Climate

UN Antalya AI Pledge Opens at COP31 With Nine Countries and Alibaba Alone

The voluntary Antalya Artificial Intelligence Pledge drew about 90 endorsements but only nine governments, and the absence of American hyperscalers leaves the UN-backed effort without the firms it most needs.

T
By TechQuire Daily Staff TechQuire Daily Staff
November 17, 2026 / 7 min read

Antalya has spent the past two weeks hosting the 31st Conference of the Parties, and the agenda inside the negotiating halls has been dominated by finance, adaptation and the familiar quarrel over fossil fuel language. Outside those halls, a newer argument has been forcing its way onto the programme: the environmental cost of the artificial intelligence boom. Data centres built to train and serve large models are now among the fastest growing sources of electricity demand in the world, and the governments meeting in Turkey are being asked to decide what, if anything, to do about it.

The COP31 Presidency and the International Telecommunication Union chose the summit's second week to answer that question with a voluntary instrument rather than a binding one. The Antalya Artificial Intelligence Pledge was launched on 16 November 2026, and its organisers presented it as a key moment in aligning AI innovation with climate action. The text acknowledges that AI cuts both ways: it can help grids, forecast weather and model emissions, and it can also drive up energy and water consumption at exactly the moment those resources need to fall.

The political weather around the launch was not helpful. The United States has stayed away from the UN climate talks for a second year, after President Donald Trump withdrew the country from the UN climate process, and Washington's absence shapes what any voluntary pledge can realistically achieve. The European Union and China, the two other jurisdictions whose rules will determine how data centres are built and powered, did not endorse the text either. What remained was a coalition of the willing, thin on the governments that own the largest computing footprints.

The numbers attached to the launch told their own story. Accounts differed slightly on the total, with Reuters putting endorsements at about 90 and Agence France-Presse reporting 80, but both agreed on the part that mattered: only nine countries were among the signatories, while the rest were companies, industry bodies and civil society groups. One of those companies was Alibaba, the Chinese e-commerce and cloud group, and it was the only major technology company on the list.

Key Facts

The pledge rests on four pillars. Signatories commit to powering and building AI infrastructure and hardware sustainably, including low- and zero-emission energy for data centres and better energy and water efficiency. They commit to designing, managing or procuring AI responsibly across its lifecycle, including circular management of electronic waste. They agree to support assessment and reporting of AI's environmental impacts using standardised methodologies. And they undertake to strengthen financial support and technical infrastructure for AI for climate action in developing countries, least developed countries and small island developing states.

The countries that signed were Andorra, Brazil, Colombia, Luxembourg, Sierra Leone, Somalia, South Korea, Turkey and Uganda, according to New Scientist, which reported on November 17 that the list had barely grown since the launch. South Korea is a significant AI user, but the other governments represent small economies. Germany said it would sign, according to Agence France-Presse, which reported on November 17 that the United States, China and the European Union were all missing from the endorsement list.

The corporate picture was starker still. Reuters reported on November 17 that Meta, Microsoft and Google, which operate global data-centre networks at enormous scale, and chipmaker Nvidia did not respond to requests for comment. Amazon declined to comment, but pointed to a target to cut the carbon intensity of its electricity consumption by 65% by 2030 against a 2019 baseline, alongside a commitment to net-zero carbon emissions by 2040. Nvidia, OpenAI and Anthropic were also absent from the list, according to Agence France-Presse. The World Business Council for Sustainable Development, whose members include Amazon, Google and Meta, did sign.

Governance is deliberately light. The ITU announced on November 16 that progress will be tracked annually at UNFCCC COP summits through 2030, starting at COP32, with ITU standards, including its guidelines for measuring AI's environmental footprint, supporting implementation. Signatories may add their own commitments on green data-centre requirements, AI skills and computing capacity. The ITU said it is engaging with major tech companies and will keep the pledge open to signatures until COP32 at the end of next year.

Analysis

The gap between the endorsement count and the governmental weight behind it is the story. Ninety signatures sounds like momentum until you read the list. The pledge arrived without the United States, without the European Union, without China and without the four American companies whose capital expenditure is shaping the physical geography of computing. What this really means is that the voluntary layer of AI climate governance is being built by the countries least able to alter the trajectory of the industry, while the jurisdictions that could alter it stay outside the tent.

The text also leaves considerable room for interpretation. New Scientist reported on November 17 that the phrase low- and zero-emission energy permits continued reliance on fossil gas rather than committing signatories to renewable electricity, and the pledge sets no binding targets for cutting emissions or reducing energy consumption. That flexibility is what allowed nine governments to sign in the first place, and it is also what makes the commitment hard to audit. A pledge that cannot fail is a pledge that cannot prove success.

Alibaba's solitary presence carries its own signal. A Chinese cloud and e-commerce group endorsing a UN-backed sustainability text at a summit the United States has abandoned is a reminder that AI's supply chain is not aligned with any single diplomatic bloc. It also gives the pledge a genuine hyperscale name, which the organisers needed, while raising the question of whether one company's participation is being asked to carry more legitimacy than it can bear.

The scale of what is being discussed makes the soft approach harder to defend. Agence France-Presse reported on November 17 that more than 400 hyperscale data centres in the United States emit roughly as much carbon dioxide as Denmark, and that about 30% of the electricity consumed by data centres comes from coal, followed by renewables at 27%, natural gas at 26% and nuclear at 15%, according to the International Energy Agency. Against those numbers, an annual reporting commitment is a starting point, not a solution.

Why It Matters

AI's electricity demand is rising faster than the grids that supply it can decarbonise. New Scientist reported on November 17 that, according to the International Energy Agency, electricity consumption from AI-focused data centres increased by 50 per cent in 2025. A UN study found that data centres consumed more electricity than all but 10 countries in 2025, and that their water consumption could meet the annual basic water needs of the entire population of sub-Saharan Africa by 2030. Those are the figures behind Simon Stiell's warning that AI is a double-edged sword.

The pledge's defenders argue that a voluntary framework with an annual reporting cycle is how international climate governance usually starts. Its critics are less patient. Julie Cerqueira, chief program officer at the Natural Resources Defense Council, told Reuters that voluntary standards alone are not enough. Jill McArdle, a campaigner at Beyond Fossil Fuels, called the pledge a distraction whose promises are vague and warned it risks giving Big Tech a green stamp without requiring it to address the real climate impacts of its infrastructure.

The deeper issue is sequencing. Data centres are being financed, sited and powered now, on multi-year timelines, and the environmental accounting that would let regulators scrutinise them is still being standardised. A pledge that asks for better measurement and leaves the build-out untouched does not slow anything down, but it does create a paper trail. Whether that trail becomes accountability or decoration depends on who signs at COP32.

Next Up

The immediate test is Germany. Berlin's stated intention to sign would give the pledge its first large European economy, and the ITU has said the door stays open until COP32 at the end of next year. The ITU also says it is engaging with major tech companies, which means the coming months will show whether the American hyperscalers treat Antalya as a document to ignore or a baseline to negotiate against.

Watch three things. Whether any of Meta, Microsoft, Google, Amazon or Nvidia signs before COP32. Whether the ITU's measurement guidelines become the reference point for corporate environmental reporting on AI. And whether the annual review through 2030 produces numbers that can be compared year on year. If it does, the Antalya pledge may yet turn out to be the scaffolding for something binding. If it does not, it will be remembered as the summit where the AI industry was invited to regulate itself and most of it did not reply.

Tagged

Comments (0)

No comments yet. Be the first to share your thoughts.