Key points
- AI boosts efficiency but drives power demand
- Data centres strain grids and emissions goals
- Net-zero pledges clash with AI expansion
ISLAMABAD: As 2025 closes, artificial intelligence has become both a climate tool and a climate headache — boosting efficiency in energy systems while driving a fresh surge in electricity demand, data-centre construction and emissions scrutiny.
The tension is now hard to ignore. In April, the International Energy Agency (IEA) warned that global electricity demand from data centres is on track to more than double by 2030 to about 945 terawatt-hours, with AI the most significant driver. The agency said electricity demand from AI-optimised data centres could more than quadruple over the same period.
Those long-range projections collided with very real, very local pressure in 2025 — from grid bottlenecks to renewed reliance on fossil “peaker” plants. A Reuters report in December described how the rise of AI data centres in the United States is helping keep older peaker plants online, reversing planned retirements and raising concerns about pollution and environmental justice in nearby communities.
In the UK, analysis cited by The Times reported electricity demand grew 3% in 2025, the fastest rise in more than two decades, with heat pumps, EVs and “increasingly energy-intensive AI data centres” among the drivers.
The broader picture is similar across advanced economies: the IEA says data centres could account for more than 20% of electricity-demand growth through 2030 in those markets.
Data centres, emissions — and a growing backlash
For the biggest tech firms, the public climate narrative has become trickier. Microsoft’s latest environmental disclosures have fuelled debate about whether AI expansion is compatible with corporate net-zero pledges. Coverage of Microsoft’s 2025 Environmental Sustainability Report noted the company’s emissions have increased 23.4% since 2020, with new data-centre build-outs tied to AI and cloud growth cited as a key factor.
Google, meanwhile, has highlighted efficiency improvements — but also disclosed the rising resource demands behind the scenes. Its 2024 Environmental Report (covering 2023 data) said Google’s data centres consumed 6.1 billion gallons of water in 2023, up 17% year-on-year, alongside similar growth in electricity use.
Water use has become a flashpoint in regions where drought risk is rising and communities are questioning whether new data-centre permits reflect local priorities.
The build-out itself is also part of the climate equation. The Guardian reported that investment in data centres worldwide hit a record $61bn in 2025, describing a “construction frenzy” tied to AI demand.
More construction means more embedded carbon — from concrete and steel to supply chains — and more long-term demand on grids that are already being stretched by electrification.
In some markets, developers are turning to stop-gap power solutions to get capacity online faster. The Financial Times reported that grid-connection delays are pushing some data centres towards on-site generation — including aeroderivative turbines — raising fresh questions about fossil fuel lock-in even as companies promise cleaner operations.
Climate tech wins — and why they still matter
Yet focusing only on AI’s footprint risks missing the other side of the ledger: the climate benefits that have made AI so attractive to policymakers and energy executives in the first place.
The IEA argues AI can help cut costs, improve system efficiency and potentially reduce emissions — for example by optimising power grids, predicting equipment failures, improving renewable integration and accelerating energy innovation.
These aren’t abstract promises. Grid operators increasingly use advanced analytics to balance variable wind and solar supply; industrial firms deploy machine learning to reduce waste; and scientists use AI to speed discovery in batteries and materials.
There is also evidence that efficiency gains in computing are real — even if they are not guaranteed to outweigh demand growth. Research cited by the MIT Press’s Harvard Data Science Review discusses how better hardware and practices can reduce the carbon intensity of training large models over time, while cautioning that the full lifecycle footprint (including data-centre construction and broader infrastructure) is often undercounted in headline comparisons.
That accounting gap is one reason the debate remains heated: critics argue the industry is too quick to talk about “net” benefits without transparent, standardised measurements of energy use, emissions and water across training, inference and deployment.
So… did AI help or hurt in 2025?
The honest answer is: both, but not evenly — and not automatically.
In 2025, AI’s direct climate cost became harder to dismiss, with increased electricity demand and infrastructure expansion colliding with grid constraints and, in some regions, a return to dirtier generation at the margins.
At the same time, the most convincing climate-tech gains tended to be targeted and practical: optimisation tools that save energy, reduce downtime, and help integrate renewables — rather than open-ended claims that “AI will solve climate change”.
The direction of travel for 2026 may depend less on the models themselves and more on the policy and procurement choices around them: where data centres are built, what powers them hour by hour, how water is sourced, and whether firms are required to disclose real-world impacts in comparable ways.
The IEA has called for faster investment in power generation and grids, greater data-centre efficiency and stronger coordination between governments, the tech sector and the energy industry.
Without that alignment, AI could become another driver of emissions growth. With it, proponents argue, the technology could still prove a meaningful lever in the climate transition — even as the world learns to pay its energy bill.
