In brief

The Antalya Artificial Intelligence Pledge is a voluntary commitment launched at COP31 on 16 November 2026 by Türkiye's COP31 Presidency and the International Telecommunication Union. Governments and organisations can sign by picking at least one commitment from its annex and setting their own target year. Three of the annex's eleven lines apply to any leader who buys and uses AI: ask your suppliers for energy and environmental data, match the model to the task, and train your people.

You approved an AI tool for your team this year. It drafts emails, answers customers and prepares reports. Do you know how much electricity and water it uses, or where that electricity comes from?

This week that question reached the UN climate talks in Antalya. On 16 November, at COP31, Türkiye's presidency and the International Telecommunication Union launched the Antalya Artificial Intelligence Pledge. Its aim fits in one line: AI that is "sustainably powered, efficiently designed, transparently measured and available to all."

The journey of one question

Think about the last question you typed into an AI tool. It left your phone or laptop and travelled, perhaps thousands of kilometres, to a data centre. Inside, rows of servers that run day and night worked out the answer. Those servers need electricity to run and water to stay cool. A few seconds later, the answer travelled back to you.

One question uses very little. In June 2025, OpenAI's chief executive, Sam Altman, put an average ChatGPT question at about 0.34 watt-hours of electricity and about a third of a millilitre of water. But by that summer, ChatGPT alone was receiving around 2.5 billion questions a day. By my calculation, every day that's as much electricity as about 124,000 homes use in a day, and enough water to fill an Olympic swimming pool roughly every three days.

Why a climate summit talks about AI

The International Energy Agency says data centres used 485 TWh of electricity in 2025, 17% more than the year before. In data centres built for AI, consumption grew by 50%. The agency expects the total to roughly double, to about 950 TWh, by 2030.

Then there's water. A June 2026 report from the United Nations University estimates that data centres used 4.5 trillion litres of water in 2025, counting both the water that cools the servers and the water used to produce their electricity. By 2030, that could reach 9.3 trillion litres.

That's why the pledge sits on a climate agenda. AI can help manage energy systems and predict climate risks, while the infrastructure behind it needs more power, water and hardware every year. The pledge is written around both sides of that story.

What the pledge asks

The text has four commitments, and their wording is worth reading as written:

  1. Power and build AI infrastructure and hardware sustainably.
  2. Design, manage or procure AI responsibly across its lifecycle.
  3. Support assessment and reporting of the environmental impacts of AI systems.
  4. Strengthen financial support and technical infrastructure for AI for climate action.

It's voluntary and creates no new obligations. Governments can sign, and so can organisations such as businesses, universities and civil society groups. An organisation picks at least one commitment from a list in the annex and sets its own target year. The first review of collective progress happens at COP32.

When I read that annex, I noticed that most of its eleven lines are written for people who build models or run data centres. Three of them are written for anyone who buys and uses AI.

Three lines that belong to you

  1. Ask your suppliers for the numbers. Commitment A2 asks signatories to require their AI providers to disclose energy and environmental impact data. You can bring that to your next contract review: how much energy and water does this service use, where does that energy come from, and do you publish the figures?
  2. Match the model to the task. Commitment B3 asks for guidelines on designing models that use fewer resources. For most leaders this starts with one conversation with the team: which tasks really need the largest model, and which ones can a smaller tool handle?
  3. Train your people. Commitment C1 is about building skills for sustainable AI. A team that understands what each request costs makes different choices about when and how to use AI.

Then name an owner. Someone should hold the AI energy and water question, the same way someone holds data protection, with a date to report back.

Five questions to ask your supplier

  1. How much energy and water does this service use for our usage? Can you share these figures with us in a form we can use in our sustainability reporting?
  2. Which international standard do you use to measure these figures, and do you publish them?
  3. What share of the energy for the data centres that run this service comes from zero- or low-emission sources?
  4. Which cooling method do your data centres use, and which metrics do you use to track water and energy use?
  5. Which model does this service use, and can we choose a model that uses fewer resources for simple tasks?

Where the pledge stops

On 27 October, the World Resources Institute said the pledge points in the right direction but lacks accountability mechanisms. I think that tells leaders something useful. With a voluntary text, the measuring is up to you, and your own AI governance is where it starts.

These three lines work as a checklist for your own AI policy, starting this quarter.

If you'd like to talk it through, bring the problem you're stuck on to a free 30-minute call: meltemgunyuzlu.com/call

Over the next two years, the leaders who can say what their AI runs on will have an easier time with their boards, their customers and their regulators.

Share this with the leader who signed off an AI tool this year and never asked what it runs on.

Sources

  • COP31 Presidency of Türkiye and ITU, Antalya Artificial Intelligence Pledge, text released 27 October 2026; launch at the High-level Green Digital Action Roundtable on AI, 16 November 2026. Annex B commitments A2, B1, B2, B3 and C1.
  • COP31 Presidency, invitation to participate in the Antalya AI Pledge.
  • International Energy Agency, Key Questions on Energy and AI, April 2026.
  • United Nations University Institute for Water, Environment and Health (UNU-INWEH), report on data centres' electricity and water use, June 2026.
  • World Resources Institute, statement on the Antalya AI Pledge, 27 October 2026.
  • Sam Altman, The Gentle Singularity, blog post, 10 June 2025 (no method published).
  • OpenAI, as reported by Axios and TechCrunch, 21 July 2025: about 2.5 billion prompts a day.
  • Daily comparisons are my own calculation from the two figures above (about 850 MWh and 800,000 litres a day), using Ofgem's typical household electricity use of 2,500 kWh a year and an Olympic pool of about 2.5 million litres.
  • Cover photo: İsmail Enes Ayhan on Unsplash.

Meltem Günyüzlü Ateş, CFCIM is the founder of MG Consultancy and Academy, an AI coach and fractional CMO. She trains and advises leaders and organisations on AI and marketing, drawing on over 25 years of marketing, strategy and transformation experience across consultancy, creative agencies and global corporates. She is an Oxford AI Programme graduate, a Global Council for Responsible AI ambassador and a mentor at Tech Istanbul, and writes the LinkedIn newsletter Lead AI Responsibly.

← All insights