Turkey wants AI's power bill inside the COP31 pledge

01

WHAT ACTUALLY CHANGED

Turkey announced the Antalya Pledge on Artificial Intelligence on the sidelines of the United Nations General Assembly on September 21. Reuters reported that the initiative is intended to align the construction of AI infrastructure with global climate goals ahead of COP31.

The International Telecommunication Union is helping develop the pledge, and the incoming Turkish COP presidency said a draft would be shared before the conference. The absence of a public draft means the announcement establishes an agenda, not the final obligations, definitions or measurement system.

Murat Kurum, Turkey's environment and urban-planning minister and COP31 president-designate, called for open discussion of AI's growing energy use. Reuters reported his expectation that companies disclose energy use and power their operations with clean energy.

Turkey also said COP31 would include a day dedicated to technology and AI. The presidency plans a portfolio of AI-enabled clean-technology pilots connecting developers, industry partners and investors in areas including cars and heat pumps. No project list, selection method, budget or evaluation design was public.

The AI pledge sits beside Turkey's separate 35x35 initiative, which asks countries to increase electricity's share of final global energy demand from roughly 20 percent today to 35 percent by 2035. Reuters said the electrification commitment is voluntary and would cover buildings, transport and industry.

The 35x35 proposal calls for clean power, storage, modern grids, support for developing economies and a mechanism to track collective progress. Electrification can reduce emissions when it replaces direct fossil-fuel use with low-emission electricity. It can also add pressure to a dirty or constrained grid if supply and infrastructure do not change with demand.

UN Climate Change confirms that COP31 will take place in Antalya from November 9 through November 20. The UNFCCC page describes Turkey and Australia working in partnership as the incoming presidency and emphasizes an open process with Parties and observers. It does not yet list the Antalya AI pledge as an adopted conference outcome.

The ITU's existing Green Digital Action program brings governments, technology companies, international organizations and civil society together around industry commitments, digital climate solutions and monitoring frameworks. That program is relevant institutional machinery, but its existence does not settle the content of the new pledge.

The IEA's 2025 Energy and AI report estimated that data centers accounted for about 1.5 percent of global electricity use in 2024, or 415 terawatt-hours. It projected demand to more than double to roughly 945 terawatt-hours by 2030, with AI the largest driver alongside other digital services.

The same IEA base case said renewables would meet about half of global data-center demand growth through 2035, while natural gas and nuclear generation would also expand. Clean additions are therefore important, but the expected system is not automatically fossil-free.

The IEA also found substantial uncertainty. Its 2035 data-center demand cases range from 700 to 1,700 terawatt-hours, and a high-efficiency case is 20 percent below the base case. A credible pledge must preserve those ranges rather than build one precise infrastructure plan around a single forecast.

No public evidence before publication showed that countries or companies had signed the Antalya AI pledge. Nor was there a defined coverage threshold for models, cloud providers, data centers, chip manufacturing, network traffic or downstream devices. Those boundaries will determine whether the eventual promise describes the whole system or only its tidiest room.

02

WHY THIS MATTERS

Energy transparency begins with an absolute number. An efficiency ratio can improve while total electricity use rises because a facility adds more servers or runs them harder. The pledge should require annual megawatt-hours, peak megawatts and year-over-year change, not only a claim that each computation became leaner.

The clock matters as much as the total. A data center can buy enough renewable electricity over a year to match its annual use while drawing fossil-heavy power during evening peaks or low-wind periods. Hourly matching reveals when the load and the claimed supply actually meet.

Additionality separates accounting from construction. Purchasing certificates from an existing wind farm may change the paper trail without adding generation for a new twenty-four-hour load. Operators should show which contract financed new clean supply, when it enters service and what powers the facility before then.

Location determines consequence. The same server workload can create very different emissions, water demand, grid congestion and public cost depending on the country, power market, weather, cooling design and transmission network. Global averages are useful for diplomacy and poor guides to a local permit.

Marginal emissions matter because a new load is served by the generator that changes output, not by the average mix printed in an annual brochure. A grid may be mostly renewable and still call on gas or coal when a concentrated data-center load arrives. The pledge needs a method for attributing that incremental effect.

Grid upgrades are part of the bill. New substations, lines, transformers, storage and balancing services can improve a region or simply socialize the cost of one customer. Public disclosure should show who paid, who owns the asset, which customers benefit and whether household reliability or rates changed.

Water cannot remain outside an energy promise. Cooling and electricity generation can draw from stressed basins. A facility should disclose total and freshwater use, source, season, discharge and local water conditions rather than translate every environmental impact into carbon and quietly close the folder.

Backup generation is real infrastructure. Diesel or gas generators may run during outages, maintenance and tests. Their fuel, hours, emissions and neighborhood effects belong in the ledger, especially where grids are unreliable and backup equipment is more than an emergency decoration.

Embodied emissions arrive before the first prompt. Chips, servers, batteries, concrete, steel, cooling equipment and transmission infrastructure require materials and energy. Operational electricity dominates some decisions, but a pledge that excludes construction and equipment can reward rapid replacement under a clean-looking meter.

Efficiency needs a rebound clause. Cheaper and more efficient inference can lower energy per task while total tasks expand even faster. The IEA's wide demand range is a warning that hardware improvement alone does not guarantee a lower system total.

Developing economies face an unfair trade. They may want local compute for language, research, public services and economic autonomy while lacking spare grid capacity or cheap capital for clean generation. Climate rules should not reserve AI infrastructure for rich countries that built their systems earlier.

Finance therefore belongs beside disclosure. If the pledge expects cleaner data centers, efficient cooling, storage and grid modernization, it should identify grants, concessional finance, technical support and risk-sharing rather than handing lower-income countries a standard they cannot afford to meet.

AI can also help the energy system. The IEA identified uses in grid forecasting, fault detection, industrial efficiency, buildings and transport. Those benefits require measured baselines and rebound checks. A data center should not subtract a hypothetical avoided emission elsewhere from the electricity it definitely used today.

A voluntary pledge can still change procurement and permitting if its metrics are specific, public and comparable. It becomes climate theater when every participant chooses its own boundary, year, certificate and counterfactual, then announces success before anyone can reproduce the calculation.

03

WHERE IT COULD HELP

• Publish the complete Antalya Pledge draft, definitions, consultation record, revision history and the institutions responsible for implementation before COP31
• Define which training runs, inference services, data centers, cloud operators, chip facilities, networks and devices fall inside each commitment
• Require facility-level annual electricity, peak demand, load profile and year-over-year change using a consistent reporting boundary
• Report electricity sources hourly, not only through annual certificate matching, and identify the residual grid mix when contracted supply is unavailable
• Show which clean generation and storage projects are additional, their commercial-operation dates, contract terms and treatment before they come online
• Publish location-based and marginal emissions with the method, grid region, time resolution, uncertainty and independent verifier
• Disclose grid-connection capacity, substation and transmission upgrades, queue position, public subsidies, rate treatment and cost allocation
• Report total water withdrawal, freshwater use, source, basin stress, seasonal profile, discharge and cooling technology
• Measure backup-generator fuel, testing and operating hours, emissions, noise, incidents and replacement options
• Add embodied emissions and material provenance for chips, servers, batteries, cooling equipment, buildings and major grid assets
• Separate efficiency per computation from total system consumption and publish both so rebound growth cannot hide behind a better ratio
• Create a public registry of signatories, covered assets, baselines, targets, progress, verification, exceptions, misses and corrective action
• Pair developing-country commitments with technical assistance, concessional finance, workforce development and access to clean power and efficient equipment
• Evaluate clean-technology pilots against a measured baseline, disclose failures and prohibit unverified avoided-emissions claims from canceling measured infrastructure emissions