What exactly is an AI data center?

WASHINGTON, D.C. - An AI data center is essentially a building — or sometimes a sprawling campus — packed with powerful computers that train and run artificial intelligence systems, but the infrastructure required to keep those machines operating has turned the once-obscure facilities into a growing political issue.

What is a data center?

A data center is a physical facility that houses the computers and related equipment used to store information, run applications and deliver online services.

The facility may support anything from email and banking systems to streaming platforms, cloud storage and online shopping. Some are operated by individual companies, while major cloud providers run large facilities that serve many customers at once.

In its simplest form, a data center is a secure building filled with servers. Those servers are powerful computers that process requests, store information and communicate with users over high-speed networks.

What makes an AI data center different?

An AI data center is not an entirely different type of building. It is a data center designed and configured for the unusually demanding work involved in artificial intelligence.

Traditional data centers often depend heavily on central processing units, or CPUs, to run websites, databases and business applications. AI facilities rely more heavily on graphics processing units and other specialized chips that can perform many calculations simultaneously.

That parallel processing is important when developers train an AI model on large amounts of data or operate, or perform "inference" with, a finished model after it is released to users.

Cisco describes AI data centers as facilities optimized for model training and deployment, with high-performance GPUs or tensor processing units, high-speed networking and systems built to minimize delays when data moves between machines.

What happens when someone uses AI?

When someone sends a question to a chatbot or asks an AI program to create an image, that request travels through a network to servers running the AI model.

The computers process the request, calculate a response and send the result back to the user. A single request may happen quickly, but popular AI services can process enormous numbers of requests while developers simultaneously train newer and more advanced models.

Data centers also provide the storage and networking needed to move information between the servers performing that work. Cisco lists model training, retrieval-augmented generation and inference among the AI workloads supported by modern data centers.

What is inside an AI data center?

The computing equipment is only part of the facility.

A modern data center generally includes servers, data-storage systems, routers, switches, fiber connections and security equipment. It also needs power systems, backup batteries, generators, ventilation, fire suppression and cooling equipment.

Redundant systems are also important. If one power supply, cooling system or network connection fails, another may take over so the services running inside do not go offline.

Some cloud companies spread applications across physically separate data centers so an outage or natural disaster affecting one location does not shut down the entire service.

Why do AI data centers need so much power?

AI requires large groups of specialized processors working together. Those chips consume electricity while also producing heat that must be removed to prevent equipment from overheating.

That means electricity is needed both to operate the computers and to run the cooling systems supporting them.

Cisco says GPUs powering AI facilities can consume significantly more electricity per processing cycle than the CPUs typically associated with traditional data centers. The exact demand varies depending on the facility, equipment and work being performed.

Data centers therefore need access to reliable electrical capacity, high-speed internet connections and cooling infrastructure before they can operate at scale.

Why do data centers need cooling?

Servers generate heat continuously while they are operating. Without cooling, temperatures inside the facility could damage equipment or interrupt service.

Some data centers cool entire rooms using air-conditioning systems. Others use liquid systems that move coolant closer to processors or place equipment in specialized fluids.

The cooling method — and the local climate — can affect how much electricity or water a particular facility uses. Not every data center uses the same system, and its environmental footprint depends heavily on its design and energy source.

Are all data centers used for AI?

No. Data centers existed decades before the current AI boom.

Companies have long used them for financial transactions, databases, websites, medical records, business software, cloud storage and other digital services.

Some existing facilities can add equipment to support AI. Other projects are built specifically for high-density AI workloads and may require more advanced chips, networking and cooling than a conventional facility.

Data centers also come in different sizes. A company may operate a private facility for its own systems, rent space inside a shared colocation center or purchase computing capacity from a large cloud provider. Smaller edge data centers can be placed closer to users to reduce the time it takes information to travel.

Why are AI data centers controversial?

Supporters argue that data centers are necessary for cloud computing, AI development, secure domestic data storage and growth in the technology industry.

Opponents and nearby residents have raised concerns about electricity prices, pressure on the power grid, water use, noise, pollution and the amount of land required for large projects.

Those concerns do not mean every proposed data center will produce the same effects. The impact depends on its size, location, cooling method, power source and the infrastructure already available in the surrounding community.

However, the concerns are increasingly influencing public opinion. A POLITICO Poll conducted with Public First found that electricity prices and the risk of blackouts ranked among respondents’ leading worries about data centers. Support also declined sharply when respondents were told a project could increase their monthly electricity bills.

How did data centers become a political issue?

For years, data center policy was largely handled through local zoning decisions, utility planning and economic development negotiations.

The AI boom has made the issue more visible. Facilities are becoming larger, companies are requesting more electrical capacity and residents are paying closer attention to how projects could affect their communities.

An earlier POLITICO Poll found that nearly half of Americans expected data centers to become a local campaign issue within five years. The survey also exposed a significant knowledge gap: Only 21% of respondents said they were very confident in their understanding, while roughly one-quarter either did not know what a data center was or identified its main purpose incorrectly.

More recent polling found attitudes had become increasingly negative and politically divided over six months. Opposition was growing among voters in both parties but was rising faster among Democrats, according to POLITICO’s July survey with Public First.

The debate leaves elected leaders weighing the demand for AI infrastructure against questions about who pays for new electricity generation, transmission lines and other upgrades — and how much control local communities should have over projects proposed near their homes.

What's next:

Demand for AI computing is expected to keep data centers at the center of debates over technology, energy and economic development.

The facilities make many online services possible, but their rapid expansion means decisions once treated as technical infrastructure questions are increasingly being made at public hearings, state legislatures and on the campaign trail.

Technology