Web Experts · AI Footprint Report · webexperts.com

The AI footprint meter

What is your AI usage costing the planet?

Every prompt draws real electricity, water, and carbon. Enter your usage below for an honest estimate, shown as a range, not a fake-precise number, because the true figures genuinely vary.

⚖️ These are estimates, deliberately shown as ranges. Published figures for AI's footprint vary by 10–100× depending on the model, data center, and power grid. We use peer-reviewed and first-party sources, cite every number, and show you the assumptions. See the methodology.

01 · Estimate

The footprint calculator

Don't know your exact tokens? That's normal, most chat apps hide them. Switch to an estimate mode below.

Units

What are you measuring?

Total tokens

tokens

A heavy Claude Code or API day can run into millions of tokens.

Which model?

Which chip is it running on?

Power grid

Carbon depends more on the grid than the model, the same query is ~15× dirtier on coal.

Runs entirely in your browser. Change an input and press Calculate again to update.

Enter your usage on the left, then press Calculate footprint to see your estimate.

Energy , Wh

range ,

Carbon (CO₂e) , g

range ,

Water , mL

Cooling only

,

Incl. power generation

,

Two scopes shown because sources measure different boundaries.

In everyday terms (central estimate)

Energy is like…

Where does the energy actually go?▸

Almost all of it turns into heat, almost immediately.

The chips do math and give off heat. Each token is billions of calculations on a GPU. Nearly 100% of the electricity a chip draws becomes heat within seconds, which is exactly why a data center needs so much cooling.
Overhead adds more. Cooling, power conversion, and networking mean the building pulls more than the chips alone, roughly 10 to 30% on top (the PUE factor).
It does not come back. Unlike water, spent electricity dissipates as low-grade heat into the environment; there is nothing to recover. The only levers are using less (a smaller model, an efficient chip) or drawing from cleaner power.

Carbon is like…

Where does the carbon come from?▸

Not from the data center itself, but from making the electricity it uses.

It is released at the power plant. The building emits little directly. The CO₂ comes from wherever the power is generated: a lot on coal or gas, close to nothing on hydro, nuclear, or solar.
It goes into the air and stays. CO₂ released today lingers in the atmosphere for centuries, trapping heat the whole time. That is why it is counted as a lasting footprint.
The grid decides almost everything. The same computation can be about 15× dirtier on a coal grid than a clean one. Some providers offset with renewable purchases, which is part of why reported numbers vary so much.

The "trees to absorb it" figure reflects that forests pull CO₂ back out slowly, over years.

Water is like…

Where does the water actually go?▸

It is not destroyed, but most of it leaves, which is why it counts as "consumed" rather than just used.

Most of it evaporates. Data centers usually cool with evaporative towers: water is trickled over hot surfaces and a portion turns to vapor, carrying the heat away. That vapor drifts off and falls as rain elsewhere, so from the local town's supply it is simply gone. In a dry region that often means treated drinking water going up into the sky.
Some is drained as wastewater. As water evaporates, minerals concentrate in what is left. To stop scaling, the plant periodically dumps that salty, chemically treated water (they add anti-scaling agents and biocides) to the sewer, where it needs treatment.
The hidden half is at the power plant. The electricity itself is often made by boiling water into steam and evaporating more water to cool it back down. That is the "including power generation" figure above, and it is usually the larger share.

So a little returns as dirty water and the rest evaporates into the atmosphere. It rejoins the global water cycle, but it is taken out of the local supply, which is what matters in a drought. Closed-loop and air-cooled designs use far less water, but usually trade that for more electricity.

02 · Look it up

How to find your usage in each app

The honest truth: only developer and API tools show real token counts. Consumer chat apps hide them, and image tools don't use tokens at all. Here's where each one stands.

App	Shows tokens?	What it reports	Where to look

📣

Notice how many apps hide this?

You have a right to know what your AI use actually costs. If a platform will not show you your token usage or its energy and water footprint, ask them to. The more people request it, the sooner transparency becomes the norm. Tell your providers that showing this should be standard, not optional.

03 · Show your work

Methodology & the numbers we use

Every constant below is editable in spirit, we picked mid-range, inference-only defaults and show the documented spread. Figures compiled from the cited sources on August 7, 2026; AI efficiency is improving quickly, so expect these to fall over time.

Energy & images

Constant	Low	Default	High
Text, per 1k tokens (by model)	0.1	0.6	34
Image, per image (Wh)	0.5	2.9	11.5
Data-center PUE	1.1	1.15	1.3

Water & carbon

Constant	Low	Default	High
Cooling water (mL/Wh)	0.9	1.1	1.8
Total water (mL/Wh)	33	60	167
Grid (g CO₂/kWh)	40	400	650

Everyday-comparison factors

Equivalent	Factor used	Equivalent	Factor used
Phone charge	11 Wh	Web search	0.2 g CO₂
LED bulb	10 W	Car (petrol)	170 g CO₂/km
Kettle boil	110 Wh	Mature tree	21 kg CO₂/yr
Average home	30 kWh/day	NYC–London flight	~1 t CO₂
Fridge	1.3 kWh/day	Beef	60 kg CO₂/kg
Water bottle	500 mL	10-min shower	62 L

The "trees" figure is the honest version of the forest metaphor: how many mature trees, working a full year, would absorb this much CO₂, not "forests destroyed," which isn't a real conversion.

Inference, not training. These are per-use figures. Training a model is a one-time cost amortized over billions of queries, we don't add it per prompt.

Model choice dominates energy. A reasoning model (o3, DeepSeek-R1) can use 30–100× a small model for the same visible answer, because of hidden "thinking" tokens.

Water is the most contested metric. The "cooling only" and "including power generation" numbers aren't contradictory, they draw the system boundary in different places. We show both.

The grid decides the carbon. The same query is ~15× dirtier on a coal grid than a nuclear/hydro one. Pick your region for a truer number.

The model matters more than the chip. Switching to efficient silicon (Groq, TPU, Cerebras) is a real but modest ~1.5–3× win; the "10–35×" vendor claims compare against badly-run GPUs. The bigger lever is architecture, an efficient open MoE model can be 5–15× lighter per token. Chip-level per-token energy is mostly unmeasured in production, so treat those numbers as rough.

The numbers are falling fast. Google reported a 33× energy drop per prompt in 12 months. The old "one query = 3 Wh / 10× a web search" meme is now considered ~10× too high. Constants dated Aug 2026.

Primary sources

Epoch AI (2025), How much energy does ChatGPT use?, epoch.ai
Jegham et al. (2025), How Hungry is AI? arXiv 2505.09598, arxiv.org
Google (2025), Environmental impact of AI at Google scale, arXiv 2508.15734, arxiv.org
Li, Ren et al. (2023/2025), Making AI Less Thirsty, arXiv 2304.03271 / CACM, arxiv.org
Luccioni, Jernite & Strubell (2023), Power Hungry Processing, arXiv 2311.16863, arxiv.org
Ember Global Electricity Review 2024, ember-energy.org
Our World in Data, carbon intensity of electricity, ourworldindata.org

04 · Our take

Why Web Experts is different

Web Experts advocates for open-source, low-power, and offline AI. They are far more environmentally friendly, and for most real tasks they are more than good enough. The biggest model is rarely the right one. Web Experts are Environmentally Conscious, Ethical AI Experts.

◍

Open source

Efficient open models (DeepSeek V4, Llama 4, Qwen3, Kimi K2) are transparent, ownable, and cheap to run, in dollars and in energy.

⚡

Low power

A right-sized model on efficient hardware can do the same job for a fraction of the energy, water, and carbon of a frontier model.

⬒

Offline first

Models that run on your own machine keep your data private, work without a connection, and draw only the power already in the room.

Thinking about AI for your office, app, or workflow?

We help teams put the right-sized AI in the right place, the efficient way. If that is something you want to explore, we would be glad to talk it through.

Get in touch