Artificial Intelligence (AI) workloads, with their breathtakingly rapid progress on large language models and generative AI, are housed in specialized data centers that display energy consumption patterns markedly different from traditional facilities or even a typical household. Understanding the scale and speed of data center demand—and crucially, its impact on power grids, a much-lamented sore point in energy transition—has become important in navigating the climate transition, although it doesn’t fully account for diverse AI applications outside data centers. Likewise, it is helpful to imagine how AI could aid the energy transition.

The Age of Electricity

Global Electricity Demand

We live in an age increasingly defined by electricity. Global electricity demand surged by an unprecedented 4.3% in 2024, growing faster than wider energy demand and global GDP (3.2%), according to the IEA’s Electricity 2025 report. The absolute increase of nearly 1,100 terawatt-hours (TWh) was more than twice the annual average increase over the past decade and the largest ever outside economic rebound years. This rapid growth is forecast to continue, with the IEA projecting annual growth rates of 3.9% through 2027. Over the next three years alone (2025-2027), global electricity consumption is forecast to rise by an unprecedented 3,500 TWh.

Power Sector Share of CO2 Emissions

This rising demand has direct consequences for CO2 emissions. Power generation remains the largest single source of emissions globally, according to the IEA. Data from the IEA’s Electricity 2025 and Global Energy Review 2025 reports show power generation accounted for approximately 13.8 billion metric tons (Gt) of CO2 in 2024, representing roughly 36.5% of total global energy-related CO2 emissions (37.8 Gt). While global CO2 emissions from the power sector rose by 1.0% in 2024, after increasing by 1.4% in 2023, the IEA forecasts that these emissions are expected to stay relatively flat (-0.1% annual average growth rate) between 2025-2027, thanks to substantial growth in clean energy sources, even amidst rising demand.

Data Centers: Exponential Growth

AI Energy Footprint

Within this picture of rising global power demand, data centers are an (as yet) insignificant but rapidly growing component, especially when compared to powerhouse sectors like electric vehicles (EVs) and air conditioning. In 2024, data centers accounted for just 1.5% of the world’s electricity consumption, or 415 TWh, according to the IEA’s Energy and AI report. This overall consumption is comparable to the airline industry’s energy use, or roughly half that of all household IT appliances.

But this has the potential to skyrocket. IEA pegs the growth rate (conservatively) at 15% per year, more than four times faster than all other sectors. Data center demand will double to 945 TWh by 2030, representing just under 3% of total demand. AI is a key driver of this growth. Even at the typical user’s end, a ChatGPT request consumes an estimated 2.9 watt-hours (Wh), roughly 10 times the 0.3 Wh for a standard Google search.

The energy sector is paying attention. It is prompting a significant expansion of fossil fuel power generation in the U.S., with scores of new gas-fired power plants a direct response to the recent surge in demand after a period of stability.

AI Carbon Footprint

Pinpointing the exact CO2 emissions attributable solely to AI within the broader data center footprint, or the above planned fossil fuel expansion, is challenging. However, the overall trend for data centers is clear. The IEA’s Energy and AI report projects emissions from total data center electricity use alone rising from approximately 180 million metric tons (Mt) of CO2 today (implied 2025) to 300 Mt CO2 by 2035 in its Base Case, potentially hitting 500 Mt CO2 in a high-growth scenario. While a fraction of total global energy emissions, the growth is significant.

AI Grid Footprint

What’s more, the growth impact isn’t evenly spread. The U.S. dominated and accounted for the largest share of global data center electricity consumption in 2024 (45%), followed by China (25%) and Europe (15%).

Nearly half of U.S. capacity is clustered in just five hubs, with Northern Virginia alone boasting 5.9 gigawatt (GW). This concentrated, high-density demand strains electricity grids, a sore point in the world’s fight for climate justice.

Most grid infrastructure wasn’t designed for a 100-1000 megawatt (MW) data center-sized load. A major bottleneck is grid connection and expansion. Planning and building new high-voltage infrastructure is notorious for taking 5-10 years or more, depending on location, far slower than the 1-2 year deployment cycle for data centers. This lag risks delaying not only data centers (up to 20% of projects, per IEA) but, more crucially, the integration of necessary renewable generation. Global grid investment needs to nearly double by 2030 to over $600 billion annually to cope.

Ireland: Where All Pressures Collide

Data centers thus leave an impact on energy consumption, CO2 emissions and the grid. Ireland exemplifies these pressures. Data centers, supported by foreign investment in IT hubs dotted around the country, consumed a staggering 21% of Ireland’s electricity in 2023, up from 5% in 2015. Projections from EirGrid, the national grid operator, suggest this could hit 31% by 2027. Critically, data center demand growth since 2017 has absorbed all additional wind energy generation, preventing fossil fuel displacement. Grid strain is evident: “amber alerts” (tight supply warnings) have increased, and EirGrid warns of a “significant power ‘generation’ deficit”. This has led to a de facto moratorium on new Dublin-area connections until at least 2028.

Sustainable Data Centers

Although diverting new renewable power to data centers might not be ideal, powering data centers sustainably is still paramount. The EU’s recast Energy Efficiency Directive (EED) policy places significant emphasis on the energy performance and sustainability of data centers. Leading tech companies are pursuing 100% carbon-free energy (CFE) goals, often through Power Purchase Agreements (PPAs).

Google’s Hamina data center in Finland uses cold seawater for cooling, famously cutting its energy needs. It achieves very high hourly CFE (98% in 2023 in its grid region) through PPAs and Finland’s clean grid mix. However, replicating Hamina globally faces challenges: its seawater cooling is geographically limited, and achieving high CFE relies on regional renewable availability and grid conditions.

But AI could come to the rescue, and there’s a clear push for AI applications to optimize energy use in data centers and for more precise, that is, hourly, CFE matching.

AI to the Rescue

Paradoxically, AI itself offers solutions to the energy challenges it exacerbates. While AI data centers strain resources, AI technology can optimize renewable energy infrastructure and grid operations. This is especially true in predictive maintenance, supply/demand forecasting and grid integration.

AI, especially machine learning, is transforming offshore wind operations and maintenance (O&M) by enabling predictive maintenance through sensor data analysis, optimizing drone inspections for blade damage, and ultimately increasing turbine availability, reducing costs for companies facing harsh offshore conditions and high maintenance costs.

The application of AI in solar energy could also provide tangible improvements when it better accurately predicts PV power output in real time than a typical model, and when it detects faults in PV panels early. It has the potential to maximize energy yield and grid integration, but all this requires high-quality, real-time data input.

Yet, what excites energy market participants the most is its ability to enable smart grid capabilities. AI can ease grid congestion through Dynamic Line Rating (DLR), a tech that unlocks unused latent capacity, for example. AI can also group together local, dispersed clean energy sources like solar panels and batteries into virtual power plants (VPPs) that can help the electricity grid by reducing strain or adding power where and when needed.

However, replicating these solutions faces hurdles that have yet to be resolved.

AI Energy Paradox

AI presents what the World Economic Forum calls a stark “energy paradox.” Its data centers are huge power consumers, straining electricity grids and potentially hindering decarbonization. Yet, AI simultaneously offers powerful tools to optimize renewable energy generation and manage grids more efficiently.