AI data centers are becoming an important part of America’s infrastructure story. The headlines focus on increasingly powerful computer chips, new AI models, enormous investments by technology companies, and the race to develop the next generation of computing. All of that is important, but there is another side of the story that deserves considerably more attention: AI requires an extraordinary amount of electricity, and supplying that electricity is beginning to reshape America’s physical infrastructure.
The computers running today’s AI systems have to be housed somewhere. Those facilities require power, cooling, backup systems, and highly reliable electrical distribution. As computing density increases, so does the electrical demand of the facilities supporting it. What began as a technology investment is therefore becoming a major infrastructure investment, with consequences reaching well beyond the walls of the data center.
The effects are already being felt throughout the electrical industry. Utilities are being asked to accommodate enormous new loads. Transmission and distribution systems have to be expanded. Substations and transformers are becoming increasingly important. New generation will eventually be required to meet growing demand, bringing nuclear power and small modular reactors back into the discussion. If nuclear generation expands substantially, the country will also have to consider where the uranium needed to fuel those reactors will come from.
At the end of that chain are the people who have to build everything: electrical contractors, electricians, engineers, project managers, and estimators. The opportunities could be enormous, but so could the challenges.
The easiest way to understand what is happening is to follow the megawatts.
AI Data Centers Are Becoming Industrial Electrical Loads
For many years, data centers were thought of as specialized commercial buildings. That description is becoming less useful as artificial intelligence drives computing requirements higher.
AI systems rely heavily on high-performance processors, particularly GPUs, capable of performing enormous numbers of calculations simultaneously. Concentrating that much computing power into a relatively small physical space creates two related problems: the equipment requires substantially more electricity, and nearly all of that electricity eventually becomes heat that has to be removed.
Industry material reviewed for this article estimates that AI-oriented racks can require approximately 50 to 150 kilowatts, compared with roughly 10 to 15 kilowatts for more traditional data-center racks. The same material notes that newer generations of GPUs can consume more than 1,000 watts per chip.
Those figures help explain why the electrical systems supporting these facilities are becoming so substantial. A large AI data center may require major transformers, medium-voltage distribution, switchgear, busway, UPS systems, generators, and battery systems, along with extensive electrical infrastructure serving sophisticated cooling equipment. Because these facilities are expected to operate continuously and reliably, redundancy is also a major consideration.
The result is a building that behaves much more like an industrial electrical load than a conventional commercial facility. The question is no longer simply how to wire the building. The larger question is how to deliver and maintain the enormous amount of power the building requires.
That brings us to the first major challenge facing the industry.
The Numbers Are Getting Bigger
The scale of the planned construction is difficult to ignore. Goldman Sachs Research projects U.S. data-center power demand to increase from approximately 31 gigawatts in 2025 to 66 gigawatts in 2027. It also projects data centers’ share of total U.S. peak summer electricity demand to increase from 4.1% in 2025 to 8.5% in 2027.
Goldman Sachs also estimates that scheduled U.S. data-center capacity additions could reach approximately 13.6 gigawatts in 2026 and 36.3 gigawatts in 2027, compared with realized additions of 6.4 gigawatts in 2024 and 8.5 gigawatts in 2025.
There is an important qualification to those numbers. Not every announced project will be completed on schedule. Goldman Sachs estimates that roughly 60% of capacity scheduled for the following year may actually come online on time, with that percentage falling to approximately 50% for projects farther into the future.
That distinction is important because announced demand and actual demand are not necessarily the same thing. Projects can be delayed, redesigned, or canceled. Nevertheless, even after allowing for those uncertainties, the projected growth is substantial enough to create a major challenge for the electrical infrastructure serving the United States.
The country is going to need more generation, transmission, distribution equipment, substations, transformers, and construction capacity if even a portion of this projected demand becomes reality.
The Electricity Has to Come From Somewhere
It is tempting to describe the situation simply as an energy shortage, but that doesn’t fully explain the problem. The United States has substantial energy resources and an enormous electrical generating system. The difficulty is that electricity has to be available in the right amount, at the right location, and at the right time.
A developer can find suitable land, secure financing, obtain permits, and begin designing a data center, but none of that guarantees that the local utility can provide the required electrical capacity. A location that looks ideal from a real-estate perspective may not be nearly as attractive once the availability of electricity is considered.
Transmission lines have limits. Substations have limits. Large power transformers have long procurement periods, and connecting a massive new load to the utility system can require years of planning and construction.
Goldman Sachs has identified significant regional differences in the ability of the electrical grid to absorb additional data-center demand. Some regions are facing greater reliability concerns because planned generation additions may not keep pace with the amount of new data-center demand being contemplated, while other regions have more generation capacity under development.
That changes the traditional development equation. The question for a data-center developer is increasingly not simply, “Where can we build?” It is, “Where can we get the power?”
That may ultimately become one of the most important factors determining where the next generation of data centers is built.
The Grid Wasn’t Designed for This Kind of Growth
The American electric grid has been developed over generations. Power plants feed transmission networks, transmission systems feed substations, and distribution systems deliver electricity to homes, businesses, and industrial customers. The system is remarkably large and complicated, but it was not designed around the possibility that enormous computing campuses would appear in concentrated areas and require hundreds of megawatts of continuous power.
That doesn’t mean the grid cannot support this growth. It means the grid has to be expanded and modified to accommodate it.
A large data center can create electrical requirements that extend well beyond its property line. Depending on the location and the available capacity, the project may require a new substation, additional transmission capacity, high-voltage equipment, large transformers, and substantial medium-voltage infrastructure before electricity ever reaches the data-center building.
Industry analysis reviewed for this article identifies grid interconnection as a major constraint on data-center development and notes that some large markets are facing multi-year timelines for new connections.
This creates an interesting conflict between two very different timelines. The technology industry is moving at extraordinary speed, while electrical infrastructure is governed by engineering, permitting, manufacturing, procurement, and physical construction. A new generation of computer processors can be introduced in a relatively short period of time. A new transmission line or substation cannot.
That difference may become one of the defining infrastructure challenges of the AI era. The demand for computing power can grow much faster than the electrical infrastructure required to serve it.
More Demand Eventually Means More Generation
There is only so much that can be accomplished by moving existing electricity from one place to another. If the country continues adding enormous new electrical loads, additional generating capacity will eventually be required.
There is no single technology that necessarily provides the answer. Natural gas will remain an important part of the generation mix, renewable generation will continue to expand, battery storage can play an increasing role, and existing nuclear plants can continue to provide large quantities of reliable electricity. At the same time, the scale and reliability requirements of some new loads are causing renewed interest in building additional nuclear generation.
That is where small modular reactors, commonly called SMRs, have entered the conversation.
Why SMRs Are Getting Attention
The concept behind an SMR is to develop nuclear reactors in smaller configurations that may be capable of being manufactured and deployed more incrementally than traditional large nuclear facilities. The U.S. Department of Energy describes advanced SMRs in sizes ranging from tens of megawatts to hundreds of megawatts and identifies electricity generation and industrial applications among their potential uses. The DOE also points to smaller footprints and the possibility of incremental additions of generating capacity as potential advantages.
The concept becomes particularly interesting when considered alongside a large data center. A facility supporting artificial intelligence does not simply need a large amount of electricity; it needs that electricity continuously and with an extremely high degree of reliability. An interruption can be extraordinarily expensive, which is why the electrical systems serving these facilities are designed with substantial redundancy and backup capability.
Nuclear generation naturally enters that discussion because of its ability to provide continuous power. NuScale, one company developing SMR technology, describes a 77-MW power module and a 12-module configuration capable of producing up to 924 MW. Its published material specifically identifies data centers as one potential application.
That does not mean SMRs are going to become the standard solution for data-center power. Significant questions remain concerning licensing, economics, manufacturing, construction, fuel supply, and deployment. The Department of Energy recognizes technology-development and licensing challenges and anticipates that some U.S. SMR deployments could occur in the late 2020s and early 2030s.
What matters for the electrical industry is that the conversation is changing. Instead of asking only how a massive new load can connect to an existing utility system, developers and energy companies are beginning to consider whether new generation can be developed closer to, or specifically for, large industrial and computing loads.
If that approach becomes commercially viable, it could create another significant category of electrical construction.
The Nuclear Story Doesn’t End With the Reactor
There is another part of the nuclear discussion that is easy to overlook. Reactors require fuel, and if nuclear generation becomes a substantially larger source of American electricity, the country will need a dependable supply of uranium and the infrastructure required to turn that uranium into nuclear fuel.
The United States has uranium resources, but domestic production has historically supplied only a portion of the uranium required by American nuclear utilities. At the same time, the uranium market is dealing with long development timelines and increasing expectations for future reactor demand.
A July 2026 uranium-market analysis from Sprott describes a significant gap developing between future U.S. reactor requirements and uranium already covered by existing contracts. According to the report, maximum contracted coverage was approximately 98% of U.S. requirements in 2026, but that figure falls to about 60% in 2030, approximately 25% in 2032, and only 9% in 2033.
Those numbers do not mean that American nuclear utilities will literally have only 9% of their required uranium in 2033. They indicate how much of their projected requirements are already covered by contracts. Utilities can continue to enter into additional contracts, but the shrinking level of contracted coverage illustrates the need to secure future supplies.
That is not something that can be solved overnight.
A new uranium mine requires exploration, permitting, financing, development, construction, and ultimately a workforce capable of producing the material. The Sprott analysis also points to production disruptions and the geographic concentration of uranium supply as additional risks to the market.
The report also notes that U.S.-origin uranium accounted for only approximately 7% of utility deliveries in 2025.
That makes domestic uranium production part of a much larger energy-security discussion. If the United States wants nuclear power to play a greater role in meeting future electricity demand, it also has to consider the supply chain supporting that nuclear power, from mining through the processing and fuel systems required to put uranium into a reactor.
In that sense, the data-center story can eventually lead all the way back to the uranium mine.
Follow the Megawatts and You Find the Electrical Contractor
This is where the larger story becomes particularly important for electrical contractors.
When billions of dollars are announced for new data centers, it is easy to focus on the buildings themselves. But the electrical work associated with those projects can extend much farther. The data center needs its own electrical distribution system. The utility may need a new substation. Transmission capacity may have to be expanded. New generation may be required, and that generation will require its own electrical infrastructure.
Every step of that chain involves construction, and a significant portion of that construction is electrical.
Inside the data center, contractors may be responsible for medium-voltage distribution, transformers, switchgear, busway, UPS systems, generators, battery systems, controls, lighting, and the electrical infrastructure supporting cooling equipment. Outside the building, other electrical contractors may be involved in substations, transmission facilities, and utility distribution systems. If new generation is constructed, another major electrical scope enters the picture.
This is why the current data-center boom should not be viewed simply as a commercial construction opportunity. It is more accurately viewed as part of a much larger power-infrastructure buildout.
And that distinction matters because the work may continue long after the first generation of data centers has been completed.
The Industry Will Need More Than Electricians
The obvious labor concern is the number of electricians required to build all of this infrastructure, and that concern is legitimate. The data-center boom is already creating competition for skilled construction workers.
A July 2026 report from Quartz described more than $265 million in combined commitments from Meta, Google, and BlackRock toward recruiting and training construction workers, including electricians, for data-center development.
But the labor issue extends beyond electricians. Large electrical projects require experienced foremen, superintendents, project managers, engineers, estimators, and commissioning personnel. Those skills are developed over years of working on real projects, dealing with real problems, and learning how decisions made in an office affect what happens in the field.
That experience becomes especially valuable as projects become larger and more complicated. A mistake on a small commercial project can be expensive. A mistake on a major data center, substation, or generation project can be enormously expensive.
The industry therefore faces a challenge that cannot be solved simply by adding more people to a training program. It needs people with the right combination of technical knowledge, field experience, and judgment.
More Work Does Not Automatically Mean More Profit
There is another issue that electrical contractors should consider as this market expands.
More work does not automatically mean more profit.
A contractor can win a very large project and still lose money if the estimate is wrong, labor productivity is unrealistic, equipment costs are underestimated, scope is missed, or the project schedule creates costs that were never properly accounted for.
That risk becomes more significant as project size and complexity increase.
Data-center projects can involve complicated specifications, extensive redundancy, demanding schedules, large equipment purchases, and extremely high expectations for reliability. A contractor considering one of these projects has to look beyond the contract value and ask whether the company has the workforce, equipment, financial capacity, and management experience necessary to execute the work successfully.
And before any of those questions can be answered, somebody has to determine what the electrical work is actually going to cost.
That is where the estimator becomes particularly important.
The Estimator May Become More Important, Not Less
Electrical estimating has never been simply a matter of counting fixtures, measuring conduit, and multiplying quantities by unit prices. The larger and more complicated these projects become, the more important that distinction becomes.
An estimator working on a major data center has to understand the drawings and specifications, but also the relationships between the various parts of the project. Who furnishes the transformer? Who installs it? What equipment is owner-furnished? What temporary power is required? What testing and commissioning are included? What are the utility responsibilities? What happens when a critical piece of switchgear has a long lead time? What assumptions are being made about labor and productivity?
Those questions aren’t necessarily answered by a takeoff.
They require judgment, experience, and an understanding of how electrical construction actually gets performed.
As the industry moves toward larger and more technically complicated projects, the ability to accurately estimate, plan, and manage electrical work becomes increasingly valuable. The industry will need more electricians, but it will also need more people who understand how to turn complicated drawings and specifications into a realistic construction plan and a profitable bid.
A New Kind of Electrical Construction Market
If the current trends continue, the electrical construction industry could look considerably different a decade from now.
Contractors that traditionally focused on commercial construction may move into mission-critical facilities. Industrial electrical contractors may pursue data centers. Utility contractors may find opportunities associated with new transmission and generation. Engineering firms will be called upon to design increasingly complicated power systems, while manufacturers face growing demand for transformers, switchgear, and other electrical equipment.
The boundaries between commercial, industrial, utility, and power-generation construction may become less distinct as the country works to accommodate increasing electrical demand.
The common denominator will be power.
That is perhaps the most important point in this entire discussion. The AI industry may be driving the demand, but the physical infrastructure required to support it belongs to a much larger part of the American economy.
The AI Boom May Be Bigger Than AI
It is tempting to think of all of this as simply a response to the rapid development of artificial intelligence. It may eventually turn out to be exactly that. But there is another possibility: AI could be the catalyst that accelerates a much broader transformation of America’s electrical infrastructure.
AI is not the only source of increasing electricity demand. Electric vehicles, industrial electrification, manufacturing growth, cloud computing, semiconductor production, and other forms of digital infrastructure are all contributing to an economy that requires more electricity and increasingly sophisticated methods of delivering it.
Meeting that demand will require changes throughout the electrical system, from generation and transmission to substations, distribution equipment, and the facilities that ultimately consume the power.
The infrastructure being built today will also outlast the individual technologies driving the current investment cycle. Transmission lines, substations, power plants, and electrical distribution systems do not become obsolete simply because a new generation of computer processors is introduced. The knowledge developed by the people who design, estimate, build, and maintain those systems will carry forward as well.
That is why the current investment in AI may leave behind something much more significant than another generation of software.
It may accelerate a long-term expansion of America’s electrical infrastructure.
Follow the Megawatts
When you look at the AI revolution from a technology perspective, you see processors, software, data, and algorithms. When you look at it from the perspective of the electrical industry, you see something else: enormous new electrical loads that have to be supplied, protected, distributed, and maintained.
Those loads affect the data center, but they also affect the utility system serving it. They influence decisions about transmission, substations, and generation. They are contributing to renewed interest in nuclear power and SMRs, which in turn raises questions about uranium supply and domestic energy security. And every new piece of infrastructure creates work for the contractors, engineers, electricians, project managers, and estimators who have to build it.
That is why the impact of AI on the electrical industry may ultimately be much larger than the data-center construction boom itself. The real story is the infrastructure being built behind it.
The chain extends from the uranium mine to the nuclear reactor, from the power plant to the transmission line, from the substation to the data center, and from the electrical contractor’s estimating department to the job site.
For electrical contractors, the opportunity could be enormous. But so will be the responsibility. These projects will require accurate estimates, disciplined project management, experienced workers, and the ability to execute complicated electrical systems on demanding schedules.
The AI revolution may have started with computers, but computers don’t run without electricity. And electricity doesn’t get to those computers without an enormous physical infrastructure built by people.
If you want to understand where the electrical construction industry is going, follow the megawatts.
Sources & Further Reading
Goldman Sachs Research — US Data Center Power Demand Projected to Double by 2027
Hanwha Data Centers — What Are the Power Requirements for AI Data Centers?
U.S. Department of Energy — Advanced Small Modular Reactors (SMRs)
NuScale Power — The NuScale Power Module
Sprott — Uranium market analysis, July 2026
Quartz — AI companies are recruiting electricians and carpenters by the thousands to build data centers
U.S. Energy Information Administration — Uranium Marketing Annual Report
