Absolute footprint, every quarter
Scope 1, 2 and 3 published beside the per-product figures, against a declared path.
Impact deck · For Wind Capital · October 2026Software, made in machine-minutes instead of person-days.
OhMyUnicorn turns one sentence into a working app or game in about 40 minutes, on a few shared servers. We measured what that does to the footprint of making software, and we show the case least favourable to us first.
Every app has an energy meterUnder an hour instead of weeks of workOpen method, independent review funded by the round
What is measured: products made by outside users on OhMyUnicorn (apps) and OhMyGames (games), the two public front doors of our factory. The factory itself, the agents, rules and servers that make them, is the company; its own development is counted inside every product as the research share.
One app, one meter readingmodelled
1.1kWh
AI computing to build a typical app, counted from its own logged calls. A game takes about 2 kWh.
01 · Measured
Every app has a meter.
Each product we build logs every AI call: what was read, what was written, by which model. So the computing behind one app is counted per unit; its energy is then estimated from published coefficients, with the range shown. A typical app takes about 1.1 kWh of AI computing to build, about one dishwasher cycle; a game about 2 kWh. No studio or agency can show that for a single product.
Calls, logged
Each product logs every AI call: what was read, what was written, by which model. Failed attempts and the first two weeks of edits are counted on the product.
Tokens read and written (measured)
Fresh input, cached input, cache writes and output, by model. This part is counted, not estimated.
kWh (estimated)
Energy per token from the EcoLogits family, hardware made included, cross-checked against Google's and Mistral's published figures. The range is shown.
1.1kWh app2kWh gamekg CO2e (estimated)
One extra booking app: its AI with failed attempts, edits, hardware and a training share, plus the customer's screen time; before hosting and our research share (modelled median).
0.71kg CO2eA real customer build, call by call
tokens measured, energy estimatedcentral coefficients
Booking site with deposits and admin for a tattoo studio (real customer build, 44 min sentence to live)
Tokens read and written10.7Mmeasured
→Energy0.85 kWhestimated
→Footprint0.49 kg CO2eestimated
| Model | Tokens read | Written to cache | Tokens written (output) | Energy (estimated) |
|---|---|---|---|---|
| Claude Sonnet 5.5 | 9.3M | 413k | 206k | 0.5 kWh |
| Claude Opus 5.5 | 485k | 167k | 54k | 0.35 kWh |
| This product | 9.8M | 579k | 260k | 0.85 kWh |
Tokens measured (build window); energy estimated, central coefficients; plan stage not split.
02 · Machine-minutes
Under an hour of machine time, instead of weeks of work.
The conventional way, in hands-on person-days (low to high): a simple website 3 to 15; a booking or business app like those our users make 6 to 60; a web game like those our users publish 12 to 65; a studio-grade casual game with progression 120 to 880. We deliver each in about 40 minutes from sentence to live (median: apps 39, games 40).
Hands-on person-days, conventional route
task-based rangelow end = AI-assisted professional or site builder
central effortlow to high of the task-based rangeOhMyUnicorn: about 40 minutes
Simple small-business website (brochure)
7 person-daysrange 3 to 15; with us about 40 min
Booking or business web app like those our users publish
25 person-daysrange 6 to 60; with us about 40 min
Hyper-casual web game like those our users publish, small team
25 person-daysrange 12 to 65; with us about 40 min
Same web game, one developer at home
25 person-daysrange 12 to 65; with us about 40 min
Studio-grade casual game with progression (upper scenario)
400 person-daysrange 120 to 880; with us about 40 min
Table view
Conventional hands-on effort, task-based range
| Class | Low (person-days) | Central | High |
|---|---|---|---|
| Simple small-business website (brochure) | 3 | 7 | 15 |
| Booking or business web app like those our users publish | 6 | 25 | 60 |
| Hyper-casual web game like those our users publish, small team | 12 | 25 | 65 |
| Same web game, one developer at home | 12 | 25 | 65 |
| Studio-grade casual game with progression (upper scenario) | 120 | 400 | 880 |
The violet tick is OhMyUnicorn's 40 minutes, drawn at a width you can see; to scale it would be far thinner than a pixel. This is a physical ratio from logs and effort ranges; no carbon model is needed for it.
Equivalence by class and written checklist, checked on 4 real customer products: see Method.
03 · Where the footprint is
Making software has a footprint. Most of it is not the code.
Every day of work brings computers and screens (made mostly in Asia, replaced every four to six years), an office to light and heat, a trip to get there, business travel. Thirteen company reports, from Ubisoft and Unity to Capgemini, put it at a median of 6.5 kg CO2e per employee per working day, before counting their computers. We count it two ways: equipment and energy only (strict), and the whole project including trips (wide). The headline numbers in this deck use the strict count.
kg CO2e per person-day of conventional work
modelledParistypical weekcentral values
Strict count: equipment and energyWide count only: trips, commuting, heating, home
Commute to the officewide only
1.4 kg
Workstation and screens, made in Asiastrict
0.64 kg
Office heatingwide only
0.34 kg
Home office energywide only
0.34 kg
Cloud and software toolsstrict
0.2 kg
Business travelwide only
0.2 kg
Office electricitystrict
0.093 kg
Devices, electricity usedstrict
0.016 kg
Table view
kg CO2e per person-day (modelled, central values)
| Component | Boundary | kg CO2e per person-day |
|---|---|---|
| Commute to the office | project | 1.4 |
| Workstation and screens, made in Asia | strict | 0.64 |
| Office heating | project | 0.34 |
| Home office energy | project | 0.34 |
| Cloud and software tools | strict | 0.2 |
| Business travel | project | 0.2 |
| Office electricity | strict | 0.093 |
| Devices, electricity used | strict | 0.016 |
3.2kgper person-day, all components (modelled, central values)
50%of it is commuting and business travel
0.95kgfalls inside the strict count (equipment and energy)
Cross-checks sit beside this bottom-up figure and are never averaged with it: 13 company reports and a spend-based estimate. We frame equipment, buildings and trips; never people.
04 · One game, two ways
A studio-grade game: about 1,100 kg CO2e to make. With us: about 7.2 kg.
About 150× less (modelled median; range 70× to 310×), 29× in our least favourable scenario. Upper scenario: a casual game with progression made by a Paris studio, hybrid week, equipment and energy only, the studio's unbilled time included; on our side the build, failed attempts, edits, hardware, model training and half of our research computing (the share we attribute to production; the whole fleet is a stress case).
For the web games our users publish today (about 25 person-days of conventional work), the gap is about 11× (82 kg vs 7.3 kg). Each extra game costs us about 1.2 kg; the rest is research shared across products.
Studio: 1,100 kg, edge 8.4 mOhMyUnicorn: 7.2 kg, edge 1.6 m
Each cube is the CO2 of making one studio-grade game, at street scale: 8.4 m and 1.6 m on a side (upper scenario, modelled).CO2 gas at 15 °C is 1.9 kg per cubic metre.
modelledupper scenarioequipment and energy onlyParismedian of 20,000 draws
Modelled, internal study, not yet critically reviewed. Medians of paired draws unless marked central values. Method and sources in the sheets.
05 · Size matters
The bigger the software, the bigger the saving.
For a simple website, a freelancer's week on a laptop is light and our own research computing is not zero: today the gap is small, about 1.8× (13 kg vs 6.7 kg), and we say so. A booking or business app: 6.7× (45 vs 6.7 kg). A web game: 11× (82 vs 7.3 kg); made by one developer at home: 11×. A studio-grade game: 150×. For small websites our impact is access, not carbon.
Ratio conventional / OhMyUnicorn, by product class
modelledParisequipment and energy onlylog scale
Strict count, typical (median); whisker = 10th to 90th percentileWide count (whole project), median
Simple small-business website (brochure)7 person-days (3 to 15)
1.8× (0.87 to 3.8)13 kg vs 6.7 kg
Booking or business web app like those our users publish25 person-days (6 to 60)
6.7× (2.8 to 14)45 kg vs 6.7 kg
Hyper-casual web game like those our users publish, small team25 person-days (12 to 65)
11× (5.3 to 23)82 kg vs 7.3 kg
Same web game, one developer at home25 person-days (12 to 65)
11× (4.9 to 22)77 kg vs 7.3 kg
Studio-grade casual game with progression (upper scenario)400 person-days (120 to 880)
150× (70 to 310)1,100 kg vs 7.2 kg
1×10×100×1,000×no benefit below 1×
Table view
Ratio conventional / OhMyUnicorn, modelled, Paris, typical work pattern
| Class | Strict p10 | Strict p50 | Strict p90 | Whole project p50 | Conventional kg (strict, whole company) | OhMyUnicorn kg | Modelled chance of no benefit |
|---|---|---|---|---|---|---|---|
| Simple small-business website (brochure) | 0.87 | 1.8 | 3.8 | 7.3 | 13 | 6.7 | 14.8% |
| Booking or business web app like those our users publish | 2.8 | 6.7 | 14 | 24 | 45 | 6.7 | 0.2% |
| Hyper-casual web game like those our users publish, small team | 5.3 | 11 | 23 | 34 | 82 | 7.3 | 0% |
| Same web game, one developer at home | 4.9 | 11 | 22 | 17 | 77 | 7.3 | 0% |
| Studio-grade casual game with progression (upper scenario) | 70 | 150 | 310 | 450 | 1,100 | 7.2 | 0% |
The modelled chance of no benefit for a simple website is 14.8% of draws. Absolute kg are on each row: the conventional route (whole company) versus ours.
Modelled, internal study, not yet critically reviewed. Medians of paired draws unless marked central values. Method and sources in the sheets.
06 · Sixteen hubs
From Paris to Bangalore: a studio-grade game is 150× to 490× heavier to make there than with us.
US hubs drive: 7 to 8 kg of CO2e per office day for the commute alone. Paris, Berlin and Tokyo ride trains, but offices on carbon-heavy grids (Warsaw, Bangalore, Tokyo) make up for it. The leanest studio we could model works from home in Montreal, on hydro power.
Footprint of one studio-grade game, by hub
modelledstrict counttypical work patternmedian
conventional studio, equipment and energy only, whole company, kg CO2e per productOhMyUnicorn: 7.2 kg (violet hairline, same scale)
Paris
1,100 kg150× vs OhMyUnicorn
Lyon
1,100 kg160× vs OhMyUnicorn
Montpellier
1,100 kg150× vs OhMyUnicorn
Stockholm
1,100 kg150× vs OhMyUnicorn
Montreal
1,100 kg150× vs OhMyUnicorn
Helsinki
1,200 kg170× vs OhMyUnicorn
London
1,800 kg250× vs OhMyUnicorn
San Francisco Bay Area
2,200 kg300× vs OhMyUnicorn
Los Angeles
2,200 kg300× vs OhMyUnicorn
Berlin
2,300 kg310× vs OhMyUnicorn
Seattle
2,500 kg340× vs OhMyUnicorn
Seoul
2,600 kg350× vs OhMyUnicorn
Austin
2,700 kg360× vs OhMyUnicorn
Tokyo
2,800 kg380× vs OhMyUnicorn
Warsaw
3,300 kg450× vs OhMyUnicorn
Bangalore
3,600 kg490× vs OhMyUnicorn
Table view
Studio-grade casual game with progression (upper scenario): footprint by hub (modelled, typical work pattern)
| Hub | Conventional kg (strict, whole company) | Ratio (strict, median) | Conventional kg (whole project) | Ratio (whole project, median) |
|---|---|---|---|---|
| Paris | 1,100 | 150 | 3,200 | 450 |
| Lyon | 1,100 | 160 | 3,600 | 500 |
| Montpellier | 1,100 | 150 | 3,800 | 530 |
| Stockholm | 1,100 | 150 | 3,100 | 430 |
| Montreal | 1,100 | 150 | 4,000 | 550 |
| Helsinki | 1,200 | 170 | 3,000 | 410 |
| London | 1,800 | 250 | 4,200 | 580 |
| San Francisco Bay Area | 2,200 | 300 | 6,300 | 870 |
| Los Angeles | 2,200 | 300 | 6,500 | 900 |
| Berlin | 2,300 | 310 | 4,400 | 610 |
| Seattle | 2,500 | 340 | 6,500 | 900 |
| Seoul | 2,600 | 350 | 5,300 | 730 |
| Austin | 2,700 | 360 | 6,800 | 940 |
| Tokyo | 2,800 | 380 | 4,800 | 660 |
| Warsaw | 3,300 | 450 | 6,200 | 860 |
| Bangalore | 3,600 | 490 | 6,200 | 860 |
The violet hairline at the left edge is OhMyUnicorn's 7.2 kg on the same scale. Every hub is built bottom-up (commute, office, home, devices, tools, trips); sources are graded in sheet S2.
Modelled, internal study, not yet critically reviewed. Medians of paired draws unless marked central values. Method and sources in the sheets.
07 · Our side, all of it
We count what flatters us least.
Our footprint per product includes the AI that builds it, failed attempts, edits, the hardware, an estimate for model training, the user's own screen time, and half of our research computing spread over the products we ship (the whole fleet is a stress case on slide 8). That research is the biggest part today: our own R&D spends about nine times what customer builds spend on AI, about five times the electricity. Each extra product costs about 0.71 kg for an app and 1.2 kg for a game; research adds the rest, and it shrinks as volume grows.
If research stayed at today's level while volume reached 2,000 products a month (central values): a web game about 54× lighter to make, a studio-grade game about 860×, even a simple website about 13×.
Parts of one studio-grade game, both routes
modelledcentral valuesParis
Conventional studio, whole project1,856 kg, central values
- Devices (made, then used)503 kg
- Commuting550 kg
- Cloud and software tools80 kg
- Office (electricity strict, heating project)177 kg
- Home office130 kg
- Trips and client meetings416 kg
OhMyUnicorn, same scale3.6 kg: the violet sliver
OhMyUnicorn, magnified 510×the same 3.6 kg, all parts
- Allocated R&D (our own AI fleet, by session purpose)2.9 kg
- AI compute for the build (failures and edits included)0.47 kg
- Share of model training0.21 kg
- The customer's own device time0.031 kg
- Hardware the AI runs on (made)0.028 kg
12 months in service (hosting), modelled from measured watts
Conventional host20 kg
OhMyUnicorn2.6 kg
Table view
Parts of one casual game, central values, kg CO2e
| Route | Part | kg CO2e |
|---|---|---|
| Conventional | Devices (made, then used) | 503 |
| Conventional | Commuting | 550 |
| Conventional | Cloud and software tools | 80 |
| Conventional | Office (electricity strict, heating project) | 177 |
| Conventional | Home office | 130 |
| Conventional | Trips and client meetings | 416 |
| OhMyUnicorn | Allocated R&D (our own AI fleet, by session purpose) | 2.9 |
| OhMyUnicorn | AI compute for the build (failures and edits included) | 0.47 |
| OhMyUnicorn | Share of model training | 0.21 |
| OhMyUnicorn | The customer's own device time | 0.031 |
| OhMyUnicorn | Hardware the AI runs on (made) | 0.028 |
The conventional bar is the whole project (wide count); 620 kg of it falls inside the strict count. Allocated R&D is 80% of our production footprint today. The headline kg elsewhere in the deck are medians of 20,000 draws; the parts here are central values, so they differ a little.
The more products a month, the lighter each one
modelledcentral valuesstrict countresearch effort spread over more products
Simple small-business website (brochure)Booking or business web app like those our users publishHyper-casual web game like those our users publish, small teamStudio-grade casual game with progression (upper scenario)1× (no benefit)
1×10×100×1,000×
Studio-grade game 1,100×Web game 69×Booking app 68×Website 19×
2005001,0002,0005,00010,000
Products made a month (log scale) → ratio vs a conventional build (log scale)
Table view
Ratio conventional / OhMyUnicorn as volume grows (strict count, Paris, modelled)
| Products a month | Website: ratio (kg per product) | Booking app: ratio (kg per product) | Web game: ratio (kg per product) | Studio-grade game: ratio (kg per product) |
|---|---|---|---|---|
| 200 | 2.9x (3.3 kg) | 10x (3.3 kg) | 15x (3.6 kg) | 250x (3.6 kg) |
| 500 | 6x (1.6 kg) | 21x (1.6 kg) | 29x (1.9 kg) | 470x (1.9 kg) |
| 1,000 | 9.4x (1 kg) | 34x (1 kg) | 42x (1.3 kg) | 680x (1.3 kg) |
| 2,000 | 13x (0.72 kg) | 47x (0.72 kg) | 54x (1 kg) | 860x (1 kg) |
| 5,000 | 17x (0.55 kg) | 61x (0.55 kg) | 65x (0.86 kg) | 1,000x (0.86 kg) |
| 10,000 | 19x (0.49 kg) | 68x (0.49 kg) | 69x (0.8 kg) | 1,100x (0.8 kg) |
Modelled, internal study, not yet critically reviewed. Medians of paired draws unless marked central values. Method and sources in the sheets.
08 · The case against us
Our least favourable modelled scenario.
We built the case against us, Scenario F: the leanest studio we could model (fully remote, in Montreal, on hydro power), AI energy at the top of published ranges, everything we attribute counted, whole company on both sides. In that scenario the 10th percentile is 29× for a studio-grade game and 2.1× for a web game made by one developer at home. For a booking app it is about even (1.1×) and for a simple website it falls below 1: there, our case is access, not carbon.
Scenario F, class by class
modelled10th percentile within our least favourable scenario
Our least favourable modelled scenario: leanest studio, AI energy at the top of published ranges (modelled)
| Class | Strict, 10th percentile | Strict, median | Whole project, 10th percentile | Per extra product (direct pair), strict 10th percentile | Where |
|---|---|---|---|---|---|
| Simple small-business website (brochure) | 0.34×below 1: no benefit | 0.67× | 1.1× | 2.5× | Montreal, fully remote |
| Booking or business web app like those our users publish | 1.1× | 2.4× | 3.1× | 8.1× | Montreal, fully remote |
| Hyper-casual web game like those our users publish, small team | 2.1× | 4.2× | 5.1× | 9.6× | Montreal, fully remote |
| Same web game, one developer at home | 2.1× | 4.2× | 3.5× | 9.6× | Montreal, fully remote |
| Studio-grade casual game with progression (upper scenario) | 29× | 57× | 68× | 130× | Montreal, fully remote |
What moves the studio-game ratio
modelledstrict countcentral values
ratio when this input moves from its low to its high valuecentral ratio, 250×
Conventional hands-on effort (person-days)
74× to 540×
Energy of re-reading cached text, large model
86× to 390×
Carbon of making a workstation
160× to 380×
Share of model training counted
170× to 310×
Grid carbon at the data centre
210× to 350×
Cloud and software tools per day
220× to 310×
Device lifetime (years)
220× to 290×
Energy of re-reading cached text, mid-size model
190× to 260×
Table view
Tornado, studio game, strict ratio at central values
| Input | Ratio at low input | Ratio at high input |
|---|---|---|
| Conventional hands-on effort (person-days) | 74 | 540 |
| Energy of re-reading cached text, large model | 390 | 86 |
| Carbon of making a workstation | 160 | 380 |
| Share of model training counted | 310 | 170 |
| Grid carbon at the data centre | 350 | 210 |
| Cloud and software tools per day | 220 | 310 |
| Device lifetime (years) | 290 | 220 |
| Energy of re-reading cached text, mid-size model | 260 | 190 |
Four stress cases
modelledcentral valuesstudio-grade game
Studio-grade casual game with progression (upper scenario): ratio conventional / OhMyUnicorn at central values under stress (modelled)
| Case | Strict count | Whole project | Studio and week |
|---|---|---|---|
| Central values | 250× | 740× | Paris, typical work pattern |
| AI ten times dirtier | 28× | 84× | Paris, typical work pattern |
| Our R&D overhead doubled | 150× | 440× | Paris, typical work pattern |
| All of it, in the leanest city | 15× | 34× | Montreal, fully remote |
Two modelling choices
modelledstudio-grade gamemedians
Studio-grade casual game with progression (upper scenario): ratio conventional / OhMyUnicorn, medians, under the two modelling choices; shown beside the draws, never inside them (modelled)
| Modelling choice | Studio-grade game |
|---|---|
| Base: everything counted | 150× |
| Without model training | 230× |
| Without input tokens | 510× |
Modelled, internal study, not yet critically reviewed. Medians of paired draws unless marked central values. Method and sources in the sheets.
09 · The balance rule
Production comes out ahead if about 1 game in 13, or 1 app in 10, replaces a conventional build.
Most of our users would never have hired a studio. So the honest question is: how many of our products must replace a real conventional build for the whole production to emit less, counting only the emissions that really stop and the extra hosting of everything else? Today about 1 in 13 for games and 1 in 10 for apps (illustrative until the share is measured). 16% of requests already carry a business signal (an existing site, a registered business or a quote): signals, not proof. From now on we ask every publisher, and we credit a replacement only when the answer is backed by an outside signal.
Central values: at 2,000 products a month, if 1 in 10 replaces a conventional build, about 192 t CO2e a year of conventional production stops in the short run, against about 39 t for our whole company at that volume.
How many products must replace a conventional build?
modelledillustrative until the share is measuredshort-runhosting counted on both sides
break-even share (median)unfavourable tail (90th percentile)to the right: production comes out ahead
Booking or business web app like those our users publish
1 in 10break-even share 10.3%; unfavourable tail (90th percentile) 1 in 5
Hyper-casual web game like those our users publish, small team
1 in 13break-even share 8%; unfavourable tail (90th percentile) 1 in 7
0.4%1%10%100%
Table view
Break-even share of products that must replace a conventional build (consequential, modelled)
| Class | Median share | One product in | 90th percentile share | One product in (tail) |
|---|---|---|---|---|
| Booking or business web app like those our users publish | 10.3% | 10 | 18.6% | 5 |
| Hyper-casual web game like those our users publish, small team | 8% | 13 | 14.2% | 7 |
What stops, at different volumes and shares
modelledcentral valuesshort-runillustrative until the share is measured
Conventional production that stops in the short run, t CO2e a year, against our whole-company footprint at that volume (central values, modelled, illustrative until the share is measured)
| Products a month | Our whole company, t CO2e a year | If 2% replace a conventional build | If 5% replace a conventional build | If 10% replace a conventional build | If 16% replace a conventional build |
|---|---|---|---|---|---|
| 260 | 17 | 5 | 13 | 25 | 40 |
| 1,000 | 26 | 19 | 48 | 96 | 154 |
| 2,000 | 39 | 39 | 96 | 192 | 308 |
| 5,000 | 75 | 96 | 240 | 481 | 769 |
The question we now ask at publish
Option order is randomised. A credit needs the answer and a behavioural signal; "not sure" earns none; a third party samples respondents.
Modelled, internal study, not yet critically reviewed. Medians of paired draws unless marked central values. Method and sources in the sheets.
10 · Rebound
Cheap software means more software. Here is how we keep each one light.
An app online uses about 0.4 W, about 3.5 kWh a year. Intensity alone never suffices, so each guardrail is dated honestly: what runs today, what is planned.
Footprint per product, by class, at a fixed mix
Published with the mix, so the average cannot be flattered by changing what we build.
One rulebook, every app, kept after every edit
Weight budgets, eco-design and accessibility rules written once into the factory and re-checked on each change.
Idle apps go to sleep
Not built today. Apps nobody visits will hibernate and wake on demand.
Energy-aware model routing
Each job goes to the lowest-energy model that passes our quality checks. Cost is not energy; today's routing saves money.
11 · People
Real software for the people agencies never serve.
Between four and five in ten app requests are for an own business or trade, in 12 languages. Outside accounts grew 14 times in 60 days. In France, 35% of small and mid-sized firms (about 840,000) have no website, and 1.17 million companies were created in 2025, two thirds of them by micro-entrepreneurs who start with no tech team and little money: two thirds of French firms spend under EUR 300 a year on their web presence. In the EU, 73% of SMEs reach basic digital intensity against a 90% target for 2030.
47%Own business or trade
26%Personal or fun
9.5%Creative
6.5%Education
6.5%Job or team
2%Health or care
Share of 170 requested apps. Apps requested by outside users, classified one by one by Gemini 3.8 Flash, two passes agreeing 92%; midpoints shown.
Booking portal and AI assistant for a tattoo studio
Back-office tool for independent physio clinics
Marketing and booking website for a golf club
Stock tracker for informal market traders
Workflow tool for energy-performance diagnostic requests
Peer-to-peer student tutoring and skill swap
Financial-literacy game for kids and teenagers
Gamified task manager for people with ADHD
Board where game makers find teammates
App that turns photos of sweaters into knitting patterns
45%of requested apps are meant to earn money
14×outside accounts in 60 days
12languages people write their requests in
16%of requests already carry a business signal (an existing site, a registered business or a quote)
12 · The fund's test
Contribution. No significant harm. Good governance. Three numbers.
The three prongs of a sustainable investment, with the honest status of each: evidence collection is underway on all three, with an owner and an open condition.
a
Contribution
An enabling solution for lower-carbon software production; footprint per product estimated per unit; displaced production tracked with a validated substitution share.
Status: evidence collection underway; share measured from now; the fund's methodology to confirm.
Owner: OhMyUnicorn
b
No significant harm
Footprint by scope (inventory v0 done, preliminary; v1 next); hosting about 0.4 W per app; AI energy counted with failures and edits; rebound guardrails.
Status: evidence collection underway; water and materials not assessed yet.
Owner: OhMyUnicorn
c
Good governance
French SAS, BSA AIR round, quarterly KPI note, data hosted in the EU.
Status: see the round structure.
Owner: OhMyUnicorn
Organisations equipped
Verified organisations whose app serves someone other than its creator every month for 3 months.
baseline pending: telemetryFootprint per product
kg CO2e per published product passing our quality floor, direct and allocated, on a dated glide path.
6.7kg
Typical app, modelled, today's baseline.
Production displaced
Substituting products credited only with an outside signal (registered business, previous site, quote) × conventional person-days, reported as estimated effort substituted; short-run CO2e only for evidenced cases; reviewed externally.
credited share pendingOur own absolute footprint. Inventory v0 (preliminary, internal estimate, location-based): about 17 t CO2e a year, within 3.5 to 125 t. Scope 1 is 0 kg and scope 2 is 0 kg a month; scope 3 carries the rest. Detail in sheet S7.
13 · Method
Open model. Independent review funded by the round.
Monte Carlo on graded, sourced inputs (16 cities, 13 company reports, manufacturer footprints, EcoLogits with Google and Mistral as cross-checks, our own logged tokens). Two boundaries, two consistent pairs (per extra product; whole company), modelling choices kept apart from uncertainty. Each class has a written acceptance checklist; 4 real customer products were scored against it from their source code and priced the conventional way for that exact feature list; an observed-effort study is funded by the round. Scope: what the factory makes through its two public front doors, OhMyUnicorn and OhMyGames; the factory's own development sits inside every number as the research share; vertical products built on the same factory (such as Escale or AdStudio) are not in this study yet. Before any public comparison: an ISO 14044 critical review by an independent panel. Until then, we do not use these comparisons in marketing.
What we do not say
- Not "carbon neutral", "net zero" or "climate positive".
- Not "at least", "floor" or "proven".
- Not "we replace developers"; we serve people who never had one.
How a claim gets out
- Every sentence sits in a claims register with its boundary, evidence and approver.
- The qualifier ("modelled, internal study") stays beside the number, in private too.
- Public pages keep the absolute footprint until the review is done.
Review
- Now: one external reviewer for the data-room version.
- Before any public comparison: an ISO 14044 panel.
- Reviewer conflicts are disclosed.