Understanding the Energy-AI Nexus

IEA (2025), Energy and AI, IEA, Paris https://www.iea.org/reports/energy-and-ai, Licence: CC BY 4.0

Artificial intelligence (AI) has a long history, dating back to at least the 1950s. Over time, it has seen a series of alternating periods of optimism and pessimism (so-called “AI winters”). In recent years, however, AI has been dramatically boosted by several developments and breakthroughs in techniques, costs and technology that have led to the rise of AI in its modern form that we are familiar with today, in particular generative AI. These developments include the massive increase in computing power and decline in cost due to exponential improvements in computing hardware performance; the exponential increase in the availability and quality of data used to train AI models due to the rise of the Internet and connectivity; and breakthroughs in the architectures and algorithms behind AI models, enabling the development of models that are exponentially larger and more capable.

Capitalising on the perceived potential of AI, technology companies have come to dominate the stock market – notably in the United States, which hosts some of the world’s largest technology companies. From November 2022 – when ChatGPT launched – to the end of 2024, 65% of the growth in market capitalisation of the S&P 500 came from companies that either deploy AI or integrate AI into their core operations. That is, of the USD 16 trillion rise in market capitalisation of S&P 500 companies, USD 12 trillion came from AI-related companies alone. This period was marked by a surge in AI-related investor expectations before the recent volatility in financial markets. AI-focused start-ups in the United States have also grown in value faster than non-AI start-ups. In 2024, by the time start-ups reached their fourth round of funding, AI-focused start-ups had an average valuation five times higher than that of other start-ups. 

AI capabilities have been evolving rapidly. AI models and applications have been steadily adding new capabilities, giving users access to tools that approach or even exceed human-level capabilities on some tasks in limited contexts. Ultimately, energy demand from AI will depend on, among other factors, the speed and scale of uptake, which in turn depends on AI’s usefulness and impact. The energy sector therefore needs to grapple with the capabilities of AI systems as it considers the outlook for AI adoption. While AI related benchmarks have limitations that necessitate a cautious approach to assessing their implications, they offer a window into the evolving capabilities of AI models that can complement data on real-world deployment.

As a result, in regions where data centres are concentrated, the share of electricity demand going to data centres is disproportionately high. In Ireland, for example, data centres consume around 20% of the metered electricity supply. There are six states in the United States where data centres already consume over 10% of the electricity supply, with Virginia leading at 25%. 

Power reliability is an important barrier to address. Many emerging market and developing economies face electricity supply challenges that complicate local hosting. In regions with frequent outages, maintaining a data centre often demands costly backup power systems, making overseas hosting or cloud services more appealing for businesses.

There is also consideration of the impacts of data centre energy demand on the broader electricity system. In some Latin American and African countries, for example, a stark contrast exists between large-scale data centre investments and everyday energy challenges. In some such countries, it is not unusual for remote communities to experience severe power scarcity, even as new data centre investments intensify competition for local energy demand. This reality underscores the critical need for reliable, locally sourced electricity to bridge the digital and energy divides in emerging markets.