How four countries are tackling AI-driven load growth

AI is becoming a core driver of innovation, competitiveness, and global economic growth. According to the International Energy Agency’s April 2026 analysis, electricity consumption from data centres is expected to double by 2030, with power use from the ones focused on AI expected to triple in that same period. Yet this digital transformation rests on energy-intensive data centres that are planned, built, and operated within electricity systems not designed for such concentrated, fast-growing demand.  

At the same time, rapid expansion of data centre infrastructure is increasingly visible at the community level—raising questions about electricity affordability, local grid capacity, land use, water resources, and environmental impact.  

The challenge facing energy providers isn’t simply how to scale infrastructure quickly but how to do so in a way that balances reliability, affordability, sustainability, and public trust. 

The complexity of competing demands 

AI-related electricity demand from data centres differs fundamentally from traditional industrial or commercial loads. While conventional data centre computing tasks feature baseload power demand, AI inference and query-driven workloads exhibit higher volatility, sharper ramping behavior, and more workload distribution uncertainty.  

AI data centres are characterised by: 

• High maximum load densities concentrated at specific network hubs that don’t always overlay with grid nodes 
• The potential need for near-continuous operational availability with stringent requirements for reliability, redundancy, and power quality 
• Deployment timelines that frequently outpace grid planning and reinforcement cycles 
• Power electronics-based loads that increase sensitivity to voltage deviations, frequency disturbances, outages, and curtailment 

Globally, policymakers, utilities, and regulators are confronting the same fundamental question: 

How can electricity systems accommodate AI-driven load growth while balancing reliability, affordability, decarbonisation objectives, and the expectations of communities directly affected by new infrastructure? 

3 critical challenges 

Energy providers and regulators are facing several challenges from AI-driven load growth, including:  

• Grid access, connection queues, and grid congestion 
• Power system stability and supply quality issues 
• Peak load management and system adequacy 

Grid access is a major constraint for data centre development. Data centres seek fast, high-capacity connections, but network reinforcement, permitting, and planning can take years. Grid connection queues for new data centres can reach seven to 10 years in certain jurisdictions. Additional concerns include opaque queue management, capacity reservation, cost uncertainty, post commissioning curtailment exposure, and stakeholder scrutiny.  

Concentration of large, power electronics-dominated loads alters dynamics at local and system levels, affecting voltage control, frequency response, fault behavior, and resilience to disturbances. Data centre loads during AI training cycles can swing from load spikes to idle states within seconds, leading to local or regional grid oscillation if not mitigated. And system operators increasingly face stability constraints that emerge before conventional operational safeguards trigger a response. 

Data centre electricity demand can also amplify power supply adequacy risks. These risks are most acute where system flexibility is limited or reserve margins are already tight. 

Below is a snapshot of how four countries experiencing rapid data centre development—the United States, Germany, the United Kingdom, and the United Arab Emirates—are managing these challenges. Each represents a different path toward a common goal: transforming policy and planning to meet an unprecedented moment for the energy sector. 

Challenge 1: Grid access, connection queues, and grid congestion

United States: Long connection queues affect new loads and the generation required to power data centres. With the median time from interconnection request to commercial operation doubling nationwide for generation and storage projects, connection queue timelines create an obstacle for data centre deployment. First-come, first-served rules and network upgrade cost uncertainty have contributed to queue congestion, speculative applications, multi-year delays, and complicated investment decisions. To meet timelines, data centre developers are pursuing self-funded upgrades or behind-the-meter generation, which some policymakers are encouraging. This accelerates deployment but risks reducing system efficiency, intensifying emissions, and weakening power system operation and planning integration. 

Germany: Data centre capacity is expected to double by 2030. To reduce speculative queue entries and improve planning reliability, the country’s four transmission system operators (TSOs) have introduced a maturity-based grid connection procedure for large assets. Under consideration are flexible connection agreements (FCAs), which can be refined with guaranteed and non-guaranteed shares, and phased application. 

United Kingdom: Designated AI Growth Zones (AIGZs) are being introduced to accelerate large-scale AI data centre development. AIGZs must provide access to at least 500 MW of power capacity by 2030 and demonstrate existing planning permission (or a pathway to securing full planning consent by 2028). They’re also encouraged to strive for strong digital connectivity and low-carbon power generation and energy storage. The UK government is bringing data centre connections into the TMO4+ grid connection reform framework, applying a “first ready and needed, first connected” approach, and aligning connections with regional infrastructure capacity to keep system load growth manageable.  

United Arab Emirates: Connection sequencing for data centre grid access happens through centralised planning and bilateral engagement with emirate-level utilities instead of a published interconnection queue. This aligns strategic priorities with national AI ambitions and lets emirates tailor delivery timelines and risk allocation to their local conditions. Where grid capacity is constrained, formal regulatory pathways allow data centre operators to connect onsite generation to avoid dependency on wider network reinforcement. For example, Abu Dhabi granted Khazna Data Center Limited an Electricity Generation Self-Supply License for a 7 MW solar photovoltaic self-supply facility. 

Challenge 2: Impact on power system stability and supply quality 

United States: Data centre growth is exposing stability risks linked to voltage-sensitive, power electronics-dominated loads. When multiple data centres in Northern Virginia experienced a simultaneous, 3,800-MW disconnect of power after a routine transmission fault, an immediate imbalance occurred where generation exceeded demand. This caused frequency and voltage spikes that required operators to re-dispatch generation and voltage-control resources to maintain grid stability. The incident highlighted how rapid, correlated load loss can stress frequency and voltage control. In April 2026, the North American Electric Reliability Corporation (NERC) confirmed that load-loss events had triggered the preparation of a Level 3 “essential actions” alert—the highest level—focused on large loads such as data centres. 

Germany: As data centre demand grows, policymakers, regulators, and system operators are examining how to integrate large loads without compromising stability. Existing grid connection rules impose technical requirements on large grid users, while grid code discussions are considering how demand-side resources can contribute more actively to system operation. One option is grid-interactive uninterruptible power supply (UPS) systems. These systems enable data centres to remain connected during short-term grid disturbances while temporarily absorbing or supplying power without affecting IT operations. International experience, including Ireland’s DS3 programme, has demonstrated that UPS and battery-based systems can provide very fast frequency support.  

United Arab Emirates: Power system stability risks from AI-driven data centre growth are compounded by rapid expansion of inverter-based solar generation, which is projected to grow from 4% of total generation in 2020 to over 20% by 2040. While the UAE has technology-neutral transmission and security standards for large demand connections, no publicly available data centre-specific technical connection requirements have been published. 

Challenge 3: Managing peak load and system adequacy 

United States: AI-driven data centre demand is tightening reserve margins, raising capacity prices, and intensifying affordability concerns. For example, regional grid operator PJM has experienced record capacity price, with a 22% increase in one year partly driven by higher forecasted load and accelerating generation retirements. Gas-fired generation is presently the dominant near-term response to new data centre demand in the US, with about 252 GW of gas-fired power capacity under development—and more than a third of that total is intended for data centre onsite electricity supply. Longer-term options such as advanced nuclear or geothermal remain largely developmental.  

Germany: Data centre capacity is expected to at least double by 2030, with a quadrupling in connected load for high-performance computing and AI. To address system adequacy amid rising, concentrated electricity demand, TSOs are considering prognosis-based curtailment windows as part of FCA arrangements for new consumers such as data centres. 

United Kingdom: The connection queue has surged to about 125 GW—more than double the current UK peak electricity demand—with nearly half (50 GW across 140 projects) attributable to data centres. Customers facing high energy costs have expressed concerns about concentrated AI loads in constrained regions such as urban centres. For example, over one-fourth of UK data centre capacity planned through 2034 is located in Greater London. 

United Arab Emirates: The adequacy challenge is further exacerbated by extreme summer peaks, where cooling demand from both buildings and data centres coincides. While the Barakah Nuclear Energy Plant supplies about 25% of the UAE’s electricity, critical shoulders remain when solar generation output drops during high demand. 

Turning data centres into assets for a stable grid 

To mitigate these challenges, system operators and policymakers are exploring how AI-driven demand can support flexibility, congestion management, and system adequacy. Key flexibility levers include: 

• Shifting workloads that aren’t time-critical 
• Reallocating computing tasks across sites 
• Deploying onsite battery storage and backup generation 

Duke University’s Nicholas Institute suggests that these programmes could unlock up to 100 GW of US capacity if there’s close coordination between IT workload management and energy management systems to help data centres respond dynamically to grid conditions.  

United States: In the US, system operators and regulators are trying to manage reliability risks associated with rapid load growth. PJM is exploring proposals such as curtailment-first load categories and “bring your own generation” models. In California, the electric utilities PG&E and SCE offer large-load demand response and emergency curtailment programmes. And Texas Senate Bill 6 pairs mandatory curtailment with voluntary demand-response programmes, authorising utilities and the Electric Reliability Council of Texas to require large-load curtailment and switching to onsite generation during grid emergencies. 

Germany: According to Germany’s 2026 national data centre strategy, data centres are expected to integrate into a flexible power system featuring high energy efficiency, renewable sourcing, and waste-heat use without demanding short-term load shedding or real-time demand response. 

United Kingdom: The UK government is considering measures for restricting or reducing large-load power usage during periods of grid stress. In the UK’s first live trial of AI technology that adjusts data centre power usage in real time, electricity demand dropped by more than a third in under one minute without disrupting critical compute workloads. 

United Arab Emirates: The UAE is moving toward formalised demand-side mechanisms to manage grid stress, with centralised utility structures providing a platform to scale over time. Abu Dhabi’s Demand Response Project, which has reduced peak demand by an average of 106 MW across 10 events, is targeting 200 MW of contracted demand response capacity by 2030 with long-term potential of up to 1,000 MW. While the programme currently targets industrial and commercial loads, extending equivalent mechanisms to data centres could be a next step as peak demand grows. 

Navigating the AI-grid nexus 

Data centre growth needs to be strategically aligned with industrial and energy policy objectives rather than rely on first-come, first-served connections. System planning should prioritise loads that contribute to economic value, resilience, and decarbonisation, while project-level processes allocate scarce grid capacity to deliverable, welfare-enhancing projects. Dedicated zones or clusters for specific load types are proving to be a “best of both worlds” approach, enabling speed and scale while preserving system efficiency. 

Managing AI demand at scale requires treating data centres as system-relevant infrastructure. Policymakers should explicitly embed AI-driven load growth into grid planning, adequacy assessments, and stability frameworks. System operators need granular operational tools and differentiated connection concepts, not blunt constraints. And corporate stakeholders need energy strategies that integrate flexibility, system compatibility, and long-term resilience alongside capacity procurement. 

When all these elements are successfully in place, the AI-grid nexus can accelerate more robust, efficient, decarbonised power systems.