Microsoft Fairwater Atlanta
AI data centre in Atlanta with liquid cooling and no backup generators
Microsoft operates this AI-focused facility in Atlanta, Georgia, as part of its Fairwater design series. It uses closed-loop liquid cooling and connects to other sites via dedicated fibre to form a distributed supercomputer for large-scale model training.
- Uses closed-loop liquid cooling for high-density GPU racks
- No on-site generators or UPS systems due to grid reliability
- Connects to Wisconsin site via dedicated fibre for distributed AI
- Supports 140 kW per rack with Nvidia Blackwell GPUs
- Two-storey design minimises distance between hardware racks
What it is
This facility is the second site in Microsoft鈥檚 Fairwater series, designed specifically for artificial intelligence workloads. Located in Atlanta, Georgia, it forms part of a distributed super-cluster connected by dedicated fibre optics. The building spans two storeys to minimise physical distance between hardware racks. It houses hundreds of thousands of Nvidia GPUs, including GB200 and GB300 models, arranged in pods. Microsoft describes the setup as an AI superfactory that runs single complex jobs across millions of hardware pieces rather than hosting separate applications for multiple clients.
How it came about
Microsoft brought this Atlanta site online in 2026. It follows the first Fairwater facility located in Wisconsin. The company plans to link all future Fairwater sites into a single network acting as a virtual supercomputer. This approach allows the infrastructure to tackle large-scale challenges that exceed the capacity of any single building. The design represents a shift from traditional multi-tenant data centres to unified systems supporting massive AI model training tasks.
The building and the halls
The structure consists of two stories, a choice made to reduce the three-dimensional distance between racks. This layout supports high-density computing with up to 72 Blackwell GPUs per rack. The racks connect via NVLink and form clusters using a two-tier ethernet backend. Microsoft employs its own SONiC operating system for network switches to avoid vendor lock-in. The design omits traditional uninterruptible power supplies and diesel generators, relying instead on grid reliability.
Power and cooling
The facility draws power directly from the local grid, which Microsoft cites as sufficiently reliable to eliminate on-site generation. There are no backup generators or UPS systems installed. Each rack supports approximately 140 kilowatts, while each row handles up to 1,360 kilowatts. Cooling is provided by a closed-loop liquid system, which helps manage the heat generated by the dense GPU arrays. This configuration reduces construction time and operational costs compared to traditional designs.
What runs inside
The primary function is AI training and inference using Nvidia hardware. It connects to the Wisconsin site via an AI Wide Area Network. Microsoft collaborated with OpenAI and Nvidia to create the Multi-Path Reliable Connected protocol for optimizing network routes. The system provides 800 Gbps connectivity between GPUs. It serves as part of Azure infrastructure, enabling distributed computing tasks that require coordination across multiple geographic locations.
The place
Atlanta was selected partly due to the reliability of its electrical grid. This stability allowed Microsoft to remove redundant power infrastructure. The urban location supports the high-band fibre connections needed for the distributed super-cluster. The site contributes to Microsoft鈥檚 broader strategy of placing AI infrastructure in regions with robust power and network capabilities.
Environment and impact
The design reduces the need for diesel fuel storage and on-site combustion engines. By removing generators and UPS systems, the facility lowers its physical footprint and construction complexity. Microsoft states this approach lowers costs and speeds up deployment. The use of liquid cooling aims to improve energy efficiency for high-density AI workloads.
Network and links
The site links to the first Fairwater facility in Wisconsin through dedicated fibre optic cables. This connection creates an AI Wide Area Network. Internal networking uses an ethernet-based backend with custom protocols. Microsoft utilizes its SONiC switch operating system to manage traffic and maintain flexibility in hardware sourcing.
Questions people ask about Microsoft Fairwater Atlanta
Why does this data centre have no backup generators?
Microsoft relied on the reliability of the Atlanta electrical grid to remove the need for on-site generation. This decision eliminates uninterruptible power supplies and diesel generators, reducing construction time and operational costs while maintaining service availability.
How are the Atlanta and Wisconsin sites connected?
The two facilities are linked by an AI Wide Area Network using dedicated fibre optic cables. This connection allows them to function as a distributed supercomputer, sharing workloads and resources across geographic distances for large-scale AI training tasks.
What type of cooling system is used here?
The facility uses a closed-loop liquid cooling system. This method is necessary to manage the high heat output from the dense arrays of Nvidia GPUs, which can draw up to 140 kilowatts per rack, ensuring efficient thermal management without traditional air conditioning.
Good reading and sources
There is no English Wikipedia article on Microsoft Fairwater Atlanta yet, so the facts on this page come from the sources listed above: datacenterdynamics.com/en/news/microsoft-launches-atlanta-fairwater-da. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryDatacenter wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.