How much CO₂ and water does an AI query served from a data centre in Spain use?

About 0.022 g CO₂e per exchange, as SOMA AI estimates it on the average Spanish grid (170 g CO₂e per kWh, 2023 data). That is one exchange of 1,210 tokens on a mid-size (class B) model, including electricity, hardware, the building and amortised training. Spain ranks 12 of the 27 EU countries, from lowest to highest. It is about 2.4 times Sweden's, Poland's is about 3.1 times Spain's, and Spain sits below the EU average (0.029 g) and below the world average (0.053 g).

A query is emitted where the data centre is, not where the user sits. The SAEF dataset maps 3 cloud regions to the Spanish grid:

• AWS eu-south-2, with three availability zones in Villanueva de Gállego, El Burgo de Ebro and Huesca since November 2022, per the AWS fact sheet "AWS en Aragón", 2026 (Spanish)
• Azure spaincentral, in Madrid per the dataset
• Google Cloud europe-southwest1, in Madrid per the dataset

All three use the same Spanish grid factor, so SAEF gives them the same estimate, 0.022 g per exchange. Among the 46 grids that host an AWS, Azure or Google Cloud region, Spain ranks 12 from lowest to highest. The AWS region has its own factor page.

SOMA gives no per-query water figure for Spain. The SAEF dataset does not yet publish water factors for the Spain region, and we would rather print nothing than a partial figure. Site-level figures are published. AWS states that its Aragón data centres used about 68 million litres of water in 2025 (AWS fact sheet "AWS en Aragón", 2026 (Spanish)). On 23 October 2026 the Aragón environmental agency (Inaga) authorised AWS to go from 36,461 to 53,961 m³ a year at El Burgo de Ebro, 48% more, which the company attributes to the effect of climate change in the area (elDiario.es Aragón, Luis Faci, 9 November 2026 (Spanish)). These are figures for specific sites and cannot be divided per query without knowing how much AI load each one serves. The AI water footprint guide explains what each water figure counts.

The SpainDC association's 2025 annual report puts installed IT power in commercial (colocation and hyperscale) data centres at 355 MW in 2024 and 439 MW at the end of 2025, and projects 2,537 MW by 2030 if the trend holds (Energías Renovables on the SpainDC 2025 annual report, 26 March 2026 (Spanish)). Citing the same association, El Independiente reports that data centres in Spain used about 2 TWh of electricity, 0.8% of the national total, and counts 97 data centres in operation (El Independiente, Mikel Segovia, 30 March 2026 (Spanish)). The government's draft royal decree estimates about 2.5 GW of computing power by 2030, needing 3.5 to 4 GW of electrical demand, and notes that more than 12 GW of grid access and connection rights have been granted since 2021 (MITECO, draft royal decree on data centres, 27 August 2026 (Spanish)). When we checked on 11 November 2026 the decree had not been approved.

Simple arithmetic on those published figures gives an illustrative scenario, not a forecast.

• CO₂ in 2025. 2 TWh × 0.1697 kg CO₂e per kWh = about 0.34 million tonnes CO₂e. The factor is the SAEF Spanish grid average for 2023, the latest year in the dataset, applied to 2025 consumption.
• Assumption for 2030. Each MW of IT power uses as much electricity as in 2025, that is 2 TWh over 439 MW, about 4,556 MWh a year per MW, the same as running each MW at about 52% of its rated power all year, cooling included. The 2 TWh covers all data centres in the country and the 439 MW only the commercial ones, so this assumption tends to inflate the result.
• Electricity in 2030 on that assumption. 2,537 MW × 4,556 MWh = about 11.6 TWh a year.

We give no 2030 tonnes. Spain's grid is decarbonising, so applying today's factor to 2030 would overstate them, and the 2030 grid factor is not known. SpainDC's 2030 figure is itself a trend projection, and the sector reports renewable contracts that a location-based figure does not capture. Data centres serve many workloads, from email and websites to banking and storage, not only AI, so none of this is the footprint of AI in Spain. And a location-based grid average is not the electricity mix of any specific site.

Because AI bought from a provider falls in the company's Scope 3, and the region the model runs in is an input to the estimate. The same class B exchange is estimated at 0.022 g on the Spanish grid and 0.029 g on the EU average, and model size weighs even more, from 0.006 g on a small model to 0.086 g on a frontier one in the Spanish region. Ask the provider where your usage is served, record it and use that region's factor, not your office's country. The page on cutting the AI footprint ranks the levers, the free calculator applies the region for you, and the Scope 3 guide has the method.

• That per-query figures are measurements. SOMA estimates them from the SAEF factors and a typical exchange, and real exchanges vary.
• That any provider serves any given model from Spain. Regions are dataset inputs, not claims about providers.
• That the Spanish grid average is the mix of a specific data centre, or that it reflects renewable contracts.
• That Spanish data centres serve only AI. They serve many workloads.
• That the 2030 arithmetic is a forecast. It is an illustrative order of magnitude for electricity with a stated assumption, and it gives no 2030 emissions.
• A per-query water figure for Spain, because SAEF has no water factors for that region yet.

The SAEF method was accepted for publication in Sustainable Production and Consumption. The equations are on the methodology page.

• Where does an AI query emit the least? The carbon footprint of AI by country. The grids of 210 countries and the EU-27 ranked
• What is the carbon footprint of one ChatGPT prompt? Published per-prompt figures and their basis
• Is AI bad for the environment? Per query against the total
• Why should AI providers say where your prompt runs? Why providers should disclose the serving region