AI Is Reindustrializing Computing

For much of the digital era, computing appeared to be separating economic activity from the physical constraints of the industrial economy. Software could be replicated almost without cost, businesses could move operations into the cloud, and digital platforms could serve global markets without building factories in every country they entered. Some of the world's most valuable companies consequently seemed increasingly detached from the steel, concrete, turbines, pipelines and power stations that had defined earlier periods of economic expansion.

Artificial intelligence is complicating that picture because producing machine intelligence at scale requires an extensive industrial system beneath the software through which users access it. Advanced semiconductors, enormous data centers, cooling systems and transmission infrastructure must be manufactured, financed and built by skilled workers, then supplied with reliable electricity throughout their operating lives. The apparent abstraction of the service conceals a growing dependence on physical capacity, with electrical power becoming an input into the production of cognitive work.

The economic relationship can be expressed as Energy → Compute → Intelligence → Productivity, provided that the sequence is understood as a description of productive capability. Machines supplied with electricity can perform increasingly useful analysis, classification, prediction, optimization, generation, simulation and automated reasoning. As organizations incorporate those capabilities into their operations, the availability and cost of the infrastructure producing them begin to influence how much cognitive work an economy can afford to deploy.

From Industrial Power to Cognitive Power

Industrial societies have long converted energy into productive capability through electricity that runs factories, petroleum that moves people and goods, natural gas that supplies heat and industrial feedstocks, and nuclear fission that produces continuous electrical generation from small quantities of fuel. These energy systems have amplified physical labor by allowing individual workers to command machinery capable of shaping materials, moving thousands of tons of freight or excavating ground on a scale far beyond unaided human effort. Their economic value comes from the productive capacity they make available throughout the economy.

Artificial intelligence extends that principle into activities previously associated almost entirely with human cognitive labor. A GPU cluster supplied with electricity can analyze images, generate computer code, search large information spaces and identify patterns, supporting work as varied as engineering design, financial analysis, biological modeling and logistics optimization. As these capabilities improve, organizations gain a means of expanding portions of their analytical and operational capacity through investment in computational infrastructure.

A company seeking additional expertise has traditionally needed to recruit, educate, compensate and organize more people, each of whom brings knowledge and judgment that remain essential. Machine intelligence introduces a complementary route: organizations can purchase computational capacity and use it to multiply the output of their existing workforce. When that approach spreads across an economy, electricity availability begins to influence the scale and cost of the machine cognition available to its businesses and institutions.

The Physical Economy Beneath AI

The vocabulary surrounding computing often obscures its material requirements, particularly when remote infrastructure is described as “the cloud.” Those services depend on physical facilities filled with server racks, connected through telecommunications networks and supplied continuously with electricity. AI intensifies that dependence by increasing the computational work those facilities must support and the electrical capacity required to sustain it.

The International Energy Agency estimates that global data-center electricity consumption reached approximately 485 terawatt-hours in 2025, with its current central projection placing consumption at roughly 950 TWh by 2030. Within that total, electricity consumption at AI-focused data centers is projected to triple over the same period. The implications depend heavily on where this demand develops, because large AI facilities can impose loads comparable to major industrial operations while clustering within the same regional power market. The IEA has estimated that roughly 20 percent of planned data-center projects could face delays if grid constraints are not addressed.

In the United States, data-center expansion is contributing to renewed electricity-demand growth after years of comparatively stagnant consumption. The U.S. Energy Information Administration identified that expansion as an important driver in 2026, while its September forecast projected national electricity consumption reaching new records in both 2026 and 2027. A 2026 Lawrence Berkeley National Laboratory update prepared for the Department of Energy estimates that data centers could account for approximately 11.8 percent of total U.S. electricity consumption by 2030 under its central estimate, with scenarios ranging from 9.5 to 15.3 percent. At that scale, decisions about computing investment have consequences for infrastructure planning well beyond the technology sector.

AI Is Reindustrializing the Digital Economy

The internet encouraged an economic narrative centered on dematerialization as information displaced physical media, software replaced physical processes, and digital communications allowed more commerce to take place across geographic boundaries. AI services retain that digital character, but their production requires an increasingly capital-intensive industrial foundation. The expansion of machine intelligence therefore connects the apparent weightlessness of software to the much slower work of building factories, electrical systems and computing facilities.

That dependence extends through several layers of production: data centers require servers built around advanced chips, whose manufacture depends on semiconductor fabrication facilities supported by supply chains for chemicals, precision equipment, minerals, water and energy. The data centers themselves need transformers, substations, transmission connections, cooling and backup systems, networking equipment, suitable land and construction capacity. AI consequently adds demand throughout the industrial economy, linking the deployment of new software capabilities to the availability of equipment and infrastructure that cannot be replicated at software's speed.

This changes the economics of expansion for services whose additional users once imposed relatively modest marginal infrastructure requirements. AI inference, particularly advanced reasoning, agentic systems and video generation, can require substantial computation each time the service is used. Although hardware efficiency is improving rapidly, the IEA reported that data-center electricity use increased 17 percent during 2025 while consumption at AI-focused facilities increased approximately 50 percent. More efficient equipment has therefore accompanied continued growth in aggregate electricity demand, reflecting a familiar industrial pattern in which lower costs make more applications economically viable and encourage wider use.

Electricity Becomes a Competitiveness Variable

The growing importance of power availability changes how regions compete for technology investment. Two metropolitan areas may offer comparable universities, educated workforces, available land, transportation access and business incentives, yet present very different development prospects if one can deliver hundreds of megawatts of reliable power while the other faces transmission congestion, transformer shortages and years of interconnection delays. A project that appears attractive on conventional economic-development measures may remain impractical because its electrical requirements cannot be met within a workable schedule.

Available electrical capacity thus begins to function like another form of industrial real estate, with the value of a potential site depending on whether adequate power can reach it at a competitive price. Economic-development agencies, municipalities and states seeking AI infrastructure must increasingly account for generation capacity, transmission availability, interconnection schedules, grid resilience and access to dispatchable power. These conditions help determine whether investment can proceed and whether the resulting facility can operate economically once it is built.

Energy and infrastructure policy consequently become part of the environment in which technology businesses develop. A region's ability to attract computational investment depends partly on whether its utilities, permitting institutions and infrastructure plans can support the scale and timing of the proposed demand. Regional competitiveness increasingly includes the practical ability to deliver electricity where companies need it, when they need it and at prices that support the underlying business.

Intelligence Scarcity Can Become an Economic Constraint

The implications extend to organizations that use AI without owning the facilities that produce it. Cloud computing allows businesses to purchase computational resources remotely, just as international trade allows countries to obtain technological inputs they do not manufacture domestically. Local generation capacity therefore does not directly determine every organization's access to machine intelligence, but constraints across the infrastructure supplying those services can still affect their cost and availability.

As machine intelligence becomes embedded in ordinary operations, those differences can influence productivity across a broad range of activities. Manufacturers may apply it to predictive maintenance and quality control, financial institutions to fraud detection and risk analysis, and healthcare organizations to documentation, scheduling and clinical decision support. Comparable uses in transportation routing, autonomous operation, government administration and infrastructure management make machine cognition another input into production and service delivery. Organizations facing expensive or constrained access may find it harder to achieve the gains available to competitors that can deploy the same capabilities more extensively.

Energy scarcity can restrict the expansion of compute, raising the cost of machine intelligence and limiting its use in processes where it could improve productivity. If those constraints persist, the resulting differences in productive capacity can affect wealth and competitiveness. Institutions, capital, talent, regulation, entrepreneurship, technological development and market structure continue to shape economic outcomes, but electrical capacity is becoming one of the physical constraints governing how extensively economies can deploy machine cognition.

A New Kind of Industrial Geography

The concentration of talent, capital and research institutions that shaped traditional technology clusters remains important, particularly for model development and high-value engineering. Computational infrastructure adds a different locational requirement because it depends on access to abundant power. Regions with natural gas resources, nuclear or hydroelectric generation, strong transmission systems or favorable electricity economics may therefore acquire advantages that carried less weight during earlier stages of the software economy.

These requirements can bring technology investment into places whose strengths developed around manufacturing and other industrial activity. Regions seeking to rebuild their economies may already possess some of the electricity supply, transportation networks, industrial land, engineering expertise and utility systems that computational infrastructure requires. Their prospects depend on how well those assets can support new investment, making an assessment of existing capacity and constraints more useful than an assumption that AI activity will remain concentrated in established technology centers.

Economic-development strategies will need to account for the connections among manufacturing, technology, energy and infrastructure that such investment creates. A region's ability to support an advanced industrial operation may also help it accommodate a computing facility, even when it lacks the concentration of software companies associated with a conventional technology hub. Policymakers and investors should evaluate where the physical foundation of the AI economy can be built alongside their assessment of where its research and development will occur.

Capital Will Follow the Bottlenecks

Semiconductor designers, cloud platforms and model developers have been among the most visible beneficiaries of AI expansion because they supply the products and services most directly associated with the technology. The infrastructure required to sustain that expansion broadens the field of economic activity supporting it. Demand for computing can create demand for electrical equipment, generation, construction and engineering capacity, connecting the investment cycle to businesses far beyond the software industry.

Transformers and transmission equipment are essential to connecting new facilities, while markets adding dispatchable generation may also require gas turbines or investment in nuclear plants and their supply chains. Cooling equipment, electrical switchgear, backup systems, copper, engineering services and construction labor support the same expansion at different points in the production process. The IEA has identified tightening supplies of gas turbines, transformers, advanced chips and other equipment as actual constraints on the pace of data-center deployment, demonstrating how shortages in established industrial markets can limit the growth of advanced computing.

When physical bottlenecks delay otherwise viable projects, investment has an incentive to expand the capacity needed to relieve them. Evaluating the AI investment cycle therefore requires attention to the availability and delivery schedules of industrial equipment alongside developments in models and software. Power generation, electrical manufacturing and regional infrastructure form part of the same economic system, with constraints in any of them capable of affecting how quickly new computational capacity becomes productive.

The Strategic Dimension

The industrial requirements of AI also connect economic competition to national capability. The United States and China are expected to account for nearly 80 percent of the increase in global data-center electricity consumption through 2030, while the IEA projects U.S. consumption increasing approximately 130 percent from 2024 levels by the end of the decade and China's increasing around 170 percent. Those projections reveal the scale of the physical expansion associated with competition between the two countries and the infrastructure each must be able to support.

Developing capable models is one element of a broader effort that includes semiconductor production, electrical generation, grid expansion, data-center construction, financing, engineering talent and telecommunications networks. National capability also depends on how effectively those resources can be organized to turn machine intelligence into productive and military applications. A country able to develop advanced models but unable to build the power infrastructure needed to operate them at scale faces a constraint that model improvements alone cannot resolve.

The competition consequently has the character of an industrial race extending across the institutions and supply chains that make large-scale computing possible. The capacity to expand generation, connect facilities, obtain equipment and finance construction influences how much of a technical advantage can be put to work. Understanding AI within that larger system provides a fuller account of national capability than comparisons confined to laboratory performance.