Microsoft is attempting to address one of the most persistent criticisms surrounding the rapid growth of AI: the environmental impact of the data centers that support it.
During Microsoft Build 2026, CEO Satya Nadella outlined the company’s latest approach to AI infrastructure, focusing on a new generation of facilities called Fairwater. The design seeks to boost computing capacity while dramatically reducing water consumption and improving energy efficiency.
Nadella touted the advantages of a 315-acre Fairwater site in Mount Pleasant, Wisconsin. Remarkably, he claimed the facility’s cooling architecture allows it to consume roughly the same amount of water annually as a neighborhood restaurant. The claim is significant because water usage is a major point of contention as hyperscale AI facilities are constructed worldwide. Even mid-size facilities need hundreds of thousands of gallons per day.
Unlike traditional evaporative water cooling systems, Fairwater uses a closed-loop cooling design. The system is filled once and then operates without ongoing water consumption under normal conditions.
The need for more resource-efficient data center technology is, to be sure, urgent. Microsoft now operates more than 500 data centers across 80 regions through its Azure cloud platform. And that number is expanding: Nadella said the company added more capacity during the past 18 months than it deployed during Azure’s first decade.
That rapid growth has greatly heightened pushback from local communities and environmental groups concerned about pressure on water supplies, electricity grids and noise levels.
Nadella acknowledged those concerns, arguing that large tech companies must earn community support before expanding infrastructure projects. He pointed to Microsoft commitments around water replenishment, local job creation, workforce training programs and investments in surrounding communities.
Vertical Design, Interconnected Platform
Fairwater benefits from a number of data center design improvements. Rather than spreading computing hardware across large single-floor layouts, Fairwater uses a two-story architecture that enables equipment racks to be arranged vertically. The design increases compute density while shortening network connections between systems.
Microsoft says each Fairwater facility can support hundreds of thousands of NVIDIA GPUs operating as a unified computing cluster. The architecture accommodates the expanding range of AI workloads now entering production, including model training, reinforcement learning, synthetic data generation and large-scale inference.
Together with Microsoft’s overall Azure infrastructure, the Fairwater facilities form what Microsoft promotes as an interconnected AI computing platform spanning multiple geographic regions.
Supporting that design is Microsoft’s continent-scale networking infrastructure, which links newer AI facilities with previous generations of AI supercomputing systems. The goal is to allow workloads to move dynamically across locations while treating distributed computing resources as a single pool of capacity.
AI systems require massive concentrations of GPUs, networking equipment and power infrastructure. Higher-density deployments can improve performance and reduce latency, but they also boost demands on cooling systems and local utilities.
While Microsoft is emphasizing sustainability improvements, skepticism remains. Many data centers still rely on traditional cooling technologies that consume vast amounts of water. Critics argue that a few examples of efficient facilities do not fully address environmental concerns created by industry-wide AI expansion.
Those concerns were visible during Build 2026, where protesters gathered outside the event calling attention to the environmental and community impacts of large-scale data center construction.
