The explosive growth of artificial intelligence is creating an environmental dilemma that is proving difficult to quantify, even as its resource demands skyrocket. While proponents argue AI can be a powerful tool against the climate crisis, the spiralling energy and water costs of powering vast datacentres are leaving scientists and campaigners deeply concerned.
The Unseen Pollution from AI's Engine Rooms
In May, environmental campaigner Sharon Wilson used a thermal imaging camera to document what she described as "jaw-dropping" methane emissions from the gas-fired turbines powering Elon Musk's xAI Colossus supercomputer datacentre in Memphis, Tennessee. Wilson, a former oil and gas worker and director of Oilfield Witness, estimated the facility was releasing more of the potent planet-heating gas than a large power plant. "Just an unbelievable amount of pollution," she said, questioning the purpose when the facility's output included the Grok chatbot repeating harmful conspiracy theories.
This incident highlights a broader unease. Scientists are watching the AI boom with alarm as it pollutes the natural world with carbon and saturates the digital landscape with risks ranging from health misinformation to deepfake pornography. Some fear datacentres may add a significant, unnecessary hurdle to the critical goal of limiting global heating to 1.5C.
Soaring Demand: A Cautionary Tale from Ireland
The trajectory of growth is what worries experts like Professor Hannah Daly of University College Cork. In Ireland, datacentres now consume a staggering one-fifth of the country's electricity, a figure projected to rise to nearly one-third within years. This explosive demand led to a de facto ban on new datacentre grid connections in 2021. Daly warns Ireland may be a harbinger of a global trend.
Globally, datacentres currently use about 1% of electricity, but their share is set to balloon. In the US, their portion of electricity demand is projected to more than double to 8.6% by 2035. The International Energy Agency (IEA) estimates datacentres will account for at least 20% of the growth in electricity demand in advanced economies this decade. While tech firms are signing deals for renewable and nuclear power, fossil fuels like natural gas and coal are expected to dominate supply in the near term, potentially locking in emissions.
Weighing the Carbon Debt Against Potential Savings
Tech companies have been reluctant to provide detailed AI energy data, but estimates suggest a simple text query may use between 0.2 to 3 Watt-hours. OpenAI's Sam Altman has claimed a ChatGPT query uses energy comparable to running a lightbulb for a few minutes. While negligible compared to a flight or a steak, the sheer scale of adoption—ChatGPT boasts hundreds of millions of weekly users—and its integration into every digital service creates a significant cumulative burden.
However, a compelling counter-argument exists. An IEA report in April suggested existing AI applications could cut global emissions by far more than datacentres produce. Research from the London School of Economics and Systemiq in June modelled AI accelerating clean tech deployment in energy grids, alternative proteins, and electric vehicle batteries. Real-world examples include Google using AI to cut datacentre cooling by 40%, and Iberdrola in Spain boosting wind turbine efficiency by 25%.
Yet, there is a darker side to this transformative potential. The AI expertise of tech giants is also being deployed to expand fossil fuel extraction. Microsoft, for instance, has partnerships with ExxonMobil and Chevron aimed at increasing oil production and streamlining drilling. The American Petroleum Institute has hailed AI as "the next fracking boom," while Saudi Aramco's CEO says AI is embedded "in everything," boosting productivity and well numbers.
Furthermore, AI's prowess in creating hyper-effective, low-cost advertising threatens to drive increased consumption, while AI agents could automate more resource-intensive lifestyles.
Calls for Guardrails and Frugal Design
In response, some are calling for a pause. The UN's special rapporteur on water and a coalition of over 230 US environmental groups have demanded moratoriums on new datacentres. Others advocate for regulation, such as Spain's climate law which promotes decarbonisation-through-digitalisation, or a tax on AI to fund climate action.
Former Microsoft employees Holly and Will Alpine, who left to campaign on tech's "enabled emissions," are pushing for the EU's AI Act to classify fossil fuel expansion as a high-risk use of the technology. Sasha Luccioni, climate lead at Hugging Face, argues the focus should be on creating efficient, "frugal by design" AI tools for genuine good, like tracking deforestation or designing better batteries, rather than powering endless "AI slop."
The industry's response emphasises balance. Microsoft stated the energy transition requires "moving forward in a principled manner," with technology helping to decarbonise existing industries. OpenAI said it supports sustainability goals and believes AI will be crucial in tackling climate change.
The central question remains unresolved: Will the AI revolution become a net contributor to the climate crisis, or can its innovative power be harnessed decisively to solve it? The answer may depend less on the technology itself, and more on the rules and priorities humans choose to set for its development.
