By Professor Arthur GO Mutambara
Artificial Intelligence’s Threat to Energy, Water and the Environment.
Artificial Intelligence (AI) systems need to be powered.
They require compute resources, comprising hardware and infrastructure components – primarily Central Processing Units (CPUs), Graphics Processing Units (GPUs), Tensor Processing Units (TPUs), Neural Processing Units (NPUs), memory, storage, and networking – that provide the processing power, data handling, and parallel-computation capabilities required to train, run, and scale AI models and other intensive workloads.
These compute resources reside in energy-hungry data centres and AI factories.
A data centre is a physical facility that houses IT infrastructure, such as servers and storage, to manage, process, and store data.
An AI Factory (which leverages data centres) is designed to develop, train, and deploy AI models at scale, focusing on computational resources and AI workflows.
The power requirements of data centres and AI factories are at the gigawatt level. For example, a typical AI factory’s energy requirement is 2 GW (2000 MW), equivalent to that of a city like San Francisco or a country like Zimbabwe.
Furthermore, there is a demand for large volumes of fresh water for cooling (to prevent hardware overheating), which puts pressure on the available water supply for human consumption, agriculture, and other industrial uses.
Hence, the large carbon footprint of AI systems primarily stems from the significant compute resources required for energy-intensive tasks such as training and deploying Large Language Models (LLMs).
These tasks require high-performance hardware, such as GPUs and TPUs, which consume substantial electricity.
For example, training a single LLM can emit hundreds of tonnes of carbon dioxide equivalent, comparable to the emissions of several cars over their lifetime.
Furthermore, as explained earlier, these models are trained and run in energy-hungry data centres, which must also be cooled.
Larger AI models require exponentially more energy.
A model with billions of parameters demands far more computational power than a smaller one, resulting in a significantly larger carbon footprint.
Once trained, AI models continue to consume energy during deployment, particularly for real-time or large-scale applications.
Common examples include AI in content recommendation systems, search engines, and real-time translation, which require continuous processing power.
The iterative nature of AI research, where models are trained and retrained to optimise performance, means that significant energy is expended in the development phase before a final model is deployed.
Hence, a country’s participation in driving the AI revolution (through data centres and AI factories) carries high environmental costs.
According to the International Energy Agency, total global electricity consumption by data centres could reach the level of Japan’s energy intake by 2026.
Another projection is that, in 2030, if all data centres worldwide were considered as one country, their overall energy demand would rank only fourth, behind China, the United States, and India.
AI companies in highly industrialised economies are even exploring the establishment of private nuclear power plants to meet their energy requirements.
Some countries are ramping up their fossil-fuel-driven power supplies to meet AI energy demands – a direct reversal of clean energy transition commitments.
US President Donald Trump revealed on 23 January 2025, while addressing the World Economic Forum, that the United States would have to double its annual electricity production to lead and drive the AI revolution.
That is the extent of the enormous energy demand exerted by the technology.
President Trump intends to use the Executive Order (which he signed on 20 January 2025) declaring a national energy emergency to address this challenge.
This legal instrument directs US agencies to utilise their statutory emergency powers to speed up the development and authorisation of energy projects.
Unfortunately, with his slogan—“drill, baby, drill”—Trump’s AI energy plan will be anchored by boosting fossil fuel production to the detriment of global climate policies and regulations.
This is an adapted excerpt from the book “Deploying Artificial Intelligence to Achieve the UN Sustainable Development Goals: Enablers, Drivers and Strategic Framework”
Larger AI models require exponentially more energy.
