Climate negotiations (COP30) begin on November 10th. We arrive after a year of record emissions, which the development of AI risks worsening.
COP30, the Conference of the Parties on climate which will revive negotiations to reduce greenhouse gas emissions, kicks off today, November 10, in the Brazilian city of Belém in the Amazon, on the tenth anniversary of the Paris Agreement (signed in December 2015).
The work starts “burdened” by an annual record of carbon dioxide and methane emissions that the world has set in its suitcase in 2024. A balance that in the future, according to a study just published on Nature Sustainabilitycould be exacerbated by AI’s growing demand for energy.
CO2 emissions: a 2024 to forget
Based on measurements by Copernicus, the European Union’s Earth observation programme, carbon dioxide concentrations reached a record value of 422 parts per million in December 2024. From the year of the Paris Agreements to today’s COP30 they have grown by 5.51%, but it is also the speed of increase that is worrying: for the World Meteorological Organization, it has tripled since the 1960s.
In the same period of time (2015-2025) methane emissions, a greenhouse gas with a heating power 80 times more powerful than CO2 in the short term, grew by 4.86%, reaching 1,897 parts per billion. The last 10 years have also been the 10 warmest ever.
One element to add to the forecast: energy for AI
Since there will be a lot of discussion about energy at COP30, it is useful to consider the results of the Cornell University study published today, November 10, which estimated how the environmental impact of IT infrastructures that support artificial intelligence will grow in the United States (which is not participating in the COP30 negotiations and which no longer adheres to the Paris Agreement).
By 2030, if it maintains current growth rates, artificial intelligence will emit from 24 to 44 million tons of CO2 into the atmosphere every year – as much as 5 to 10 million more cars added to the roads of the United States would emit – and will consume from 731 to 1,125 million cubic meters of water per year, the equivalent of the annual water consumption of 6-10 million US citizens (the water is used in fact to cool the systems and to produce energy in the first place). A footprint that is a long way from enabling the AI sector’s net-zero emissions goal.
The solution? Install data centers in the right location
The researchers explain: “The AI infrastructure choices we make this decade will determine whether AI accelerates climate progress or becomes a new burden on the environment.” In short, it is better to plan logistics and technologies in advance to limit emissions rather than focus on decarbonisation later, with systems already installed.
For example, by choosing to install AI data centers in regions with less water stress, or where renewable energy production is more widespread, reductions of around 73% in carbon emissions and around 86% in water consumption could be achieved. The choice of site is therefore crucial, together with strategies to improve the efficiency of server cooling and better use of them.
Decarbonization needs to keep pace with the increase in demand for AI energy: “The solution is to accelerate the transition to clean energy in the same places where artificial intelligence is expanding” explains Fengqi You, professor of Energy Systems Engineering at Cornell University who coordinated the study.
