- UN climate chief Simon Stiell has accused AI developers of increasing fossil fuel use, grid pressure and consumer energy costs.
- Global data centre electricity consumption is projected to more than double to around 945 TWh by 2030.
- For Britain, the central policy question is increasingly who supplies new data centre capacity, where it is located and who pays for the network reinforcement.
AI companies must disclose their environmental impact and secure genuinely clean electricity for their data centres or risk losing public support, the UN’s senior climate official has warned in New York.
UN Climate Change executive secretary Simon Stiell used a Climate Week speech to argue that the rapid construction of AI infrastructure is increasing fossil fuel generation and placing additional pressure on electricity systems.
“Energy-guzzling artificial intelligence is driving up planet-heating pollution,” Stiell said, adding that it is also increasing energy costs for households and businesses. He called on technology companies to set credible climate targets, disclose their energy and water use, improve efficiency and power data centres with renewable electricity.
The intervention is significant because it moves AI’s environmental impact from a corporate reporting issue into mainstream energy and climate diplomacy. The Turkish COP31 presidency is also preparing an Antalya Pledge on AI, intended to address how the technology is designed, powered, measured and managed, although detailed commitments have yet to be published.
Clean procurement falls short
The International Energy Agency expects global data centre electricity consumption to more than double to about 945 TWh by 2030, slightly more than Japan’s present annual consumption. Data centres are expected to account for roughly one tenth of global electricity demand growth over the period, with AI the most important driver.
Renewables are forecast to meet about half the additional demand, supported by storage and network investment. But the IEA also expects natural gas generation serving data centres to increase by around 175 TWh by 2035.
This is the tension underlying Stiell’s warning: technology companies may procure large quantities of renewable energy while their round-the-clock loads still increase fossil generation during periods of low wind or solar output.
Traditional annual power-purchase agreements and renewable certificates do not necessarily demonstrate that consumption is matched with clean electricity at the same location and time. The policy debate is therefore moving towards hourly matching, additional generation, storage and flexible computing loads.
There is also the question of distribution. New transmission, substations and firm generating capacity must be financed even where one large customer is responsible for much of the additional demand. Public resistance is likely to grow if households believe data centres are increasing bills, consuming scarce water or delaying connections for other businesses.
The issue is particularly acute in the UK, where the government wants to attract AI investment while simultaneously electrifying transport, heating and industry. Official modelling suggests cumulative emissions associated with UK AI computing over ten years could range from 34 million to 123 million tonnes of carbon dioxide, depending heavily on the pace of demand growth and grid decarbonisation.
Britain’s emerging response is to align data centres more closely with available generation and constrained electricity. That approach will require location to become a more important part of planning and connection decisions. A data centre situated near surplus renewable production, equipped with storage and able to vary some workloads could strengthen the system; the same facility in a heavily constrained area could add considerable cost.
AI may ultimately deliver substantial climate benefits through better forecasting, industrial optimisation and grid management. But those potential savings are not automatic offsets for the sector’s measurable electricity consumption.
Stiell’s intervention establishes a tougher test: developers will increasingly be expected to prove that AI’s benefits exceed its energy system costs, rather than promising that they eventually will.
