55 AI Environmental Impact Statistics for 2026

ByShane Tran Updated on

AI is growing fast — and so is its environmental footprint. Data centers now consume more electricity than some countries, and their demand for energy, water, and raw materials keeps climbing.

Here are 55 statistics that show the real scale of AI’s environmental impact.

Top AI Environmental Impact Statistics

  • Global data centers consumed approximately 415 TWh of electricity in 2024, accounting for about 1.5% of total global electricity consumption (IEA, Energy and AI).
  • Global data center electricity consumption is projected to double to approximately 945 TWh by 2030 in the IEA’s base case, reaching just under 3% of total global electricity demand (IEA, Energy and AI).
  • Google’s total greenhouse gas emissions reached approximately 14.5 million metric tons of CO2e in 2025, an 18% increase year-over-year and 81% higher than its 2019 baseline (Google, 2026 Environmental Report).
  • Microsoft’s total greenhouse gas emissions reached approximately 20.3 million metric tons of CO2e in fiscal year 2025, a 25% increase from the previous year (Microsoft, 2026 Sustainability Report).
  • Global AI demand is projected to account for 4.2 to 6.6 billion cubic meters of water withdrawal in 2027 (Li, Yang, Islam & Ren, 2025).
  • Generative AI could produce a cumulative 1.2 to 5.0 million tons of electronic waste between 2020 and 2030 (Wang et al., Nature Computational Science, 2024).
  • 72% of Americans say they are at least somewhat concerned about AI’s environmental impact, with 41% very or extremely concerned (AP-NORC/EPIC, 2025).

How Much Energy Does AI Use?

1. Global data centers consumed approximately 415 TWh of electricity in 2024, accounting for about 1.5% of total global electricity consumption (IEA, Energy and AI).

2. Global data center electricity consumption is projected to double to approximately 945 TWh by 2030 in the IEA’s base case, reaching just under 3% of total global electricity demand (IEA, Energy and AI).

AI Energy Consumption Statistics

The growth in AI workloads is reshaping electricity demand at both national and global scales, with projections consistently revised upward.

3. Electricity demand from AI-optimized data centers is projected to more than quadruple by 2030, as AI drives surging data center electricity demand (IEA, Energy and AI press release).

4. Electricity consumption in accelerated servers, mainly driven by AI adoption, is projected to grow by 30% annually through 2030 (IEA, Energy and AI).

5. Data center electricity consumption in the United States is expected to increase by up to 240 TWh by 2030, a 130% jump compared to 2024 levels (IEA, Energy and AI).

6. Data centers could consume 9% to 17% of U.S. electricity generation by 2030, more than double their current share (EPRI, Powering Intelligence 2026).

7. EPRI’s 2026 data center load estimates are 60% higher than its prior 2024 projections, driven by the accelerated pace of data center development (EPRI, Powering Intelligence 2026).

8. U.S. data centers consumed approximately 176 TWh of electricity in 2023, about 4.4% of total U.S. electricity consumption (LBNL, 2024 US Data Center Energy Usage Report).

9. Google consumed approximately 42 million MWh of electricity across its data centers in 2025, with total electricity consumption rising 37% year-over-year (Google, 2026 Environmental Report).

10. In Virginia, data centers already consume around 25% of the state’s electricity, and that share could rise to between 39% and 57% by 2030 (EPRI, Powering Intelligence 2026).

How much energy does a single AI query use?

11. A typical ChatGPT query using GPT-4o consumes approximately 0.3 watt-hours of electricity, roughly ten times less than the widely cited 2023 estimate of 3 Wh (Epoch AI, 2025).

12. OpenAI CEO Sam Altman stated that the average ChatGPT query uses approximately 0.34 Wh of electricity (Sam Altman, The Gentle Singularity).

13. Training GPT-4 is estimated to have consumed between 51,773 and 62,319 MWh of electricity, using approximately 25,000 NVIDIA A100 GPUs over 90-100 days (Ludvigsen, Towards Data Science, 2023).

AI Carbon Footprint Statistics

Data centers currently represent a small share of global emissions, but that share is growing quickly as AI infrastructure scales up.

How much CO2 do data centers produce?

14. Data centers currently account for approximately 0.5% of global CO2 emissions (Carbon Brief analysis of IEA data, 2025).

15. Global CO2 emissions from data center electricity use currently stand at approximately 180 million tonnes, projected to grow to 300 Mt by 2035 under the IEA’s base case (IEA, Energy and AI, 2025).

16. Google’s total greenhouse gas emissions reached approximately 14.5 million metric tons of CO2e in 2025, an 18% increase year-over-year and 81% higher than its 2019 baseline (Google, 2026 Environmental Report).

17. Microsoft’s total greenhouse gas emissions reached approximately 20.3 million metric tons of CO2e in fiscal year 2025, a 25% increase from the previous year (Microsoft, 2026 Sustainability Report).

18. Microsoft’s Scope 2 emissions jumped from roughly 2% to 13% of its total carbon footprint in fiscal year 2025, after the company stopped using unbundled renewable energy certificates (Microsoft, 2026 Sustainability Report).

19. Meta reported total greenhouse gas emissions of 8.2 million metric tons of CO2e in 2024, with 99% attributed to Scope 3 supply chain sources (Meta, 2025 Sustainability Report).

How much CO2 does training an AI model produce?

20. Training OpenAI’s GPT-3 consumed an estimated 1,287 MWh of electricity and emitted approximately 552 tonnes of CO2e (Patterson et al., 2021).

21. Training the 176-billion-parameter BLOOM model produced 50.5 tonnes of CO2e across its full lifecycle, including embodied hardware emissions (Luccioni, Viguier & Ligozat, 2022).

22. Training Meta’s Llama 3.1 405B required 30.84 million GPU hours and produced an estimated 8,930 tonnes of CO2e in location-based emissions (Meta, Llama 3.1 Model Card, 2024).

23. In Luccioni et al.’s comparison of location-based training emissions, GPT-3 produced 502 tonnes of CO2e, OPT produced 70 tonnes, and BLOOM just 25 tonnes — roughly 20 times less than GPT-3, largely because France’s grid carbon intensity is 57 gCO2eq/kWh versus 429 gCO2eq/kWh for GPT-3’s U.S.-based training (Luccioni, Viguier & Ligozat, 2022).

AI Model Training Carbon Emissions

Source: Luccioni, Viguier & Ligozat, 2022 (arXiv:2211.02001), Table 4

AI Water Consumption Statistics

AI infrastructure is a growing consumer of freshwater for cooling. For a deeper look at this topic, see our AI water usage statistics.

24. Global AI demand is projected to account for 4.2 to 6.6 billion cubic meters of water withdrawal in 2027 (Li, Yang, Islam & Ren, 2025).

25. Training GPT-3 in Microsoft’s U.S. data centers can directly evaporate 700,000 liters of clean freshwater (Li et al., 2025).

26. Google’s water consumption climbed 34% in 2025 to 10.9 billion gallons, more than double its 2021 level (Axios, citing Google 2026 Environmental Report).

27. Google’s water stewardship projects replenished approximately 7.7 billion gallons of water in 2025, roughly 78% of its freshwater consumption (Google, 2026 Environmental Report).

AI Hardware and E-Waste Statistics

The physical infrastructure behind AI creates its own environmental burden, from semiconductor manufacturing to the disposal of rapidly obsolete hardware.

28. Generative AI could produce a cumulative 1.2 to 5.0 million tons of electronic waste between 2020 and 2030 (Wang et al., Nature Computational Science, 2024).

AI E-Waste Projection 2030

Source: Wang et al., Nature Computational Science, 2024

29. Implementing circular economy strategies could reduce generative AI e-waste by 16% to 86% (Wang et al., Nature Computational Science, 2024).

30. The world generated a record 62 million tonnes of e-waste in 2022, an 82% increase from 2010 (UNITAR/ITU, Global E-Waste Monitor 2024).

31. Only 1% of global rare earth element demand is currently met through e-waste recycling (UNITAR/ITU, Global E-Waste Monitor 2024).

32. Semiconductor giant TSMC consumed 101 million cubic meters of water in 2023 (IDTechEx, 2025).

33. Water usage across semiconductor manufacturing is forecast to double by 2035 as chip demand continues to rise (IDTechEx, 2025).

34. By 2030, data center demand for gallium could equal up to 10% of today’s global supply, while China accounts for 95% of gallium refining (IEA, Energy and AI, 2025).

AI Efficiency and Mitigation Statistics

While AI’s resource demands are growing, efficiency improvements in hardware, software, and data center operations are partially offsetting the increase.

How efficient are AI data centers?

35. Google operates its data centers at a fleet-wide average PUE of 1.09 (Google Cloud Blog, August 2025).

36. The global average data center PUE stood at 1.54 in 2025, virtually unchanged for the sixth consecutive year (Uptime Institute, 2025).

37. Between May 2024 and May 2025, the energy consumed per median Gemini Apps text prompt fell by 33 times, and its carbon footprint dropped by 44 times (Google Cloud Blog, August 2025).

Gemini Energy Efficiency Improvement

Source: Google Cloud Blog, August 2025

38. The median Gemini Apps text prompt uses 0.24 watt-hours of energy (Google Cloud Blog, August 2025).

39. Google reduced its data center energy emissions by 12% in 2024 even as electricity demand grew 27% year-over-year (Google, 2025 Environmental Report).

40. Pre-training compute efficiency for AI models is improving at roughly 3 times per year (Epoch AI, Trends dashboard, 2026).

How Are Tech Companies Addressing AI’s Environmental Impact?

Major technology companies are pursuing renewable energy procurement and nuclear power agreements to meet the growing electricity demands of AI infrastructure.

41. Google increased its 24/7 carbon-free energy percentage from 64% to 66% in 2024 (Google, 2025 Environmental Report).

42. Microsoft matched 100% of its global electricity with renewable energy in 2025, backed by 40 GW of new renewable energy contracted across 26 countries (Microsoft Blog, February 2026).

43. Amazon matched 100% of its electricity with renewable energy for the second consecutive year in 2024, with over 600 renewable energy projects (Amazon, 2024 Sustainability Report).

Are tech companies investing in nuclear power for AI?

44. Microsoft signed a 20-year power purchase agreement with Constellation Energy to restart the 835 MW Crane Clean Energy Center (formerly Three Mile Island Unit 1) (Microsoft Blog, February 2026).

45. Amazon secured a 1,920 MW nuclear power purchase agreement with Talen Energy from the Susquehanna nuclear plant (Talen Energy, June 2025).

46. Google signed the world’s first corporate agreement to buy power from multiple small modular reactors, contracting up to 500 MW from Kairos Power by 2035 (Google Blog, October 2024).

AI Environmental Impact Projections

Forecasts for AI-driven electricity demand continue to be revised upward as the pace of data center construction accelerates.

47. Goldman Sachs Research estimates global data center power demand will grow 160% by 2030, pushing their share of electricity from 1-2% to 3-4% (Goldman Sachs, 2024).

48. U.S. data centers are expected to consume 6.7% to 12% of total national electricity by 2028 (U.S. DOE / LBNL, 2024).

49. Roughly 20% of planned data center projects worldwide could face delays due to electricity grid constraints (IEA, Energy and AI, 2025).

50. In the United States, data centers are expected to account for nearly half of all electricity demand growth through 2030 (IEA, Energy and AI, 2025).

Public Opinion and AI Environmental Regulation

Concern about AI’s environmental footprint is growing among the public, and regulators are beginning to impose transparency requirements on AI developers.

Are Americans worried about AI’s environmental impact?

51. 72% of Americans say they are at least somewhat concerned about AI’s environmental impact, with 41% very or extremely concerned (AP-NORC/EPIC, October 2025).

52. American concern about AI’s environmental impact (41% very/extremely concerned) surpasses concern about cryptocurrency (29%), meat production (29%), and air travel (23%) (AP-NORC/EPIC, 2025).

Americans’ Concern About Environmental Impact by Industry

Source: AP-NORC/EPIC, October 2025 (n=3,154)

53. 50% of U.S. adults now say they are more concerned than excited about AI in daily life, up from 37% in 2021 (Pew Research Center, September 2025).

54. Under the EU AI Act, providers of general-purpose AI models must disclose energy consumption, with non-compliance penalties reaching up to the greater of EUR 15 million or 3% of worldwide annual turnover (White & Case analysis of EU AI Act, 2025).

55. The EU AI Act requires the European Commission to publish its first report on AI energy efficiency standards by August 2028, with the potential to recommend legally binding measures (White & Case analysis of EU AI Act, 2025).

Conclusion

AI’s environmental costs are growing alongside the technology itself. Hardware and software efficiency are improving, but so far those gains haven’t kept pace with surging demand. What happens next depends on the infrastructure choices, energy investments, and transparency standards that take shape over the coming years.