How big is AI’s carbon footprint, really?
A plain-language breakdown of what we know, what we don’t, and how AI emissions compare to the industries you already understand.
Sources: IEA, IATA, UNEP, and peer-reviewed research.
Source: IEA, Energy and AI (2025) — data centre electricity-related CO₂ emissions, all uses. AI-specific share estimated at 32–44% of total by 2030.
Before any numbers.
Transparency in this space is still limited. Tech companies are not legally required to disclose the energy use of their AI systems, so every figure on this page is an estimate derived from methodology, not corporate self-reporting.
That is exactly why the methodology matters. The numbers below are not guesses. They are bounded estimates built from public data, peer-reviewed research, and the most credible institutional sources available.
How we measure it.
Three steps. Each one sourced. The result is a range, not a single number, because the range is the honest answer.
Total data center electricity
The IEA reports that global data centers consumed 415 TWh of electricity in 2024, around 1.5% of total global electricity demand. This is the starting point.
Source: IEA, Energy and AI Report, 2025
Isolate the AI share
AI currently accounts for an estimated 5–15% of data center electricity use, rising to 35–50% by 2030 as workloads scale. That gives us the electricity attributable specifically to AI.
Source: IEA-4E TCP Critical Review, 2025
Convert electricity to carbon
We multiply by the carbon intensity of the grids powering those data centers, around 395g CO₂ per kWh. This is slightly below the global average because data centers are concentrated in the US and Europe, which run on cleaner grids.
Source: IEA Global Energy Review, 2024
32.6–79.7
Million tonnes of CO₂ from AI
The range reflects genuine data uncertainty, not imprecision. Smoothing it into a single number would be misleading.
Source: The Carbon and Water Footprints of Data Centers, Cell Press / ScienceDirect, December 2025
AI vs five industries.
Annual greenhouse gas emissions, in million tonnes of CO₂. Bars are scaled to the largest value on the chart.
Agriculture
Farm-gate production of crops and livestock emits roughly 8.1 billion tonnes of CO₂-equivalent each year — the largest single source on this list, and that is before land use change is counted.
Source: FAO — farm-gate production only (crops + livestock)
Oil & gas operations
Extracting, processing and transporting oil and gas releases about 5.1 billion tonnes of CO₂-equivalent a year — and this figure excludes the much larger emissions from burning the end products.
Source: IEA — scope 1 & 2 operations; combustion of end products excluded
Steel
Producing the world’s steel releases about 4.1 billion tonnes of CO₂-equivalent a year on a full scope 1, 2 & 3 basis — the backbone material of buildings, vehicles, and infrastructure.
Source: World Steel Association / WEF — scope 1, 2 & 3
Cement
Cement manufacturing emits roughly 1.6 billion tonnes of CO₂ a year, mostly from the chemistry of making clinker — emissions that exist whether the grid is clean or not.
Source: IEA / Global Cement and Concrete Association
Aviation
Every commercial flight taken worldwide in 2024 produced a combined 942 million tonnes of CO₂ — roughly eleven to twenty-eight times current AI emissions, depending on which AI estimate you use.
Source: IATA — gross total, all commercial aviation
AI.
Estimated emissions from AI workloads in 2025: operational electricity (IEA) plus the lifecycle emissions of the hardware itself (~1.8 Mt). The range reflects genuine uncertainty about AI’s share of total data centre electricity.
Source: IEA + Luccioni et al. (arxiv 2025) hardware lifecycle
All figures CO₂e. Industry totals reflect full production-scope emissions. AI figure includes operational electricity plus hardware manufacturing. Agriculture figure covers farm-gate only; full agrifood system including land use: 16,500 Mt CO₂e (FAO 2023).
AI’s current footprint is real but relatively small compared to legacy industries. The concern is not today’s number. It is the trajectory.
The IEA projects data center emissions could reach 300–500 million tonnes by 2035, depending on how fast AI scales and how quickly grids decarbonise.
Source: IEA, Energy and AI, executive summary, 2025
For your organisation.
Sustainability Teams
Your Scope 3 reporting may already include AI. Do you know how to measure it?
Talk methodologyLeadership & Strategy
AI adoption decisions are now climate decisions. Your strategy needs to reflect both.
Build the strategyMarketing & Communications
Greenwashing risk is rising. Getting the numbers right protects your brand.
Sharpen the messageEvery number, sourced.
- 01IEA, Energy and AI, 2025
- 02IEA, Global Energy Review, 2025
- 03IEA-4E TCP, Data Centre Energy Use: Critical Review of Models and Results, 2025
- 04IEA, Emissions from Oil and Gas Operations in Net Zero Transitions, 2022
- 05IATA, Net Zero Progress Report, 2024
- 06FAO, Greenhouse Gas Emissions from Agrifood Systems, Global Trends 2001–2023
- 07World Steel Association / WEF, Net-Zero Industry Tracker, 2024
- 08Global Cement & Concrete Association
- 09UNEP, Emissions Gap Report, 2024
- 10Luscombe et al., The Carbon and Water Footprints of Data Centers, Cell Press / ScienceDirect, December 2025
- 11Luccioni et al., Life-Cycle Emissions of AI Hardware: A Cradle-To-Grave Approach, arxiv 2025
The numbers exist. Most organisations haven’t looked.
The AI emissions debate moves fast. The gap between what’s being reported, what’s being measured, and what’s actually happening is where reputational and strategic risk lives.
This index is a starting point — not a verdict.
Bring this thinking into your organisation.
I work with sustainability teams, leadership groups, and marketing functions to build strategy, fluency, and communications around AI and sustainability. If this resource was useful, the work I do goes much deeper.