AI and ESG governance addresses the two-way relationship between artificial intelligence and environmental, social, and governance reporting. On one side, AI systems create ESG impacts that need to be measured and disclosed: energy consumption from model training and inference, water usage for data centre cooling, electronic waste from AI hardware, workforce displacement, algorithmic bias affecting communities, and governance structures for AI oversight. On the other side, AI is increasingly used to improve ESG measurement, reporting, and performance, from satellite-based environmental monitoring to supply chain transparency analytics. For organisations subject to ESG reporting obligations, the governance challenge is ensuring that AI-related ESG impacts are accurately captured in disclosures, and that AI systems used for ESG reporting are themselves governed, validated, and transparent.

Environmental impact of AI

The environmental footprint of AI is material and growing. Training large AI models requires significant computational resources and energy. Use the current vintage. The International Energy Agency's Key Questions on Energy and AI, published 16 April 2026, supersedes the 2024 baseline that most commentary still quotes: data centre electricity demand "grew by 17% in 2025, in line with IEA projections", and "electricity consumption from AI-focused data centres grew even faster, surging 50% in 2025". The IEA's updated projections "see electricity consumption from data centres roughly doubling from 485 TWh in 2025 to 950 TWh in 2030, accounting for around 3% of global electricity demand by that date", and it describes the trajectory as remaining close to the one set out in its 2025 report.

Three things to carry with those numbers. First, 950 TWh is a projection, and the IEA labels it a central projection rather than a certainty. Second, the widely quoted range of roughly 700 to over 1,700 TWh belongs to 2035, across the named cases in the IEA's 2025 report, and should never be attached to 2030. Third, the IEA does not publish a figure for AI's share of total data centre electricity, so anyone who gives you one is estimating. A number repeated without its year, its scenario and its boundary is the most common failure in this subject, and it is worth distrusting even when it points the direction you expect. Water consumption for data centre cooling is an emerging concern, a single large data centre can consume millions of litres of water annually. For organisations subject to climate disclosure requirements, principally the EU CSRD and the ISSB standards as adopted in each jurisdiction, AI-related emissions must be reported, but which scope depends on who owns the machines, and most organisations get this backwards. The GHG Protocol confines Scope 2 to electricity "purchased or otherwise brought into the organizational boundary of the company". A cloud contract does not do that: it buys a service, and the Scope 3 standard is explicit that products "include both goods (tangible products) and services (intangible products)". So AI workloads run in a third-party cloud are the provider's Scope 2 and the customer's Scope 3 Category 1, purchased goods and services. Scope 2 is the right home only where you run the hardware yourself, or where a colocation arrangement has you buying the power directly; otherwise colocation lands in Scope 3 Category 8, upstream leased assets. Getting this wrong does not just misplace a number, it puts it in a category with different assurance expectations. The United States is not one of those jurisdictions today: the SEC adopted climate-related disclosure rules in March 2024 but stayed them the following month, they never took effect, the Commission ended its defence of them in March 2025, and on 3 June 2026 it published a proposed rule to rescind them in their entirety. US-listed organisations should plan against state-level regimes such as California's climate disclosure laws and against their own investors' expectations, not against the SEC rules. Organisations using cloud-based AI services should obtain emissions data from their cloud providers and include it in their reporting.

Social impact of AI

AI's social impacts include workforce effects (job displacement, skill requirements, working conditions under algorithmic management), consumer impacts (algorithmic bias in credit, insurance, housing, and employment decisions), and community effects (surveillance, content moderation, access to services). ESG reporting frameworks increasingly expect disclosure of how organisations manage these impacts. The EU AI Act's fundamental rights impact assessment requirement for high-risk AI directly intersects with the social dimension of ESG reporting.

Governance dimension

The governance pillar of ESG maps directly to AI governance: board oversight of AI, accountability structures, risk management frameworks, ethics policies, audit mechanisms, and stakeholder engagement. For investors and analysts evaluating ESG performance, AI governance maturity is becoming a proxy for broader governance quality, organisations that govern AI well are likely governing other complex risks well too.

Practical integration

Integrate AI into your ESG materiality assessment, determine which AI-related ESG impacts are material to your stakeholders. Include AI energy consumption in environmental reporting. Disclose workforce AI impacts in social reporting. Report AI governance structures, policies, and oversight mechanisms in governance disclosures. Ensure AI systems used for ESG reporting and measurement are themselves governed, validated, and transparent, an AI system that produces inaccurate ESG data creates securities law exposure for misleading disclosures.

Related reading

Further reading: IEA, Energy and AI | OECD AI Principles