Climate Tech and the Green Digital Transition

Executive Summary

The digital economy is simultaneously the greatest enabler of decarbonisation and one of the fastest-growing sources of energy demand. Digital technologies — smart grids, precision agriculture, remote work infrastructure, and AI-optimised logistics — have the potential to reduce global emissions by 15–20% according to the Global e-Sustainability Initiative (GeSI). Yet the digital infrastructure itself consumes approximately 1.5–2% of global electricity, a figure that the International Energy Agency (IEA) projects could double or triple by 2030 as artificial intelligence workloads drive unprecedented data centre expansion.

This research report analyses the tension between digital transformation and decarbonisation through the lens of externality theory and green growth economics. The central paradox — that the tools needed for climate action are themselves becoming a significant climate burden — demands policy frameworks that internalise the environmental costs of digital infrastructure while preserving its decarbonisation potential. We estimate that without policy intervention, AI-driven data centre growth could add 300–600 million tonnes of CO₂ annually by 2030, equivalent to the current emissions of France and Germany combined. Conversely, well-designed green digital policies could ensure that the net climate impact of digitalisation remains strongly positive.

The Energy Footprint of Digital Infrastructure: Scale and Trajectory

Global data centre electricity consumption reached approximately 460 TWh in 2024, according to the IEA — roughly equivalent to France's total electricity consumption. This figure has grown at 20–25% annually since 2022, driven overwhelmingly by AI training and inference workloads. Goldman Sachs projects that data centre power demand will reach 1,000–1,200 TWh by 2030, requiring approximately $250 billion in new power generation investment in the United States alone.

The energy intensity of AI is qualitatively different from previous digital workloads. Training a single large language model (such as GPT-4-class systems) requires an estimated 50–100 GWh of electricity — comparable to the annual consumption of 15,000–30,000 European households. Inference — the ongoing operational use of trained models — scales linearly with usage and now accounts for approximately 60% of AI-related energy consumption. As AI becomes embedded in search, productivity software, autonomous systems, and scientific research, inference demand is projected to grow at 40–50% annually through the end of the decade.

The geographic distribution of this demand is highly concentrated. The United States hosts approximately 40% of global data centre capacity, followed by the EU (16%), China (15%), and the UK (5%). Within these regions, data centres cluster in areas with reliable power, low latency, and favourable tax regimes — often in regions where grid electricity is carbon-intensive. Northern Virginia, the world's largest data centre market, draws primarily from the PJM Interconnection, where the carbon intensity of electricity remains approximately 350 gCO₂/kWh — well above the level consistent with net-zero trajectories.

The Jevons Paradox in Digital Energy Efficiency

Proponents of digital growth often cite improvements in computational efficiency: energy per computation has fallen by approximately 50% every 2.5 years, consistent with improvements in chip design and data centre operations. Power Usage Effectiveness (PUE) — the ratio of total data centre energy to IT equipment energy — has improved from approximately 2.0 in 2010 to 1.2–1.3 in modern hyperscale facilities. Google, Microsoft, and Amazon have all committed to operating on 100% renewable energy.

However, these efficiency gains are overwhelmed by demand growth — a manifestation of the Jevons Paradox, first articulated in 1865 by William Stanley Jevons regarding coal consumption. As computational efficiency improves and costs decline, total computational demand increases faster than efficiency gains, driving net energy consumption upward. The IEA observes that despite a 40% improvement in average data centre PUE between 2015 and 2025, total data centre energy consumption increased by approximately 80% over the same period.

The AI era intensifies this dynamic. Unlike traditional cloud computing — where workloads are largely elastic and can be scheduled during periods of renewable energy availability — AI training runs are continuous, multi-week computations that require sustained power. AI inference workloads, while individually small, are latency-sensitive and must be served from facilities near end users, limiting the ability to locate AI infrastructure exclusively in regions with abundant renewable energy.

Corporate Renewable Energy Claims: Additionality and Temporal Matching

Major technology companies' claims of operating on "100% renewable energy" require careful scrutiny. The standard approach involves purchasing Renewable Energy Certificates (RECs) or Power Purchase Agreements (PPAs) equivalent to annual electricity consumption. However, this accounting methodology — annual matching — permits a data centre to consume fossil-fuel-generated electricity at night while claiming renewable status based on daytime solar generation credits. The temporal and spatial mismatch between renewable generation and data centre consumption means that annual matching significantly overstates the actual decarbonisation impact.

Google has pioneered a shift toward "24/7 carbon-free energy" (CFE) matching — ensuring that every hour of electricity consumption is matched by carbon-free generation in the same grid region. This more rigorous standard reveals the gap: Google reported 64% 24/7 CFE across its global operations in 2023, compared to its 100% annual renewable energy claim. The EU Energy Efficiency Directive (recast 2023) now requires data centres above 500 kW to report energy consumption and carbon footprint data, a step toward greater transparency but not yet requiring temporal matching.

The additionality question is equally important. RECs from existing renewable installations do not drive new clean energy deployment. Only PPAs for new-build renewable projects create genuine additionality. BloombergNEF estimates that approximately 45% of corporate renewable energy procurement in 2024 involved new-build projects — meaning that over half of corporate renewable claims rest on existing renewable capacity that would generate clean energy regardless.

The Decarbonisation Potential: Digital as Climate Solution

Against the energy cost of digital infrastructure, digital technologies offer substantial decarbonisation potential across sectors. The GeSI estimates a potential 15–20% reduction in global emissions through digital enablement, concentrated in:

• Energy systems (5–7% reduction): Smart grids, demand response, distributed energy resource management, and AI-optimised grid operations. The IEA estimates that AI-enabled grid optimisation could reduce curtailment of renewable generation by 30–40%, unlocking 150–200 TWh of clean energy annually.
• Transport and logistics (3–5% reduction): Route optimisation, autonomous vehicle platooning, shared mobility platforms, and virtual alternatives to physical travel. McKinsey estimates that AI-optimised logistics could reduce freight emissions by 20–25% by 2030.
• Industry and manufacturing (2–3% reduction): Digital twins, predictive maintenance, process optimisation, and supply chain transparency. The World Economic Forum estimates that digital manufacturing solutions could reduce industrial emissions by 1.5 GtCO₂e annually.
• Buildings and cities (2–3% reduction): Smart building management, urban planning optimisation, and remote work infrastructure reducing commuter emissions.
• Agriculture (1–2% reduction): Precision agriculture, satellite-enabled monitoring, and AI-optimised input management reducing fertiliser use and land conversion.

The net climate equation — digital emissions versus digital-enabled emission reductions — remains strongly positive. We estimate a net benefit ratio of approximately 5:1 to 10:1 under current trajectories. However, this ratio is declining as AI energy demand grows, and policy intervention is needed to ensure it remains favourable.

Policy Frameworks: Internalising the Digital Carbon Externality

The environmental costs of digital infrastructure represent a classic negative externality: data centre operators and their customers do not bear the full social cost of the carbon emissions associated with their energy consumption. Standard externality theory prescribes Pigouvian taxation or cap-and-trade mechanisms to align private incentives with social costs.

1. Sectoral Carbon Pricing for Data Centres. Including data centres explicitly within emissions trading systems (the EU ETS already covers electricity generation, indirectly pricing data centre emissions in Europe) with transparent reporting requirements. Singapore's carbon tax, extended to data centre operators in 2024 at S$25/tCO₂ (rising to S$50–80 by 2030), provides a pioneering model. Direct carbon pricing for data centres would incentivise both efficiency improvements and renewable energy procurement with genuine additionality.

2. Mandatory 24/7 Carbon-Free Energy Standards. Moving beyond annual renewable energy matching to require hourly temporal and spatial matching for new data centre developments. Ireland's moratorium on new Dublin data centre connections (2022–2024) and Singapore's data centre pause (2019–2022) demonstrated that capacity constraints can be effective, but blanket moratoria risk driving investment to less regulated jurisdictions. Conditional permitting — granting data centre approvals contingent on 24/7 CFE commitments and local grid capacity — offers a more targeted approach.

3. Green Digital Procurement Standards. Government digital procurement — representing approximately $600 billion annually across OECD countries — should incorporate carbon intensity requirements, creating demand-side incentives for green data centre operations. The US Federal Sustainability Plan and EU Green Public Procurement criteria provide frameworks, but neither currently includes data centre-specific carbon intensity thresholds.

4. AI Efficiency Standards. Just as energy efficiency standards (such as ENERGY STAR) transformed appliance markets, AI-specific efficiency standards — measuring computational output per unit of energy — could drive industry-wide efficiency improvements. The Green Software Foundation's Software Carbon Intensity specification and the MLPerf benchmark suite provide technical foundations for such standards.

Implications for GDEF's Technology & Transformation Working Group

The green digital transition requires policy frameworks that are technologically informed, internationally coordinated, and adaptive to rapidly evolving digital technologies. GDEF's Technology & Transformation Working Group will develop a Green Digital Infrastructure Charter, proposing multilateral standards for data centre sustainability reporting, renewable energy procurement, and AI efficiency, for endorsement at the 2026 Annual Summit.

References & Sources

1. IEA, World Energy Outlook 2025. International Energy Agency. iea.org/reports/world-energy-outlook-2025
2. Goldman Sachs, AI, Data Centres and the Coming US Power Demand Surge. Global Investment Research, 2024. goldmansachs.com/insights
3. IPCC, Sixth Assessment Report: Mitigation of Climate Change. Working Group III, 2022. ipcc.ch/report/ar6/wg3
4. BloombergNEF, Energy Transition Investment Trends 2025. about.bnef.com
5. GeSI, Digital with Purpose: Delivering a SMARTer2030. Global e-Sustainability Initiative. gesi.org
6. Google, 24/7 Carbon-Free Energy: Methodology and Results. Google Sustainability Report 2024. sustainability.google
7. Jevons, W.S. (1865). The Coal Question: An Inquiry Concerning the Progress of the Nation, and the Probable Exhaustion of Our Coal-Mines. London: Macmillan and Co.
8. McKinsey & Company, The Net-Zero Transition: What It Would Cost, What It Could Bring. McKinsey Global Institute, 2022. mckinsey.com
9. World Economic Forum, The Future of Industrial Decarbonisation. WEF Centre for Energy and Materials, 2025. weforum.org/publications
10. Green Software Foundation, Software Carbon Intensity Specification. greensoftware.foundation