Artificial intelligence is moving rapidly into corporate strategy, customer service, financial analysis, supply-chain management, research, manufacturing, and operational decision-making. As businesses increase their reliance on AI-enabled systems, they also need to understand the physical infrastructure that supports this digital expansion.
AI infrastructure energy demand is becoming a material sustainability issue because advanced computing depends on data centres, high-performance servers, cooling systems, backup capacity, grid connections, water resources, and long-term energy procurement. The quality of an organisation’s AI strategy will increasingly depend on whether these infrastructure requirements are assessed with the same discipline applied to financial, operational, and regulatory risks.
The sustainability discussion around AI therefore needs to move beyond software governance. Responsible AI adoption requires a clear understanding of electricity use, energy sourcing, cooling efficiency, water exposure, supplier transparency, and resilience planning.
AI is often discussed through the lens of productivity, automation, innovation, and competitive advantage. These benefits are significant, but they depend on a growing network of physical assets.
AI models are trained and deployed through data centres. These facilities contain servers, storage systems, networking equipment, cooling systems, uninterruptible power supplies, backup generators, and grid connections. Each component affects the environmental footprint and operational resilience of AI-enabled services.
The International Energy Agency’s 2026 analysis shows that electricity consumption from data centres increased by 17% during 2025. Electricity use from AI-focused data centres increased by 50% during the same period. The IEA expects data-centre electricity consumption to approximately double from 485 TWh in 2025 to 950 TWh by 2030.
This growth does not mean that every company using AI needs to operate its own data centre. It does mean that organisations should understand the infrastructure implications of the AI services they procure, develop, finance, host, or integrate into their operations.
AI workloads require substantial computing capacity. Large-scale model training, inference, image generation, video processing, AI agents, simulation, and real-time analytics can increase the demand placed on data-centre infrastructure.
The IEA’s Energy and AI report explains that servers account for approximately 60% of electricity demand in modern data centres, although this varies by facility. Accelerated servers, including systems using graphics processing units and specialised AI chips, are becoming more important as AI adoption expands.
The IEA reported that electricity consumption from accelerated servers is projected to grow by approximately 30% annually in its base case. These systems account for almost half of the expected increase in global data-centre electricity consumption between 2024 and 2030.
The issue is therefore connected to the design of the entire AI infrastructure ecosystem.
Infrastructure Component | Role in AI Operations | Sustainability Relevance |
Servers and accelerators | Process AI training, inference, and data-intensive workloads | Electricity consumption, embodied carbon, equipment lifecycle, and resource efficiency |
Cooling systems | Maintain safe operating temperatures for high-performance equipment | Electricity use, water demand, climate suitability, and operational efficiency |
Grid connections | Supply electricity to the facility | Energy security, local grid pressure, carbon intensity, and connection delays |
Backup systems | Maintain service continuity during disruption | Fuel use, battery requirements, resilience, and emissions exposure |
Storage and networking | Support data access, movement, and availability | Electricity demand, equipment efficiency, and infrastructure optimisation |
Facility location | Determines the operating environment and access to resources | Renewable availability, water exposure, grid capacity, heat conditions, and regulatory requirements |
A responsible AI strategy should consider how these components affect the organisation’s sustainability commitments and operating model.
The IEA’s updated outlook provides a useful perspective on the scale of expected growth. Global data-centre electricity consumption is projected to rise from 485 TWh in 2025 to approximately 950 TWh by 2030. AI-focused data centres are expected to grow considerably faster than the wider market.
The IEA has also highlighted the increasing concentration of power demand within advanced facilities. By 2027, an individual AI server rack could have peak power demand comparable to the electricity needs of 65 households. Between 2020 and 2025, the power density of AI servers increased elevenfold, and the IEA expects a further fourfold increase by 2027.
These changes create practical implications for infrastructure planning. Data centres need access to electricity systems capable of supporting high and concentrated loads. They also require equipment that can respond to rapid changes in demand.
The pressure is particularly visible in markets with significant data-centre investment. The Lawrence Berkeley National Laboratory reported that data centres consumed approximately 4.4% of total United States electricity during 2023. It estimated that this share could increase to between 6.7% and 12% by 2028.
The global and national figures should not be applied mechanically to every market. They provide an important signal for companies, investors, utilities, and policymakers: AI growth requires infrastructure planning that is grounded in realistic energy assessments.
Electricity demand within a data centre extends beyond computing equipment. Cooling systems are required to manage the heat produced by servers and maintain reliable operating conditions.
The IEA reports that cooling and environmental-control systems can account for approximately 7% of electricity consumption in efficient hyperscale data centres and more than 30% in less-efficient enterprise facilities.
This variation demonstrates why facility design matters. A data centre’s environmental performance depends on server efficiency, cooling technology, local climate, operating practices, equipment utilisation, and infrastructure management.
As AI workloads increase power density, heat management becomes more demanding. Businesses should therefore consider whether their technology partners have credible systems for monitoring and improving cooling efficiency.
Common areas for assessment include:
A procurement decision based solely on computing performance can overlook material environmental and operational risks.
Water is becoming an important part of the AI infrastructure discussion because some data-centre cooling systems require water for heat management. The level of water use can vary significantly by facility, climate, technology, and electricity source.
A 2025 Berkeley Lab study found that workload-level water use can vary by more than 10,000 times across different operating conditions. The researchers identified several factors that influence water performance, including server efficiency, grid-related water consumption, server utilisation, cooling-system type, infrastructure efficiency, climate zone, inactive-server levels, and equipment-refresh cycles.
This finding has practical importance. There is no single cooling design or technology choice that produces the best environmental outcome in every location. Businesses need site-specific analysis.
The water issue is particularly relevant for data-centre planning in water-stressed regions. The World Bank has highlighted that data centres can compete for scarce water and energy resources in some markets, which means governments and developers should consider these pressures in planning and land-use decisions.
Organisations assessing AI infrastructure energy demand should therefore review the relationship between electricity use, cooling systems, water availability, and local environmental conditions.
AI infrastructure can affect several areas of ESG performance.
The environmental dimension includes electricity consumption, greenhouse gas emissions, water use, facility efficiency, equipment lifecycle, and renewable-energy sourcing. The social dimension includes community concerns around affordability, water access, infrastructure development, and local impacts. The governance dimension includes supplier oversight, data quality, investment controls, transparency, and accountability for technology procurement.
The issue is relevant even where a company relies on cloud services or external technology providers. Scope 3 emissions, supplier engagement, procurement criteria, and digital-service dependencies can all influence sustainability reporting.
A business may need to ask:
ESG Area | Core Assessment Question |
Electricity consumption | How much energy is associated with the AI services, workloads, or facilities used by the organisation? |
Emissions | What is the carbon intensity of the electricity supporting the relevant data-centre operations? |
Water use | Do the cooling systems rely on water, and are the facilities located in water-stressed regions? |
Renewable energy | Are credible renewable-energy procurement arrangements in place? |
Resilience | Can the infrastructure remain reliable during grid stress, extreme weather, or supply constraints? |
Supplier oversight | Do providers disclose meaningful data on PUE, water use, emissions, and renewable sourcing? |
Governance | Who reviews the infrastructure implications of AI-related procurement and investment decisions? |
Companies should avoid assuming that digital services have an immaterial physical footprint. AI procurement decisions can carry significant energy and resource implications.
Data centres can place concentrated pressure on electricity systems because they require large and reliable power connections. This creates challenges for utilities, regulators, developers, and communities.
The IEA’s 2026 report notes that data-centre deployment is increasingly encountering physical bottlenecks. These include constrained grid connections, supply-chain limitations for transformers and gas turbines, advanced-chip availability, planning approvals, and regulatory processes.
These constraints affect project timelines and investment decisions. A technology strategy that assumes unlimited access to electricity may face practical limitations.
The IEA also notes that AI data centres can experience rapid shifts in electricity load. Some facilities may have swings of more than 50% of rated capacity within a second. This can increase the importance of battery storage, flexible operations, grid coordination, and resilient infrastructure design.
For companies, the strategic implication is clear. AI infrastructure should be treated as part of energy planning and business-continuity management.
Rapid growth in AI workloads does not remove the importance of energy efficiency. Improvements in hardware, software, cooling systems, server utilisation, and workload management can reduce the energy required for specific tasks.
The IEA reported in April 2026 that energy efficiency per AI task is improving rapidly. However, wider AI adoption and the growth of energy-intensive applications continue to increase total electricity consumption.
This creates an important management principle: efficiency improvements need to be measured alongside absolute energy use.
A company can improve the efficiency of its AI systems while its overall electricity footprint continues to rise because more applications, users, and workloads are introduced. Sustainability reporting should therefore include both intensity metrics and absolute-performance indicators.
Metric Type | Example | Why It Matters |
Absolute electricity use | Total electricity consumed by relevant facilities or services | Shows the overall infrastructure footprint |
Energy intensity | Electricity consumed per workload, transaction, model run, or computing output | Measures efficiency improvements |
PUE | Total facility electricity divided by IT-equipment electricity | Assesses data-centre infrastructure efficiency |
Water Usage Effectiveness | Water use relative to IT-equipment energy | Helps assess cooling-related water pressure |
Renewable-energy share | Percentage of electricity matched with credible renewable sourcing | Supports emissions-management analysis |
Carbon intensity | Emissions associated with electricity consumption | Links energy use with climate impact |
Utilisation rate | Productive computing use relative to available server capacity | Identifies inefficient infrastructure deployment |
A balanced dashboard gives management a clearer view of performance.
Renewable-energy procurement is becoming increasingly relevant for data-centre operators and businesses that rely on AI infrastructure.
The IEA reported that the technology sector accounted for approximately 40% of corporate renewable power-purchase agreements signed during 2025. This indicates that major technology companies are using long-term energy procurement as part of their infrastructure strategy.
Renewable-energy sourcing can reduce emissions exposure, but it should be assessed carefully. Companies need to consider the quality of procurement arrangements, the timing of electricity consumption, local grid conditions, additionality, contractual structure, and reporting methodology.
Renewable procurement should sit within a broader infrastructure strategy that includes energy efficiency, load management, storage, resilience, and transparent performance reporting.
The UAE and the wider GCC are advancing digital-economy, cloud-computing, AI, smart-city, and infrastructure priorities. Businesses operating in the region are increasingly adopting AI-enabled tools across financial services, energy, logistics, real estate, construction, government services, education, healthcare, and retail.
The regional operating environment creates several considerations.
High temperatures can increase the importance of efficient cooling systems. Water scarcity requires careful assessment of cooling technologies and water-management practices. Grid capacity, renewable-energy availability, facility location, resilience planning, and power-purchase arrangements can influence long-term infrastructure performance.
Companies in the region may also rely on international cloud providers and external data-centre operators. This creates a need for supplier engagement and reporting clarity.
A GCC-based company should consider:
These questions can help organisations connect AI adoption with sustainability governance.
Businesses do not need to wait for new reporting rules before assessing their AI infrastructure footprint. A phased approach can create a clear starting point.
Step 1: Map AI Use Across the Organisation
Identify where AI systems are being used, tested, procured, or developed. The assessment should cover internal applications, external platforms, cloud services, embedded AI tools, analytics systems, and supplier-provided solutions.
The purpose is to understand which activities may create material infrastructure demand or supplier dependencies.
Step 2: Identify Infrastructure Providers
Determine which cloud providers, data-centre operators, software platforms, and technology vendors support the organisation’s AI workloads.
Request information on:
This information may be incomplete during the first assessment. The data gaps should be documented and addressed through procurement and supplier-engagement processes.
Step 3: Define Materiality
Assess whether AI infrastructure energy demand could materially affect the company’s emissions profile, sustainability commitments, operating costs, resilience, procurement decisions, or stakeholder expectations.
Materiality will vary by sector and business model. A company developing AI products may face a different level of exposure from a company using a limited number of third-party applications.
Step 4: Review Energy and Water Risks
Analyse the energy and water implications of relevant facilities and providers. The assessment should consider electricity sources, cooling design, water stress, grid constraints, backup systems, and climate conditions.
Location-specific analysis is important because the environmental impact of AI infrastructure can vary substantially across regions.
Step 5: Add ESG Criteria to Procurement
Technology procurement policies should include sustainability criteria for AI-related products and services.
Relevant requirements may include:
Procurement teams should work with sustainability, finance, technology, legal, and risk professionals to define proportionate criteria.
Step 6: Establish Metrics and Reporting Boundaries
Decide which metrics should be tracked at company, supplier, platform, or facility level.
The reporting boundary should be clear. Companies should explain whether metrics cover owned facilities, colocated infrastructure, cloud services, contracted platforms, or selected material providers.
Step 7: Integrate AI Infrastructure Into Sustainability Governance
Assign accountability for monitoring AI-related energy and resource risks.
Boards and senior management should receive relevant information where AI infrastructure could affect investment decisions, sustainability commitments, business continuity, or reputational risk.
A mature governance model connects AI strategy with sustainability, energy management, risk oversight, and financial planning.
Companies should use careful language when describing AI as efficient, low-carbon, sustainable, or environmentally beneficial.
An AI application may improve efficiency in one area while increasing electricity use elsewhere. A cloud provider may report renewable-energy procurement while operating within a grid that faces capacity or timing constraints. A data centre may reduce water use while increasing electricity requirements for cooling.
Credible communication should explain:
Transparent reporting strengthens credibility and supports better decision-making.
AI models depend on computing infrastructure for training, inference, data storage, networking, and real-time processing. Advanced AI applications often require high-performance servers and specialised chips, which increase electricity demand and cooling requirements.
The International Energy Agency’s April 2026 central projection estimates that global data-centre electricity consumption could increase from 485 TWh in 2025 to approximately 950 TWh by 2030, representing around 3% of global electricity demand.
The appropriate level of assessment depends on materiality. Companies with significant AI adoption, technology products, large cloud workloads, data-centre investments, or sustainability commitments should assess the infrastructure implications of their AI strategy. Smaller users can begin by identifying material providers and requesting available sustainability information.
Power Usage Effectiveness is a common data-centre efficiency metric. It compares total facility electricity consumption with the electricity used by IT equipment. A lower PUE generally indicates more efficient infrastructure, although the metric should be considered alongside water use, carbon intensity, utilisation, and local conditions.
Some data-centre cooling systems use water to remove heat from computing equipment. Water performance varies significantly depending on facility design, cooling technology, climate, electricity source, and server efficiency. Water use should be assessed carefully in water-stressed regions.
Renewable-energy sourcing is an important part of the solution, but it does not remove the need for energy efficiency, storage, grid planning, resilience, cooling optimisation, water management, and transparent reporting.
Climate risk is increasingly affecting how companies protect assets, manage supply chains, assess insurance exposure, safeguard employees, and maintain business continuity…
European sustainability reporting is entering a more focused phase. The reporting landscape still requires companies to understand their material environmental…
Artificial intelligence is moving rapidly into corporate strategy, customer service, financial analysis, supply-chain management, research, manufacturing, and operational…
Typically replies within a day
