The Data Center Fight Is Mostly About AI Scale

Ten years ago, I worked in data centers that few people noticed. Now one proposal can fill a town meeting. Residents want to know how much electricity it will consume, how much water it will use, whether it will hum all night and who will pay for the substation.

When I was inside those facilities, the work was practical and mostly invisible: replacing failed drives, tracing cables, watching temperature alerts and testing backup power. The goal was not to invent a new digital economy. It was to keep existing services online when a disk, switch or power feed failed. A good day was quiet because customers never knew anything had gone wrong.

The backlash followed the AI building boom. The AI race has produced plans for enormous computing campuses, often announced before the public can see credible estimates of their costs. In a 2026 Pew Research Center survey, 38% of Americans said data centers were mostly bad for home energy costs. Just 6% said they were mostly good. Views of their environmental impact were similarly negative.

The International Energy Agency expects worldwide data-center electricity use to roughly double from 2025 to 2030. AI-focused facilities are growing much faster than conventional ones.

Yet the label “data center” covers very different operations. Using it as shorthand for “giant AI facility” confuses the debate.

The internet has to live somewhere

A data center is a building that houses servers, storage equipment and network connections. Some facilities train large AI models. Many simply run the services people already use every day.

Websites are stored on servers. Email is routed through servers. Card payments, video calls, cloud files, streaming services and business software all depend on computers running somewhere. Hospitals need systems for appointments and patient records. Banks need transaction processing and fraud checks. Schools, local governments and small businesses rent computing capacity because they cannot maintain it all themselves.

Consider a hospital portal. When a patient checks a test result, the request travels across networks to an application and a database. Those systems may run in a regional data center. Backup equipment keeps them available when hardware fails. Security controls limit who can enter the building or reach the machines. None of that is glamorous, and none of it requires training a giant language model.

Internet exchanges and local hosting facilities also shorten the distance data must travel. That can make services faster and more resilient. The “cloud” is still physical: cables, routers, servers, cooling equipment, backup power and buildings.

Smaller providers also give organizations a place to keep equipment close to the people using it. That matters when latency, legal requirements or unreliable long-distance connections make a faraway hyperscale cloud a poor fit. A local facility may host dozens of businesses rather than one technology giant. A regional hosting facility and an AI campus may share a zoning label. They do not place the same demand on the grid.

A server room is not an AI campus

A regional facility hosting ordinary websites is not equivalent to a hyperscale campus packed with AI accelerators. Their demands on electricity, cooling, land and public infrastructure can differ by orders of magnitude. A shared industry label does not erase those differences.

Local officials should start with basic facts. What will the site compute? What are its peak electricity and water demands? Who pays for transmission lines and substations? Can the operator reduce its load during a grid emergency? How many permanent jobs will remain after construction? Will the answers be published and enforced?

An AI developer should not answer those questions with a speech about how the internet works. Essential hosting does not justify unlimited expansion for speculative AI demand. Companies should identify what they are building and accept rules that reflect its actual impact.

Regulate the impact, not the buzzword

Permits should follow the actual load, not the industry label. The useful dividing lines are measurable: power demand, water use, noise, emissions, land requirements and the cost shifted to ratepayers.

Judge projects by what they consume and provide, not by the language in a corporate press release.

Require large projects to pay for the infrastructure they need, reduce water use during drought, meet noise limits near homes and disclose how much capacity is reserved for AI. “Cloud” is too vague for a project asking a community for power and water.

Communities are right to challenge projects built on vague promises and enormous utility requests. They should also know whether a proposal will train AI models or host the payroll, medical records, messages and websites they already depend on.

The internet needs data centers. It does not need every data center on a corporation’s wish list.

Sources

• Pew Research Center: How Americans view data centers’ impact
• International Energy Agency: Key Questions on Energy and AI
• Internet Society: Who Makes the Internet Work
• Cloudflare: What Is a Data Center?

Need Remote Hands in Los Angeles?

When your engineers need someone at the equipment, share the location and task list with FOO.XYZ. Our Los Angeles Remote Hands services cover agreed physical work, with access and scheduling confirmed before a visit.

Phone: (213) 212-8317

Email: hello@foo.xyz

FOO.XYZ serves businesses in Los Angeles. Scope and scheduling are agreed before work begins.