THE WORLD IS NOT STANDING STILL.RSS
BIG CHANGE.
BIG CHANGE, one tap away
BIG CHANGE, one tap away

Make room for the big picture.

Add BIG CHANGE to your home screen or desktop. It opens in its own window, with the latest stories when you’re online.

iPhone & iPad
Open bigchange.ai in Safari. Tap Share, then Add to Home Screen. Turn on Open as Web App if shown, then tap Add.
Android
Open bigchange.ai in Chrome. In the browser menu, choose Install app or Add to Home screen.
Desktop
In Chrome or Edge, use the install icon in the address bar or the browser’s app menu. In Safari on Mac, choose File → Add to Dock.

Markdown edition

# COP31 launches voluntary Antalya AI climate pledge

> Participants choose commitments, with progress reviews planned from COP32.

By BIGCHANGE.AI Editorial

Published: 2026-11-20T01:39:38.087Z
Updated: 2026-11-20T01:39:38.087Z
Canonical: https://bigchange.ai/blog/cop31-antalya-ai-climate-pledge

![A flanged water pipe and meter sit outside a utility building, with a separate electrical cabinet farther along the wall.](https://bigchange.ai/api/media/file/cop31-utility-metering-v1.png)
AI-generated conceptual illustration for BIG CHANGE.

Türkiye's COP31 Presidency and the International Telecommunication Union launched the Antalya Artificial Intelligence Pledge on November 16. [ITU announcement](https://www.itu.int/en/mediacentre/Pages/PR-2026-11-16-COP31-Antalya-AI-Pledge.aspx) The conference's scheduled dates are November 9–20. [UNFCCC conference page](https://unfccc.int/cop31)

## The big change

- The pledge creates no new obligations. Participants choose voluntary commitments, with Annex B signers registering their own targets and dates. Understanding an endorsement therefore means examining which actions and coverage it actually includes.
- A footprint report needs a defined system and task. ITU's assessment method covers the lifecycle around AI, including hardware and operation, and asks reporters to identify boundaries, data sources and assumptions so the result can be interpreted.
- A new Harvard-led study illustrates the evidence gap. It estimates electricity and emissions for a defined set of large facilities, but cannot separate their AI workloads from general computing. A data-centre estimate is therefore broader than an AI-only measurement.

## What participants select

The pledge covers power, lifecycle choices, environmental reporting and support for climate applications. Annex A invites countries to request provider data and build public energy and emissions inventories for information and communications technology. Annex B lets other stakeholders select disclosure coverage for recently released AI systems, supplier-data requirements and their own target dates. [Pledge and annexes](https://cop31.tr/assets/docs/action-agenda/cop31-antalya-ai-pledge-text.pdf)

Endorsers select at least one applicable commitment. Joining remains open between COP31 and COP32, with implementation cooperation planned after COP31 through Green Digital Action. Collective stocktakes start at COP32 and continue through 2030. The form lets participants choose whether commitments may be publicized. [ITU endorsement details](https://www.itu.int/initiatives/green-digital-action/form-pledges/)

## What goes inside a footprint

ITU-T L.1801, approved in February, is a methodology for assessing AI systems. Its lifecycle approach includes raw materials, hardware production, use and retirement. Initial training belongs to production, while inference and continued learning sit within use. The standard calls for training impacts to be reported separately, making that energy visible when interpreting the system's footprint. [Assessment methodology](https://www.itu.int/rec/T-REC-L.1801-202602-I/en)

The method also requires a functional unit, meaning a quantified description of the service being provided. For AI, that includes what the function does, how much is delivered, its duration and its quality. Defining the service makes it possible to compare alternatives that provide equivalent functionality, with comparable boundaries and methods. [Assessment methodology](https://www.itu.int/rec/T-REC-L.1801-202602-I/en)

Its reporting guidance asks for the system description, included lifecycle stages, data quality, assumptions, allocations and emission factors. It distinguishes assessment of an AI system from national or worldwide aggregation, which falls outside this recommendation's scope. [Assessment methodology](https://www.itu.int/rec/T-REC-L.1801-202602-I/en)

## Keep a rate distinct from a total

The Green Software Foundation's separate Software Carbon Intensity specification provides an example of structured software reporting. It combines operational emissions with an allocated share of hardware emissions, then relates them to a functional unit such as a request or training run. Its procedure includes disclosing the software boundary and calculation method. [SCI specification](https://sci.greensoftware.foundation/)

That rate answers a particular question about emissions per unit of service. For a hypothetical application, a lower per-request figure can coexist with greater total emissions if request volume grows enough. The specification requires comparisons with a baseline to keep the methodology, assumptions and functional unit consistent. That helps a team interpret a change in its software rather than a change in the accounting. [SCI specification](https://sci.greensoftware.foundation/)

## What an existing estimate can show

The Harvard-led research, published November 16 in Nature Sustainability, examined 403 US hyperscale data centres operating between May 2024 and April 2025. Across facility-load scenarios, the authors estimated roughly 68–99 terawatt-hours of electricity and 21–31 million tonnes of attributable carbon dioxide. The central electricity estimate was about 82 terawatt-hours. The model uses EPA's 2023 regional electricity-generation data to attribute emissions to the sites. These are scenario-based estimates for that sample and period. [Original study](https://www.nature.com/articles/s41893-026-01985-7) · [Authors' public materials](https://github.com/gianguidi/hyperscale-emissions)

The authors' university account says the data could not distinguish AI-specialized operations from general-purpose computing and did not cover every US hyperscale facility. Their repository makes aggregate outputs and reproducibility materials public while withholding restricted facility-level records. Those limits matter when discussing the sector, because the sample does not establish an AI-only footprint. [University account](https://hsph.harvard.edu/news/hyperscale-data-centers-energy-consumption-carbon-emissions-newly-estimated-by-researchers/) · [Authors' public materials](https://github.com/gianguidi/hyperscale-emissions)

BIG CHANGE reviewed the pledge, annexes and cited methods. We did not submit an endorsement or measure a participant's implementation.

## Sources

- [ITU launch](https://www.itu.int/en/mediacentre/Pages/PR-2026-11-16-COP31-Antalya-AI-Pledge.aspx) — Launch.
- [Pledge text](https://cop31.tr/assets/docs/action-agenda/cop31-antalya-ai-pledge-text.pdf) — Objectives and annexes.
- [ITU endorsement](https://www.itu.int/initiatives/green-digital-action/form-pledges/) — Joining and implementation.
- [ITU-T L.1801](https://www.itu.int/rec/T-REC-L.1801-202602-I/en) — AI-system assessment method, February 2026.
- [Software Carbon Intensity](https://sci.greensoftware.foundation/) — Separate software reporting method.
- [Hyperscale data-centre study](https://www.nature.com/articles/s41893-026-01985-7) — November 16 estimates and scope.
- [UNFCCC conference](https://unfccc.int/cop31) — Scheduled dates.