{
  "version": "2026-08-30.2",
  "reviewed_on": "2026-08-30",
  "license": "https://creativecommons.org/licenses/by/4.0/",
  "license_scope": "Inzonex compilation, classifications, unit conversions and original chart only. Upstream publications, site code, fonts and logos retain separate rights.",
  "title": "AI water consumption: what the numbers measure",
  "scope": "A boundary-first evidence table for selected public AI and data-centre water claims. It separates withdrawal, consumption, replenishment and avoided use; onsite, electricity-related and operational boundaries; and measured, estimated and modelled values.",
  "sources": [
    {
      "id": "google-gemini-2025",
      "publisher": "Google",
      "title": "Measuring the environmental impact of delivering AI at Google Scale",
      "published": "2025-08-21",
      "url": "https://services.google.com/fh/files/misc/measuring_the_environmental_impact_of_delivering_ai_at_google_scale.pdf",
      "locator": "Sections 3.3-3.4; Table 2, comprehensive approach; May 2025",
      "rights": "Google and author rights retained; original paper is not redistributed in downloads."
    },
    {
      "id": "li-2025",
      "publisher": "Communications of the ACM",
      "title": "Making AI Less 'Thirsty'",
      "published": "2025-06-17",
      "url": "https://doi.org/10.1145/3724499",
      "locator": "Estimating AI's Water Footprint; GPT-3 example; Water Withdrawal versus Water Consumption",
      "rights": "Underlying article rights remain with ACM and the authors. Values are quoted with attribution."
    },
    {
      "id": "microsoft-2025",
      "publisher": "Microsoft",
      "title": "2025 Environmental Sustainability Report announcement",
      "published": "2025-05-29",
      "url": "https://blogs.microsoft.com/on-the-issues/2025/05/29/environmental-sustainability-report/",
      "locator": "Our progress, Water positive",
      "rights": "Microsoft source rights retained."
    },
    {
      "id": "google-2025",
      "publisher": "Google",
      "title": "2025 Environmental Report",
      "published": "2025",
      "url": "https://www.sustainability.google/google-2025-environmental-report/",
      "locator": "2024 water replenishment highlight",
      "rights": "Google source rights retained."
    },
    {
      "id": "meta-2025",
      "publisher": "Meta",
      "title": "2025 Environmental Data Index",
      "published": "2025",
      "url": "https://sustainability.atmeta.com/asset/2025-esg-data-index/",
      "locator": "Sections 3.1 Water Withdrawal and PUE/WUE; 2024 data",
      "rights": "Meta source rights retained."
    }
  ],
  "rows": [
    {
      "id": "google-gemini-median-consumption",
      "claim": "Gemini Apps text prompt: reported median water consumption",
      "metric": "Consumption",
      "boundary": "Operational onsite cooling; comprehensive serving approach",
      "geography": "Google data centres supporting Gemini Apps",
      "data_year": "May 2025; WUE from 2024",
      "status": "Derived",
      "native_value": 0.26,
      "native_unit": "mL per median text prompt",
      "normalised_unit": "L per source-derived median text prompt",
      "litres_min": 0.00026,
      "litres_max": 0.00026,
      "source_id": "google-gemini-2025",
      "comparison_warning": "Derived from energy and prior-year consumptive WUE, not water metered per prompt. Excludes electricity-generation and embodied water. The statistic uses model-average energy weighted by prompt counts, not individual-prompt water measurements. Not a universal factor or a mean for estimating totals."
    },
    {
      "id": "gpt3-training-onsite-consumption",
      "claim": "GPT-3 training in Microsoft US data centres: onsite water consumption",
      "metric": "Consumption",
      "boundary": "Onsite cooling",
      "geography": "United States",
      "data_year": "Model example",
      "status": "Modelled",
      "native_value": 700000,
      "native_unit": "litres per training run",
      "normalised_unit": "L per modelled training run",
      "litres_min": 700000,
      "litres_max": 700000,
      "source_id": "li-2025",
      "comparison_warning": "Modelled GPT-3 example, not a measured OpenAI or Microsoft disclosure and not transferable to another model or site."
    },
    {
      "id": "gpt3-training-operational-consumption",
      "claim": "GPT-3 training example: onsite plus electricity-related water consumption",
      "metric": "Consumption",
      "boundary": "Operational onsite + electricity",
      "geography": "United States",
      "data_year": "Model example",
      "status": "Modelled",
      "native_value": 5400000,
      "native_unit": "litres per training run",
      "normalised_unit": "L per modelled training run (includes onsite)",
      "litres_min": 5400000,
      "litres_max": 5400000,
      "source_id": "li-2025",
      "comparison_warning": "Includes modelled offsite electricity water; do not compare directly with onsite-only WUE or corporate withdrawal."
    },
    {
      "id": "gpt3-medium-response-consumption",
      "claim": "Medium-length GPT-3 responses: modelled water consumption range",
      "metric": "Consumption",
      "boundary": "Operational onsite + electricity",
      "geography": "Location and time dependent",
      "data_year": "Model example",
      "status": "Modelled",
      "native_value": "0.5 L per 10-50 responses",
      "native_unit": "litres per 10-50 medium-length responses",
      "normalised_unit": "L per modelled medium-length response",
      "litres_min": 0.01,
      "litres_max": 0.05,
      "source_id": "li-2025",
      "comparison_warning": "Derived range for the paper's GPT-3 scenario, not a universal ChatGPT, image-generation or reasoning-model factor."
    },
    {
      "id": "global-ai-2027-withdrawal",
      "claim": "Global AI demand in 2027: projected water withdrawal",
      "metric": "Withdrawal",
      "boundary": "Operational onsite + electricity",
      "geography": "Global",
      "data_year": "2027 projection",
      "status": "Modelled",
      "native_value": "4.2-6.6 billion m3",
      "native_unit": "m3 per year",
      "normalised_unit": "L per year (2027 projection)",
      "litres_min": 4200000000000,
      "litres_max": 6600000000000,
      "source_id": "li-2025",
      "comparison_warning": "Projection of withdrawal, not consumption and not a measured 2027 total."
    },
    {
      "id": "microsoft-zero-water-design",
      "claim": "Microsoft zero-water cooling design: estimated water use avoided",
      "metric": "Avoided use",
      "boundary": "Onsite cooling",
      "geography": "Not specified",
      "data_year": "Design estimate",
      "status": "Estimated",
      "native_value": 125000,
      "native_unit": "m3 per facility-year",
      "normalised_unit": "L per facility-year (avoided-use estimate)",
      "litres_min": 125000000,
      "litres_max": 125000000,
      "source_id": "microsoft-2025",
      "comparison_warning": "Avoided-use design estimate, not observed withdrawal or consumption for Microsoft's fleet."
    },
    {
      "id": "google-2024-replenishment",
      "claim": "Google 2024 water replenishment",
      "metric": "Replenishment",
      "boundary": "Global operations",
      "geography": "Global",
      "data_year": "2024",
      "status": "Reported",
      "native_value": 4500000000,
      "native_unit": "gallons per year (gallon type not stated in highlight)",
      "normalised_unit": "L per year (assuming US gallons)",
      "litres_min": 17034353028,
      "litres_max": 17034353028,
      "source_id": "google-2025",
      "comparison_warning": "The source highlight reports a rounded 4.5 billion gallons without defining gallon type. The litre conversion assumes US gallons and is approximately 17 billion L, not additional source precision. Replenishment is not withdrawal or consumption and cannot automatically be subtracted."
    },
    {
      "id": "google-2024-replenishment-ratio",
      "claim": "Google 2024 freshwater-consumption replenishment ratio",
      "metric": "Replenishment ratio",
      "boundary": "Google-reported freshwater-consumption baseline",
      "geography": "Global",
      "data_year": "2024",
      "status": "Reported",
      "native_value": 64,
      "native_unit": "percent",
      "normalised_unit": "Not a standalone volume",
      "litres_min": null,
      "litres_max": null,
      "source_id": "google-2025",
      "comparison_warning": "A percentage, not a water volume and not an AI-specific metric."
    },
    {
      "id": "meta-2024-datacenter-withdrawal",
      "claim": "Meta data-centre water withdrawal",
      "metric": "Withdrawal",
      "boundary": "Data-centre operations",
      "geography": "Global reported facilities",
      "data_year": "2024",
      "status": "Reported / estimated mix",
      "native_value": 4145,
      "native_unit": "megalitres per year",
      "normalised_unit": "L per year (2024 operations)",
      "litres_min": 4145000000,
      "litres_max": 4145000000,
      "source_id": "meta-2025",
      "comparison_warning": "Corporate data-centre withdrawal, not AI-only consumption; incomplete activity data can use proxies under Meta's method."
    },
    {
      "id": "meta-2024-wue",
      "claim": "Meta annual data-centre water usage effectiveness",
      "metric": "Withdrawal intensity",
      "boundary": "Data-centre withdrawal / IT electricity",
      "geography": "Global data-centre fleet",
      "data_year": "2024",
      "status": "Reported / estimated mix",
      "native_value": 0.19,
      "native_unit": "litres per IT kWh",
      "normalised_unit": "Not a standalone volume",
      "litres_min": null,
      "litres_max": null,
      "source_id": "meta-2025",
      "comparison_warning": "Withdrawal intensity, not consumption per query and not a workload-level AI benchmark."
    },
    {
      "id": "meta-2024-construction-withdrawal",
      "claim": "Meta data-centre construction water excluded from operational withdrawal total",
      "metric": "Withdrawal",
      "boundary": "Data-centre construction",
      "geography": "Global reported projects",
      "data_year": "2024",
      "status": "Reported",
      "native_value": 1019,
      "native_unit": "megalitres per year",
      "normalised_unit": "L per year (2024 construction)",
      "litres_min": 1019000000,
      "litres_max": 1019000000,
      "source_id": "meta-2025",
      "comparison_warning": "Reported separately and excluded from Meta's 2024 operational withdrawal table."
    }
  ]
}