.svg){kind=spacer}
Details
- Microsoft has launched the new Fairwater AI datacenter in Atlanta, Georgia, creating a critical hub connected to its original Wisconsin site as part of an advanced, planet-scale AI supercomputing network.
- The Atlanta facility incorporates hundreds of thousands of NVIDIA Blackwell GPUs, unified into an immense compute cluster that supports leading customers and partners including OpenAI, Mistral AI, xAI, and Microsoft's own AI models.
- Key innovations feature a sustainable closed-loop liquid cooling system that reuses coolant over a 6-year cycle, a two-story building architecture minimizing latency and maximizing rack placement efficiency, and a proprietary Multi-Path Reliable Connected (MRC) networking protocol for seamless supercomputer-scale GPU operation.
- The site delivers high resiliency, offering 99.99% power availability at the cost level of 99.9% by relying on robust grid power and eliminating traditional on-site backup systems such as generators and UPS units.
- With the deployment of over 120,000 new miles of fiber across the United States within the past year, Microsoft has expanded its dedicated AI optical network, allowing for dynamic, cross-site allocation of immense compute resources to handle emerging workloads requiring trillions of model parameters.
Impact
This milestone places Microsoft at the forefront of the global AI infrastructure race, leveraging custom engineering and massive scale to cut training costs and accelerate AI development. The Atlanta Fairwater site underscores a fierce battle among cloud giants as tens of billions pour into next-generation compute, connectivity, and cooling—reshaping the economics and reach of AI innovation worldwide.
