Executive summary

Data centre investment is surging as artificial intelligence has become central to economic policies across key markets. Globally, the largest data centre operators – so-called hyperscalers – are set to spend more than USD 700 billion as capital expenditure in 2026, mostly for the development of new data centres. Many jurisdictions want to attract digital infrastructure as the foundation of domestic innovation in artificial intelligence and digital services, while policy makers are making data centres and artificial intelligence key pillars of economic security and digital sovereignty.

Policy makers are increasingly sensitive to growing concerns, particularly among local communities, about data centre impacts on electricity prices, emissions and water. The unprecedented artificial intelligence-driven investment boom into digital infrastructure is straining local grids and stretching the supply chains for key electrical equipment. At the same time, it is driving rapid innovation in energy technologies, procurement strategies, and policy and regulatory frameworks. Many jurisdictions also face constraints in terms of available network capacity, long connection queues and slow permitting timelines.

Requested by France’s G7 Presidency, this report takes stock of evolving practices in data centre integration into electricity systems. It surveys the current state of policy and regulatory activity and corporate strategy, looking across the data centre value chain in more than 34 policy jurisdictions covering about 90% of global data centre capacity. It identifies key challenges, constraints and opportunities in the policy and regulatory domain. This report builds on previous IEA analysis, notably the 2025 Energy and AI special report and 2026 Key Questions on Energy and AI report.

Our analysis finds 134 policies and regulations guiding the integration of data centres into electricity systems globally. Nearly 40% of these policies were announced or implemented in the last 12 months, highlighting the dynamic nature of the field.

The most common area of policy activity focuses on improving the visibility of data centre demand in terms of scale, timing and location. This highlights the key challenge that is proactively planning network capacity for data centres and connecting new large data centres to existing networks. Policies aiming to promote the smooth operational integration of data centres into electricity networks are rarer, although notable examples have emerged. Some of these mandate that data centres contribute to generation adequacy (so-called “bring-your-own-power,” or BYOP), or to network stability and flexibility.

Data centre operators are also innovating in how they procure electricity and interact with the grid. Traditionally, these companies have been large procurers of renewable electricity; that has continued, with operators signing 45 terawatt-hours (TWh) of power purchase agreements (PPAs) in the first half of 2026, up by nearly 20% on the same period in 2025. Of this, nearly 70% was in the United States. Additional innovations are also emerging, with a small but growing number of power purchase agreements for emerging technologies such as advanced geothermal, small modular nuclear reactors, long-duration energy storage, virtual power plant arrangements, or battery storage. A large number of projects for onsite power have been announced, especially for natural gas in the United States.

In several jurisdictions, network planners and operators are seeing connection queues for data centres with cumulative capacity that is larger than today’s peak demand. Examples of jurisdictions with large pipelines include the US states of Texas and Virginia, Finland, and the United Kingdom. Connection queues contain duplicative or speculative projects, and policy makers must therefore prepare for potentially very large new loads without clear visibility of how much capacity will ultimately be built. Overestimating demand can risk creating stranded or underutilised grid investments, while underestimating it can lead to bottlenecks and delays.

This “queue inflation” is not unique to data centres, but several of their characteristics sharpen the challenge. Firstly, data centres can be very large projects, and they are often being developed in areas that were not previously major sources of electricity demand; this can substantially shift network topography that had been structured around historical load centres, such as industrial clusters and cities. Secondly, electricity demand for data centres is hard to assess. Even when projects proceed, demand typically ramps up gradually as buildings, servers and customer workloads come online, creating uncertainty over when and how much power consumption will materialise. More fundamentally, AI-focused data centres are huge capital investments for a novel, potentially transformational technology whose development, uptake, and profitability are uncertain.

Many jurisdictions are therefore undertaking queue reform with three broad objectives. The first is to ensure that connection queues reflect credible projects, using stronger queue entry requirements, maturity assessments, financial commitments and milestone-based capacity release mechanisms. A second objective is prioritising projects effectively within queues; several jurisdictions are moving from first-come-first served approaches to criteria that combine project maturity with broader considerations, such as strategic value, economic contribution and sustainability. However, ensuring that criteria remain transparent and predictable remains a key challenge. Several jurisdictions are also experimenting with methods to accelerate the connection studies needed to ensure projects can be safely connected to the grid.

Traditionally, data centres have been passive actors in electricity networks; in some cases, that is now changing. In the face of growing pressures and bottlenecks, data centre operators, power project developers and policy makers are redefining how data centres interact with the grid. Importantly, not all data centres are the same, and their ability to interact proactively with the grid depends on factors such as their customer mix; the operator’s size, financial position and operational sophistication; and local regulatory and policy conditions.

Policy makers and regulators are increasingly offering non-firm grid connections as a means of connecting data centres faster to electricity networks. Periods of system stress typically occur for relatively few hours per year, and non-firm grid connections could enable an additional 750-900 gigawatts (GW) of grid hosting capacity globally, according to IEA analysis. Jurisdictions ranging from Norway, South Australia, Spain, and the Southwest Power Pool in the United States are experimenting with different models of non-firm connection agreements. These include conditional non-firm or curtailable access; time-dependent, phased or ramped access; and dynamic consumption limits adjusted to real-time or forecast network conditions. A key condition for the greater use of non-firm connections is the need to provide standardised and transparent conditions so data centre operators can understand and mitigate potential risks to their operations. Japan’s Watt-Bit Collaboration is an example of a public-private initiative that aims to support more flexibility from compute infrastructure over time.  

Data centre operators are increasingly exploring diverse approaches to provide flexibility to the grid. Active management of computing loads remains quite rare and is constrained by the widespread unbundling of compute demand and compute supply. Many data centre operators provide compute services to third-party actors, and therefore have little visibility and control over workloads that take place in their data centres. Compute prices are also astronomical relative to the prices of other flexibility sources: AI-optimised data centre capacity can cost as much as USD 20-30 billion per GW to rent for a year, which is 100 times more than the equivalent capacity cost of a battery. This makes curtailing compute operations very expensive. Nonetheless, the industry is exploring compute flexibility, especially if it enables faster grid connections. Frameworks to standardise the exchange of information across compute demand, compute supply and electricity networks will be critical to bridging a fragmented value chain and unlocking more flexibility. Recent stability-related events – including the abrupt disconnection of a large data centre load (over 3 GW) from the grid and switch to backup supply in the United States – have prompted several system operators to implement or consider technical performance requirements for data centres to support grid stability.

Bring-your-own-power (BYOP) is becoming a more common policy requirement and business strategy but takes different forms. Some jurisdictions, such as Ireland, have stipulated that certain kinds of new data centres need to contribute to system adequacy by building or contracting with new dispatchable capacity. A large number of onsite gas power projects have been proposed in the United States. While many remain at the proposal stage, some are moving forward with credible customers in the technology sector. However, onsite projects are technically complex and depend on installing generation and power equipment with long wait times. For these reasons, many onsite projects are integrating battery storage into their design and are still seeking an eventual grid connection; policy makers will therefore need to understand how they impact the broader grid if and when they connect. In other jurisdictions, including China and Spain, policy makers have placed renewables purchase obligations on new data centres or are considering them. Amid these innovations, the key point for policy makers is that data centre procurement choices offer an opportunity for the sector to contribute to broader policy goals in terms of energy security, adequacy or sustainability.

Data centres are currently associated with around 150 million tonnes of CO2 emissions due to their electricity consumption. Our central scenario sees this growing to around 350 million tonnes by 2035, with recent IEA projections increasing due to the higher use of onsite natural gas. This is set to be around 1.5% of global energy-related emissions in 2035. IEA analysis of data centres dedicated to AI reveals that nearly 40% of identified capacity is in areas with high or extremely high baseline water stress.

Policy makers are increasingly implementing energy and water efficiency standards for data centres, but the nature of the challenge is complex. Today, there are energy efficiency standards for equipment, but not AI models. While there is significant potential to improve the efficiency of models, extremely rapid technical improvements, heterogenous AI services and uncertain baselines make standard-setting at the model level very difficult to implement. Moreover, while market pressures to optimise the use of compute – an expensive production input – are leading to rapid improvements in model efficiency, leading AI labs also face market pressures, and an ample supply of capital, to grow customer numbers and customer use. This can mean offering the best-performing models even if smaller, more efficient ones could satisfactorily complete many tasks. There is also a trade-off between water-use efficiency and energy efficiency in data centres, as cooling equipment that minimises water use tends to consume more energy, and vice versa. Additionally, the largest share of water use in the data centre value chain takes place upstream in power generation.

Data centres concentrate an enormous amount of economic value in a small space geographically, and they are exposed to extreme weather and other physical risks. Industry players are indicating that the high costs of data centres are leading to gaps in insurance coverage. Given the tight linkages between AI, data centres and financial markets, policy makers could consider the broader implications of risks and the extent to which individual data centres, or clusters of them, are exposed.

The effect of data centres on electricity bills is not clear-cut. Where a new large load requires additional network or generation investment, and that cost is recovered from all users, it can raise bills for other consumers. Where it makes fuller use of capacity that already exists and pays its share of the cost, a data centre can spread fixed costs over more consumption and ease pressure on bills. Either way, how the costs of connecting large loads are defined and allocated is a choice for policy makers.

Tariff frameworks allocate risks as well as costs. Network tariffs recover the largely fixed cost of building and maintaining the grid, and regulators design them to balance several principles, among them cost reflectivity (or the idea that each user’s charges should reflect the true costs it places on the system). For large loads, this usually means charging for the capacity they reserve rather than the energy they consume. Networks are also planned and built on forecasts of loads that may not materialise, and the size and uncertainty of data centre loads make this risk particularly pronounced. Large-load tariffs can reduce other consumers’ exposure to these risks through minimum-payment commitments, long-term contracts matched to the life of the assets, exit fees, collateral and phased connections, while financial commitments at the connection stage screen out projects that are unlikely to proceed. Together this addresses three distinct risks: that a load uses less capacity than it reserved, that it leaves before the assets built for it are paid off, and that it never connects at all.

Effective policy packages involve balancing cost reflectivity and consumer protection against simplicity, investment certainty and competitiveness. Because risk mitigation instruments interact, assessing them together helps avoid both shifting unjustified costs onto existing consumers and charging data centres more than once for the same risk. This depends on transparent coordination among policy makers, regulators, utilities and system operators.

The data centre buildout is increasingly facing local opposition in some jurisdictions. In the face of this, policy makers have multiple tools, and company strategies are also adapting. Data centre grid connections are increasingly being made conditional on being “good grid citizens”: contributing to flexibility, efficiency, fairness, stability and capacity adequacy. The industry itself is innovating to give back more to the grid, too, although it is important to recognise that data centres are a diverse asset class, with different impacts, incentives and capacities. As the sector develops rapidly, sharing information about best practices, tested approaches and new innovations will be critical. This report, and the dataset of policies and regulations that accompany it, aims to support these efforts. 

1. Strive for clarity and transparency in the geographic and electricity consumption attributes of proposed data centres.

2. Establish transparent and predictable grid connection processes that prioritise credible and mature projects, reflecting strategic priorities where appropriate.

3. Approach energy and water use efficiency holistically, factoring in the complexity of when, where and how energy and water are used.

4. Design transparent and coordinated tariff frameworks that allocate the costs and risks of large loads fairly.

5. Create clear frameworks that enable data centres to contribute actively to grid flexibility, security, stability and sustainability.

6. Make dialogue and experience sharing central to inform decision making in this fast-moving field.