Building Responsibly: How AI Data Centers Are Navigating the Power and Water Crunch

As AI development accelerates, attention is increasingly turning to the physical infrastructure needed to support it. The past few years have been marked by rapid development of AI models and platforms from major technology companies.

As AI deployment expands, there is increasing demand for the data centers, electrical infrastructure and power capacity required to support computing workloads.

Power and water use associated with AI infrastructure have also become important considerations for developers, utilities and communities. Some developers, from hyperscalers to specialized infrastructure companies, are responding with alternative power strategies and designs intended to reduce water consumption.

The race for AI power

One of the major obstacles standing in the way of new AI data center development is access to enough power.

As AI computing is being rolled out, there is increased need for data center infrastructure

One way to describe the core requirements for data-center development is land, power and shell (LPS):

• Land: Geographic area required for development of the infrastructure, including zoning and site development.
• Power: Significant amount of electricity required to operate the AI data center. Access to large amounts of power can be constrained by capacity, cost, generation and environmental considerations.
• Shell: Physical data center building equipped with cooling and electrical infrastructure ready for hardware deployment.

Each of these requirements can affect the timing and feasibility of new data-center projects, with available power capacity frequently cited as a significant constraint.

The issue with AI power and water

AI infrastructure depends on reliable access to substantial amounts of electricity.

Competition for available grid capacity is influencing where and how quickly some data-center projects can move forward. One professional services firm reported that 72% of respondents characterized power availability and grid capacity as very or extremely challenging.

In some markets, developers may face grid-interconnection timelines measured in several years.

Cooling requirements add another consideration. Water use varies substantially by facility design, location, climate and cooling technology. One estimate puts annual cooling-water use for a medium-sized facility at roughly 100 million gallons.

Those concerns are contributing to greater interest in cooling systems designed to reduce dependence on municipal water supplies.

Modular builds and closed-loop cooling

Giga Energy has developed infrastructure for computing applications, including bitcoin mining and AI data centers. As demand for AI computing increased, the company expanded its focus to high-density data-center infrastructure, including its GigaBase platform.

GigaBase uses a closed-loop cooling system designed to limit ongoing use of municipal water for equipment cooling. According to Giga Energy, cooling fluid is supplied to the closed-loop system rather than being continuously replenished with municipal water. Municipal water may still be required for domestic uses such as plumbing, but Giga says the cooling system is designed not to require a continuous municipal-water supply.

Giga Energy also says the closed-loop system is designed to operate without discharging cooling water into municipal wastewater systems.

Beyond its cooling design, GigaBase uses standardized, pre-engineered components intended to reduce deployment time. According to Giga Energy, GigaBase is designed to move from site selection to energization in approximately nine months. Giga says assembling key equipment off-site can reduce the amount of work required during on-site installation. The platform can be developed alongside dedicated power infrastructure, depending on the site and project.

Colocating compute with stranded energy

While modular engineering can help reduce water demand, some infrastructure providers are addressing power constraints by changing where computing infrastructure is located. One approach places mobile data center modules near certain energy sources.

This type of deployment approaches some land and power constraints differently from conventional data-center development. Locating computing infrastructure near energy assets can allow certain workloads to use power that may otherwise be difficult to bring to market. For some workloads that are less sensitive to location and latency, colocating computing equipment near generation may offer another way to add capacity without relying on a conventional grid-connected data-center model.

Scaling closed-loop designs and water replenishment

Some major cloud providers are also changing data-center cooling designs in response to water-use and efficiency concerns. Newer designs can use closed-loop, chip-level cooling systems intended to reduce water consumed for cooling.

Some large data-center operators are also investing in water-replenishment projects alongside changes to cooling systems. Closed-loop designs recirculate cooling fluid, reducing the need for continuous water evaporation during cooling. Water-replenishment efforts can also include watershed projects.

Addressing power and water needs in AI infrastructure

Growing demand for AI computing is increasing attention on the infrastructure required to support it. That expansion is increasing demand for power capacity while also drawing scrutiny to data-center water and energy use.

The examples above illustrate that developers are pursuing several different approaches to power and cooling rather than relying on a single infrastructure model. Some operators are reducing their reliance on municipal water for cooling, while others locate computing infrastructure near available energy sources or adopt different approaches to thermal management. How effectively those approaches reduce resource use, grid constraints and community impacts will depend on the design, location and operation of individual projects.