Measuring and Standardizing AI’s Energy and Environmental Footprint to Accurately Access Impacts – Federation of American Scientists

Measuring and Standardizing AI’s Energy and Environmental Footprint to Accurately Access Impacts – Federation of American Scientists

Report on Measuring and Standardizing AI’s Energy and Environmental Footprint with Emphasis on Sustainable Development Goals (SDGs)

AI Data Center

The rapid expansion of artificial intelligence (AI) has led to a significant increase in data center energy consumption, water usage, carbon emissions, and electronic waste. These environmental impacts remain largely unquantified due to inconsistent metrics and opaque reporting. This situation challenges policymakers and grid operators in managing AI’s growing resource footprint effectively. Current measures, such as Power Usage Effectiveness (PUE), are outdated and narrow, often obscuring the true emissions through renewable energy credits. Studies indicate that actual carbon footprints may be up to 662% higher than reported figures. Furthermore, hyperscale AI data centers consume hundreds of thousands of gallons of water daily and contribute substantially to electronic waste, with only about a quarter of operators tracking retired hardware.

This report proposes a comprehensive set of congressional and federal executive actions to establish standardized metrics for AI’s energy and environmental impacts across all stages including model training, inference, and data center infrastructure. The recommendations align with multiple Sustainable Development Goals (SDGs), particularly SDG 7 (Affordable and Clean Energy), SDG 9 (Industry, Innovation, and Infrastructure), SDG 12 (Responsible Consumption and Production), and SDG 13 (Climate Action).

Challenge and Opportunity

Inconsistent metrics and opaque reporting create uncertainty in forecasting AI’s future power demand, jeopardizing grid planning and climate targets.

AI’s Opaque Environmental Footprint

Generative AI and large-scale cloud computing have caused an unprecedented surge in energy demand. AI systems require substantial computing power during both training and inference phases. Data centers consumed approximately 415 Terawatt hours (TWh) of electricity in 2024, representing about 1.5% of global power demand. The International Energy Agency (IEA) forecasts this could more than double to 945 TWh by 2030, comparable to the electricity consumption of countries like Sweden or Germany. Estimates vary widely due to differing methodologies, assumptions on AI query volumes, hardware supply, workload mix, and efficiency gains.

This uncertainty complicates energy supply planning, as electricity grid operators traditionally anticipate gradual demand increases. The rapid build-out of AI data centers, combined with other growing demands such as electric vehicles, threatens to disrupt this model, potentially causing increased costs, power shortages, and reliability crises.

Moreover, the surge in power demand risks undermining climate progress. Many AI data centers require 100–1000 megawatts (MW), equivalent to a medium-sized city’s consumption. Grid connection lead times for clean energy can exceed two years, leading some utilities to restart retired coal plants, counteracting local climate goals. Major technology companies have reported significant increases in carbon emissions largely attributable to cloud computing and AI.

Transparency issues exacerbate these challenges. Companies often claim carbon neutrality through renewable energy credits, while actual emissions remain unreported or underestimated. For example, Meta’s reported emissions were found to be over 19,000 times higher when calculated using actual grid emissions. Water consumption for cooling AI data centers is also substantial but rarely disclosed, with each AI query indirectly consuming about half a liter of fresh water.

Outdated and Fragmented Metrics

Legacy metrics like Power Usage Effectiveness (PUE) do not capture critical aspects such as water consumption, hardware manufacturing impacts, or electronic waste.

PUE measures the ratio of total facility energy to IT equipment energy but does not assess the efficiency of the IT equipment itself. Consequently, data centers with good PUE scores may still be inefficient in their AI compute operations. Furthermore, only 28% of operators track hardware beyond its use phase, and a mere 25% measure e-waste, leading to significant environmental harm through improper disposal.

Opportunities for Action

Standardizing AI’s energy and environmental footprint metrics presents a strategic opportunity to align AI growth with sustainability goals. Transparent measurement enables policymakers to incentivize efficiency innovations and optimize grid investments. Industry benefits include identifying inefficiencies and reducing operational costs. Legislative and international efforts, such as the Artificial Intelligence Environmental Impacts Act and the European Union’s AI Act, support this momentum. A recent U.S. Executive Order directs the Department of Energy (DOE) to develop reporting requirements covering AI data centers’ full lifecycle, including embodied carbon, water usage, and waste heat.

Plan of Action

Congress should authorize DOE and the National Institute of Standards and Technology (NIST) to lead an interagency working group and collaborate with public, private, and academic stakeholders to develop and implement standardized AI environmental metrics.

Recommendation 1: Identify and Assign Agency Mandates

  • Department of Energy (DOE): Co-lead agency coordinating AI-related efforts through the Office of Critical and Emerging Technologies (CET), promoting energy-efficient technologies, and addressing grid integration challenges.
  • National Institute of Standards and Technology (NIST): Co-lead responsible for developing and standardizing metrics, convening experts, and publishing guidelines.
  • White House Office of Science and Technology Policy (OSTP): Coordinating body ensuring multi-agency alignment and integration with broader climate and technology policies.
  • Environmental Protection Agency (EPA): Lead environmental data collection and oversight, including lifecycle emissions, water use, and e-waste.
  • Federal Energy Regulatory Commission (FERC): Support grid and market integration by streamlining interconnection processes and updating reliability assessments.
  • Congressional Committees: Provide legislative oversight, champion relevant legislation, and ensure agency accountability.

Recommendation 2: Develop a Comprehensive AI Energy Lifecycle Measurement Framework

A multi-stakeholder process led by NIST, DOE, and EPA should create standardized metrics covering the entire AI lifecycle, including training, inference, data center operations, and hardware manufacturing and disposal.

  1. Data Center Efficiency Metrics: Measure power usage effectiveness and related efficiency indicators.
  2. AI Hardware & Compute Metrics: Include performance per watt and compute utilization rates.
  3. Cooling and Water Metrics: Quantify energy and water consumption for cooling systems.
  4. Environmental Impact Metrics: Assess carbon intensity per AI task and e-waste generation.
  5. Composite or Lifecycle Metrics: Develop holistic sustainability scores combining multiple factors.

Design and Governance

NIST should coordinate metric development through an open process, leveraging existing standards and updating metrics as technology evolves. A Metrics Review Committee should oversee continuous improvement and stakeholder engagement.

Recommendation 3: Operationalize Data Collection, Reporting, and Policy Integration

  1. Initiate a six-month voluntary reporting program involving AI developers, data center operators, and utilities.
  2. Develop standardized reporting templates and data sharing protocols.
  3. Transition to mandatory reporting legislation modeled after EPA’s Greenhouse Gas Reporting Program and DOE’s energy data collection forms.
  4. Integrate collected data into energy outlooks, grid planning, and regulatory frameworks.
  5. Ensure coordination with the National Telecommunications and Information Administration (NTIA) and the Census Bureau to streamline reporting and incorporate AI sector data into national statistics.

Roles and Responsibilities to Measure AI’s Environmental Impact

Agency/Entity Role Key Responsibilities
Department of Energy (DOE) Co-lead
  • Coordinate AI-related efforts across DOE programs.
  • Promote energy-efficient data center technologies.
  • Address grid integration and planning challenges.
National Institute of Standards and Technology (NIST) Co-lead for metrics and standards
  • Lead development and standardization of AI environmental metrics.
  • Convene experts and industry stakeholders.
  • Publish technical standards and guidelines.
White House Office of Science and Technology Policy (OSTP) Coordinating body
  • Coordinate multi-agency efforts.
  • Align AI metrics with climate and technology policies.
  • Integrate metrics into federal sustainability requirements.
Environmental Protection Agency (EPA) Environmental oversight
  • Lead environmental data collection and lifecycle impact studies.
  • Apply greenhouse gas accounting expertise.
  • Quantify carbon intensity and water usage.
Federal Energy Regulatory Commission (FERC) Grid and market support
  • Streamline interconnection processes for AI data centers.
  • Incorporate AI load growth into reliability assessments.
Congressional Committees Legislative oversight
  • Champion legislation and hold hearings on AI energy demands.
  • Support R&D funding and standards development.
  • Ensure agency accountability and progress reporting.

Example Metrics to Illustrate Shared Information

Metric Category Metric Name Definition Purpose/Benefit
Data Center Efficiency Metrics Power Usage Effectiveness (PUE) Ratio of total facility energy to IT equipment energy, refined for AI workloads. Measures overall data center energy efficiency for AI-specific operations.
Data Center Infrastructure Efficiency (DCIE) IT power versus total facility power (inverse of PUE). Alternative perspective focusing on IT equipment power proportion.
Energy Reuse Factor (ERF) Quantifies waste heat reused on-site. Measures ability to utilize waste heat, reducing energy needs.
Carbon Usage Effectiveness (CUE) Links energy use with carbon emissions (kg CO₂ per kWh). Provides holistic view of facility carbon intensity.
Environmental Metrics Energy Intensity Energy consumed per unit of data volume processed (kWh/GB). Reveals energy cost per data unit, useful for tuning AI models.
Annual Water Consumption Total liters of water used annually at data center level. Tracks water consumption for sustainability reporting.
AI Hardware & Compute Metrics Performance per Watt (PPW) Throughput of AI computations per watt of power. Encourages energy-efficient model training and inference hardware.
Compute Utilization Average utilization rates of AI accelerators (GPUs/TPUs). Ensures hardware is efficiently used rather than idling.
Training Energy per Model Total kWh or emissions per training run, normalized by model size or training hours. Quantifies energy cost of model development.

Conclusion

Artificial intelligence’s transformative potential must be balanced with commitments to energy security and environmental sustainability, in line with the United Nations Sustainable Development Goals (SDGs). This report outlines a strategic framework to measure AI’s environmental footprint through standardized metrics and coordinated federal agency action. The principle “what isn’t measured cannot be effectively managed” underscores the urgency of this initiative.

Standardized metrics and transparent reporting will enable responsible AI growth, ensuring data center expansion is matched by clean energy adoption, grid modernization, and efficiency improvements. This approach supports SDG 7 (Affordable and Clean Energy), SDG 9 (Industry, Innovation, and Infrastructure), SDG 12 (Responsible Consumption and Production), and SDG 13 (Climate Action).

The benefits extend to policymakers, communities, and the AI industry. Policymakers gain tools to balance innovation with environmental goals; communities are protected from unanticipated resource strain; and the AI sector benefits from a predictable regulatory environment that fosters innovation and cost savings.

This report is part of the AI & Energy Policy Sprint, a policy project focused on shaping U.S. policy at the intersection of AI and energy. For further information, visit AI & Energy Policy Sprint 2025.

1. Sustainable Development Goals (SDGs) Addressed in the Article

  1. SDG 7: Affordable and Clean Energy
    • The article discusses the surge in energy consumption by AI data centers and the need for clean energy integration and grid upgrades.
  2. SDG 9: Industry, Innovation, and Infrastructure
    • Focus on developing standardized metrics, promoting energy-efficient technologies, and innovation in AI hardware and software.
  3. SDG 12: Responsible Consumption and Production
    • Addressing e-waste, lifecycle emissions, and resource use transparency in AI data centers.
  4. SDG 13: Climate Action
    • Mitigating carbon emissions from AI data centers and ensuring alignment with climate goals.
  5. SDG 6: Clean Water and Sanitation
    • Managing water consumption for cooling AI data centers and promoting water efficiency.
  6. SDG 17: Partnerships for the Goals
    • Multi-agency coordination and collaboration between government, industry, and international bodies.

2. Specific Targets Under the Identified SDGs

  1. SDG 7: Affordable and Clean Energy
    • Target 7.2: Increase substantially the share of renewable energy in the global energy mix.
    • Target 7.3: Double the global rate of improvement in energy efficiency.
  2. SDG 9: Industry, Innovation, and Infrastructure
    • Target 9.4: Upgrade infrastructure and retrofit industries to make them sustainable, with increased resource-use efficiency.
    • Target 9.5: Enhance scientific research and upgrade technological capabilities.
  3. SDG 12: Responsible Consumption and Production
    • Target 12.4: Achieve environmentally sound management of chemicals and all wastes throughout their life cycle.
    • Target 12.6: Encourage companies to adopt sustainable practices and sustainability reporting.
  4. SDG 13: Climate Action
    • Target 13.2: Integrate climate change measures into national policies, strategies, and planning.
  5. SDG 6: Clean Water and Sanitation
    • Target 6.4: Increase water-use efficiency across all sectors.
  6. SDG 17: Partnerships for the Goals
    • Target 17.16: Enhance the global partnership for sustainable development.
    • Target 17.17: Encourage and promote effective public, public-private, and civil society partnerships.

3. Indicators Mentioned or Implied in the Article to Measure Progress

  1. Energy and Efficiency Indicators
    • Power Usage Effectiveness (PUE) – ratio of total facility energy to IT equipment energy.
    • Data Center Infrastructure Efficiency (DCIE) – IT power versus total facility power.
    • Energy Reuse Factor (ERF) – quantifies waste heat reuse on-site.
    • Carbon Usage Effectiveness (CUE) – kg CO₂ emitted per kWh of energy used.
    • Performance per Watt (PPW) – throughput of AI computations per watt of power.
    • Compute Utilization – average utilization rates of AI accelerators.
    • Training Energy per Model – total kWh or emissions per AI training run.
  2. Water Usage Indicators
    • Annual water consumption – total liters of water used annually at data centers.
    • Water Usage Effectiveness (WUE) – liters of water used per kWh of IT compute.
    • Cooling Energy Efficiency Ratio (EER) – output cooling power per watt of energy input.
  3. Environmental Impact Indicators
    • Carbon intensity per AI task – CO₂ emitted per training or per 1000 inferences.
    • Greenhouse Gas emissions per kWh – linked to actual grid emissions factors.
    • E-waste metrics – total hardware weight decommissioned annually and recycling ratio.
  4. Composite or Lifecycle Indicators
    • AI Sustainability Score – composite rating incorporating energy efficiency, renewables use, water efficiency, and recycling.
    • AI Energy Star rating – certification for AI hardware or services meeting efficiency and transparency criteria.

4. Table: SDGs, Targets and Indicators

SDGs Targets Indicators
SDG 7: Affordable and Clean Energy
  • 7.2: Increase share of renewable energy
  • 7.3: Double rate of energy efficiency improvement
  • Power Usage Effectiveness (PUE)
  • Data Center Infrastructure Efficiency (DCIE)
  • Energy Reuse Factor (ERF)
  • Performance per Watt (PPW)
SDG 9: Industry, Innovation, and Infrastructure
  • 9.4: Upgrade infrastructure for sustainability
  • 9.5: Enhance scientific research and technology
  • Training Energy per Model
  • Compute Utilization
  • AI Sustainability Score
SDG 12: Responsible Consumption and Production
  • 12.4: Environmentally sound management of waste
  • 12.6: Encourage sustainability reporting
  • E-waste metrics (hardware decommissioned, recycling ratio)
  • Carbon Usage Effectiveness (CUE)
  • AI Energy Star rating
SDG 13: Climate Action
  • 13.2: Integrate climate measures into policies
  • Carbon intensity per AI task
  • Greenhouse Gas emissions per kWh
  • Carbon Usage Effectiveness (CUE)
SDG 6: Clean Water and Sanitation
  • 6.4: Increase water-use efficiency
  • Annual water consumption
  • Water Usage Effectiveness (WUE)
  • Cooling Energy Efficiency Ratio (EER)
SDG 17: Partnerships for the Goals
  • 17.16: Enhance global partnerships
  • 17.17: Promote public-private partnerships
  • Multi-agency coordination and reporting frameworks
  • International standards alignment (ISO, EU AI Act)

Source: fas.org