
Only nine countries have signed an international pledge to make AI greener, despite warnings at the COP31 climate summit that the technology’s growing energy demands threaten climate goals.
The Antalya Artificial Intelligence Pledge was launched on 16 November at the summit in Turkey. Developed by the summit’s presidency and the International Telecommunication Union (ITU), it intends to make AI more sustainable by encouraging cleaner energy use, more efficient data centres and greater transparency about the technology’s environmental footprint.
“AI and other frontier technologies are a double-edged sword,” said Simon Stiell, the executive secretary of UN Climate Change, at a press conference on 16 November. “The task ahead is to blunt their dangerous edges, sharpen their catalytic ones and put these technologies responsibly to work.”
But Ketan Joshi, an independent analyst researching AI’s climate impacts, questions that analogy. He says it conflates traditional machine learning, already used in applications such as weather forecasting, with energy-intensive generative AI. “You don’t need to have a giant, energy-hungry generative AI chatbot if you want to have better wind forecasting for a wind power developer,” he says.
The initiative comes as AI’s rapid expansion is placing growing pressure on electricity grids. According to the International Energy Agency, electricity consumption from AI-focused data centres increased by 50 per cent in 2025.
Joshi says there is little evidence that technology companies are building enough renewable energy capacity to meet this growing demand. Even companies that have invested heavily in renewables, such as Amazon and Google, are increasingly turning to fossil fuels to support their rapid expansion, he says.
The only countries to have endorsed the pledge so far are Andorra, Brazil, Colombia, Luxembourg, Sierra Leone, Somalia, South Korea, Turkey and Uganda.
The pledge is voluntary, with no binding targets for cutting emissions or reducing energy consumption. It sets out four areas for action, including increasing the use of low- and zero-emission energy to power AI infrastructure, improving energy and water efficiency, and measuring the environmental impacts of AI systems. It also calls for greater financial and technical support to help developing countries use AI to tackle climate change.
However, the agreement leaves considerable room for interpretation. The call for using “low- and zero-emission energy”, for example, allows continued reliance on fossil gas rather than committing to renewable electricity.
Countries can also choose additional commitments, such as requiring AI companies to disclose their energy consumption and environmental impacts, or introducing sustainability requirements for data centres.
“I’m surprised that more countries haven’t signed up, given the pledge’s limited demands,” says Joshi. “It’s not a pledge that really strikes at the heart of the problem.” He argues that voluntary promises are unlikely to be sufficient without restrictions on the expansion of data centres and stronger AI regulation.
COP31 President Murat Kurum defended the initiative, saying the rapid growth of AI must be “powered and deployed responsibly, and used for the benefit of climate actions”. The ITU says progress on the pledge will be reviewed at annual UN climate summits until 2030, starting at COP32 next year.
Joshi fears the pledge could provide cover for continued expansion via fossil fuel sources. “I think we’re going to be going into an era of a return to greenwashing, a return to sort of empty promises,” he says. “What we really need is stronger and stricter regulation.”