In the heart of the American Midwest, a modern marvel is rising. We’re in the final phases of building Fairwater, the world’s most powerful AI datacenter in Mount Pleasant, Wisconsin — part of a region forged by generations of hard work and ingenuity. This facility is more than a technological feat. It’s a promise to grow responsibly, invest deeply, and create opportunities for Wisconsin and for the nation.

We are on track to complete construction and bring this AI datacenter online in early 2026, fulfilling our initial $3.3 billion investment pledge. We’ve already begun hiring full-time employees to support its operation.

And we’re not done. Today, we’re committing an additional $4 billion to be spent in the next three years to build our second datacenter of similar size and scale — bringing our total investment in Wisconsin to more than $7 billion.

Building the most advanced AI datacenter in the world

Engineered to train the next decade of artificial intelligence, our Mount Pleasant facility will house hundreds of thousands of the world’s most powerful NVIDIA GPUs, operating in seamless clusters connected by enough fiber to wrap the planet four times over. These processors will handle training for frontier AI models — delivering ten times the performance of today’s fastest supercomputers.

This datacenter is designed to help AI researchers and engineers build the world’s most advanced models, test ideas faster, and do it all more efficiently. It’s not just about running AI — it’s about creating it. This is where the next generation of AI will be trained, setting the stage for breakthroughs that will shape the future. New discoveries in medicine, science, and other critical fields will start right here, with the models we train in Wisconsin.

But what does that mean for the average Wisconsinite? It means new jobs, new skills, and new opportunities — right here at home. From union construction roles to long-term careers in operations and IT, this facility is creating pathways for Wisconsinites to be part of the future of technology. It means students at Gateway Technical College can train for high-demand roles through Wisconsin’s first Datacenter Academy. It means local companies — from manufacturers to startups — can partner with Microsoft engineers to turn AI ideas into real solutions.

As someone who spent almost five years as a kid going to school and delivering the morning newspaper by bicycle in Mount Pleasant, this moment means more than just personal nostalgia. It shows that Wisconsin has not just a longstanding and proud industrial past — it’s helping define the future of American innovation.

Designed and built for sustainability

What makes this datacenter distinctive isn’t just its scale or speed — it’s how thoughtfully it’s built, with both people and the planet in mind. From day one, sustainability has been central to its design.

More than 90 percent of the facility will rely on a state-of-the-art closed-loop liquid cooling system, filled during construction and recirculated continuously. The remaining portion of the facility will use outside air for cooling, switching to water only on the hottest days, minimizing environmental impact and maximizing operational efficiency. The result is a technological milestone — a datacenter with enough fiber cable to circle the Earth four times, yet its annual water use is modest, requiring roughly the amount of water a typical restaurant uses annually or what an 18-hole golf course consumes weekly in peak summer.

We appreciate that energy prices are increasing across the country and have worked hard to ensure our datacenter will not drive-up costs for our neighbors. That’s why we’re pre-paying for the energy and electrical infrastructure that we’ll use — ensuring prices remain stable and protecting consumers from future cost increases because of our datacenter. We will match every kilowatt hour we consume that comes from a fossil fuel source one for one with carbon-free energy we put back onto the grid. This includes a new 250 MW solar project in Portage County that is under construction to support this commitment. And our partnership with WE Energies ensures we will continually explore and add energy transmission, generation, and usage — under transparent tariffs that support grid reliability.

Our commitment to be a good neighbor in Racine County is one of the reasons we joined forces with Root-Pike Watershed Initiative Network (WIN) to restore prairie and wetland habitats in Racine and Kenosha counties — funding 20 ecological restoration projects, including Cliffside Park along the shores of Lake Michigan, Lamparek Creek in Mount Pleasant, Kirkorian Park in Village of Sturtevant, and Shagbark Restoration Area in Kenosha.

Investing in people

This facility is a catalyst for economic opportunity. At its peak, we have employed more than 3,000 construction workers during daily peak activity, including electricians, plumbers and pipefitters, carpenters, structural iron and steel workers, concrete workers, and Earth movers. Once our first datacenter is fully operational, we will employ around 500 full-time employees, with that number growing to around 800 once the second datacenter is complete.

We are committed to Wisconsin and other communities that host our cutting-edge datacenters, creating hubs for AI innovation where local businesses, nonprofits, students, and workers can all benefit from the growing AI economy. That’s why we partnered in Racine with Gateway Technical College to launch Wisconsin’s first Datacenter Academy, training more than 1,000 students in five years for high-demand datacenter roles. Across the state, Microsoft and more than 40 partners like the United Way, the University of Wisconsin, and the Wisconsin Technical College System have worked together and with Gener8tor to train 114,000 people in AI — including 1,400 people living in Racine County.

We’ve also sponsored and helped to open the nation’s first manufacturing-focused AI Co-Innovation Lab on the campus of the University of Wisconsin-Milwaukee. It results from a partnership with UW-Milwaukee, WEDC, Gateway Technical College, and TitletownTech (a partnership between Microsoft and the Green Bay Packers). The lab has already helped 23 Wisconsin companies — including Regal Rexnord, Renaissant, and BW Converting — turn AI ideas into real solutions.  A few of these companies are headquartered right in Racine County, like Wiscon Products, a family-owned precision machining company founded in 1945.

Finally, to help more Wisconsinites access the world’s most advanced technology being built in their home state, we’ve expanded broadband access to more than 9,300 rural residents and delivered next-generation service to 1,200 homes and businesses in Sturtevant — bringing faster, more reliable internet to the places where people live, work, and learn.

***

Mount Pleasant isn’t just becoming a hub for AI — it’s becoming a blueprint for how innovation can serve everyone. We’re not just investing in an AI datacenter; we are investing in a community. And we are investing in a powerful idea: that innovation is for everyone, and that we can build the future together — with care for people, place, and planet.

The post Made in Wisconsin: The world’s most powerful AI datacenter appeared first on Microsoft On the Issues.

This week we have introduced a wave of purpose-built datacenters and infrastructure investments we are making around the world to support the global adoption of cutting-edge AI workloads and cloud services.

Today in Wisconsin we introduced Fairwater, our newest US AI datacenter, the largest and most sophisticated AI factory we’ve built yet. In addition to our Fairwater datacenter in Wisconsin, we also have multiple identical Fairwater datacenters under construction in other locations across the US.

In Narvik, Norway, Microsoft announced plans with nScale and Aker JV to develop a new hyperscale AI datacenter.

In Loughton, UK, we announced a partnership with nScale to build the UK’s largest supercomputer to support services in the UK.

These AI datacenters are significant capital projects, representing tens of billions of dollars of investments and hundreds of thousands of cutting-edge AI chips, and will seamlessly connect with our global Microsoft Cloud of over 400 datacenters in 70 regions around the world. Through innovation that can enable us to link these AI datacenters in a distributed network, we multiply the efficiency and compute in an exponential way to further democratize access to AI services globally.

So what is an AI datacenter?

The AI datacenter: the new factory of the AI era

An AI datacenter is a unique, purpose-built facility designed specifically for AI training as well as running large-scale artificial intelligence models and applications. Microsoft’s AI datacenters power OpenAI, Microsoft AI, our Copilot capabilities and many more leading AI workloads.

The new Fairwater AI datacenter in Wisconsin stands as a remarkable feat of engineering, covering 315 acres and housing three massive buildings with a combined 1.2 million square feet under roofs. Constructing this facility required 46.6 miles of deep foundation piles, 26.5 million pounds of structural steel, 120 miles of medium-voltage underground cable and 72.6 miles of mechanical piping.

Unlike typical cloud datacenters, which are optimized to run many smaller, independent workloads such as hosting websites, email or business applications, this datacenter is built to work as one massive AI supercomputer using a single flat networking interconnecting hundreds of thousands of the latest NVIDIA GPUs. In fact, it will deliver 10X the performance of the world’s fastest supercomputer today, enabling AI training and inference workloads at a level never before seen.

The role of our AI datacenters – powering frontier AI

Effective AI models rely on thousands of computers working together, powered by GPUs, or specialized AI accelerators, to process massive concurrent mathematical computations. They’re interconnected with extremely fast networks so they can share results instantly, and all of this is supported by enormous storage systems that hold the data (like text, images or video) broken down into tokens, the small units of information the AI learns from. The goal is to keep these chips busy all the time, because if the data or the network can’t keep up, everything slows down.

The AI training itself is a cycle: the AI processes tokens in sequence, makes predictions about the next one, checks them against the right answers and adjusts itself. This repeats trillions of times until the system gets better at whatever it’s being trained to do. Think of it like a professional football team’s practice. Each GPU is a player running a drill, the tokens are the plays being executed step by step, and the network is the coaching staff, shouting instructions and keeping everyone in sync. The team repeats plays over and over, correcting mistakes until they can execute them perfectly. By the end, the AI model, like the team, has mastered its strategy and is ready to perform under real game conditions.

AI infrastructure at frontier scale

Purpose-built infrastructure is critical to being able to power AI efficiently. To compute the token math at this trillion-parameter scale of leading AI models, the core of the AI datacenter is made up of dedicated AI accelerators (such as GPUs) mounted on server boards alongside CPUs, memory and storage. A single server hosts multiple GPU accelerators, connected for high-bandwidth communication. These servers are then installed into a rack, with top-of-rack (ToR) switches providing low-latency networking between them. Every rack in the datacenter is interconnected, creating a tightly coupled cluster. From the outside, this architecture looks like many independent servers, but at scale it functions as a single supercomputer where hundreds of thousands of accelerators can train a single model in parallel.

This datacenter runs a single, massive cluster of interconnected NVIDIA GB200 servers and millions of compute cores and exabytes of storage, all engineered for the most demanding AI workloads. Azure was the first cloud provider to bring online the NVIDIA GB200 server, rack and full datacenter clusters. Each rack packs 72 NVIDIA Blackwell GPUs, tied together in a single NVLink domain that delivers 1.8 terabytes of GPU-to-GPU bandwidth and gives every GPU access to 14 terabytes of pooled memory. Rather than behaving like dozens of separate chips, the rack operates as a single, giant accelerator, capable of processing an astonishing 865,000 tokens per second, the highest throughput of any cloud platform available today. The Norway and UK AI datacenters will use similar clusters, and take advantage of NVIDIAs next AI chip design (GB300) which offers even more pooled memory per rack.

The challenge in establishing supercomputing scale, particularly as AI training requirements continue to require breakthrough scales of computing, is getting the networking topology just right. To ensure low latency communication across multiple layers in a cloud environment, Microsoft needed to extend performance beyond a single rack. For the latest NVIDIA GB200 and GB300 deployments globally, at the rack level these GPUs communicate over NVLink and NVSwitch at terabytes per second, collapsing memory and bandwidth barriers. Then to connect across multiple racks into a pod, Azure uses both InfiniBand and Ethernet fabrics that deliver 800 Gbps, in a full fat tree non-blocking architecture to ensure that every GPU can talk to every other GPU at full line rate without congestion. And across the datacenter, multiple pods of racks are interconnected to reduce hop counts and enable tens of thousands of GPUs to function as one global-scale supercomputer.

When laid out in a traditional datacenter hallway, physical distance between racks introduces latency into the system. To address this, the racks in the Wisconsin AI datacenter are laid out in a two-story datacenter configuration, so in addition to racks networked to adjacent racks, they are networked to additional racks above or below them.

This layered approach sets Azure apart. Microsoft Azure was not just the first cloud to bring GB200 online at rack and datacenter scale; we’re doing it at massive scale with customers today. By co-engineering the full stack with the best from our industry partners coupled with our own purpose-built systems, Microsoft has built the most powerful, tightly coupled AI supercomputer in the world, purpose-built for frontier models.

Addressing the environmental impact: closed loop liquid cooling at facility scale

Traditional air cooling can’t handle the density of modern AI hardware. Our datacenters use advanced liquid cooling systems — integrated pipes circulate cold liquid directly into servers, extracting heat efficiently. The closed-loop recirculation ensures zero water waste, with water only needed to fill up once and then it is continually reused.

By designing purpose-built AI datacenters, we were able to build liquid cooling infrastructure into the facility directly to get us more rack-density in the datacenter. Fairwater is supported by the second largest water-cooled chiller plant on the planet and will continuously circulate water in its closed loop cooling system. The hot water is then piped out to the cooling “fins” on each side of the datacenter, where 172 20-foot fans chill and recirculate the water back to the datacenter. This system keeps the AI datacenter running efficiently, even at peak loads.

Over 90% of our datacenter capacity uses this system, requiring water only once during construction and continually reusing it with no evaporation losses. The remaining 10% of traditional servers use outdoor air for cooling, switching to water only during the hottest days, a design that dramatically reduces water usage compared to traditional datacenters.

We’re also using liquid cooling to support AI workloads in many of our existing datacenters; this liquid cooling is accomplished with Heat Exchanger Units (HXUs) that also operate with zero-operational water use.

Storage and compute: Built for AI velocity

Modern datacenters can contain exabytes of storage and millions of CPU compute scores. To support the AI infrastructure cluster, an entirely separate datacenter infrastructure is needed to store and process the data used and generated by the AI cluster. To give you an example of the scale — the Wisconsin AI datacenter’s storage systems are five football fields in length!

We reengineered Azure storage for the most demanding AI workloads, across these massive datacenter deployments for true supercomputing scale. Each Azure Blob Storage account can sustain over 2 million read/write transactions per second, and with millions of accounts available, we can elastically scale to meet virtually any data requirement.

Behind this capability is a fundamentally rearchitected storage foundation that aggregates capacity and bandwidth across thousands of storage nodes and hundreds of thousands of drives. This enables scale to exabyte scale storage, eliminating the need for manual sharding and simplifying operations for even the largest AI and analytics workloads.

Key innovations such as BlobFuse2 deliver high-throughput, low-latency access for GPU node-local training, ensuring that compute resources are never idle and that massive AI training datasets are always available when needed. Multiprotocol support allows seamless integration with diverse data pipelines, while deep integration with analytics engines and AI tools accelerates data preparation and deployment.

Automatic scaling dynamically allocates resources as demand grows, combined with advanced security, resiliency and cost-effective tiered storage, Azure’s storage platform sets the pace for next-generation workloads, delivering the performance, scalability and reliability required.

AI WAN: Connecting multiple datacenters for an even larger AI supercomputer

These new AI datacenters are part of a global network of Azure AI datacenters, interconnected via our Wide Area Network (WAN). This isn’t just about one building, it’s about a distributed, resilient and scalable system that operates as a single, powerful AI machine. Our AI WAN is built with growth capabilities in AI-native bandwidth scales to enable large-scale distributed training across multiple, geographically diverse Azure regions, thus allowing customers to harness the power of a giant AI supercomputer.

This is a fundamental shift in how we think about AI supercomputers. Instead of being limited by the walls of a single facility, we’re building a distributed system where compute, storage and networking resources are seamlessly pooled and orchestrated across datacenter regions. This means greater resiliency, scalability and flexibility for customers.

Bringing it all together

To meet the critical needs of the largest AI challenges, we needed to redesign every layer of our cloud infrastructure stack. This isn’t just about isolated breakthroughs, but composing multiple new approaches across silicon, servers, networks and datacenters, leading to advancements where software and hardware are optimized as one purpose-built system.

Microsoft’s Wisconsin datacenter will play a critical role in the future of AI, built on real technology, real investment and real community impact. As we connect this facility with other regional datacenters, and as every layer of our infrastructure is harmonized as a complete system, we’re unleashing a new era of cloud-powered intelligence, secure, adaptive and ready for what’s next.

To learn more about Microsoft’s datacenter innovations, check out the virtual datacenter tour at datacenters.microsoft.com.

Scott Guthrie is responsible for hyperscale cloud computing solutions and services including Azure, Microsoft’s cloud computing platform, generative AI solutions, data platforms and information and cybersecurity. These platforms and services help organizations worldwide solve urgent challenges and drive long-term transformation.

The post Inside the world’s most powerful AI datacenter appeared first on The Official Microsoft Blog.

---

We’re excited to share that Docker Model Runner is now generally available (GA)! In April 2025, Docker introduced the first Beta release of Docker Model Runner, making it easy to manage, run, and distribute local AI models (specifically LLMs). Though only a short time has passed since then, the product has evolved rapidly, with continuous enhancements driving the product to a reliable level of maturity and stability.

This blog post takes a look back at the most important and widely appreciated capabilities Docker Model Runner brings to developers, and looks ahead to share what they can expect in the near future.

What is Docker Model Runner?

Docker Model Runner (DMR) is built for developers first, making it easy to pull, run, and distribute large language models (LLMs) directly from Docker Hub (in an OCI-compliant format) or HuggingFace (if models are available in the GGUF format, in which case they will be packaged as OCI Artifacts on-the-fly by the HuggingFace backend).

Tightly integrated with Docker Desktop and Docker Engine, DMR lets you serve models through OpenAI-compatible APIs, package GGUF files as OCI artifacts, and interact with them using either the command line, a graphical interface, or developer-friendly (REST) APIs.

Whether you’re creating generative AI applications, experimenting with machine learning workflows, or embedding AI into your software development lifecycle, Docker Model Runner delivers a consistent, secure, and efficient way to work with AI models locally.

Check the official documentation to learn more about Docker Model Runner and its capabilities.

Why Docker Model Runner?

Docker Model Runner makes it easier for developers to experiment and build AI application, including agentic apps, using the same Docker commands and workflows they already use every day. No need to learn a new tool!

Unlike many new AI tools that introduce complexity or require additional approvals, Docker Model Runner fits cleanly into existing enterprise infrastructure. It runs within your current security and compliance boundaries, so teams don’t have to jump through hoops to adopt it.

Model Runner supports OCI-packaged models, allowing you to store and distribute models through any OCI-compatible registry, including Docker Hub. And for teams using Docker Hub, enterprise features like Registry Access Management (RAM) provide policy-based access controls to help enforce guardrails at scale.

11 Docker Model Runner Features Developers Love Most

Below are the features that stand out the most and have been highly valued by the community.

1. Powered by llama.cpp 

Currently, DMR is built on top of llama.cpp, which we plan to continue supporting. At the same time, DMR is designed with flexibility in mind, and support for additional inference engines (such as MLX or vLLM) is under consideration for future releases.

2. GPU acceleration across macOS and Windows platforms 

Harness the full power of your hardware with GPU support: Apple Silicon on macOS, NVIDIA GPUs on Windows, and even ARM/Qualcomm acceleration — all seamlessly managed through Docker Desktop.

3. Native Linux support 

Run DMR on Linux with Docker CE, making it ideal for automation, CI/CD pipelines, and production workflows.

4. CLI and UI experience 

Use DMR from the Docker CLI (on both Docker Desktop and Docker CE) or through Docker Desktop’s UI. The UI provides guided onboarding to help even first-time AI developers start serving models smoothly, with automatic handling of available resources (RAM, GPU, etc.).

Figure 1: Docker Model Runner works both in Docker Desktop and the CLI, letting you run models locally with the same familiar Docker commands and workflows you already know

5. Flexible model distribution 

Pull and push models from Docker Hub in OCI format, or pull directly from HuggingFace repositories hosting models in GGUF format for maximum flexibility in sourcing and sharing models.

6. Open Source and free 

DMR is fully open source and free for everyone, lowering the barrier to entry for developers experimenting with or building on AI.

7. Secure and controlled 

DMR runs in an isolated, controlled environment that doesn’t interfere with the main system or user data (sandboxing). Developers and IT admins can fine-tune security and availability by enabling/disabling DMR or configuring options like host-side TCP support and CORS.

8. Configurable inference settings 

Developers can customize context length and llama.cpp runtime flags to fit their use cases, with more configuration options coming soon.

9. Debugging support 

Built-in request/response tracing and inspect capabilities make it easier to understand token usage and framework/library behaviors, helping developers debug and optimize their applications.

Figure 2: Built-in tracing and inspect tools in Docker Desktop make debugging easier, giving developers clear visibility into token usage and framework behavior

10. Integrated with the Docker ecosystem 

DMR works out of the box with Docker Compose and is fully integrated with other Docker products, such as Docker Offload (cloud offload service) and Testcontainers, extending its reach into both local and distributed workflows.

11. Up-to-date model catalog 

Access a curated catalog of the most popular and powerful AI models on Docker Hub. These models can be pulled for free and used across development, pipelines, staging, or even production environments.

Figure 3: Curated model catalog on Docker Hub, packaged as OCI Artifacts and ready to run

The road ahead

The future is bright for Docker Model Runner, and the recent GA version is only the first milestone. Below are some of the future enhancements that you should expect to be released soon.

Streamlined User Experience 

Our goal is to make DMR simple and intuitive for developers to use and debug. This includes richer response rendering in the chat-like interface within Docker Desktop and the CLI, multimodal support in the UI (already available through the API), integration with MCP tools, and enhanced debugging features, alongside expanded configuration options for greater flexibility. Last but not least, we aim to provide smoother and more seamless integration with third-party tools and solutions across the AI ecosystem.

Enhancements and better ability to execute 

We remain focused on continuously improving DMR’s performance and flexibility for running local models. Upcoming enhancements include support for the most widely used inference libraries and engines, advanced configuration options at the engine and model level, and the ability to deploy Model Runner independently from Docker Engine for production-grade use cases, along with many more improvements on the horizon.

Frictionless Onboarding 

We want first-time AI developers to start building their applications right away, and to do so with the right foundations. To achieve this, we plan to make onboarding into DMR even more seamless. This will include a guided, step-by-step experience to help developers get started quickly, paired with a set of sample applications built on DMR. These samples will highlight real-world use cases and best practices, providing a smooth entry point for experimenting with and adopting DMR in everyday workflows.

Staying on Top of Model Launch 

As we continue to enhance inference capabilities, we remain committed to maintaining a first-class catalog of AI models directly in Docker Hub, the leading registry for OCI artifacts, including models. Our goal is to ensure that new, relevant models are available in Docker Hub and runnable through DMR as soon as they are publicly released.

Conclusion

Docker Model Runner has come a long way in a short time, evolving from its Beta release into a mature and stable inference engine that’s now generally available. At its core, the mission has always been clear: make it simple, consistent, and secure for developers to pull, run, and serve AI models locally,. using familiar Docker CLI commands and tools they already love!

Now is the perfect time to get started. If you haven’t already, install Docker Desktop and try out Docker Model Runner today. Follow the official documentation to explore its capabilities and see for yourself how DMR can accelerate your journey into building AI-powered applications.

Learn more

• Read our quickstart guide to Docker Model Runner.
• Visit our Model Runner GitHub repo! Docker Model Runner is open-source, and we welcome collaboration and contributions from the community!
• Learn how Compose works with Model runner, making building AI apps and agents easier
• Learn how to build an AI tutor
• Explore how to use both local and remote models in hybrid AI workflows 
• Building AI agents made easy with Goose and Docker
• Using Model Runner on Hugging Face
• Powering AI generated testing with Docker Model Runner
• Build a GenAI App With Java Using Spring AI and Docker Model Runner
• Tool Calling with Local LLMs: A Practical Evaluation
• Behind the scenes: How we designed Docker Model Runner and what’s next
• Why Docker Chose OCI Artifacts for AI Model Packaging
• What’s new with Docker Model Runner 
• Publishing AI models to Docker Hub
• How to Build and Run a GenAI ChatBot from Scratch using Docker Model Runner 

Learn how to generate images in .NET 9 using OpenAI’s DALL·E models. Step-by-step guide with C# minimal API examples, streaming results, and best practices for production-ready AI image generation.

Learn how to use OpenAI in .NET 9 with C#. Explore examples for ChatGPT, embeddings, image generation, and streaming. Includes best practices for dependency injection, AOT, minimal APIs, and production deployment.