As tech firms and investors race to frame artificial intelligence (AI) as key to solving the climate crisis, the deployment of AI tools to curb the energy intensity of the technology is not keeping pace with the sector’s voracious appetite for power, a new report from the International Energy Agency (IEA) suggests.
The report, Key Questions on Energy and AI, finds that while tech firm are pouring billions into data centres with Southeast Asia emerging as a critical hub for AI infrastructure, the energy sector is still struggling to adopt AI tools that could improve efficiency, reduce emissions and strengthen grid reliability.
The deployment gap comes as electricity demand from AI-focused data centres surged 50 per cent in 2025 alone, driven by explosive growth in energy-intensive applications such as video generation, AI reasoning and autonomous “agentic” systems.
IEA said that capital expenditure by the world’s largest tech firms exceeded US$400 billion in 2025 and is expected to rise by another 75 per cent in 2026. Spending by just five tech companies now exceeds total global investment in oil and gas production, it noted.
However, the agency also found that deployment of AI within the energy industry remains slow and fragmented. While AI has the potential to improve energy security and sustainability, a survey of energy companies by IEA found that a lack of digital skills, fragmented data systems, cybersecurity concerns and weak policy support are holding back adoption.
Globally, less than half of energy demand is covered by policy frameworks aimed at promoting AI uptake in the energy sector, while only 10 per cent of global electricity consumption falls under open electricity data policies.
The report highlights a widening gulf between the rapid growth of AI infrastructure and the slower pace of reform in the physical, social and economic systems needed to support it.
Energy use by data centres is projected to grow continually beyond the mid 2030s, driven by AI [click to enlarge]. Source: IEA
The IEA estimates that global electricity demand from data centres will nearly double from 485 terawatt-hours (TWh) in 2025 to 950 TWh by 2030, equivalent to around 3 per cent of global electricity demand. Electricity consumption from AI-focused data centres — which are designed to support much higher loads than traditional facilities — is expected to triple over the same period.
Using satellite tracking, IEA found that “AI factories” — specialised data centres designed for training and running advanced AI models — have more than tripled in capacity over the last 18 months.
While the sector has made strides to improve efficiency — the energy consumed per AI task has fallen dramatically, with simple AI text queries now consuming less electricity than running a television over the same period of time — these gains are being overwhelmed by soaring demand and the emergence of more energy-intensive AI applications.
Video generation, advanced reasoning models and autonomous AI agents can consume hundreds or even thousands of times more electricity per query than simple chatbot interactions. An individual server rack in an advanced data centre could, by 2027, have peak power demand equivalent to 65 households, the report projected.
Energy bottlenecks and ESG pushback
The report warns that physical energy infrastructure and supply chains are struggling to keep up with demand. Grid connection delays, shortages of transformers and power electronics, limited supplies of high-bandwidth memory chips and a surge in demand for gas turbines are all creating bottlenecks.
Orders for gas turbines to power data centres rose 70 per cent in 2025, highlighting mounting pressure on energy equipment supply chains. In the United States, some developers are moving ahead with onsite natural gas generation because of slow grid connection timelines.
The IEA estimates that between 15 and 27 gigawatts of onsite gas-fired power capacity could be supplying data centres globally by 2030.
AI is also driving growth in battery storage; data centres could host between 20 and 25 gigawatts of battery storage capacity globally by 2030, potentially allowing them to act as flexible grid assets if the right market incentives are introduced, IEA said.
As data centre construction accelerates, public opposition is also growing in some jurisdictions over electricity prices, water consumption, environmental impacts and land use. In February, communities in Johor, one of the world’s largest hubs for data centre infrastructure, protested against the water intensity of AI-based IT infrastructure.
The report noted that data centres have become a “highly visible flashpoint” for anxieties about AI’s impact on jobs, affordability and sustainability. Emissions from data centres are projected to double by 2035 to around 350 million tonnes, although the IEA estimated that they would still account for only around 2 per cent of global electricity-sector emissions.
The agency warned that poor planning could increase electricity prices in some regions if grid investments and data centre demand become misaligned. “Data centres are large, concentrated, rapidly developed infrastructure that are likely to trigger a need for new generation and grid investments in systems that host them,” the report said.
The uncertainty surrounding future AI demand further complicates planning, IEA said. While some data centre projects may never be completed, investment pipelines continue to expand rapidly, fuelled by expectations that AI will transform economic growth, productivity and industrial competitiveness.
Despite questions around the huge and growing environmental and societal impact of AI, the IEA argued that the tech could still play a key role in accelerating the energy transition — if adoption barriers are addressed. AI technologies are already being used to monitor transformers and electricity grids, optimise industrial processes and improve energy efficiency.
The agency estimates that existing AI applications could save more than 13 exajoules of energy by 2035 — equivalent to around 3 per cent of global final energy consumption — if deployment scales up. In energy-intensive industries, AI-enabled optimisation could reduce energy costs by between 3 and 10 percentage points.
