Myth vs Fact: Is AI Really Bad for the Environment?
Separating real AI environmental impacts from exaggerations and wishful thinking.
Reading Time
6 min
Article Sections
6
Share Links
3
On This Page
Article Section
Introduction
Is AI wrecking the climate, or is its footprint too small to matter? Both claims circulate widely, and both are easy to support with a number from somewhere.
The stakes are practical. Overstating AI's impact can slow useful projects, while dismissing it can leave electricity, water and local planning risks unmanaged. Readers in business, policy and investment need a balanced way to weigh the evidence.
This article tests four common myths about AI environmental impact against International Energy Agency (IEA) data and company disclosures, and shows how to judge future claims.
Article Section
Context: What Counts as AI's Environmental Footprint?
AI's footprint has several parts. Training builds a model, and inference runs it each time someone uses it. Direct impacts occur inside the data centre, mainly electricity for computing and cooling, plus water for cooling. Indirect impacts include the emissions of the grid supplying the power, water used at power plants, and the manufacturing of chips and servers.
Estimates differ because boundaries differ. One study may count only the chip's power, another the whole data centre including idle capacity and cooling overhead. Location matters too, since the same workload has different emissions on different grids.
Misinformation usually enters at this point. A viral per-prompt figure may combine direct and indirect water, while a company average may cover only a median task. Neither is necessarily false, but they cannot be compared without knowing what was counted.
Article Section
Myth vs Fact: AI Environmental Impact, Four Myths Tested
Myth 1: AI is a major drain on the world's electricity.
Why People Believe It: Headlines about gigawatt campuses and rising power demand suggest AI is reshaping global energy use. Fact: Data centres, which run AI alongside other digital services, are a small but fast-growing share of global electricity. Evidence: The IEA estimates data centres used about 415 TWh in 2024, around 1.5% of global electricity, and projects about 945 TWh by 2030 in its base case, just under 3%. That is growth of roughly 15% a year, yet only around one-tenth of global electricity demand growth to 2030. Practical Takeaway: Judge impact locally as well as globally, because the IEA notes that demand is concentrated, with the United States accounting for about 45% of data centre consumption in 2024.
Myth 2: Every AI prompt uses a bottle of water and huge amounts of energy.
Why People Believe It: Early academic estimates for large models were high, and figures of tens of millilitres of water per response spread widely. Fact: Operator disclosures for a typical text prompt are far smaller, though they depend on how impacts are counted. Evidence: Google reports that its median Gemini text prompt in May 2025 used 0.24 Wh of energy and about 0.26 mL of water. This is a company disclosure for a median prompt, indirect water from power generation is not counted, and heavier tasks such as long reasoning or video generation sit well above the median. Practical Takeaway: Treat per-prompt figures as ranges tied to a method and a task, not as universal constants.
Myth 3: AI is getting more efficient, so its footprint will shrink.
Why People Believe It: Companies report large per-task gains, which suggests total impact must fall. Fact: Efficiency per task and total demand can move in opposite directions when use grows and heavier uses appear. Evidence: Google reports a 33-fold reduction in energy per median prompt over one year, and the IEA notes that energy per AI query has fallen sharply. Even so, the IEA projects data centre electricity use roughly doubling by 2030, as AI use expands and more energy-intensive applications become popular. Practical Takeaway: Ask for total energy and emissions as well as per-task figures.
Myth 4: AI will cancel out its own footprint by cutting emissions elsewhere.
Why People Believe It: AI is used in grid planning, building controls and industrial optimisation, so savings seem automatic. Fact: Potential savings exist, but they depend on deliberate adoption and are not guaranteed. Evidence: The IEA says widespread adoption of existing AI applications could cut emissions by more than data centres emit, but by far less than climate goals require. It projects data centre electricity emissions rising from about 180 Mt today to about 300 Mt by 2035 in its base case, still below 1.5% of energy sector emissions. Practical Takeaway: Treat expected savings as a hypothesis to measure, not a credit to claim in advance.
Article Section
Broader Lessons: Judging Claims About AI's Footprint
Ask what is being measured
Check whether a figure covers training or inference, direct or indirect impacts, and a median or an average task.
Match the model to the task
Where a smaller model meets the need, it uses less energy than a larger one, so reserving heavy models for heavy tasks cuts both energy and cost.
Check the power source and location
The same workload has different emissions and water effects depending on the grid supplying it and the local water stress.
Treat savings as a hypothesis
Measure claimed reductions against a baseline before reporting them, and keep reporting total energy and water use alongside them.
Article Section
Conclusion
The evidence supports neither alarm nor dismissal. AI's electricity use is small globally but growing fast and concentrated locally, per-prompt figures depend on method, efficiency gains do not guarantee lower totals, and emission savings have to be earned.
As AI use spreads, the most reliable test of any claim is what was counted and over what boundary.
Article Section
Frequently Asked Questions
Is AI bad for the environment?
It has real and growing impacts on electricity, water and emissions, but its global share is still small, so the answer depends on scale, location and how it is used.
How much energy does AI use, and is it more energy-intensive than traditional software?
The IEA estimates data centres used about 1.5% of global electricity in 2024, and AI tasks generally need more computing than simple software, with the gap depending on the model and task.
Does AI use water?
Yes, mainly for data centre cooling and indirectly through power generation, though the amount per prompt depends on cooling design, location and how it is measured.
What creates AI's carbon footprint, and can AI help reduce emissions?
Electricity use is the main source, along with hardware manufacturing and construction, and AI can help cut emissions elsewhere, but the IEA says those savings are not guaranteed.
How can companies use AI more sustainably?
By matching models to tasks, choosing locations and suppliers with cleaner power and efficient cooling, and tracking total energy and water use.
Keep Reading
Related Blogs
Next Step
Ready to turn ESG complexity into strategic advantage?
Talk to ESG Astraa about disclosures, climate strategy, governance controls, and execution support for your team.