Microsoft: Inside the World’s First AI Superfactory

The race to train frontier AI models is pushing data centre infrastructure to its physical limits – and the constraints are increasing. 

So much so, now the speed of light is determining how tightly processors can be packed together – and heat dissipation governs how much power can be pumped through a rack. 

These are the hard boundaries influencing where and how AI gets built.

Tackling the challenge, Microsoft has opened its second Fairwater AI data centre in Atlanta, Georgia, connecting it to the existing Wisconsin site through a dedicated AI wide area network.

What’s unique about the factory is that the company has designed the facility to house hundreds of thousands of Nvidia GB200 and GB300 graphics processing units (GPUs) in a single flat network architecture – abandoning the traditional cloud data centre model in favour of something purpose-built for the demands of modern AI training.

Satya Nadella, CEO at Microsoft, says: “Today we announced our new Fairwater data centre in Atlanta, connected with our first Fairwater site in Wisconsin and our broader Azure footprint to create the world’s first AI superfactory.”

So what makes an AI superfactory? And what’s the importance of this being the first in history? 

How does the AI superfactory work and what are its benefits?

The Atlanta site shows how AI workloads have evolved beyond simply training massive models. 

Satya says: “AI workloads have evolved beyond large-scale pre-training. Today, they encompass fine-tuning, reinforcement learning, synthetic data generation, evaluation pipelines and more.”

As a result, the Fairwater data centres use facility-wide liquid cooling systems to manage the intense heat output from AI servers. 

Each rack draws approximately 140kW of power, with entire rows consuming 1,360kW. 

The closed-loop cooling approach then reuses water continuously after the initial fill, which requires the equivalent of what 20 homes consume in a year.

The shift to liquid cooling isn’t just about sustainability. It changes how densely Microsoft can pack computing power. 

Air cooling simply can’t remove heat fast enough at these power levels, so liquid cooling becomes the only viable path forward.

Microsoft has also implemented a two-storey building design to minimise cable run lengths between GPUs. 

When every GPU needs to connect to every other GPU in the cluster, physical distance matters. 

Placing racks in three dimensions reduces how far signals must travel, directly improving latency and bandwidth.

Satya explains: “Fairwater’s two-story design and liquid cooling system lets us place racks in three dimensions and pack them with GPUs as densely as possible, minimising cable runs and improving latency and effective bandwidth.”

Each rack houses up to 72 Nvidia Blackwell GPUs connected via NVLink, Nvidia’s proprietary interconnect technology. 

Furthermore, the Blackwell accelerators support FP4, a four-bit floating point format that increases operations per second while reducing memory requirements. Each rack provides 1.8TB of GPU-to-GPU bandwidth.

How Microsoft’s AI superfactory grid power delivers 99.99% availability

The Atlanta location was chosen partly for its access to utility power with four nines availability at three nines cost – 99.99% uptime at a price point typically associated with 99.9% reliability. 

This reliable grid connection allows Microsoft to eliminate traditional backup infrastructure including on-site generation and uninterruptible power supplies for the GPU fleet.

However, managing power at this scale presents its own challenges.

Large-scale training jobs create power oscillations that can affect grid stability, so Microsoft has developed solutions with industry partners including software approaches that smooth demand and hardware solutions where GPUs enforce their own power thresholds.

Satya says: “Each Fairwater DC can integrate hundreds of thousands of the latest Nvidia GPUs into a single coherent cluster. 

“This provides flexible infra that can support the full spectrum of workloads and ensure no GPU is left unnecessarily idle.”

Microsoft deployed more than 120,000 new fibre miles across the US last year to build the dedicated AI wide area network connecting these facilities. 

Satya says: “Every Fairwater DC will connect through our continent-spanning AI WAN to prior generations of AI supercomputers, forming a truly fungible pool of compute.”

The company is bringing more than 100,000 GB300 GPUs online this quarter for inference across its broader fleet. 

Satya concludes: “For us, it’s all about turning every gigawatt into the maximum number of useful tokens. Not every GW is created equal.”