Part 2 of 13

What Data Centers Do and Why We Need Them

By Randy Salars · Source-review edition: September 7, 2026

When we check a bank balance, send a document, or watch a video, we are using equipment somewhere beyond the screen. “The cloud” has buildings, wires, machines, and a power bill. Understanding that physical reality helps us see both why data centers matter and why their host communities deserve a clear account of what they require.

A data center is a place that houses computers and the systems needed to keep them working. Servers perform calculations and run software. Storage equipment keeps information available. Networking equipment moves it between machines and users. Power, cooling, fire protection, and security support that work. A company can own a facility for itself, rent space for its equipment, or purchase computing services from a cloud provider. Those arrangements have different owners and customers, even when the buildings look similar. AWS explains the equipment and business arrangements.

For a small business or nonprofit, buying a service can be much more practical than assembling a computer department. Consider a hypothetical local organization that needs payroll, donor records, appointments, and reliable backups. It needs those functions to work, but it may have neither the staff nor the money to operate every underlying system. Shared computing services let organizations purchase capabilities in manageable pieces. Many ordinary needs, added together, create demand for substantial computing infrastructure.

The word “data” can also create a misleading picture of a giant warehouse that merely collects information about people. Storage is one function. Computing is another. The question “Why store so much?” therefore needs a companion: “What calculations and services will these machines perform?” Privacy remains an appropriate concern, but the amount of electricity a facility uses does not tell us whether it holds personal records, analyzes scientific information, runs business applications, or performs some combination.

AI adds two terms worth learning: training and inference. Training is the process of adjusting a model using examples and feedback so it becomes better at a task. Inference is using the resulting model to produce an answer, prediction, image, or other output. The first builds a capability; the second puts it to work. Both require computation. A finished model does not answer an unlimited number of people without equipment continuing to run. NVIDIA describes these stages in its training and inference explainer, updated in October 2025.

A company may train a model during a concentrated period and then serve requests over a much longer period. Serving many users at once requires capacity even when individual requests are small. Grant County should ask which activities are intended here because that helps establish the equipment, business plan, and operating pattern.

Some large AI training jobs require thousands of processors to exchange information rapidly while working on the same task. Distance and slow connections can make that difficult. Meta describes building a cluster of about 129,000 H100 processors across five buildings, and explains how a lagging processor can slow a synchronized training job. This is one reason a very large facility can have a technical purpose beyond simply fitting more computers under a roof. It is an example from Meta, not evidence that Meta is a customer of this proposal. Meta’s account of its infrastructure development.

The machinery also produces heat. Getting heat away from a chip is only part of the cooling job; the facility must ultimately release it outside. Water consumption depends on that whole arrangement. A circulating liquid loop inside the building does not, by itself, establish whether the outside cooling system evaporates water. The Department of Energy explains that distinction in its cooling guidance.

This is why a phrase such as “advanced cooling” cannot settle a local water question. Residents need an understandable description of the complete system and quantities for its actual operating conditions. The same applies to claims of efficiency. A design might perform very well by one measure while still requiring a substantial amount of a resource at its proposed scale. We need both the efficiency figure and the total.

Large remote facilities are also only one part of computing’s future. Apple describes an approach that handles many AI functions on a personal device and uses cloud computing for more demanding requests. That example shows how local and centralized computing can work together. Apple’s description of its device and cloud approach.

The scale of the industry’s electricity demand is already significant. The International Energy Agency’s April 2026 outlook projects global data-center consumption rising from 485 terawatt-hours in 2025 to roughly 950 in 2030. A terawatt-hour is a billion kilowatt-hours. These figures cover all data centers, and the later figure is a projection. They are not a measurement or forecast for Grant County. IEA’s updated outlook · Official PDF, page 10.

Better chips and software can reduce the electricity needed for a particular task. Total demand can still grow if many more people use the service or ask it to perform more demanding work. The IEA identifies efficiency, expanding use, and changing tasks as separate influences on demand. An efficiency improvement therefore does not automatically make a facility unnecessary, and rising industry demand does not guarantee that every proposed project will find customers. IEA’s explanation of these competing forces · Official PDF, executive summary.

I use digital services and see their value. That gives me a reason to understand the infrastructure, and a reason to expect a convincing explanation of this particular project. The places hosting the equipment deserve clear information about its physical demands and local terms.

For Site Layer 1, the next explanation should connect the machines to an actual business purpose: what customers would buy, why this location works, and how much equipment would operate during each phase. That would give residents something concrete to evaluate.

What a useful answer would include

The developer and intended operator should provide a dated schedule showing proposed computing uses, equipment, customer status, and electricity demand for each phase. An engineer should explain the full power and cooling system, including the hottest operating conditions. Forecasts and maximum permitted operation should be identified separately from signed commitments.

Questions worth asking

• What computing uses are planned, and what documents confirm customer demand for each phase?
• How much electricity would the computers use, compared with cooling and other supporting equipment?
• What technical or business requirements explain the proposed size and location?
• What operating schedule and peak demand has the engineer assumed for each phase?
• Can the developer publish a complete cooling diagram and water budget for those operating conditions?

AI assisted most of the research and initial drafting under my direction. Sources are linked; corrections are welcome.