Hidden geography of AI: Why electricity is becoming strategic advantage

The global expansion of AI is being shaped not only by software innovation and national digital strategies, but by the physical systems that generate power, host servers, and sustain large-scale computation, according to a new study published in AI & Society. The findings place energy infrastructure and data center geography at the center of the global AI economy, raising new questions about who will control the next phase of digital development and who may be left behind.

The study, “Energy Creates Intelligence: The Infrastructural Geography of Global AI,” examines how energy consumption, AI readiness, and data center deployment interact across countries and regions. The paper notes that AI is not simply a digital phenomenon, it is an infrastructural one, rooted in electricity systems, data center networks, and spatial planning decisions that determine where computation can be built, scaled, and governed.

Energy is emerging as the hidden foundation of AI power

Energy plays a dual role in shaping AI infrastructure. First, it has a direct effect. Data centers are electricity-intensive facilities, and countries with stronger energy systems are better positioned to host them. Second, energy also has an indirect effect through national AI readiness. Countries with greater energy capacity tend to be better prepared institutionally and technologically for AI deployment, which in turn helps support data center expansion. In the authors’ framework, electricity is not just an input. It is part of a broader developmental system that enables digital intelligence to scale.

This matters because discussions of AI readiness often focus on digital policy, research investment, education, and startup ecosystems. The author show that these factors do not operate in isolation. A country may pursue ambitious AI plans, but without sufficient energy infrastructure and the ability to support data centers, those ambitions may not translate into durable computational capacity. The study suggests that the path to AI leadership runs through the power grid as much as through the research lab.

The authors describe this pattern as an infrastructure divide, a concept meant to go beyond the familiar idea of the digital divide. The traditional digital divide usually refers to unequal access to devices, internet connections, or digital services. The infrastructure divide is deeper and more structural. It concerns the unequal ability to host, operate, and control the physical systems that power AI itself. In this view, countries can appear digitally connected while still lacking the infrastructural base needed to shape the AI economy on their own terms.

That distinction is significant because AI systems are becoming more energy-intensive and more centralized at the same time. Training large models, running inference at scale, and supporting growing cloud demand all increase the importance of reliable electricity and physical hosting capacity. Countries without those systems may still use AI applications produced elsewhere, but they may have limited influence over the underlying platforms, standards, and value chains. The author’s analysis suggests that the future geography of AI may therefore reproduce and intensify older infrastructure inequalities rather than simply overcoming them through digital access alone.

The findings also point to a political shift in how AI development should be understood. If energy availability shapes AI deployment, then AI policy can no longer be treated as separate from energy policy. Decisions about generation capacity, grid reliability, transmission, land use, and regional planning will increasingly influence which countries can expand computational infrastructure. This turns energy ministries, urban planners, and infrastructure regulators into important actors in the AI landscape, not just technology ministries and innovation agencies.

Data centers are mapping a new geography of inequality

The study analyses where data centers are actually located within countries. Across the 34 countries examined at the subnational level, the distribution of data centers varies sharply. In some countries, the infrastructure is relatively decentralized, spread across multiple regions. In others, it is highly concentrated in a few metropolitan hubs. These patterns matter because they reveal whether AI infrastructure is being integrated into national territory broadly or clustered in narrow corridors of power, investment, and connectivity.

The study finds that countries such as the United States and Canada tend to show more decentralized data center patterns. The authors link this to factors such as land availability, renewable energy access, cooling conditions, and cost optimization. In these cases, infrastructure can be distributed beyond a single urban core because the underlying energy and logistical systems are strong enough to support multiple nodes. That allows national AI infrastructure to scale across a wider geography rather than concentrating only in primary cities.

On the other hand, many countries at earlier stages of digital development show stronger concentration around major urban centers. This is where telecommunications links, investment, technical labor, and institutional support are most likely to be found. In these environments, data centers cluster where infrastructure is already dense, reinforcing the primacy of leading cities and leaving wider territories less integrated into the AI economy. The pattern suggests that AI infrastructure is not spreading evenly. It is following existing concentrations of economic and infrastructural advantage.

The paper makes clear that these differences are not explained by economic size alone. Countries with similar levels of income can show different data center geographies depending on their energy systems, planning institutions, and overall AI readiness. Western European countries, for example, often show more distributed structures supported by stronger regional grids and more balanced planning frameworks. Lower-income countries, on the other hand, more often face a double constraint: limited energy infrastructure and limited spatial decentralization of digital assets.

That has consequences well beyond the data center industry itself. If AI infrastructure remains concentrated in a small group of countries and subnational hubs, so too may the capacity to shape the development of large-scale AI. Control over hosting, storage, and processing confers strategic advantages in platform development, cloud services, and the governance of digital ecosystems. The author’s research suggests that the geography of data centers is becoming a geography of digital sovereignty.

The findings also complicate optimistic assumptions that the spread of data centers automatically brings inclusive digital development. Even when facilities are built in developing regions, the economic and strategic benefits may not remain local. If ownership, service provision, and data flows are controlled by external corporations, local communities may host the energy burden and land footprint of AI infrastructure without capturing equivalent gains in technological autonomy or domestic digital ecosystem growth. The paper warns that AI expansion may deepen uneven development if hosting capacity grows without broader local control.

This is especially relevant as countries court AI and cloud investment as part of economic modernization strategies. Data centers can bring construction activity, connectivity upgrades, and some employment, but they can also intensify pressure on local energy systems and reinforce external dependence if national policy does not connect hosting growth to domestic capability building. In the authors’ analysis, infrastructure matters not only because it enables AI, but because it shapes who benefits from it.

Why AI policy is becoming energy policy

The future of AI development will increasingly be shaped by decisions that sit outside the traditional boundaries of digital policy. Energy grids, data center siting, spatial planning, and infrastructure governance are becoming central to the politics of AI. The authors argue that this means countries seeking a stronger role in the AI economy must think beyond software and regulation and toward the physical conditions that allow digital systems to exist.

That reframing could prove important for national governments now drafting AI strategies. Many policy documents emphasize innovation ecosystems, education, responsible AI frameworks, and international competitiveness. Those are still critical, but the paper suggests they are incomplete if they do not address electricity supply and infrastructural capacity. A country may be able to adopt AI services without hosting the systems behind them. But if it wants to produce, control, and sustain AI domestically, it needs the power systems and spatial infrastructure to do so.

The work also suggests a new lens for evaluating global AI competition. Instead of focusing only on headline model releases or chip export controls, analysts may need to track energy buildout, grid resilience, and data center distribution as indicators of long-term AI capacity. Countries that can align energy expansion with digital planning may gain structural advantages over those whose AI ambitions outpace their infrastructural base.

There is also a sustainability dimension. AI’s growing electricity demand is already raising concerns about emissions, water use, and local energy stress. A more infrastructure-centered view of AI makes those issues harder to treat as secondary. If energy systems create intelligence, then the environmental and spatial costs of that intelligence become part of the core AI policy debate. That means future decisions about where to build data centers, how to power them, and who bears the cost of that expansion will become increasingly contested.

For lower-income countries, the paper highlights a difficult strategic choice. They risk exclusion if they do not develop the energy and digital infrastructure needed for AI. But they also risk a form of dependent inclusion if infrastructure is built mainly to serve external firms without strengthening domestic capabilities. The authors' concept of the infrastructure divide captures this tension. Participation in the AI era is not only about access to tools. It is about ownership of the systems that make those tools possible.