The thirst for energy in the age of AI

M

odern economies rely on vast networks of digital and energy infrastructure. These systems are becoming increasingly interconnected as artificial intelligence grows in importance for both the global economy and the energy sector. AI models are trained and operated in large, power-intensive data centers, which in turn rely on steady energy supplies from electricity grids and continuous data flow through fiber optic networks. Ensuring sufficient domestic computing capacity is now a key policy goal for many governments —an objective made more urgent by current geopolitical and trade tensions.

Cutting edge artificial intelligence models, such as ChatGPT, are too large and complex to run on laptops and phones. Instead, they are trained and run in large, specialized data centers. As artificial intelligence emerges as a potentially transformative general-purpose technology, investment in these facilities has surged. According to the International Energy Agency (IEA), more than USD 500 billion is expected to be invested in data centers globally in 2025 —equivalent to nearly 0.5 percent of global GDP or about 2 percent of total global investment.

Data centers are growing significantly in scale. Prior to the AI boom, a typical data center had a capacity of 10–20 megawatts (MW). Today, hyperscale facilities commonly reach 100 MW. The largest currently under construction is expected to reach 2,000 MW, with some proposed centers aiming for as much as 5,000 MW. These are enormous figures —by comparison, a typical aluminum smelter, one of the most energy-intensive industries, consumes around 700 MW.

The surge in electricity consumption

The rise of larger, more power-intensive data centers has driven a sharp increase in electricity consumption across the sector. The IEA estimates that data centers now consume around 415 terawatt-hours (TWh) of electricity globally —more than the total electricity use of Italy. While this accounts for only about 1.5 percent of global electricity consumption, it has been growing rapidly —at an average annual rate of 12 percent over the past five years, more than four times the growth rate of overall electricity demand.

Meeting the rapid growth in data center electricity demand is already straining power systems. This challenge is compounded by several characteristics that make data centers a distinct type of energy infrastructure.
First, data centers tend to cluster —often around fiber optic hubs, near concentrations of digital service demand such as financial centers, and in areas with a skilled workforce to build and operate them. Second, these clusters are typically located near cities, which have large, digitized, service-oriented economies. Proximity also helps meet the needs of customers who prioritize low latency, requiring data centers to be physically close.

Finally, data centers can be built quickly —sometimes in as little as 18 months— much faster than major energy infrastructure like power plants or transmission lines. In contrast, building a transmission line capable of serving a hyperscale data center can take 4 to 8 years in advanced economies.

For these reasons, even though data centers account for a small share of global electricity use, their local impact can be far greater. In six U.S. states, data centers already consume more than 10 percent of total electricity. In Ireland and Virginia, that share exceeds 20 percent. In some jurisdictions, grid constraints have led authorities to impose restrictions on new data center development.

Looking forward, the IEA projects that data center electricity consumption will more than double by 2030, reaching around 945 TWh —exceeding Japan’s current total electricity use. While data centers will still represent a relatively small share of global electricity consumption, they are expected to be major drivers of demand growth in certain regions. In the United States, for example, data centers are projected to account for nearly half of total electricity consumption growth through 2030.

Electricity grids in advanced economies are aging, with more than half of transmission and distribution lines in Europe and the United States now over 20 years old. These older grids also tend to lag in adopting smart technologies, reducing their efficiency and resilience. At the same time, supply chains for grid equipment are under pressure. Average lead times for cables and large power transformers have nearly doubled since 2021, with transformer prices rising sharply over the same period. Similar constraints are affecting generation equipment —wait times for gas turbines now stretch to 5-7 years, and prices have surged.

The complexity of supply chains for the IT components

Trade adds another layer of complexity. The supply chains for IT components used in data centers are highly globalized and intricate. The same is true for the power electronics, cables, backup batteries, and transformers that keep data centers running —these rely on international supply chains for both critical raw materials and finished products.

For instance, the IEA estimates that data center demand for gallium —a key metal used in advanced power electronics and semiconductors— could reach the equivalent of 10 percent of current global production. Yet the supply chain for gallium is extremely concentrated, with a single country responsible for 99 percent of refined output.

Policymakers, investors, the energy sector, and the tech industry now face a highly complex and uncertain landscape. Much of this uncertainty stems from the rapid evolution of a new technology. The uptake of AI could either fall short of or exceed current expectations. While AI hardware and models have already made significant gains in computational and energy efficiency —and further breakthroughs are possible— many tech firms are also deploying increasingly complex models that require greater computational power and so more energy. Broader macroeconomic and trade conditions will also play a role in shaping future investment in data centers.

The IEA’s projections for data center electricity consumption in 2030 vary widely, with the highest scenario reaching 1,265 TWh and the lowest at 670 TWh —a difference of nearly twofold. Cumulative investment in data centers through 2030 ranges from USD 2.5 trillion in the low scenario to USD 6.5 trillion in the high scenario.
If the energy sector fails to expand quickly enough, it could slow AI development or force trade-offs with other electricity-intensive industries such as manufacturing reshoring or decarbonization targets. Conversely, if investment lags behind actual demand, substantial amounts of capital could be misallocated.

A roadmap for policymakers

As policymakers pursue goals like building domestic digital infrastructure and advancing decarbonization, they must navigate an increasingly complex landscape shaped by the growing interdependence of digital and energy systems.

First, stronger dialogue is needed between the energy sector and the tech industry to support better planning and coordination of infrastructure development. Major tech firms should be more transparent about their current and projected electricity use. Second, “no-regrets” investments in modernizing electricity systems —making them more digitalized, flexible, and resilient— are more urgent than ever. These improvements will help accommodate not just data center growth, but rising demand from other sectors as well. Third, both policymakers and tech companies should explore strategies and incentives to maximize the efficiency and grid flexibility of data centers, reducing their strain on power systems. Fourth, recent steps to shorten permitting timelines for energy infrastructure will need to be expanded and reinforced. Fifth, securing local community support for data center development will require clear frameworks for benefit sharing and fair allocation of costs, along with sustained engagement.
 

The energy sector is central to one of today’s most significant technological shifts. There is no AI without energy. But no one actor can meet this challenge alone, and focusing solely on one aspect —such as ramping up power generation— will not be enough. What’s needed are smart, flexible, and collaborative strategies to navigate this new map.