mint treesresearch logby bdrop.studio
open notebook · by bdrop.studio · updated Sep 2026
written and designed with AI, editorially reviewed

What does AI actually cost the atmosphere?

Almost everyone answering that question is working from a feeling — that AI is ruinous for the planet, or that it is a rounding error. Both camps cite real numbers; the numbers are between one and two orders of magnitude apart. mint trees is an attempt to get past the feeling: measure one real AI product end to end, in the open, with the method and the error bars on the table.

No product. No credits for sale.
A measurement problem, worked in public.
bdrop.studio is the first case, not the point.
Method and factors reusable under CC BY.

415 TWh
data-centre electricity, 2024
≈ 1.5 % of world demand 1
+12 % / yr
growth since 2017 — four times
faster than electricity overall 1
0.03 g
CO₂e per median Gemini text
prompt, market-based 2
4–20×
by which widely cited per-query
energy figures are overstated 5
01 — the boundary problem

A footprint is mostly an argument about where you draw the line.

Two teams can measure the same web app and land a factor of ten apart — not because one lied, but because one counted only the server and the other counted the hardware, the network, the idle capacity and the laptops. Here is the line we draw, and what sits on each side of it.

usually counted — the drop

Runtime compute

Server electricity while requests are served
Reported as market-based scope 2

Often ~0 g on paper, thanks to PPAs
also real — the bucket

Everything around it

Idle and redundant capacity, 24/7
CI runs, preview deploys, build caches

Coding-agent inference during development

Database and object storage at rest

Network transport and the user's device

Cooling overhead (PUE) and water
rarely counted — the well

Embodied hardware

Manufacturing of servers and accelerators
Our own laptops, phones, displays

Model training, amortised over inference

Building shell, generators, grid connection
Our working rule: if we cannot measure it, we estimate it with a stated range and mark it low confidence — we never drop it to zero. A zero is a claim, not a blank.
02 — the equation everyone should know

Software carbon intensity, in one line.

SCI=(E×I)+MperR
Green Software Foundation, ISO/IEC 21031:2024 3
E
energy · kWh

What the software actually draws, including the data centre's overhead. Measurable for our own machines, modelled for managed services.

I
intensity · gCO₂e / kWh

How dirty that electricity was — which grid, which hour. The only term you can change without touching a line of code.

M
embodied · gCO₂e

The hardware's manufacturing footprint, amortised across its useful life and the share of it you occupied.

R
functional unit

Per what? Per request, per user-month, per deploy, per shipped feature. Change R and the headline number changes with it.

SCI is a rate, not a total — it goes down when your software gets better, not when you buy credits. Offsets are deliberately excluded from it.
03 — the same workload, six grids

Where you run it matters more than almost anything you write.

Pick a monthly electricity figure for a workload and watch the same kilowatt-hours land fifteen times apart, purely by region. Annual average grid intensities, 2024. 4

workload energy per month
730
kWh
5 kWhone rack server at 1 kW, 730 h2,000 kWh
In the cleanest region shown, this workload emits 30 kg CO₂e a month. In the dirtiest, 453 kg. Same code, same users — a region setting.
Sweden30 kg · 41 g/kWh
France41 kg · 56 g/kWh
EU-27 average177 kg · 242 g/kWh
Germany265 kg · 363 g/kWh
United States269 kg · 369 g/kWh
Poland453 kg · 620 g/kWh
Location-based method: the physical grid mix, not the certificates bought against it.

Why there are no car-kilometres or tree-years on this page.

Equivalences convert a figure you have measured into one you have not. A “tree-year” depends on species, age, climate and what you assume about the tree’s next fifty years; a car-kilometre depends on a fleet average that differs by a factor of two between test cycle and road. Both add uncertainty while appearing to add clarity. If a number needs an analogy to be persuasive, the honest move is to explain the number.

04 — field notes from our own stack

What we can measure today, and what we still can’t.

bdrop.studio runs on managed infrastructure, like most small teams. That means most of our footprint sits inside someone else’s reporting boundary. Here is the honest state of each line.

This is an instrumentation register, not a result. We publish no figures for our own stack yet — only what each line would take to measure and how far we are from being able to. Numbers appear here when they come from telemetry we can show.

development

AI coding agents

Inference during development: long contexts, tool calls, repeated runs. Per-prompt medians published by providers are small, but agentic sessions are orders of magnitude above a median chat prompt.

Missing: per-session token and energy telemetry from providers. We estimate from token counts and published per-prompt figures.

low confidence
ci / build

Pipelines & preview deploys

Build minutes are billable, therefore countable. Runner class, region and duration give a defensible energy estimate — and preview deploys on every push add up faster than production.

Missing: runner hardware spec and regional placement. Embodied share of the runner fleet is modelled, not measured.

medium confidence
hosting / edge

Vercel

Serverless and edge invocations, bandwidth, image optimisation. Duration × memory class is a usable proxy for energy once a regional grid factor is applied.

Missing: idle and redundancy overhead, and a location-based figure — provider reporting is market-based.

medium confidence
cloud

Google Cloud

The one provider giving us region-level carbon data per service, including a location-based view. Our best-instrumented line by a wide margin.

Missing: embodied hardware attribution per workload, and hourly rather than annual grid resolution.

higher confidence
data

Supabase / Postgres

A database is the quiet constant: it draws power at 3 a.m. with zero users. Instance size × hours is the honest baseline, not query counts.

Missing: underlying host utilisation and multi-tenant allocation. Storage at rest is estimated from disk size.

low confidence
people

Devices & the studio

Laptops, displays, phones — dominated by manufacturing, not use. Amortised over a realistic replacement cycle, this rivals everything above it for a small team.

Missing: verified per-device LCA data beyond manufacturer product sheets.

medium confidence
05 — how to read any carbon number

Five questions that expose a footprint claim in under a minute.

Use them on us too. If an answer here is missing, that is a finding, and it belongs in the log.

Per request, per user, per month, per shipped feature — a number without a denominator cannot be compared to anything. Watch for totals quoted where a rate belongs, and rates quoted where a total belongs.

06 — sources & method
[1]

IEA, Energy and AI (2025): data centres ≈ 415 TWh in 2024, ~1.5 % of global electricity, +12 %/yr, projected ~945 TWh by 2030.

iea.org/reports/energy-and-ai
[2]

Google (2025), Measuring the environmental impact of delivering AI at Google scale: median text prompt 0.24 Wh, 0.03 gCO₂e, 0.26 mL water — market-based, excludes training and network.

arxiv.org/abs/2508.15734
[3]

Green Software Foundation, Software Carbon Intensity specification, standardised as ISO/IEC 21031:2024.

sci.greensoftware.foundation
[4]

Grid intensities: Ember Electricity Review and national TSO data, annual averages 2024. Indicative — hourly values swing by a factor of three within a single day.

ember-energy.org
[5]

Wu et al. (2026), Energy use of AI inference, efficiency pathways, and test-time scaling, Joule: frontier models on H100 nodes, median 0.31 Wh/query (IQR 0.16–0.60) — widely cited estimates overstated by 4–20×.

cell.com/joule
[6]

Epoch AI (2025), How much energy does ChatGPT use? — model-based estimate of ~0.3 Wh for a typical prompt, with the calculation shown.

epoch.ai

Figures on this page are order-of-magnitude estimates for public education, not a verified inventory. Where we state a number from our own stack it is labelled with its confidence. Corrections are welcome and will be published with attribution.

07 — why this exists

One product measured properly beats a thousand opinions.

The public argument about AI and the climate is running far ahead of the evidence, in both directions. What is missing is not more opinion — it is a small number of workloads measured end to end by people willing to publish their assumptions and be wrong in public. We can supply exactly one of those. It happens to be ours.

bdrop.studio is the specimen

A small generative-AI product — text, image and audio models, managed infrastructure, a tiny team. Unremarkable, which is the useful part: most AI products look like this, and almost none of them are measured.

what should travel

The boundary definition, the factor registry with its tiers and sources, the uncertainty propagation, the error budget that names which unknown is currently costing you the most. Those are portable. Our kilowatt-hours are not, and nobody should reuse them.

what would change our mind

A measured figure that lands outside our published band, or a provider disclosure that collapses one of our wide terms. Either would be a good day. Revisions get published with the reasoning — starting with our own first guess, which was more than ten times too high and is on the record as such.

This log takes no position on whether AI is good or bad for the climate. It is trying to establish what one unit of it costs, to a precision that would let someone else have that argument with evidence instead of adjectives.

The next entry: one sprint, fully instrumented.

Two weeks of a real bdrop.studio project — every agent call, CI minute, preview deploy and idle database hour, measured and published with the raw data alongside.

read what a tonne costsread the method