There’s no doubt that artificial intelligence is revolutionizing the world, but its energy appetite is raising serious concerns. A report from the International Energy Agency (IEA), published this Thursday, April 10, reveals an exponential trajectory in data center electricity consumption, driven mainly by the meteoric rise of generative AI.
The figures are telling: from 415 terawatt-hours (TWh) in 2024—already representing 1.5% of global electricity consumption—data center needs are expected to more than double by 2030, reaching around 945 TWh.
This astronomical amount is equivalent to Japan’s current energy consumption and will bring the global share of these infrastructures to nearly 3% of the world total.
Over the past five years, this demand has already grown by 12% per year—an intense pace directly linked to the mass adoption of generative AI and its colossal computing needs.
AI data centers: energy-hungry giants concentrated
The scale of this consumption is staggering. A single 100-megawatt data center uses as much electricity in a year as one million households.
And current projects aim for facilities twenty times more powerful, capable of drawing the equivalent of two million households’ consumption. These energy behemoths are mainly concentrated in the United States, Europe, and China, which together account for 85% of global data center consumption. And projects like Stargate launched by Donald Trump in partnership with OpenAI won’t help the situation.
This geographic concentration creates major challenges for energy supply, especially near large metropolitan areas.
Read also on this topic: Does AI pollute? The answer is surprising and… complicated
Artificial intelligence has a growing impact on the climate
This growing energy thirst from data centers is not without climate consequences. Carbon dioxide (CO2) emissions linked to their operation are expected to jump from 180 million tons today to 300 million tons by 2035. We can also mention that AI consumes water indirectly, precisely to cool data centers.
Although this may seem marginal compared with the 41.6 billion tons emitted globally in 2024, data centers are among the fastest-growing sources of emissions.
In the United States, where they could account for half of the additional electricity demand, the energy issue has become a major political topic, with Donald Trump launching a council to boost production in response to China.
Which generative AIs are the most energy-hungry?
While the IEA report does not detail the specific consumption of each generative AI model, it’s clear that the most complex and most heavily used architectures are the most energy-intensive.
This includes large language models (LLMs) used for natural language processing, machine translation, text and code generation, such as:
- models from the GPT family (OpenAI), also used by our Yiaho platform,
- LaMDA (Google),
- Llama 4 (Meta).
Training them requires massive amounts of data and considerable computing power, mobilizing entire server farms for weeks, or even months.
Likewise, generative AI models dedicated to creating images, videos, and sounds—such as DALL-E 2, Midjourney, or video generation models—require significant processing power to handle and generate complex multimedia data.
Training and inference (using the trained model) for these AIs contribute significantly to overall energy consumption.
Read also: ChatGPT & Ghibli: Between revolution and ethical debates
Energy solutions to explore?
Faced with this escalation, the question of energy sources becomes crucial. Currently, coal still covers 30% of data centers’ energy needs. However, the IEA anticipates a growing dominance of renewables and natural gas, favored by their cost and availability.
The IEA also highlights AI’s potential paradox: while it is a source of increased energy consumption, it could also help reduce emissions through innovations in optimizing power grids, demand management, and discovering more efficient materials.
However, the agency warns about potential rebound effects and a continued reliance on fossil fuels, which could significantly limit this positive impact.
The IEA report is a clear wake-up call: without proactive policies and coordinated efforts to optimize AI energy efficiency and prioritize sustainable energy sources, the artificial intelligence revolution could run into an energy and climate wall.
AI will not be a miracle solution for the energy transition if it is not deployed with a clear awareness of its environmental impact.
