Artificial Intelligence:
Environmental Impact &
Economic Effects
Energy Consumption, Water Use, Carbon Footprint & Economic Impact
Created with the assistance of Claude.ai
AI's Growing Energy Footprint: The Scale
~415 TWh
Global data center electricity (2024)
IEA, 2025
~945 TWh
Projected global consumption by 2030
IEA, 2025
~183 TWh
U.S. data center use in 2024 — >4% of all U.S. electricity
IEA, 2025
Regional concentration: Virginia data centers consume 26% of state electricity; North Dakota 15%. The U.S. accounts for ~45% of all global data center capacity. (IEA, 2025; Pew Research, 2025)
U.S. Growth Trajectory
IEA projects U.S. data center electricity demand will grow 133% by 2030 — from 183 TWh to ~426 TWh. That growth exceeds the entire current electricity consumption of France. The increase is driven primarily by AI workloads, not general computing growth (IEA, 2025).
Energy Per Query: A Concrete Comparison
A modern Google search uses ~0.04 Wh. A ChatGPT query uses ~2.9 Wh — roughly 50–100x more per interaction. The widely cited "10x" figure was based on a 2009 Google baseline (0.3 Wh/search); advances in server efficiency make the actual gap significantly larger (de Vries, 2023; Google, 2009).
IEA. (2025). Energy and AI. | Pew Research Center. (2025, Oct. 24). Energy use at U.S. data centers amid the AI boom. | de Vries, A. (2023). The growing energy footprint of AI. Joule, 7(10).
AI's Specific Role & Carbon Footprint
33–80
Mt CO₂
Projected carbon footprint
of AI systems alone in 2025
Masanet et al.,
cited in ScienceDirect, 2025
For comparison: the entire aviation industry emits ~900 Mt CO₂/yr globally. AI's 33–80 Mt range reflects deep uncertainty in model size, query volume, and grid carbon intensity.
AI Is Now the Primary Demand Driver
IEA identifies AI as "the most important driver" of data center electricity growth. In the IEA Base Case, AI-optimized server electricity demand is projected to grow at 30% per year, more than quadrupling by 2030. Overall data center demand has grown at 12% per year since 2017 — AI is accelerating well above the baseline trend (IEA, 2025).
Scenario Range: Wide Uncertainty
IEA models a range of outcomes depending on adoption pace and hardware efficiency: total data center electricity demand could reach 700–1,700 TWh globally by 2035. The upper bound would equal roughly 6% of total global electricity supply at current levels — a meaningful macroeconomic energy commitment (IEA, 2025).
Training vs. Inference
Carbon footprint concentrates in two phases: training large models (a one-time high-energy event per model) and inference (every query, billions of times daily). Inference is now the dominant and fastest-growing share of AI energy use as deployment scales (de Vries, 2023; IEA, 2025).
IEA. (2025). Energy and AI. | Masanet, E., et al. (2020; updated projections cited 2025). Recalibrating global data center energy-use estimates. Science. | de Vries, A. (2023). Joule, 7(10).
AI's Hidden Water Crisis
~800B
liters
Total U.S. data center
water footprint in 2023
(direct + indirect use)
Lawrence Berkeley
National Laboratory, 2024
Direct use only: ~66B liters. The ~800B figure includes water consumed to generate the electricity powering data centers.
One conversation = one bottle
A 20–50 exchange AI conversation uses ~500 mL of water. Per-response use is small; the concern is cumulative scale at billions of daily interactions (Li, Yang, Islam & Ren, 2023, UC Riverside).
Projected 2030 footprint
U.S. AI server deployment estimated to generate 731–1,125 billion liters of water footprint annually by 2030 under current growth trajectories (Huang et al., Nature Sustainability, 2025).
1.9 liters per kWh
Data centers use an average of 1.9 liters of water per kWh of electricity consumed for cooling purposes. AI workloads are more compute-dense, increasing per-rack water demand (The Green Grid).
Evaporation, not return
Water withdrawn for evaporative cooling is largely not returned to local water supplies — a critical concern for data center clusters in already water-stressed regions (EESI, 2024).
Li, Yang, Islam & Ren. (2023). Making AI less thirsty. arXiv 2304.03271. | Huang, X., et al. (2025). Nature Sustainability. https://doi.org/10.1038/s41893-025-01681-y | Lawrence Berkeley National Lab. (2024). U.S. data center energy usage report. | The Green Grid. WUE. | EESI. (2024).
Water & Emissions: Corporate Pledges vs. Reality
The Pledge: Google, Microsoft, Meta & Amazon have all committed to being "water positive" by 2030 — returning more water than they consume globally.
Microsoft
Pledged: Water positive by 2030
Reality: Microsoft's own 2022 Environmental Sustainability Report disclosed a 34% rise in water use year-over-year, its largest single-year increase, coinciding with major AI infrastructure expansion. Electricity consumption also rose sharply in the same period.
Microsoft Environmental Sustainability Report, 2022 (reported in Food & Water Watch, 2025)
Pledged: Net-zero by 2030; water positive by 2030
Reality: Google's total greenhouse gas emissions were 48% higher in 2023 than in 2019. The company's own environmental report attributed the increase largely to growth in data center energy use driven by AI compute demands.
Noble & Berry (2024). Planet Detroit. https://planetdetroit.org/2024/10/ai-energy-carbon-emissions/
Industry-wide
Pledged: Net-zero aspirations by 2030
Reality: Huang et al. (2025) in Nature Sustainability found the AI server industry is "unlikely to meet its net-zero aspirations by 2030" without substantial reliance on highly uncertain carbon offset and water restoration mechanisms.
Huang, X., et al. (2025). Nature Sustainability. https://doi.org/10.1038/s41893-025-01681-y
Microsoft. (2022). Environmental sustainability report. | Noble & Berry. (2024, Oct. 2). Planet Detroit. | Huang, X., et al. (2025). Nature Sustainability. https://doi.org/10.1038/s41893-025-01681-y | Food & Water Watch. (2026). AI energy and water footprints.
Reasons for Optimism: Active Mitigation Efforts
Clean Energy Investment
Google signed contracts for 8 GW of new clean energy in 2024 alone, bringing 2.5 GW online, and partnered with Kairos Power for first-of-a-kind small modular reactors (SMRs) (IEA, 2025).
Microsoft committed 10.5 GW of renewable energy 2026-2030 via Brookfield Asset Management, and is repowering Three Mile Island to supply its data centers (Climate Solutions Legal Digest, 2024).
Meta wind and solar projects now add over 15 GW of clean energy to global grids; 100% of its electricity was matched with renewables in 2024 (Energy Digital, 2025).
Globally, renewables are projected to meet nearly half of new data center demand growth through 2035. In the U.S. specifically, natural gas leads — adding 130 TWh vs. renewables' 110 TWh by 2030. Decarbonization cannot be assumed; the IEA stresses that net-zero requires active policy intervention, not just efficiency gains (IEA, 2025).
Water-Saving Cooling Tech
Liquid cooling (direct-to-chip and immersion) eliminates evaporative cooling towers entirely, reducing water use by up to 70% vs. traditional air cooling (EESI, 2024).
Leading closed-loop liquid cooling facilities report a Water Usage Effectiveness (WUE) as low as 0.20 vs. an industry average of ~1.8 liters/kWh — though the 0.20 figure reflects best-in-class vendor performance, not a general industry benchmark (Cologix, 2025; The Green Grid).
The Microsoft sidekick system circulates chilled liquid in a closed loop, removing heat without large-scale water-consuming chillers (WWT, 2025).
By 2027, over 50% of new hyperscale data center capacity is projected to be liquid-cooled; liquid cooling market growing from .9B (2024) to 2.6B by 2034 (CIO.inc, 2025).
Algorithmic & Hardware Efficiency
Google DeepMind AI achieved up to 40% reduction in cooling energy in Google data centers through machine learning optimization — equal to a 15% reduction in total facility Power Usage Effectiveness (PUE). Source: Google DeepMind. (2016, Jul.). Data centre cooling. deepmind.google
NVIDIA Blackwell platform liquid cooling reduces annual energy consumption by 25% and cuts rack space by 75% vs. prior generations (NVIDIA, 2025).
DeepSeek R1 (Jan. 2025) demonstrated that models can run up to 40% more efficiently via algorithmic optimization — showing that software gains can rival hardware upgrades (Brown Advisory, 2025).
MIT researchers are developing workload-shifting tools that time AI jobs to run when grids carry more renewable power, reducing carbon intensity without new infrastructure (MIT News, 2025).
IEA. (2025). Energy and AI. | Energy Digital. (2025). How Meta is powering AI with renewable energy. | MIT News. (2025, Sept.). Responding to the climate impact of generative AI. | NVIDIA. (2025). Blackwell platform boosts water efficiency.
AI & the Economy: What Is Happening Right Now
Jobs Displaced (2025)
~54,836 AI-attributed layoffs in 2025
The highest AI-attributed total on record, from 1.21 million total U.S. layoffs. Important caveat: Oxford Economics and Yale Budget Lab (2026) both concluded companies are not replacing workers with AI at scale — AI attribution is self-reported by firms and may serve as corporate signaling. (Challenger, 2025; Oxford Economics, 2026).
Major tech layoffs citing AI explicitly
Amazon cut 14,000 corporate roles; Microsoft ~15,000; Workday 1,750 (8.5% of staff); Salesforce 4,000 customer support roles -- all named AI as a driver (AIM, 2025).
Entry-level and tech hit hardest
Big Tech cut new graduate hiring 25% in 2024. Unemployment among 20-30-year-olds in AI-exposed tech occupations rose nearly 3 percentage points since early 2025 (Goldman Sachs, 2025).
Customer service in steep decline
U.S. customer service employment fell by ~80,000 positions 2022-2024 as AI automation accelerated across the sector (AIM, 2025).
Jobs Created (2024-2025)
~119,900 direct jobs created in 2024
Includes ~8,900 AI/ML engineering and data science roles, plus over 110,000 construction jobs from data center expansion (each site requires ~1,500 workers). Net ratio: ~9 jobs created per 1 lost (ITIF, 2026).
AI skills now a hiring premium
96% of companies say AI skills benefit job candidates; 83% say AI proficiency improves existing job security. 91% of AI-using firms plan to hire more staff in 2025 (WEF, 2025).
Healthcare and trades resilient
Nurse practitioners projected +52% by 2033. Construction and skilled trades among least threatened. Food service expected to add 500,000+ roles by 2033 (National University, 2025).
Net picture is positive -- so far
A 2025 Economic Innovation Group study found no significant nationwide increase in unemployment due to AI. However, displacement is concentrated and the new roles require different skills and credentials (EIG, 2025).
Challenger, Gray & Christmas. (2025). Job cuts report. | Goldman Sachs. (2025). How will AI affect the global workforce. | ITIF. (2026). AI job impact: Gains outpace losses. | EIG. (2025). AI and jobs: The final word.
AI & the Economy: The Long-Term Outlook
92M
jobs displaced globally by 2030
WEF, 2025
170M
new roles created by 2030
WEF, 2025
$2.6–4.4T
annual GenAI value across enterprise use cases
McKinsey, 2023
+1.5%
projected U.S. productivity gain — if widely adopted over a decade
Goldman Sachs, 2023 (projection)
The Displacement-Creation Gap
WEF projects a net gain of 78 million jobs globally by 2030, but cautions these are not one-to-one swaps. New roles concentrate in AI development, cybersecurity, and green energy — requiring different skills and credentials than those displaced. 63% of employers cite skills gaps as the primary barrier, and 40% of workers' core skills will need updating by 2030 (WEF, 2025).
Who Bears the Risk?
Women face disproportionate exposure: the Kenan Institute at UNC estimates 8 in 10 women in the U.S. workforce hold jobs with high AI automation risk, vs. roughly 6 in 10 men. IMF (WP/25/76, 2025) warns that without inclusive policy, AI could sharply widen inequality both between advanced and emerging economies, and between high- and low-credential workers within countries.
World Economic Forum. (2025). Future of jobs report 2025. | Kenan Institute of Private Enterprise. (2023). Automation and the future of work. UNC. | IMF. (2025). The global impact of AI: Mind the gap (WP/25/76).
AI & the Economy: The Productivity Upside
These are projections — not current findings. All figures are contingent on adoption pace and remain actively contested.
+0.5%
Global GDP growth per year, 2025–2030
IMF WP/25/81, 2025
+1.5%
U.S. labor productivity per year — after widespread adoption
Goldman Sachs, 2023 (projection)
$2.6–4.4T
Annual GenAI value across 63 enterprise use cases
McKinsey, 2023 (projection)
IMF: GDP Growth Under Current Energy Policy
A companion IMF paper (WP/25/81, 2025) models the macroeconomic implications of AI-driven energy demand and finds that AI raises average annual global GDP growth by ~0.5 percentage points between 2025 and 2030 under current policies. This is a scenario estimate, not an empirical finding.
Goldman Sachs: The 1.5% Productivity Projection
Goldman Sachs economists Briggs & Kodnani (2023) projected AI could add ~1.5 percentage points of annual U.S. labor productivity growth over the decade following widespread adoption — contingent on full integration across the economy. GS's March 2026 research found no meaningful economy-wide impact yet.
McKinsey: $2.6–4.4 Trillion in Enterprise Value
McKinsey Global Institute (2023) estimated GenAI could generate $2.6–4.4 trillion annually across 63 business use cases — described as "equivalent to adding the UK economy each year." This is potential theoretical value, not realized productivity. The range reflects wide uncertainty in adoption pace.
IMF. (2025). Power hungry: How AI will drive energy demand (WP/25/81). | Goldman Sachs. (2023). The potentially large effects of AI on economic growth (Briggs & Kodnani). | McKinsey Global Institute. (2023). Economic potential of generative AI.
A Critical View: Could AI Harm the Human Experience?
Mainstream economists — not fringe voices — are raising structural concerns that go beyond short-term job loss.
The "Jobs Always Come Back" Assumption Is Untested at This Scale
Daron Acemoglu (MIT; 2024 Nobel laureate in Economics, shared with Simon Johnson and James Robinson, for work on how institutions shape prosperity) argues the reassurance that new jobs always replace lost ones rests on only two historical transitions — agrarian to industrial, and industrial to services. That sample is too small to treat as an economic law, and GenAI targets cognitive work in a way prior automation never did (Acemoglu & Johnson, Power and Progress, 2023).
"So-So Automation": Productivity Gains May Not Justify the Displacement
Acemoglu & Restrepo (2019) introduced the "so-so automation" concept: when AI becomes only marginally better than humans at cognitive tasks, firms automate to cut labor costs, but productivity gains are too modest to generate the new demand needed to re-employ displaced workers — leaving them worse off with no offsetting economic benefit. Acemoglu (2024, NBER WP 32512) extends this analysis specifically to generative AI.
Redistribution Is Not Automatic — And Policy Is Lagging Far Behind
Korinek and Stiglitz (2025) argue economists "set themselves too easy a goal" by demonstrating AI can benefit everyone in theory, without specifying how redistribution happens in practice. The IMF (WP/25/76) finds that high-income countries spend an average of just 0.1% of GDP on active labor market programs — far below what the pace and scale of AI-driven displacement would require to prevent rising structural unemployment (Korinek & Stiglitz, 2025; IMF, 2025).
The AGI Scenario: Wage Share Collapse
Korinek & Suh (2024) model the macroeconomic effects of AI automation and find that labor's share of income remains stable only if automation proceeds gradually. Under a scenario approaching Artificial General Intelligence — where nearly all cognitive tasks become automatable — their model projects a collapse in the wage share of GDP, structurally impoverishing workers as a class regardless of overall economic growth. "Profit-driven AI may be inherently biased against workers" (Spencer, PMC, 2025).
Acemoglu & Johnson. (2023). Power and progress. PublicAffairs. | Acemoglu. (2024). The simple macroeconomics of AI. NBER WP 32512. | Acemoglu & Restrepo. (2019). Automation and new tasks. AEA Papers & Proceedings. | Korinek & Stiglitz. (2025). Steering technological progress. INET. | Korinek & Suh. (2024). Scenarios for the transition to AGI. NBER WP 32255.
References
de Vries, A. (2023). The growing energy footprint of artificial intelligence. Joule, 7(10), 2191-2194. https://www.cell.com/joule/fulltext/S2542-4351(23)00365-3
Food & Water Watch. (2026, February). A no brainer: How AI energy and water footprints threaten our climate. https://www.foodandwaterwatch.org/wp-content/uploads/2026/02/FSW_2602_AI_Water_Energy_UPDATE.pdf
Google. (2009, January 11). Powering a Google search [Blog post]. https://googleblog.blogspot.com/2009/01/powering-google-search.html
Huang, X., et al. (2025, November 10). Environmental impact and net-zero pathways for sustainable AI servers in the USA. Nature Sustainability. https://doi.org/10.1038/s41893-025-01681-y
International Energy Agency. (2025). Energy and AI. IEA Publications. https://www.iea.org/reports/energy-and-ai
Li, P., & Ren, S. (2023). Making AI less thirsty: Uncovering the secret water footprint of AI models. arXiv. https://arxiv.org/abs/2304.03271
Noble, G., & Berry, F. (2024, October 2). AI and climate: Energy and water use from AI applications. Planet Detroit. https://planetdetroit.org/2024/10/ai-energy-carbon-emissions/
U.S. Environmental and Energy Study Institute. (2024). Data centers and water consumption. https://www.eesi.org/articles/view/data-centers-and-water-consumption
Brown Advisory. (2025). The data center balancing act: Powering sustainable AI growth. https://www.brownadvisory.com/us/insights/data-center-balancing-act-powering-sustainable-ai-growth
Climate Solutions Legal Digest. (2024, December). Renewable energy and AI data centers. https://www.climatesolutionslaw.com/2024/12/renewable-energy-and-ai-data-centers/
Cologix. (2025). Liquid cooling for data centers: Meeting the growing demand of AI. https://cologix.com/resources/blogs/liquid-cooling-for-data-centers-meeting-the-growing-demand-of-ai/
Energy Digital. (2025, December 2). How Meta is powering AI and data centres with renewable energy. https://energydigital.com/news/how-meta-is-powering-ai-data-centres-with-renewable-energy
MIT News. (2025, September 30). Responding to the climate impact of generative AI. https://news.mit.edu/2025/responding-to-generative-ai-climate-impact-0930
All citations formatted in APA 7th Edition style.
Pew Research Center. (2025, Oct. 24). Energy use at U.S. data centers amid the AI boom. https://www.pewresearch.org/short-reads/2025/10/24/what-we-know-about-energy-use-at-us-data-centers-amid-the-ai-boom/
Masanet, E., et al. (2020; updated projections cited 2025). Recalibrating global data center energy-use estimates. https://datacenters.lbl.gov/sites/default/files/Masanet_et_al_Science_2020.full_.pdf
Acemoglu & Johnson. (2023). Power and progress
Korinek & Stiglitz. (2025). Steering technological progress. INET https://www.ineteconomics.org/research/research-papers/steering-technological-progress
References
(continued)
NVIDIA. (2025, April 29). NVIDIA Blackwell platform boosts water efficiency by over 300x. https://blogs.nvidia.com/blog/blackwell-platform-water-efficiency-liquid-cooling-data-centers-ai-factories/
World Wide Technology. (2025, October 23). The overlooked water challenge in the age of AI. https://www.wwt.com/blog/the-overlooked-water-challenge-in-the-age-of-ai
AIPRM. (2024, July 11). AI replacing jobs statistics. https://www.aiprm.com/ai-replacing-jobs-statistics/
AIM Research. (2025). Top 20 predictions from experts on AI job loss. https://research.aimultiple.com/ai-job-loss/
Challenger, Gray & Christmas. (2025). Job cuts annual report 2025. https://www.challengergray.com/wp-content/uploads/2026/01/Challenger-Report-December-2025.pdf
Economic Innovation Group. (2025, August 14). AI and jobs: The final word (until the next one). https://eig.org/ai-and-jobs-the-final-word/
Goldman Sachs. (2025, August 13). How will AI affect the global workforce? https://www.goldmansachs.com/insights/articles/how-will-ai-affect-the-global-workforce
Information Technology and Innovation Foundation. (2026, January 17). AI job impact: Gains outpace losses. https://itif.org/publications/2025/12/18/ais-job-impact-gains-outpace-losses/
International Monetary Fund. (2025, April). Power Hungry: How AI Will Drive Energy Demand. https://www.imf.org/en/publications/wp/issues/2025/04/21/power-hungry-how-ai-will-drive-energy-demand-566304
McKinsey Global Institute. (2023). The economic potential of generative AI. https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier
National University. (2025, May 30). 59 AI job statistics: Future of U.S. jobs. https://www.nu.edu/blog/ai-job-statistics/
Ozkan, S., & Sullivan, N. (2025, August 26). Is AI contributing to rising unemployment? St. Louis Fed On the Economy. https://www.stlouisfed.org/on-the-economy/2025/aug/is-ai-contributing-unemployment-evidence-occupational-variation
World Economic Forum. (2025). Future of jobs report 2025. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
All citations formatted in APA 7th Edition style.
International Monetary Fund. (2025, April). The global impact of AI: Mind the gap (WP/25/76). https://www.imf.org/en/publications/wp/issues/2025/04/11/the-global-impact-of-ai-mind-the-gap-566129
Lawrence Berkeley National Lab. (2024). U.S. data center energy usage report. https://eta.lbl.gov/publications/2024-lbnl-data-center-energy-usage-report
Acemoglu. (2024). The simple macroeconomics of AI. NBER WP 32512 https://economics.mit.edu/sites/default/files/2024-04/The%20Simple%20Macroeconomics%20of%20AI.pdf
Acemoglu & Restrepo. (2019). Automation and new tasks. AEA Papers & Proceedings https://www.aeaweb.org/articles?id=10.1257/jep.33.2.3