Artificial intelligence is often discussed as software, algorithms and computing power. But behind every AI model, cloud service and digital platform is physical infrastructure—and that infrastructure requires enormous amounts of electricity, cooling equipment and water.
The scale is becoming difficult to ignore. U.S. data centers consumed roughly 176 terawatt-hours (TWh) of electricity in 2023, representing approximately 4.4% of total U.S. electricity consumption. The latest federal research suggests the industry’s share could become substantially larger before the end of the decade.
For the skilled trades building and maintaining this infrastructure, those numbers represent something else: substations, transmission equipment, generators, cooling plants, piping systems, pumps, cooling towers, electrical distribution, controls and thousands of miles of supporting infrastructure.
Figure 1. U.S. data center electricity demand has roughly tripled since 2014, with rapid AI and hyperscale expansion expected to drive substantially higher power requirements through 2028.
The Electricity Numbers
Data-center electricity consumption has changed dramatically in little more than a decade.
In 2014, U.S. data centers consumed approximately 58 TWh of electricity annually. By 2018, consumption had climbed to approximately 76 TWh. By 2023, it reached approximately 176 TWh.
That means data-center electricity consumption roughly tripled between 2014 and 2023.
And the growth is expected to continue.
Earlier federal projections estimated U.S. data centers could consume approximately 325–580 TWh annually by 2028, equivalent to roughly 6.7%–12% of total U.S. electricity consumption.
More recent federal modeling extends the outlook further. A 2026 Lawrence Berkeley National Laboratory update estimates data centers could account for approximately 11.8% of U.S. electricity consumption by 2030 under its reference case, with modeled scenarios ranging from approximately 9.5% to 15.3%.
That range is important. Nobody knows exactly how quickly AI computing demand, hardware efficiency, new construction and other factors will develop. But across the scenarios, the direction is clear: data centers are becoming a major component of U.S. electrical demand.
AI Changed the Equation
Traditional data centers were already substantial electrical consumers, but accelerated computing has changed the density of the load.
AI workloads increasingly rely on powerful GPU-based servers. Packing large numbers of these processors into a facility can create extremely high electrical loads concentrated inside relatively small physical footprints.
Electricity entering the facility does not simply power computers. Supporting systems also consume energy.
Switchgear distributes it. UPS systems protect it. Transformers change voltage. Pumps move cooling water. Chillers remove heat. Cooling towers reject heat. Fans circulate air. Backup generators provide emergency capacity. Control systems continuously monitor the entire operation.
A modern data center is therefore not simply a building filled with servers. From an industrial perspective, it is a continuously operating electrical and mechanical plant built around computing equipment.
Then Comes the Water
Figure 2. U.S. data center water demand is rising alongside electricity consumption, driven largely by cooling requirements as larger and more power-dense computing facilities come online.
Electricity is only half of the resource story.
Computers convert much of the electricity they consume into heat. That heat has to be removed continuously, and many large facilities use water-based cooling somewhere in that process.
Federal research estimated that U.S. data centers directly consumed approximately 17.4 billion gallons of water in 2023, primarily associated with cooling. That was more than three times the approximately 5.6 billion gallons estimated for 2014.
Under a high-growth projection, direct data-center water consumption could reach approximately 32.8 billion gallons annually by 2028.
And direct facility consumption doesn’t capture the entire water footprint.
Direct vs. Indirect Water Consumption
There are two important categories when discussing data-center water use.
Direct water consumption occurs at the data-center site, primarily through cooling operations where water can be lost through evaporation.
Indirect water consumption occurs elsewhere, particularly when water is consumed in producing the electricity that ultimately powers the facility.
Federal research estimated approximately 211 billion gallons of indirect water consumption associated with electricity generation for U.S. data centers in 2023.
That distinction matters because saying a data center “uses water” can mean very different things depending on what is being measured. Water withdrawal, direct consumption and indirect consumption are separate metrics and should not be treated as interchangeable.
Not Every Data Center Uses the Same Amount of Water
There is no universal number for how much water one data center consumes.
Facility size, climate, server density, cooling architecture, operating load, water source and local weather can all substantially change consumption.
Some facilities rely heavily on evaporative cooling. Others use closed-loop systems, air cooling, reclaimed water or combinations of technologies. Increasing rack densities are also accelerating deployment of direct-to-chip liquid cooling and other advanced thermal-management systems.
That means two facilities with similar computing capacity can have significantly different water requirements.
The Mechanical Infrastructure Behind the Numbers
For industrial workers, the growth in power and water demand translates directly into construction.
A hyperscale campus can require extensive chilled-water and condenser-water piping, pumps, heat exchangers, cooling towers, chillers, water-treatment equipment, chemical-feed systems, storage tanks, valves, instrumentation and controls.
On the electrical side, facilities can require utility substations, transformers, medium-voltage distribution, switchgear, UPS systems, battery systems, generators, busway and massive quantities of conduit and cable.
These are not small support systems attached to an IT building. On the largest campuses, the supporting mechanical and electrical infrastructure can resemble what industrial trades have traditionally encountered in power plants, refineries and large manufacturing facilities.
Why Power May Become the Bigger Construction Constraint
A company can purchase servers relatively quickly. Building the infrastructure necessary to continuously deliver hundreds of megawatts of dependable electrical power is considerably more complicated.
New generation, transmission lines, substations and utility interconnections can require major construction programs. The same applies to cooling infrastructure when extremely high computing densities must operate around the clock.
This is why the data-center boom reaches far beyond the data-center property itself. Increased computing demand can trigger construction throughout the surrounding power and utility network.
What This Means for the Skilled Trades
The statistics point toward a major physical infrastructure buildout.
Data centers require electricians and electrical technicians to build and maintain enormous power-distribution systems. Pipefitters and welders construct chilled-water, condenser-water, fuel and process-support systems. Millwrights install and maintain pumps, chillers and rotating equipment. Instrumentation and controls technicians commission increasingly sophisticated automated systems.
Riggers, crane operators, ironworkers, equipment operators, insulators, HVAC technicians, commissioning specialists, foremen and supervisors are also required throughout construction and startup.
The technology may be new, but much of the infrastructure required to make it work depends on familiar industrial skills.
The Bigger Picture
In 2014, U.S. data centers consumed approximately 58 TWh of electricity and 5.6 billion gallons of direct water.
By 2023, those figures had climbed to approximately 176 TWh and 17.4 billion gallons.
Federal projections indicate electricity demand and water consumption could continue climbing substantially as larger AI-oriented facilities enter service.
Those numbers reveal something important about the digital economy: it is becoming increasingly physical.
Every new AI cluster ultimately needs electricity delivered through real transformers and switchgear. Every megawatt consumed eventually becomes heat that must be removed. Cooling requires real pumps, piping, equipment and controls. Those systems must be fabricated, installed, tested, commissioned and maintained by people.
The AI revolution may happen on silicon, but the infrastructure supporting it is being built out of steel, copper, pipe, concrete, power and water.
Leave a Reply