On a spring evening in Memphis, campaigner Sharon Wilson aimed a thermal imaging camera at a gas-fired power plant supplying Elon Musk’s xAI Colossus datacentre. The device revealed plumes of invisible methane and other gases pouring into the sky from turbines that are not subject to pollution controls.
“It was jaw-dropping,” said Wilson, a former oil and gas worker who now directs the campaign group Oilfield Witness. After more than a decade documenting methane releases, she estimated the facility was emitting more planet-heating gas than a large power plant. “Just an unbelievable amount of pollution.”
At the same time, xAI’s chatbot Grok was drawing attention for repeating a conspiracy theory about “white genocide” in South Africa and generating other extremist content, including praise for Hitler and far-right talking points. Some of the offending posts were later deleted.
“It’s a horrible, horrible waste,” Wilson said, pointing to Grok-produced images such as a “Nazi Mickey Mouse” as examples of what fossil gas was being burned to create. “What useful purpose does this serve?”
Scientists and campaigners are increasingly asking similar questions about artificial intelligence. They warn that the rapid buildout of energy-hungry datacentres is fuelling demand for fossil power, while AI systems amplify social harms, from medical misinformation to deepfake abuse.
Some experts fear this growth could complicate the shift to a low-carbon economy and make it harder to keep global heating within 1.5C. Others argue AI’s energy use is relatively small compared with heavy industry and that the technology could help cut emissions across other sectors.
Datacentres push electricity demand higher
When energy modeller Hannah Daly worked at the International Energy Agency (IEA) and later became a professor of sustainable energy at University College Cork, digital emissions were a minor concern compared with cars, agriculture and heating. That changed as Ireland’s datacentre boom gathered pace.
Datacentres now use about one-fifth of Ireland’s electricity and are projected to draw nearly one-third within a few years. Their rapid expansion has outstripped grid capacity, prompting the regulator in 2021 to effectively halt new grid connections.
“The trajectory of enormous, exponential growth is what is so worrying,” said Daly. “I’m not sure whether Ireland is a real outlier or a harbinger of what’s to come. But it’s definitely a cautionary tale.”
Globally, datacentres currently consume about 1% of electricity, but this share is expected to rise sharply. BloombergNEF projects they will more than double their portion of US electricity use to 8.6% by 2035. The IEA estimates they will account for at least 20% of electricity demand growth in advanced economies by 2030.
Tech companies have signed long-term contracts to buy renewable power and, in some cases, nuclear energy. These deals can support new clean generation even when facilities still rely on fossil-heavy grids. But for now, coal, oil and gas dominate the supply mix in many key markets.
In China, most datacentres sit in coal-dependent eastern regions. In the US, natural gas is expected to provide most datacentre power for the next decade. Under Donald Trump, the administration framed the sector as a reason to keep burning coal. Announcing a $625m (£467m) package in September, energy secretary Chris Wright said: “Beautiful, clean coal will be essential to … winning the AI race.”
In Ireland, the growth of datacentres has offset emissions reductions from renewable power, according to Daly’s analysis for Friends of the Earth Ireland. Similar dynamics may emerge in poorer countries. Pakistan has begun replacing coal with cheap solar, but the government has pledged 2GW of power to AI and bitcoin, potentially giving idle coal plants new customers.
“This idea that the lower cost of renewables alone will drive decarbonisation – it’s not enough,” Daly said. “Because if there’s a huge source of energy demand that wants to grow, it will land on these stranded fossil fuel assets.”
How much energy does AI actually use?
For individual users, estimates suggest a single text query to a large AI model consumes roughly 0.2 to 3 watt-hours (Wh) of energy, with more complex tasks such as “deep research” or video generation requiring more. In a July blogpost, OpenAI chief executive Sam Altman said a ChatGPT request uses about as much energy as running a lightbulb for a couple of minutes, echoing a recent Google report on typical Gemini prompts.
These figures are tiny compared with the emissions from activities such as driving, flying or eating meat. Concern instead centres on scale and ubiquity. ChatGPT reports several hundred million weekly users less than three years after launch. Google, which handles an estimated 90% of global search traffic, has added generative AI answers to its results pages and is rolling out AI “agents” and automated services in the background.
“What I’m worried about is that we’re deploying AI in such a way that we don’t have a good idea of the energy use,” said Sasha Luccioni, climate lead at AI company Hugging Face. She criticised “selective disclosures” from major firms that, she argued, obscure the real climate impact of their products. “We’re essentially operating on the hypothesis that it’s not a problem – or that if it is a problem it will somehow be solved – instead of getting ahead of it.”
Can AI cut more emissions than it creates?
An IEA report in April argued that, in principle, AI applications could reduce emissions in other sectors by more than datacentres emit. A June paper by researchers at the London School of Economics (LSE) and consultancy Systemiq reached similar conclusions, modelling how AI could speed up the integration of renewables into power grids, help develop meat alternatives, improve electric vehicle batteries and encourage greener choices.
“AI can accelerate the deployment of those clean technologies by basically accelerating their position along the curve of innovation and adoption,” said co-author Roberta Pierfederici, a policy fellow at the LSE’s Grantham Institute.
The potential savings are uncertain and vulnerable to “rebound effects”, where efficiency gains spur more consumption. The IEA warns that self-driving technology, for example, could undermine public transport if it makes car travel easier and cheaper.
Still, some practical examples already exist:
- Google says AI has cut cooling energy use in its own datacentres by 40%.
- Spanish utility Iberdrola reports a 25% increase in the operational efficiency of wind farms through AI-driven maintenance and performance optimisation.
- French company Engie says AI-based fault detection has reduced downtime at its solar farms.
Because sectors such as power and food are so carbon-intensive, even modest efficiency improvements could in theory offset AI’s own footprint, which one recent study estimated at 0.1–0.2% of global emissions and growing. “In the power sector we are already seeing the results,” said Pierfederici. “Whereas the meat sector is not there yet.”
AI and the fossil fuel industry
Holly and Will Alpine, a married couple who worked at Microsoft on responsible AI and sustainability, saw both sides of the technology’s potential. Will was an early internal voice on the energy cost of datacentres. But the company’s work with oil and gas producers left them increasingly uneasy.
In 2019, Microsoft announced a partnership with ExxonMobil that could boost production by up to 50,000 barrels of oil a day. It also launched a digital project with Chevron, which the company says has shortened deepwater well planning by 30 days. As more deals emerged, the Alpines pressed for answers.
“The response from the company was often pointing back to their own operational footprint, which is not relevant,” said Holly Alpine. After four years of internal advocacy that they say yielded more promises than progress, the couple left to campaign for the tech sector to address “enabled emissions” – the pollution its products help other industries generate.
The IEA estimates AI could increase technically recoverable oil and gas reserves by 5% and cut the cost of deepwater offshore projects by 10%. Industry leaders are even more optimistic. “Artificial intelligence is, ultimately, within the industry, going to be the next fracking boom,” Mike Sommers, head of the American Petroleum Institute, told Axios. Saudi Aramco chief executive Amin Nasser told Bloomberg Television that the company had embedded AI “in everything”, doubling technology spending from 2023 to 2024 and helping to raise productivity and the number of wells.
Oil and gas firms argue that AI can also help them cut emissions intensity, for example by using satellite data to detect methane leaks. But Wilson, who recently documented “giant clouds of gas escaping everywhere” during a trip to the Permian Basin, said such tools have not translated into sufficient action and that deliberate releases remain a much bigger problem.
“They are using this as an excuse to delay action,” she said. “Watching methane from space is not stopping methane.”
Supercharged consumption
Another concern is the impact of AI on demand for goods and services. A study published in October found that generative AI advertising outperforms human-crafted ads, while drastically lowering production costs. Marketing firms are preparing for AI agents capable of automatically buying presents, booking flights and arranging holidays.
Tui, Europe’s largest travel operator, says it is investing heavily in AI as more customers turn to chatbots to plan trips. Critics warn these tools could encourage more high-carbon consumption, from flights to fast fashion.
“The narrative is really focused on this false comparison between the energy used to run the technology and the positive use cases,” said Alpine. “But it is dangerous to omit the negative use cases.”
Calls for limits and ‘guardrails’
These concerns have sparked growing demands for regulation. In October, the UN special rapporteur on the human right to safe drinking water called for a moratorium on new datacentres because of their environmental impacts. Two months later, more than 230 environmental organisations in the US urged a national halt on datacentre expansion until stronger rules are in place.
Ireland’s utility regulator has lifted its effective ban on new connections but now requires that 80% of a datacentre’s annual electricity use eventually come from new renewable projects. Spain, the only country to reference AI directly in climate law, instructs the government to promote digital technologies that help decarbonise the economy.
Laurence Tubiana, a key architect of the Paris climate agreement, has suggested taxing AI to raise money for climate action. The Alpines, who stress they are not anti-AI but want “reasonable guardrails”, are pushing for the EU’s upcoming AI legislation to classify fossil fuel applications as high-risk and for investors to factor enabled emissions into environmental, social and governance (ESG) ratings.
Asked about their climate impact, Google and xAI did not comment. OpenAI said it thinks carefully about how it uses computing power, supports partners’ sustainability efforts and believes AI will be crucial in tackling climate change. Microsoft said the energy transition is complex and must be managed “in a principled manner”, with technology playing a key role in decarbonising industry. “That requires balancing the energy needs and industry practices of today, while inventing and deploying those of tomorrow,” a spokesperson said.
Luccioni said the focus should be on building AI systems that are efficient by default rather than on abandoning the technology.
“Maybe I’m a little naive, but I still believe that AI can do good in terms of fighting the climate crisis – designing the next generation of batteries, tracking deforestation, predicting hurricanes,” she said. “There are so many good things for which we can be using it – and instead we’re creating social media websites filled with AI slop while datacentres are getting powered by diesel generators.”