AI that shows its numbers.
Answers from a data centre inside the European Union that runs on 100% renewable electricity, on open-weight models, with public pages on who processes what. Plus a plain list of what we do not claim.
Electricity
100%
renewable at the AI data centre, backed by Guarantees of Origin (Scaleway, 2024).
PUE
1.25
against a global average of 1.55. Cooled with outside air, no air conditioning.
Extra person
€10
a month excl. VAT, 53% less than ChatGPT Business.
Honest about the limits.
A mission-driven company is judged on its suppliers' claims too. These are ours, and the ones we will not make.
- True: Every AI inference runs in a data centre inside the European Union, on 100% renewable electricity.
- True: That data centre has a PUE of 1.25 and is cooled without air conditioning (Scaleway, 2024 data).
- True: The chat shows an estimate of the energy each answer used, and the method is public.
- True: Customer data is never used to train models.
- Not true: “Carbon-neutral AI.” The renewable supply is backed by Guarantees of Origin, and servers have a manufacturing footprint.
- Not true: “Trained on green energy.” Others trained the open-weight models we run, and we cannot verify how that was powered.
- Not true: “Our whole stack is green.” The figures cover inference. Hosting, database and payments run at other providers, under their own numbers.
- Not true: “We are a B Corp.” We are not certified.
Work that serves the mission.
Everyday writing and reading, with one tool for the whole team.
Impact reporting
Structure your impact data into a clear report, and rewrite it for each audience.
Grants and tenders
Draft applications, answer questionnaires and cut long texts down to the word limit.
Policies
Draft a supplier code of conduct, a hiring policy or an onboarding guide to adapt.
Stakeholder updates
Turn notes into newsletters, board updates and posts in your organisation's voice.
Translation
Reach partners and customers across Europe in their own language.
Research summaries
Summarise the reports, studies and regulations your team needs to keep up with.
Check a claim before you repeat it.
AI can help you write about impact. It can also make impact up.
- Verify every figure. Models can invent statistics and sources. Check the numbers in your report against your own data.
- Watch for greenwashing. Do not let a draft claim more than you can prove. EU rules on green claims are getting stricter.
- Use the lightest model that does the job. The lighter tiers handle everyday text. Keep the largest model for hard problems: it uses more energy per answer.
- Keep personal data out unless your processing agreement and your own policies cover it.
- Quote our figures with their source. They are Scaleway's 2024 data, and they cover inference only.
Every figure, with its source.
For your supplier questionnaire or your impact report.
| Measure | Figure | Source |
|---|---|---|
| Electricity at the AI data centre | 100% renewable, wind and hydro, backed by Guarantees of Origin | Scaleway Impact Report 2025 |
| PUE | 1.25, against a global average of 1.55 | Scaleway Impact Report 2025 |
| Water | 0.25 litres per kWh, no cooling towers | Scaleway Impact Report 2025 |
| Energy per answer | Estimated from published measurements, shown in the chat | Sustainability |
| Models | Open-weight, named per tier with their licence | AI transparency |
| Who processes what | Every party, its region and its legal entity | Subprocessors |
| Each extra person | €10 a month excl. VAT | Pricing |
The data centre figures are Scaleway's own, for 2024. They cover AI inference, not the rest of our stack.
The pages behind the numbers.
All public, so you can cite them rather than take our word for it.
What mission-driven teams ask us.
Short answers. The pages above have the sources.
Is there a sustainable AI chatbot for B Corps?
Look for a provider that names its data centre and publishes its energy source, its efficiency and its limits. Ours runs inference on 100% renewable electricity at a PUE of 1.25, on open-weight models, and our sustainability page lists what those figures do not cover.
Is Union AI carbon-neutral?
No, and we do not claim it. Inference runs on renewable electricity backed by Guarantees of Origin, but servers have a manufacturing footprint and we cannot verify how the open-weight models were powered while others trained them.
How much energy does one AI answer use?
It depends on the model and on the length of the answer. The chat shows an estimate under each answer, based on published measurements and the data centre's PUE. It is an estimate, not a meter reading, and the method is on our sustainability page.
Why do open-weight models matter?
You can see which model you use and under which licence, and you are not tied to one company's closed model. We publish the model behind every tier on our AI transparency page.
Is Union AI a certified B Corp?
No, we are not B Corp certified. This page is for companies that are, or that hold themselves to the same standards, because they ask the questions our transparency pages answer.
What should we ask an AI supplier about sustainability?
Four questions: which data centre runs the model, what powers it, how efficient it is and what the figures leave out. A supplier that answers all four in public is easier to report on than one that answers in a sales call.
What does Union AI cost for a team?
The chat is €13 a month excluding VAT, and each extra person is €10 a month on the same invoice, 53% less than ChatGPT Business. You can cancel at any time.
Start with AI for Europeans.
Use AI without sending your data abroad, everything on European servers, at a European price.
€13 a month for the chat, cancel any time. Or per token from €10 of credit.
Not sure yet? Take the 60-second quiz →