AI & Sustainability: a double-edged transformation

As artificial intelligence becomes increasingly embedded in the global economy, it is generating both new environmental pressures and new opportunities for efficiency. On one hand, AI has a growing footprint through its demand for electricity, water, minerals and computing infrastructure. On the other, it creates a handprint by enabling smarter decisions, cleaner technologies and more efficient systems. Whether AI becomes a net positive for sustainability will depend on whether these benefits can offset its costs on a large scale and over the long run.

The hidden footprint of progress

Energy, emissions, water and materials. AI is rapidly increasing the environmental footprint of data centers. Although there is no single audited figure for the share of data-center capacity dedicated to AI, estimates suggest AI currently accounts for around 8% of data-center energy use, a share that could rise to 36% within three years. Global data centers consumed approximately 415 TWh of electricity in 2024, representing about 1.5% of worldwide electricity demand. Their electricity use grew by 17% in 2025, far outpacing the 3% growth in global electricity demand. By 2030, data-center electricity consumption could double, with AI as the main driver, while energy use in AI-dedicated facilities could triple alongside investments expected to reach up to $7 trillion. Although AI systems are becoming more energy-efficient, growing adoption and increasingly powerful AI agents may outweigh these gains. Beyond energy consumption, AI infrastructure also places pressure on water resources through cooling requirements, while semiconductor production and frequent server replacements contribute to embodied emissions, demand for critical minerals, and rising volumes of electronic waste.

Per-prompt estimates of AI-related water consumption vary considerably. For example, PwC cites Mistral’s estimate of 45 mL for a 400-token response generated by Mistral Large 2, roughly equivalent to three tablespoons of water, whereas Google reports 0.26 mL for a median Gemini text prompt, or about five drops. These figures are not directly comparable, as they depend on factors such as the model used, response length, geographic location, cooling systems, and accounting methodologies. Moreover, image and video generation, as well as extended reasoning tasks, can require substantially more resources than a short text response.

These are scenario-dependent estimates for the United States (not observed global totals) but they illustrate why location, power mix and cooling technology matter.

Local communities. Also, community acceptance is increasingly becoming a critical factor in determining whether projects can proceed as planned. In the United States alone, 75 data center projects valued at $130 billion were blocked or delayed in the first quarter of 2026, highlighting the growing importance of local support for successful project delivery.

Economic trade-offs. AI investment can raise productivity and create demand for clean power, storage and grid upgrades. Yet concentrated data-center loads may strain grids, delay other connections and affect local electricity affordability if costs are not allocated fairly. The WEF argues that AI-related clean-energy investment should strengthen power systems for wider society, rather than simply secure supply for technology firms.

AI’s sustainability paradox

AI as an enabler of transition. AI can help forecast renewable generation and demand, balance electricity networks, manage buildings, find methane leaks, optimize industrial processes and logistics, reduce agricultural inputs, and accelerate climate or materials research. These uses can cut waste across whole systems – not only within technology companies. The WEF and IEA see meaningful potential, while noting that adoption, data quality and energy-system constraints remain barriers.

The rebound risk. Cheaper or more efficient services can trigger more consumption, eroding expected savings. AI-optimized logistics might reduce fuel per delivery but increase total deliveries; more efficient computing may also make much heavier use affordable. The long-term result therefore depends on clean electricity, low-impact hardware, efficient models and policy or company incentives that turn productivity gains into absolute resource and emissions reductions.

Navigating the ESG trade-offs

E — Environmental performance. Near term, AI can help forecast energy use, monitor emissions, improve predictive maintenance, reduce scrap and support supplier-risk screening. It can also speed up ESG data collection and reporting (because AI is making disclosure cheap and routine). But AI can add to Scope 2 emissions through electricity use and Scope 3 impacts through cloud services and hardware. Automated reporting is not automatically accurate: proxy data, missing supplier information and generated errors can create false precision or greenwashing.
S — People and decent work. AI will automate some routine tasks and reduce demand for certain roles, particularly in administration, data entry, bookkeeping and customer support. However, job exposure does not necessarily mean job loss: the ILO estimates that one in four workers is in an occupation affected by GenAI, but most jobs are likely to be transformed rather than eliminated. In the short term, firms may need fewer junior staff as tasks become faster to complete. Over time, demand may shift toward oversight, verification, complex problem-solving and client-facing judgement. The social impact will depend on reskilling efforts, worker consultation and whether organisations redeploy employees or simply reduce hiring.
G — Governance and trust. Bias in recruitment or performance management, privacy breaches, opaque decisions, cybersecurity and unclear accountability can translate into material ESG and legal risks. Negative impacts on sustainability goals commonly reported are AI-related risks, reinforcing that unmanaged adoption can undermine the benefits it promises. Human accountability, documented controls and traceable data are essential, especially where AI supports material ESG disclosures or decisions affecting people.

What it looks like in practice. A factory can use computer vision to catch defects before a whole batch is scrapped; a supermarket can forecast demand to reduce food waste; a facilities team can adjust heating and cooling to match occupancy; and procurement can read supplier invoices to estimate Scope 3 emissions. In each case, track the actual waste, energy or emissions reduction and check that savings are not cancelled by more production or consumption.

From AI ambition to responsible impact

Realizing AI’s potential requires looking beyond innovation and productivity gains to consider its broader environmental and social consequences. Embedding sustainability into the AI business case can help organizations create value while identifying and managing associated risks. A practical first step is to establish a baseline for each significant use case, assessing not only energy consumption and greenhouse gas emissions but also water use, hardware requirements and the business benefits delivered. Choosing the smallest model capable of meeting a given need can reduce unnecessary resource consumption, while comparing outcomes against a credible baseline helps determine whether efficiency improvements lead to genuine reductions in impact or simply stimulate additional demand.

At the same time, AI can strengthen ESG reporting and sustainability decision-making when supported by robust governance. Reliable reporting still depends on high-quality source data, validation processes, audit trails and human oversight. Similarly, organizations should assess how AI may affect jobs and skills, engaging employees and investing in reskilling where necessary. Clear governance frameworks also remain essential for addressing issues such as privacy, cybersecurity, fairness and accountability.

The opportunities are substantial. AI can improve the forecasting of renewable energy generation and electricity demand, optimize energy-system flexibility and support predictive maintenance of critical infrastructure. It can also help consolidate and analyze data from diverse sources, including utility bills, supplier databases, IoT sensors and financial systems. In sustainability reporting, AI can assist with data classification, quality checks and disclosure drafting, while enabling organizations to benchmark reports against regulatory requirements and identify potentially misleading ESG claims. In procurement and supply-chain management, it can support the classification of supplier spending into Scope 3 emissions categories, detect unusual patterns in energy use or emissions and strengthen the traceability of reported KPIs. Ultimately, however, the value of these applications depends not only on the technology itself but also on the quality of the underlying data, the effectiveness of governance processes and the ability to demonstrate measurable outcomes.