IEA global estimate
THE DIGITAL
FOOTPRINT.
The internet feels invisible. The infrastructure behind it isn’t. Explore the electricity, water, emissions and discarded hardware behind our connected world—from early data centers to the AI era.
IEA central projection
ITU / UNITAR global estimate
All reported Google operations
Every click has
a physical address.
Digital services depend on buildings full of computers, high-voltage electricity, cooling equipment, networks and continuously replaced hardware. A single interaction may be small; global demand changes the scale.
These are related but different indicators. Electricity, water consumption, emissions and e-waste cannot simply be added together as one measure of environmental damage.

The physical cloud
Server campuses, power lines, backup systems and cooling infrastructure.

Cooling the machines
Water and electricity are often part of the same trade-off.

After the upgrade
Devices and equipment eventually become electronic waste.
Technology through time.
Explore milestones and the data available for each era. Years without measured data are not filled with fabricated observations.
How fast is it growing?
Compare the historical global electricity estimates with IEA projections, then examine the reported evolution of Google’s own data centers.
Data-center electricity
Terawatt-hours per year · not AI alone
2035 = 1,200 TWh from the IEA’s 2025 baseline outlook; the IEA updated its 2025–2030 values in 2026.
Google data-center electricity
TWh · company-reported, 2021–2025
Global electronic waste
Million metric tonnes generated each year
Change the future.
Adjust how quickly computing grows, how much more efficient it becomes, and how it is powered and cooled. Watch the possible resource demands change in real time.
Your assumptions
Electricity starts at 485 TWh in 2025. Estimated emissions use an assumed fossil-powered intensity of 0.6 kg CO₂/kWh. Water includes modeled direct operational cooling only, not upstream electricity or chip manufacturing. All sliders represent hypothetical assumptions.
Continued expansion
Water is not
one number.
Google’s 2026 environmental disclosures let us distinguish what was withdrawn, returned, and consumed. That matters more than repeating one enormous number.

These are worldwide Google operations, not Search alone. Google’s data centers accounted for 10.523 billion gallons consumed in 2025, with the remainder from offices and other facilities. Source: Alphabet FY2025 environmental disclosures.
Follow the water
Billion US gallons per year
Local impacts differ
Million gallons consumed at each named location
Selected sites, not every facility. High total water use is not sufficient by itself to determine local environmental harm; watershed conditions and available alternatives matter.
Water replenishment ≠ water not consumed
Google reports replenishing about 7.717 billion gallons in 2025, equal to 78% of its separately defined 9.947 billion gallons of freshwater consumption. Replenishment uses estimated watershed benefits from projects; it does not undo every local withdrawal or make the original water use disappear.
What does one question cost?
Published figures for individual requests are much smaller than the infrastructure totals—but the figures use different definitions, providers, and years.
Estimated 0.20 g CO₂ per search. No reliable current per-search water figure.
About 0.32 mL water. No comparable independently documented carbon figure.
0.26 mL water and 0.03 g CO₂e, using Google’s specified operating methodology.
Multiply the requests
Exploratory arithmetic using fixed published per-request figures, not a measurement of your activity.
Field notes & photographs
Explore real facilities, energy and cooling equipment, materials and electronic waste. Documentary photos link to their Wikimedia Commons records; fallback artwork is clearly illustrative.
The questions behind the numbers
Short explainers on the biggest misconceptions around digital technology’s environmental impact.
Why does energy use rise if chips are more efficient?
Lower energy per task can make digital services cheaper and more useful. If the number and complexity of tasks grows even faster, total demand still rises. That does not negate efficiency—it changes what efficiency can accomplish.
Does cooling actually destroy water?
No. Some systems evaporate water, so it leaves the immediate watershed or supply rather than disappearing from Earth. The local consequences can still matter, especially where water is scarce.
Would renewable electricity solve everything?
Cleaner electricity can sharply reduce operating emissions, but the equipment still needs minerals, manufacturing, buildings, grid connections, cooling systems and eventual responsible disposal.
Is every AI prompt equal?
No. Short text, deep research, image creation, agentic workflows and video generation can have very different computing footprints. Claims about an “average query” rarely capture all possible workloads.
Does replenishing water cancel water consumption?
Not automatically. Projects may help restore streams, wetlands or water management in other places or times. The effectiveness and location of those benefits must be evaluated separately from the original consumption.
Can we call technology carbon neutral?
Only with a clear, verified accounting boundary and well-supported claims. Purchasing renewable electricity or donating to conservation does not prove all emissions or ecological effects have been eliminated.
Can the growth of computation be separated from the growth of environmental harm?
That is the central challenge: expand useful technology while reducing real-world electricity emissions, local water stress, material extraction and waste—not merely making individual requests look efficient.
Research library & methodology
Every primary statistic in the explorer is tied to a published report or clearly labeled as a simulation assumption. Open the original sources to check for updates.
How to read this atlas
Published data or historical estimates. Company disclosures may contain estimates even when subject to assurance.
Outlooks published by institutions such as the IEA and UN; not guarantees.
A transparent what-if calculation using assumptions you choose. Not an authoritative prediction.
Global data-center electricity, company-specific water and global electronic waste have different scopes. We never turn Google’s company-wide water total into a per-search figure.
Research cut-off: November 9, 2026. This static demonstration does not continuously refresh company or IEA data. Check the linked reports for revisions and new releases.
Photo rights: Each linked Wikimedia Commons image retains its original license and attribution. Locally bundled composites are illustrative images from this project's earlier visuals.