mlsanityData to 7 Oct 2026

How Hungry is AI? Benchmarking Energy, Water, and Carbon Footprint of LLM Inference

arXiv 2505.09598 ↗First posted 14 May 2025cs.CY cs.AIcited by 73

Not on the map. Why?

Abstract (arXiv)

This paper introduces an infrastructure-aware benchmarking framework for quantifying the environmental footprint of LLM inference across 30 state-of-the-art models in commercial datacenters. The framework combines public API performance data with company-specific environmental multipliers and statistical inference of hardware configurations. We additionally utilize cross-efficiency Data Envelopment Analysis (DEA) to rank models by performance relative to environmental cost and provide a dynamically updated dashboard that visualizes model-level energy, water, and carbon metrics. Results show the most energy-intensive models exceed 29 Wh per long prompt, over 65 times the most efficient systems. Even a 0.42 Wh short query, when scaled to 700M queries/day, aggregates to annual electricity comparable to 35{,}000 U.S. homes, evaporative freshwater equal to the annual drinking needs of 1.2M people, and carbon emissions requiring a Chicago-sized forest to offset. These findings highlight a growing paradox: as AI becomes cheaper and faster, global adoption drives disproportionate resource consumption. Our methodology offers a standardized, empirically grounded basis for sustainability benchmarking and accountability in AI deployment.

Key facts. Cited by 73 papers in the map (data through 7 Oct 2026). Builds on 16 of its 74 references that are in the map. First posted on arXiv 14 May 2025.

Builds on

16 of 74 references.

References · 16

  1. 2024The Llama 3 Herd of Modelscited by 14,192
  2. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
  3. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
  4. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  5. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,748
  6. 2024MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmarkcited by 1,758
  7. 2021Carbon Emissions and Large Neural Network Trainingcited by 561
  8. 2023Splitwise: Efficient generative LLM inference using phase splittingcited by 536
  9. 2025Humanity's Last Examcited by 467
  10. 2023Power Hungry Processing: Watts Driving the Cost of AI Deployment?cited by 186
  11. 2024SciCode: A Research Coding Benchmark Curated by Scientistscited by 172
  12. 2023From Words to Watts: Benchmarking the Energy Costs of Large Language Model Inferencecited by 168
4 more
  1. 2024LLM-Inference-Bench: Inference Benchmarking of Large Language Models on AI Acceleratorscited by 31
  2. 2024The Price of Prompting: Profiling Energy Use in Large Language Models Inferencecited by 22
  3. 2024Double-Exponential Increases in Inference Energy: The Cost of the Race for Accuracycited by 6
  4. 2024Empirical Measurements of AI Training Power Demand on a GPU-Accelerated Nodecited by 4

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 24; 2026: 49.

73 papers.

Sort:

By area:

  1. 2025TokenPowerBench: Benchmarking the Power Consumption of LLM Inferencecited by 21
  2. 2025Energy Use of AI Inference, Efficiency Pathways, and Test-Time Scalingcited by 12
  3. 2025OckBench: Measuring the Efficiency of LLM Reasoningcited by 10
  4. 2026EET: Experience-Driven Early Termination for Cost-Efficient Software Engineering Agentscited by 6
  5. 2025Otters: An Energy-Efficient SpikingTransformer via Optical Time-to-First-Spike Encodingcited by 5
  6. 2026Do Chatbot LLMs Talk Too Much? The YapBench Benchmarkcited by 5
  7. 2025ECVL-ROUTER: Scenario-Aware Routing for Vision-Language Modelscited by 4
  8. 2025Edge Deployment of Small Language Models, a comprehensive comparison of CPU, GPU and NPU backendscited by 4
  9. 2026Prompt Injection Mitigation with Agentic AI, Nested Learning, and AI Sustainability via Semantic Cachingcited by 4
  10. 2025AI Benchmark Democratization and Carpentrycited by 3
  11. 2026This Is Taking Too Long -- Investigating Time as a Proxy for Energy Consumption of LLMscited by 3
  12. 2026Stable FP4 Training via Transposition-Invariant Block Quantizationcited by 3
38 more
  1. 2025Accurate and Energy Efficient: Local Retrieval-Augmented Generation Models Outperform Commercial Large Language Models in Medical Taskscited by 2
  2. 2025Walking the Tightrope of LLMs for Software Development: A Practitioners' Perspectivecited by 2
  3. 2026The Energy Footprint of LLM-Based Environmental Analysis: LLMs and Domain Productscited by 2
  4. 2026Llamas on the Web: Memory-Efficient, Performance-Portable, and Multi-Precision LLM Inference with WebGPUcited by 2
  5. 2025Phishing Detection in the Gen-AI Era: Quantized LLMs vs Classical Modelscited by 1
  6. 2025Learning the Topic, Not the Language: How LLMs Classify Online Immigration Discourse Across Languagescited by 1
  7. 2025Dissecting Transformers: A CLEAR Perspective towards Green AIcited by 1
  8. 2025Performance Trade-offs of Optimizing Small Language Models for E-Commercecited by 1
  9. 2025Analysing Environmental Efficiency in AI for X-Ray Diagnosiscited by 1
  10. 2025David vs. Goliath: Can Small Models Win Big with Agentic AI in Hardware Design?cited by 1
  11. 2026Scaling Quantum Machine Learning without Tricks: Full-Resolution and Diverse Image Generationcited by 1
  12. 2026Multilinguality at the Edge: Developing Language Models for the Global Southcited by 1
  13. 2026Position: the Stochastic Parrot in the Coal Mine. Model Collapse is a Threat to Low-Resource Communitiescited by 1
  14. 2026Energy per Successful Goal: Goal-Level Energy Accounting for Agentic AI Systemscited by 1
  15. 2026Attention to Detail: Evaluating Energy, Performance, and Accuracy Trade-offs Across vLLM Configurationscited by 1
  16. 2025Cat and Mouse -- Can Fake Text Generation Outpace Detector Systems?cited by 0
  17. 2025The Generative Energy Arena (GEA): Incorporating Energy Awareness in Large Language Model (LLM) Human Evaluationscited by 0
  18. 2025Modular, On-Site Solutions with Lightweight Anomaly Detection for Sustainable Nutrient Management in Agriculturecited by 0
  19. 2025Smaller Models, Smarter Rewards: A Two-Sided Approach to Process and Outcome Rewardscited by 0
  20. 2025Energy Scaling Laws for Diffusion Models: Quantifying Compute in Image Generationcited by 0
  21. 2025Text2Graph: Combining Lightweight LLMs and GNNs for Efficient Text Classification in Label-Scarce Scenarioscited by 0
  22. 2025Prefill vs. Decode Bottlenecks: SRAM-Frequency Tradeoffs and the Memory-Bandwidth Ceilingcited by 0
  23. 2025Energy-Aware Routing to Large Reasoning Modelscited by 0
  24. 2025Emissions and Performance Trade-off Between Small and Large Language Modelscited by 0
  25. 2026Balancing Sustainability And Performance: The Role Of Small-Scale LLMs In Agentic Artificial Intelligence Systemscited by 0
  26. 2026AI Cap-and-Trade: Efficiency Incentives for Accessibility and Sustainabilitycited by 0
  27. 2026Grandes Modelos de Linguagem Multimodais (MLLMs): Da Teoria `a Pr'aticacited by 0
  28. 2026Addressing Climate Action Misperceptions with Generative AIcited by 0
  29. 2026Robust Heterogeneous Analog-Digital Computing for Mixture-of-Experts Models with Theoretical Generalization Guaranteescited by 0
  30. 2026Energy Efficient Software Hardware CoDesign for Machine Learning: From TinyML to Large Language Modelscited by 0
  31. 2026Where Does AI Leave a Footprint? Children's Reasoning About AI's Environmental Costscited by 0
  32. 2026LLM-generated personalized nudges for improving pro-environmental behavior: Field evidence from resource conservationcited by 0
  33. 2026Toward a Sustainable Software Architecture Community: Evaluating ICSA's Environmental Impactcited by 0
  34. 2026Scrapyard AIcited by 0
  35. 2026Environmental Footprint of GenAI Research: Insights from the Moshi Foundation Modelcited by 0
  36. 2026LLM-Guided Runtime Parameter Optimization for Energy-Efficient Model Inferencecited by 0
  37. 2026LLMSpace: Carbon Footprint Modeling for Large Language Model Inference on LEO Satellitescited by 0
  38. 2026EnergyLens: Interpretable Closed-Form Energy Models for Multimodal LLM Inference Servingcited by 0

50 most cited of 73 All 73 (JSON)