, and improve the development experience as you build AI apps. 

We’re introducing two offerings in Azure Container Apps to help transform how AI app developers work: serverless GPUs, now in preview, and dynamic sessions, available now.  

With Azure Container Apps serverless GPUs—you can seamlessly run your customer AI models on NVIDIA GPUs. This feature provides serverless scaling with optimized cold start, per-second billing, with built-in scale down to zero when not in use, and reduced operational overhead. It supports easy real-time inferencing for custom AI models, allowing you to focus on your core AI code without worrying about managing GPU infrastructure. 

Azure Container Apps dynamic sessions—offer fast access to secure sandboxed environments. These sessions are perfect for running code that requires strong isolation, such as large language model (LLM) generated code or extending and customizing software as a service (SaaS) apps. You can mitigate risks, leverage serverless scale, and reduce operational overhead in a cost-efficient manner. Dynamic sessions come with a Python code interpreter pre-installed with popular libraries, making it easy to execute common code scenarios without managing infrastructure or containers. 

These new offerings are part of our ongoing work to put Azure’s comprehensive dev capabilities within easy reach. They come right on the heels of announcing the preview of GitHub Copilot for Azure, which is like having a personal guide to Azure. By integrating with tools you already use, GitHub Copilot for Azure enhances Copilot Chat capabilities to help manage resources and deploy applications and the “@azure” command provides personalized guidance without ever leaving the code.

Updates to our intelligent data platform and Microsoft Fabric help propel AI innovation through your unique data  

While AI capabilities are remarkable, even the most powerful models don’t know your specific business. Unlocking AI’s full value requires integrating your organization’s unique data—a modern, fully integrated data estate forms the bedrock of innovation. Fast and reliable access to high-quality data becomes critical as AI applications handle increasing volumes of data requests. This is why we believe in the power of our Intelligent Data Platform as an ideal data and AI foundation for every organization’s success, today and tomorrow.   

To help meet the need for high-quality data in AI applications, we’re pleased to announce that Azure Managed Redis is now in preview. In-memory caching helps boost app performance by reducing latency and offloading traffic from databases. This new service offers up to 99.999% availability³ and comprehensive support—all while being more cost-effective than the current offering. The best part? Azure Managed Redis goes beyond standard caching to optimize AI app performance and works with Azure services. The latest Redis innovations, including advanced search capabilities and support for a variety of data types, are accessible across all service tiers_⁴.  

Just about a year ago we introduced Microsoft Fabric as our end-to-end data analytics platform that brought together all the data and analytics tools that organizations needed to empower data and business professionals alike to unlock the potential of their data and lay the foundation for the era of AI.​ Be sure to check out Arun Ulag’s blog today to learn all about the new Fabric features and integrations we’re announcing this week to help prepare your organization for the era of AI with a single, AI-powered data platform—including the introduction of Fabric Databases.  

How will you create the future? 

As AI transforms industries and unveils new opportunities, we’re committed to providing practical solutions and powerful innovation to empower you to thrive in this evolving landscape. Everything we’re delivering today reflects our dedication to meeting the real-world needs of both developers and business leaders, ensuring every person and every organization can harness the transformative power of AI.  

With these tools at your disposal, I’m excited to see how you’ll shape the future. Have a great Ignite week! 

Make the most of Ignite 2024 

• Read our Ignite announcements on how to scale your AI transformation with a powerful, secure, and adaptive cloud infrastructure.

• Tune in for can’t-miss sessions at Ignite 2024:
  • Microsoft Ignite Keynote with Satya Nadella and Microsoft leaders.
  • Empowering your AI Ambitions with Azure.
  • Azure AI unlocking the AI revolution.
  • Microsoft Fabric What’s new and what’s next.
  • Trustworthy AI Future trends and best practices.

• Do a deep dive on all the product innovation rolling out this week over on Tech Community. 

• Check out the Master GenAI Model Selection, Evaluation, and Multimodal Integration with Azure AI Foundry Plan on Microsoft Learn. With this outcome-based skilling plan, you can explore the process of selecting and applying GenAI models, how to benchmark and apply multimodal models, and complete evaluations to ensure performance and safety.

• Find out how we’re making it easy to discover, buy, deploy, and manage cloud and AI solutions via the Microsoft commercial marketplace, and get connected to vetted partner solutions today. 

• We’re here to help. Check out Azure Essentials guidance for a comprehensive framework to navigate this complex landscape, and ensure your AI initiatives not only succeed but become catalysts for innovation and growth.

References

^{1. Four futures of generative AI in the enterprise: Scenario planning for strategic resilience and adaptability.}

^{2. Microsoft Fiscal Year 2025 First Quarter Earnings Conference Call.}

^{2. Up to 99.999% uptime SLA is planned for the General Availability of Azure Managed Redis.}

^{3. B0, B1 SKU options, and Flash Optimized tier, may not have access to all features and capabilities.}

The post The next wave of Azure innovation: Azure AI Foundry, intelligent data, and more appeared first on Microsoft AI Blogs.