COP31 Antalya AI Pledge Targets Data Centre Energy, Water and Emissions
A new voluntary pledge launched at COP31 asks governments and technology companies to measure and reduce artificial intelligence’s growing demand for electricity, water and hardware. The Antalya Artificial Intelligence Pledge is not a binding emissions limit, but it creates a useful test: will the organisations building the largest computing systems publish enough comparable data for anyone to judge their promises?
The COP31 presidency and the International Telecommunication Union launched the initiative in Türkiye on November 16. It treats AI as both a climate tool and an environmental burden. Machine-learning systems can help forecast weather, manage electricity grids and analyse land use, while the data centres behind them consume energy, cooling water, land and raw materials.
The four pillars
The first pillar asks participants to power and build AI infrastructure more sustainably. Suggested actions include increasing low- and zero-emission electricity, improving energy and water efficiency, designing better cooling systems and integrating large data centres into power grids without shifting unreasonable upgrade costs to other users.
The second covers the full hardware and software lifecycle. It encourages repair, reuse and recycling instead of treating servers and accelerators as disposable. Manufacturing chips and constructing facilities have environmental impacts before a model handles its first query, so an assessment limited to operating electricity misses part of the footprint.
The third pillar focuses on measurement and reporting. The ITU has developed guidance for assessing AI’s environmental impact, but a common method is still difficult. Results depend on which emissions are counted, the electricity mix, model use, equipment life and whether companies report water consumed or merely withdrawn.
The fourth seeks finance and technical support for climate-related AI in developing countries, least-developed countries and small island states. That matters because the communities most exposed to extreme weather often have the least access to computing, reliable data and early-warning infrastructure.
Why data-centre disclosure matters
A facility can be efficient per unit of computing and still increase total emissions if demand grows faster than efficiency. This is the rebound problem: cheaper or more capable computing encourages more use. Power Usage Effectiveness, a common data-centre metric, measures overhead relative to computing equipment but says little about whether the electricity is clean, whether water is scarce locally or whether the workload is socially useful.
Better disclosure would separate at least four questions: how much energy a system uses; the emissions associated with that energy; water use in the facility and electricity supply; and the embodied impact of construction and hardware. Location matters. A litre of water used in a stressed basin is not equivalent to the same use where supplies are abundant.
A voluntary pledge has obvious limits
The document expressly creates no new legal obligations. Participants choose actions according to national circumstances and capabilities. That flexibility may attract more supporters, but it also permits shallow commitments. A company could endorse the pledge, select a modest measure and continue expanding fossil-powered capacity.
The launch announcement says collective progress will be reviewed annually at UN climate conferences through 2030. Reviews will be more persuasive if they name endorsers, list the actions each selected, publish baselines and distinguish absolute reductions from efficiency gains. Without those details, annual reporting could become a catalogue of projects rather than an account of environmental performance.
AI can help climate action—but claims need evidence
AI is already used in weather prediction, equipment maintenance, transport planning and energy forecasting. Some applications can reduce waste or improve warnings. Their benefits should be compared with simpler alternatives and with the resources needed to build and run them. Adding an AI label to an existing database does not create a climate benefit.
The pledge also calls for wider access to climate applications. That should include local knowledge, languages and public institutions, not only cloud credits that expire after a pilot. A warning model is useful when forecasts reach people in time, officials have authority to act and communities trust the message.
What would make Antalya consequential
Governments can turn voluntary principles into procurement rules, planning requirements and utility regulation. Before approving a major data centre, authorities can require credible power and water assessments, grid-upgrade plans and public reporting. Public buyers can ask vendors for lifecycle data and favour systems that deliver the required result with less computation.
Companies can publish facility-level electricity, water and emissions data; disclose the assumptions behind model-footprint estimates; extend hardware life where security and performance allow; and match new demand with genuinely additional clean power. Renewable-energy certificates alone do not guarantee that electricity is available at the same hour and place a data centre consumes it.
The Antalya pledge places the right questions on an international stage. Its success will depend on whether voluntary language becomes measurable practice before AI infrastructure expands further. Climate credibility will come from audited numbers and better local outcomes, not from the number of logos beneath a declaration.
Sources
- ITU and COP31 presidency launch announcement
- Official Antalya Artificial Intelligence Pledge text
- United Nations COP31 second-week update


