This is an extract from a recent report “Energy and AI” by IEA.
Investment in new data centres has surged, increasing by nearly 70% in the last two years at the global level. One of the main drivers of this investment has been the rise of artificial intelligence (AI), alongside the deepening digitalisation of the global economy. The rapid increase in data centre investment is raising concerns about the ability of electricity systems to meet growing demand in a timely, secure and sustainable way. Data centres – at least at the scale seen today – are relatively new actors in the energy system at the global level, and data collection and reporting on their electricity consumption remain limited. There is therefore substantial uncertainty about both their current and future consumption. Moreover, AI models are highly heterogeneous, and data on their uptake and electricity intensity are limited. As a result, it is challenging to analyse the link between AI demand and data centre electricity consumption. On the electricity supply side of the equation, the sector is facing several challenges. Electricity demand is already growing strongly in emerging markets and developing economies, driven especially by economic growth, industrialisation, increased adoption of appliances, and surging needs for cooling. Advanced economies are also returning to growth in electricity demand after two decades of stagnation. However, the electricity sector faces several bottlenecks, including permitting times and tangled supply chains.
Electricity consumption of data centres
The Internet revolution took off in the 1990s, and early growth in the demand for digital services was strong. The electricity consumption of data centres in the United States almost doubled between 2000 and 2005, raising concerns about runaway growth. An inflection point occurred around 2007-2008, when slowing growth in data centre electricity consumption indicated a decoupling from the still-booming demand for digital services. Several factors contributed to this slowdown in global data centre electricity consumption, including the migration of service demand to more efficient, larger data centres (colocation, service provider and hyperscale), but also continued improvements in hardware efficiency and operating efficiency (declining idle power ratios, for example).
However, a sharp acceleration in data centre electricity consumption took place from around 2017 onwards. Important drivers of this step change were the growth of cloud computing, the shift to online media consumption, the wider use of social media platforms and the rise of AI, which increased the demand for high-performance computing, facilitated by the rise of accelerated servers. Between 2015 and 2024, the capacity of accelerated servers grew four times faster than the total capacity of servers. While accelerated servers are much more efficient on a per-task basis, they also unlocked many new tasks, that were not possible on conventional servers. These new capabilities, among other factors, drove an increase in service demand that outstripped the pace of continued efficiency improvements.
From 2005 to 2015, global Internet Protocol (IP) traffic, mobile broadband subscriptions and active social media accounts grew by more than 25% per year. These are proxies for the rapid initial growth in demand for digital services. Growth rates moderated in the period from 2015 to 2023. In contrast, the growth rate of the total stock of servers in data centres accelerated from an annual growth rate of 4% seen in the period 2005 to 2015 to 8% per year from 2015 to 2023. Several key indicators of efficiency saw faster improvements from 2005 to 2015, including the rate of the shift from less-efficient enterprise data centres to more-efficient hyperscale, colocation and service provider data centres. As a result of these trends, data centre electricity consumption growth accelerated from 3% annually from 2005 to 2015 to 10% annually from 2015 to 2024.
Data centre electricity consumption is not spread evenly around the world
The United States, Europe and China account for around 85% of global electricity consumption from data centres today. In the United States, electricity consumption from data centres grew by around 12% a year between 2015 and 2024. Data centres accounted for around 180 TWh of electricity consumption in 2024 in the United States, nearly 45% of the global total and more than 4% of US electricity consumption from all sources. In China, the data centre sector started to expand significantly from 2015 onwards, with electricity demand growing 15% per year between 2015 and 2024, more than twice the rate seen between 2005 and 2015. Over the same period, electricity consumption across all sectors grew at an annual rate of around 7%. As of today, data centres account for approximately 100 TWh of electricity consumption, roughly equivalent to that of electric vehicles in China. The country accounts for around 25% of global data centre electricity consumption, up from less than 20% a decade ago. However, substantial data gaps make it challenging to accurately estimate China’s data centre electricity consumption.
Data centres account for slightly less than 2% of Europe’s electricity consumption, a share that is higher than China’s (1.1%). However, in absolute terms, Europe’s consumption is lower, at an estimated 70 TWh in 2024. Europe’s share of the global electricity consumption of data centres has decreased over the past decade but still represents slightly above 15%. In Japan, it is estimated that data centres account for less than 20 TWh of electricity consumption (about 2% of Japan’s total consumption, on a par with Europe). It is estimated that data centres account for around 9 TWh of consumption in India, or about 0.5% of total consumption. However, the sector appears poised for rapid growth.
Outlook for electricity consumption from data centres
Global electricity consumption by data centres is projected to reach around 945 TWh by 2030 in the Base Case, representing just under 3% of total global electricity consumption in 2030. This is more than double the estimated approximately 415 TWh for 2024, which accounted for around 1.5% of today’s global electricity demand. From 2024 to 2030, data centre electricity consumption grows by around 15% per year, more than four times faster than the growth of total electricity consumption from all other sectors. However, in the wider context, a 3% share in 2030 means that the data centre share in global electricity demand remains limited. Electricity consumption in accelerated servers, which is mainly driven by AI technology adoption, is projected to grow by 30% annually in the Base Case, while conventional server electricity consumption growth is slower at 9% per year.
The United States, China and Europe are projected to remain the largest regions for data centre electricity demand over the coming years. However, other regions are experiencing strong growth in data centre development, positioning them to play increasingly important roles in the global data centre landscape. A notable example is Southeast Asia, where electricity demand from data centres is expected to more than double by 2030, partially due to the presence of a regional hub in Singapore and southern Malaysia. China and the United States are the most significant regions for data centre electricity consumption growth, accounting for nearly 80% of global growth to 2030. Consumption increases by around 240 TWh (up 130%) in the United States, compared to the 2024 level. In China it increases by around 175 TWh (up 170%). In Europe it grows by more than 45 TWh (up 70%). Japan increases by around 15 TWh (up 80%).
Comparing data centre electricity consumption normalised per capita can give a sense of the importance of this sector in different economies. Africa has the lowest consumption at less than 1 kWh of data centre electricity consumption per capita in 2024, rising to slightly less than 2 kWh per capita by the end of the decade. However, there are strong differences within the region, with South Africa showing strong growth and per-capita consumption more than 15 times larger than the continental average in 2030, with an intensity higher than 25 kWh per capita. By contrast, the United States has the highest per-capita data centre consumption, at around 540 kWh in 2024. This is projected to grow to over 1 200 kWh per capita by the end of the decade, which is roughly as much as 10% of the annual electricity consumption of a US household. This intensity is also one order of magnitude higher than any other region in the world.
Implications of AI for ICT sector energy use
Data centres are part of the broader ICT sector, which also includes telecommunication networks and end-user devices such as laptops and smartphones. The implications of AI for energy use in data centres – and in the broader ICT sector – depend largely on how generative AI is adopted and deployed, both of which are highly uncertain. This section explores possible scenarios and their implications. Consuming around 360 TWh of electricity in 2023, data centres accounted for one-third of overall ICT sector electricity use, estimated at over 1 000 TWh2 in 2023, equivalent to 4% of global electricity use. Telecommunication networks, including fixed and mobile access and core networks, consumed around 280 TWh, while personal computers, mobile phones and other connected devices used around 440 TWh.
Cryptocurrencies and televisions are large energy users often associated with the ICT sector but are technically outside the sector scope according to definitions from the International Telecommunications Union. Cryptocurrencies – primarily from Bitcoin mining – consumed around 125 TWh in 2023 (0.5% of global electricity), while televisions, peripherals and cable television networks consumed around 500 TWh (2% of global electricity). Data centres have contributed most to ICT sector energy growth since 2020, increasing by over 90 TWh between 2020 and 2023. Energy used for cryptocurrency mining has also increased strongly, growing by over 50 TWh since 2020. Energy use by telecommunication networks has grown slightly, driven by strong growth in 5G mobile networks but partially offset by reductions in fixed networks from the switch from copper to fibre optic networks. Energy use by devices decreased in the early 2010s due to efficiency gains but has since increased, driven by the growth in the number of devices and new segments, such as the Internet of Things and surveillance cameras. There is considerable uncertainty around overall energy use by devices due to a lack of comprehensive data regarding use patterns and stocks.
Electricity supply to meet data centre demand
Procuring electricity supplies that are reliable and cost-effective is crucial to meeting the rapidly growing electricity demand from data centres. Many technology companies and large data centre operators have set ambitious goals for reducing emissions and procuring clean energy. To meet these objectives, data centre operators use various procurement strategies. These vary by company and region, with liberalised electricity markets generally offering more procurement choices than regulated markets. In addition to sourcing the grid electricity mix, procurement strategies include acquiring electricity through PPAs. Many companies also purchase renewable energy certificates to meet their clean energy targets.
The recent surge in data centre electricity demand has led to significant interest in additional natural gas-fired power generation, largely in the United States, where natural gas is a low cost fuel. Gas turbine manufacturers are reporting an uptick in orders, and several large data centre operators have announced partnerships with utilities and energy companies developing new gas-fired power capacity. In Louisiana, for example, Entergy Louisiana is planning more than 2 GW of additional gas-fired power generation to provide power for Meta data centres. NextEra Energy and GE Vernova also aim to develop natural gas-fired power generation projects across the United States, primarily to meet the growing electricity demand of data centres. At the same time, many US utilities are currently revising their integrated resource plans to account for rising data centre electricity demand, proposing additional natural gas-fired capacity to meet it. To bring down emissions, some data centre operators are considering fitting natural gas-fired plants with carbon capture in the long run.
Most renewable energy PPAs are financial agreements for annual volumes of electricity and are not tied to the hour-to-hour consumption profile of a data centre or the generation profiles of the renewable assets, which can also be located in different regions. While these PPAs help data centre operators meet their clean energy targets, the separation of renewable generation and data centre consumption often means other sources, like natural gas or coal, are used to meet physical electricity needs. This results in a physical electricity mix that differs from the procured, or “financial”, electricity mix. To enhance their sustainability strategies and further support decarbonised grids where they operate, some technology companies are concluding PPAs with hourly matching. This means that some or all of their electricity consumption is matched hour-by-hour by a portfolio of renewable energy and storage assets, or other types of low-emissions power generators located in the same region.
As part of these strategies, technology companies are also supporting the development and commercialisation of innovative low-emissions baseload technologies, such as small modular reactors (SMRs) and next-generation geothermal. To date, plans to build up to 25 GW of SMR capacity associated with supplying the data centre sector have been announced worldwide, almost all of them in the United States, although projects are at varying stages of maturity and certainty. The first projects are expected to start to materialise only towards the end of this decade. As an alternative to procuring electricity from utilities or through PPAs, some technology companies are co-locating data centres with power generation facilities, enabling them to generate some or most of their own electricity directly. The primary benefit of co-locating generation is potentially faster development times, as this approach can allow them to downsize or opt for an interruptible grid connection, saving costs and helping to alleviate grid congestion. The downsides are higher complexity, increased permitting requirements, higher investment costs, potentially lower reliability and a greater maintenance burden.
Recent years have seen rising interest in co-locating data centres and generation assets. Google is partnering with Intersect Power and TPG Rise Climate to develop co-located clean energy projects with data centres, aiming for completion by 2027. Chevron and Engine No. 1 are partnering with GE Vernova, with plans to supply up to 4 GW of natural gas capacity to co-located data centres, aiming to start operations by the end of 2027. Amazon and Talen Energy have signed a 10-year PPA for 300 MW to 960 MW of nuclear energy from the Susquehanna nuclear plant to supply a co-located data centre, although a recent Federal Energy Regulatory Commission ruling on the repurposing of existing grid-connected power plants to directly provide power to co-located loads halted plans to expand the electricity supply beyond the initially awarded 300 MW.
Matching electricity supply with data centre demand
Electricity supply to meet data centre demand can come from a wide set of sources, each with unique characteristics related to technical performance, cost, emissions, the development process and lead times. Consideration of these options, either to be developed onsite or connected through the grid, is critical to scaling up electricity supply to meet data centre demand.
As data centres are projected to grow rapidly over the years to come, the strategy to build out and ensure a stable and efficient source of electricity becomes crucial. Currently, the only reliable electricity sources that can be developed within a short timeframe – ideally one to two years – are solar PV and gas turbines, aligning with the typical construction timeline of data centres. Even in these cases, supply chain delays or tight supplies can further extend development times. Wind turbines could also be a viable option in terms of deployment speed; however, lengthy permitting processes often extend their timeline to around five years, a similar development time to conventional geothermal, or longer. Other dispatchable technologies, such as large-scale nuclear reactors or hydropower plants, typically require closer to a decade or more to complete. Once SMRs or next-generation geothermal become commercial, they may also offer medium-length development times of approximately three to five years.
Technology costs are another important factor in considering supply options to meet data centre demand. Wind and solar PV technologies are currently among the cheapest sources of electricity. Additionally, in regions where natural gas prices are low, such as the United States and the Middle East, gas turbines offer an alternative. To be comparable with dispatchable sources of electricity, solar PV and wind need to be paired with storage to increase their availability throughout the day, but the cost comparison remains valid. Coalfired power can be one of the lowest-cost sources of electricity in places where prices on CO2 emissions are low or zero, but development times for coal plants can be quite long outside China.
Emissions at the point of electricity generation are an important factor, especially in light of the sustainability targets set by many technology companies and national and international climate goals. Coal-fired power has the highest emissions intensity of the potential options (oil-fired power is of a similar level), with natural gas-fired power plants emitting roughly half as much CO2 per unit of electricity output. Excluding indirect emissions from their life cycle – such as extraction, manufacturing and decommissioning – renewable energy and other lowemissions sources like nuclear energy have no direct CO₂ emissions.
Hourly matching: What does it really take?
Hourly matching of the procured electricity supply to the data centre electricity demand is an approach pursued by several large technology companies, but achieving this ambition with variable renewables comes with challenges. Solar PV and wind generation are inherently variable. Solar PV varies across the day and seasons. Wind production is less variable on average but can vary quickly from hour to hour, with extended periods of low or high generation. However, hybrid projects combining solar PV, wind and storage offer a better match to baseload demand, with storage helping to smooth out variable output from renewables. Solar PV combined with battery storage has the advantage that it can be deployed quickly and provides a more constant supply. Combining solar PV, wind and battery storage results in an even flatter supply. To align with baseload demand on an hourly basis, the installed capacity of renewable sources must be higher than the average demand.
In order to analyse the ability of solar PV, wind and battery storage to meet baseload demand, a new analysis of over 1 000 use cases covering eight configurations in more than 100 regions was carried out. The regions include European countries, each state of the United States and each province of China. Different procurement strategy configurations were tested, resulting in different combinations of renewable and storage technologies. The remaining portion of electricity demand not covered by renewable sources was assumed to come from the grid at the average industry retail price. The analysis measures the hourly matching of supply and demand, the average cost and the associated CO2 emissions.
Looking across regions, there were several similar results, including that 80% hourly matching portfolios are comparable in cost and even more affordable than annual matching hybrid projects. Annual matching hybrid projects can be more expensive because of their lack of storage and greater reliance on grid electricity. In many countries in Europe and provinces in China, the respective average costs of USD 100/MWh and USD 70/MWh for the “80% hourly matching” configuration are below the 2023 average industry retail electricity price. The analysis also revealed several regional differences. In the cost optimal configuration for Europe, the share of demand covered by wind, solar PV and battery averages 80%, which is notably higher than for the United States due to a generally more expensive electricity price. In China, the coverage share averages 70% and ranges between 30% and more than 90% in some provinces in the most cost-effective case because of the lower investment costs compared to the United States. When full hourly matching is the target, the cost premium in China is 5% above the 2023 average grid cost. In Europe, 90% hourly matching can be achieved for USD 105/MWh and 99% coverage for less than USD 150/MWh on average.
Constant baseload demand for data centres does not necessarily imply conventional dispatchable power sources. As variable renewables are now cheaper and faster to deploy in many regions compared to other technologies, pairing them with storage can increase their alignment with baseload-type demand. Hybrid portfolios of wind, solar PV and storage can cover a relatively high share of demand on an hourly basis at a competitive price. Aiming for a very high share of hourly matching raises the costs, which can exceed the average industry retail price depending on the region. Compared with conventional annual matching PPAs, hourly matching PPAs with a high share of low-emissions sources provide a higher guarantee of covering electricity demand, reducing CO2 emissions and mitigating the volatility risk associated with electricity prices. The role of renewables should also be analysed at the broader system level to better assess the balance of the variability.
Regional outlook
With a share of over 40%, natural gas is currently the biggest source of electricity for data centres in the United States, followed by renewables – mostly solar PV and wind – at 24%, as well as nuclear and coal power with shares of close to 15% and around 20%, respectively. As demand growth is particularly rapid over the next five years, natural gas is the largest source of additional supply, adding over 130 TWh of annual generation until 2030. Utilities are revising their integrated resource plans, with the construction of additional gas-fired power plants planned across the country, some of them to support the increase in data centre loads
In China, as data centres are located mostly in the east of the country, their electricity supply is dominated by coal with a share of about 70%, followed by renewables with nearly 20%, nuclear close to 10% and natural gas accounting for the remainder. Between 2024 and 2030 both coal and renewables – mostly solar PV and wind – add about 90 TWh to the data centre electricity supply. The increase in renewables is supported by their rising share in the grid electricity mix, provincial colocation mandates and policies to prioritise the construction of data centres in renewables-rich western China. After 2030, the introduction of SMRs significantly boosts the nuclear share of the data centre electricity mix. Between 2030 and 2035, the rise in renewables and nuclear pushes coal into decline. By 2035, both sources together make up 60% of China’s data centre electricity supply
In Europe, renewables and nuclear are set to supply most of the additional electricity required, with their combined share rising to 85% by 2030. Japan and Korea together account for about 5% of global data centre electricity demand today, a share they are expected to retain to 2030. Renewables and nuclear are set to provide nearly 60% of the electricity consumed by data centres in 2030, up from 35% today. The rest of the world is responsible for about 10% of total data centre electricity generation, with Southeast Asia and India accounting for a significant portion of that. In both regions, coal remains a key pillar of the data centre electricity supply, but renewables are projected to eclipse it by 2035.
Access the report here
