Skip to content

#2 of 30 · AI lab + cloud

Google (Gemini)

The most detailed per-prompt disclosure — attached to the fastest-growing footprint.

Mountain View, California·google.com ·Verified Oct 16, 2026
AHigh confidence
Index score
High confidence

In plain words

How scores work

Google scores 71 out of 100 — leader. Discloses real numbers and runs mostly on clean power. Still has a footprint, but you can see it. Confidence in this score is high confidence (A).

What helps

  • Efficiency · Median Gemini Apps text prompt measured at 0.24 Wh, 0.03 gCO₂e and 0.26 mL water using a comprehensive method that includes idle capacity and overhead (May 2025 data).
  • Efficiency · Energy per median prompt fell 33× and carbon 44× over the twelve months to May 2025.
  • Carbon · 100% annual renewable matching for nine consecutive years and a 24/7 carbon-free energy goal for 2030; 12 GW of clean-energy contracts signed in 2025.

What hurts

  • Efficiency · Figures are self-reported and not independently verified; critics note the water figure excludes water consumed to generate electricity.
  • Carbon · Total emissions rose 18% in 2025, the largest annual increase Google has reported; electricity use rose 37% to about 43 TWh.
  • Carbon · Total emissions were 51% above 2019 levels in 2024 (ambition-based boundary); Scope 3 is roughly four fifths of the footprint and rose 25% in 2025 on chip and server manufacturing.

What nobody knows

  • Training emissions in model cards
  • Third-party verified AI figures

Google is the only hyperscaler to publish a measured, fleet-wide per-prompt footprint for its assistant: 0.24 Wh, 0.03 gCO₂e and 0.26 mL of water for the median Gemini Apps text prompt. The same company used 43 TWh of electricity in 2025 (up 37% in a year), saw total emissions rise 18%, and consumed 10.9 billion gallons of freshwater. It owns about a quarter of the world's AI compute, mostly on its own TPUs.

Products
Gemini app · Gemini 2.5 / 3 models · Vertex AI · Google Cloud TPUs
Hosting
Own global data-centre fleet (TPU v5/v6/Ironwood); hourly carbon-free energy varies widely by region.

Key figures

Per-prompt and fleet metrics. Hover a title for what it means; open “How we got this number” for the working.

How to read the labels:Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.Measured independentlyMeasured independently. Measured or estimated by researchers or analysts with no stake in the result.ecogpt.co estimateecogpt.co estimate. Our own estimate from published factors. The working is shown next to the number; treat it as an order of magnitude, not a measurement.DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.Not disclosedNot disclosed. The company publishes nothing on this and no defensible estimate exists.— hover any label.
Energy per promptElectricity used to answer one typical request. Under 0.3 Wh is good. A 10 W LED bulb uses 0.3 Wh in about two minutes.
0.24
Wh / median text prompt
≈ 20.0 smartphone charges per 1,000 prompts
Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.Google (arXiv 2508.15734)

Comprehensive method; 0.10 Wh under the narrower active-chip method.

Water per promptWater evaporated cooling the servers for one request. Scope matters: on-site cooling only, or also the water used to generate the electricity (which adds ≈1.8 mL per Wh).
0.26
mL / median text prompt
≈ 0.52 500 mL bottles per 1,000 prompts
Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.Google (arXiv 2508.15734)

On-site cooling water only (WUE-derived).

Carbon per promptGreenhouse gas from one request on the grid where it actually runs. Below 0.1 g is clean. A kilometre in a petrol car is about 170 g.
0.030
gCO₂e / median text prompt
≈ 0.18 km driven (petrol car) per 1,000 prompts
Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.Google (arXiv 2508.15734)

Market-based; uses Google's contracted clean energy.

Annual emissionsThe company's whole yearly footprint, in million tonnes of CO₂e. Clouds and device makers dwarf pure labs because they also count shops, logistics or products. Compare the trend, not the size.
11.5
MtCO₂e (2024, ambition-based)
Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.Google

Rose a further 18% in 2025.

Emissions trendDirection of travel against last year or the company's own baseline. Negative is good. Positive means the AI build-out is outrunning clean energy.
+18.0
% YoY (2025)
Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.AxiosGoogle

+51% vs 2019 as of 2024.

Water per kWh (WUE)Litres of water consumed per kilowatt-hour of server electricity, across the fleet. 0.15 is excellent, 0.3 is good, 1.8 is the US data-centre average.
1.15
L/kWh (consumptive, 2024)
Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.Google (arXiv 2508.15734)

Category-2 WUE from the inference paper; the environmental report does not print a fleet figure.

Clean electricityShare of electricity matched with renewable or carbon-free sources. Annual matching counts certificates bought anywhere; hourly matching on the same grid is far stricter.
66.0
% hourly carbon-free energy (2024)
Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.GoogleAxios

Held roughly flat in 2025 despite a 37% rise in electricity use.

Hardware & materials

The chips, power, copper and future e-waste behind the fleet — the part sustainability reports leave out

Factors we use
AI chips in service
5.1 million
H100-equivalents (Epoch AI, Q4 2025)
≈ 70,833 racks of 72 GPUs
Measured independentlyMeasured independently. Measured or estimated by researchers or analysts with no stake in the result.Network WorldEpoch AI

GPUs, TPUs or custom accelerators the company runs. Analyst figures are in H100-equivalents, company figures in physical chips.

How we got this number

About 4 million of these are Google's own TPUs; roughly a quarter of all AI compute on Earth.

Power draw
4,909
MW (average draw)
≈ 3.7 million US homes' worth of power
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.GoogleAxios

Megawatts the AI fleet pulls. One megawatt runs roughly 750 US homes.

How we got this number

43 TWh a year ÷ 8,760 hours. Average draw implied by annual electricity; peak capacity is higher.

Electricity a year
43.0
TWh (2025)
≈ 4.1 million US homes for a year
Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.GoogleAxios

Terawatt-hours per year. One TWh is a year of electricity for about 95,000 US homes.

How we got this number

Up 37% in a year; roughly 3.5× 2019.

Carbon baked into the hardware
2,621
ktCO₂e embodied in chips and servers
≈ 569,870 petrol cars driven for a year
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.Network WorldEpoch AIarXiv 2607.05408NVIDIA developer blog

CO₂e emitted making the chips and servers before they answer a single prompt.

How we got this number

5,100,000 accelerators × 514 kg CO₂e each (chip plus share of the host server). Emitted in manufacturing, before the first prompt is served.

Future e-waste
86,700
tonnes of servers to retire within ~5 years
≈ 482 empty Boeing 747s by weight
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.Network WorldEpoch AINVIDIAThe Register

Tonnes of server hardware that will be retired in three to five years.

How we got this number

5,100,000 accelerators × 17 kg of server hardware each. Tomorrow's e-waste unless reused; generative AI could produce 2.5 Mt of it a year by 2030.

Copper in the build
132,543
tonnes of copper in the build
≈ copper for 1.6 million electric cars
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.GoogleAxiosBHPS&P Global Market Intelligence

Tonnes of copper in cabling, busbars and transformers.

How we got this number

4,909 MW × 27 t of copper per MW (cabling, busbars, transformers).

Fossil generators on site
1,888
backup diesel generators (implied)
≈ 4,909 MW of backup capacity
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.GoogleAxiosLatitude MediaLatitude Media / Floodlight

Diesel backup units or gas turbines at the data centres, permitted or observed.

How we got this number

4,909 MW ÷ 2.6 MW per generator, the average size permitted at US data centres. Permit counts are not published fleet-wide; this is the number a fleet this size normally carries.

Hardware reused or recycled
8.8 million
components harvested for reuse or resale (2024)
Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.Google

Share of retired servers and parts given a second life.

How we got this number

Count, not a share — Google does not publish a fleet-wide reuse rate.

Mineral depletion
—
Only Mistral has published a mineral-depletion figure.
Not disclosedNot disclosed. The company publishes nothing on this and no defensible estimate exists.

Kilograms of antimony-equivalent, the life-cycle unit for scarce minerals consumed.

Score breakdown

Each pillar is 0–100; weights are fixed across the index

Energy efficiency — How much electricity one answer costs.weight 20% · confidence A
90
  • Median Gemini Apps text prompt measured at 0.24 Wh, 0.03 gCO₂e and 0.26 mL water using a comprehensive method that includes idle capacity and overhead (May 2025 data). Google (arXiv 2508.15734)Google Cloud blog
  • Energy per median prompt fell 33× and carbon 44× over the twelve months to May 2025. Google (arXiv 2508.15734)
  • Figures are self-reported and not independently verified; critics note the water figure excludes water consumed to generate electricity. The Verge
Carbon intensity — How clean that electricity is, and whether total emissions are rising.weight 25% · confidence B
60
  • Total emissions rose 18% in 2025, the largest annual increase Google has reported; electricity use rose 37% to about 43 TWh. AxiosGoogle
  • Total emissions were 51% above 2019 levels in 2024 (ambition-based boundary); Scope 3 is roughly four fifths of the footprint and rose 25% in 2025 on chip and server manufacturing. GoogleTrellis
  • 100% annual renewable matching for nine consecutive years and a 24/7 carbon-free energy goal for 2030; 12 GW of clean-energy contracts signed in 2025. Axios
Water — How much water the data centres consume, and whether it is replaced.weight 15% · confidence B
55
  • Freshwater consumption rose 34% to 10.9 billion gallons in 2025, more than double 2021. Axios
  • Fleet WUE of 1.15 L/kWh (consumptive, 2023–2024) — several times AWS and Microsoft, because many Google sites use evaporative cooling. Google (arXiv 2508.15734)
  • Replenishment reached 64% of freshwater consumption in 2024 and roughly 78% in 2025, toward a 120% target for 2030. GoogleAxios
  • In The Dalles, Oregon, Google's data centres used 355 million gallons in 2021 — about 29% of the city's water — figures the city fought for 13 months to keep secret. Data Center DynamicsReporters Committee for Freedom of the Press
Transparency — How much the company actually publishes.weight 20% · confidence A
82
  • Per-prompt energy, carbon and water published with methodology; annual environmental report with location- and market-based Scope 2, fleet WUE and electricity. Google (arXiv 2508.15734)Google
  • Reports 8.8 million components harvested from retired servers for reuse or resale in 2024. Google
  • No training-emissions disclosure for Gemini models and no third-party verification of the inference study. The Verge
Commitments — Whether it has dated, measurable climate and water targets.weight 10% · confidence A
80
  • Net-zero across operations and value chain by 2030, 24/7 carbon-free energy on every grid by 2030, 120% water replenishment by 2030. Google
Conduct — How it behaves: permits, lawsuits, gas turbines, community water.weight 10% · confidence B
55
  • Funded The Dalles' legal fight to withhold water records from The Oregonian before reversing course and paying the newspaper's costs. Reporters Committee for Freedom of the Press
  • Now backing public water-reporting standards for data centres amid local backlash. Axios

Models

What this company ships, and what each prompt costs

ModelWeightsEnergy / promptWaterTraining CO₂e
Gemini (Apps, median text prompt)
Frontier (undisclosed)
Closed0.24 Wh DisclosedCompany figure. Published by the company itself. We take it as given and record the scope next to it.0.260 mL DisclosedCompany figure. Published by the company itself. We take it as given and record the scope next to it.9,000 t Estimateecogpt.co estimate. Our own estimate from published factors. The working is shown next to the number; treat it as an order of magnitude, not a measurement.

Where the compute lives

Named sites from company disclosures and press reporting

OperationalConstructionPlanned
  • The DallesOregon, USA · ~29% of the city's water in 2021.
  • Council BluffsIowa, USA
  • MesaArizona, USA · 86% clean electricity per Google.
  • HaminaFinland · Seawater cooling; among the highest CFE regions.
  • St. GhislainBelgium
  • Changhua CountyTaiwan