Calculate AI emissions

Take responsibility for the CO₂ emissions from your use of AI

Even if we use artficial intelligence responsibly, emissions cannot be avoided. This is where you can take responsibility.

To use this page, JavaScript is required. You can activate it in your browser settings.

Calculate your AI emissions

Simply enter which AI you use, together with your team if applicable, and we’ll work out how many emissions are associated with it. In the next step, you can take responsibility for these emissions by making a donation to support ForTomorrow’s initiatives in Europe.

Would you like to calculate your personal or your company’s AI usage?

PersonalMy company's

Other
Multiple selections allowed.

Anthropic Google Open AI Mistral Other

Anthropic
Claude Individual Pro
Licenses
Anthropic
Claude Individual Max
Monthly costs
€
Anthropic
Claude Team Standard
Users
Anthropic
Claude Team Premium
Users
Anthropic
Enterprise
Monthly costs
€

Google
AI Plus
Licenses
Google
AI Pro
Licenses
Google
AI Ultra
Licenses
Google
Gemini API
Monthly costs
€

Open AI
Plus
Licenses
Open AI
Go
Licenses
Open AI
Pro
Monthly costs
€
Open AI
Business
Users

Mistral
Pro
Licenses
Mistral
Team
Users
Mistral
Enterprise
Monthly costs
€
Mistral
Education plan
Licenses

How much do you spend on flexible and other AI services?

€

per monthper year

Why are we asking?

AI costs are linked to usage, and usage is linked to emissions.

Choose your safety factor

What’s that?

AI companies do not publish detailed data on their emissions. The safety factor increases your contribution to climate protection. As it is impossible to calculate your AI usage accurately, we recommend a safety factor of 2 or higher.

12345

Donate €0 per month to ForTomorrow

Donate €0 per month and take responsibility for one tonne of CO₂ per year

Donate €0 per month and take responsibility for 0 tonnes of CO₂ per year

How do we do that?

We use your contribution to purchase emission rights from the European Union’s Emissions Trading System. We then cancel them, which means the allowances are permanently destroyed! In this way, we force major CO₂ emitters, such as coal-fired power plants in the EU, to emit less CO₂. This is because EU emission rights define the total amount of CO₂ that may be emitted by the biggest polluters in the EU. It’s the most effective method of reducing emissions that we know of.

You can also choose your own climate protection measures and how you want to make an impact: as a one-time donation or on a monthly basis.

Climate protection measures can be used to offset CO₂ emissions. As ForTomorrow’s climate protection activities are accelerating the energy transition, they are also reducing the CO₂ emissions from AI use in the long term.

Why does AI cause CO₂ emissions?

The problem is that the expansion of AI is happening faster than the expansion of renewable energy. To meet the enormous additional demand for energy caused by AI, gas and coal power stations have to remain connected to the grid for longer. This results in additional emissions.

How big is the AI climate problem?
Although the company once had ambitious net-zero targets, Microsoft’s emissions rose by 25 per cent last year, according to its own sustainability report.

A recent study by the Öko-Institut on behalf of Greenpeace estimates that annual emissions from AI data centers could rise from around 29 to 166 million tonnes of CO₂ by 2030 — despite a growing share of renewable energy. For context: 29 million tonnes of CO₂ is roughly equivalent to the annual emissions of Bolivia or Finland. By contrast, 166 million tonnes is more than what the whole of Nigeria, with its 229 million inhabitants, emits into the atmosphere each year, or 2.7 per cent of all US annual emissions.

The availability of electricity is so crucial that a data centers projected electricity consumption helps determine where such a facility can be built. This is why the capacity of data centers is often specified not only in terms of computing power but also in gigawatts (GW) – that is, according to how much electricity they require.

Take this for example: to supply electricity to its data centers, Amazon plans to build an off-grid power station with 35 gas turbines that are set to produce 7.65 GW of electricity. According to the planning permission, 33 million tonnes of CO₂ may be emitted annually. Based on information about capacity utilization at other plants, we estimate that in practice, around 24 million tonnes would be a realistic figure.
Can’t AI help combat climate change?
Scientists have known about the climate crisis for decades and have been warning the public about it for just as long. The solutions have also been clear for a long time: in order to preserve the foundations of life, humanity must emit fewer greenhouse gases. This means heating more efficiently, generating electricity sustainably, consuming fewer animal products and using less energy overall, so that we can meet demand with renewables.

Specialised AI models can help to increase efficiency and thus operate in a more environmentally friendly way. The more general-purpose AI models, for example those from OpenAI and Anthropic, can theoretically be used to produce such specialised models.

Unfortunately, however, economic efficiency does not always mean lower emissions. A study published in Nature describes how the fossil fuel industry is also using AI models to extract more oil, coal and gas from the ground. And it shows that the climate damage is greater than the benefits resulting from AI-enabled energy efficiency.

Because generative AI models consume a great deal of energy and are generally not used to tackle the climate crisis, using AI to save the climate can be viewed as a PR exercise for AI companies.

Not only does the operation of electronic components require a great deal of electricity, but cooling them is also extremely energy-intensive. Water is often used for cooling via evaporation. The image shows a data centre in Ohio with a capacity of 3.8 MW, which is already considered small by today’s standards.

Image: Google

How much CO₂ does your AI usage generate?

A truly accurate calculation of the climate impact of your AI usage would only be possible if you were running AI models yourself. There are models available to download for local use, such as DeepSeek, Llama and Qwenn. For such models, there are estimates of how much extra electricity your computer will consume when running them. However, these do not take into account the energy used to train the models.

Image: Google

How we calculate the CO₂ emissions from your AI usage

Emissions are directly linked to the electricity consumption of data centres. For the AI services listed in our form, it is not the number or complexity of your prompts that determines this electricity consumption, but rather how much money you pay to the companies behind them.

60 to 90 per cent of AI data centre capacity is used for training the models. AI service providers are still in start-up mode, and they are not yet profitable. All the money that comes in – including that from investors – is spent. The biggest cost for AI companies is their data centre capacity.

Do you have an AI contract with a fixed monthly fee but aren’t using up all the tokens available to you? In that case, the majority of your payments will be spent on data centre capacity: for research, training and to fund free usage (which is a form of marketing for the paid models).

You can view our detailed calculations in our spreadsheet (OpenOffice, Excel, Google Sheets). In summary, our method is based on the following:

The biggest cost for AI companies is the capacity of their data centres. Currently, no profit is made from the frontier models. This means that what you pay for AI services does not even cover the costs incurred by your usage. However, your expenditure is a useful indicator, as AI companies do not provide exact figures on the carbon footprint of your AI use.
We do have information about the companies’ data centre capacities. OpenAI, for example, has a capacity of 10 GW. We conservatively assume that 60 per cent of this capacity is actually utilised. We do not take purchased computing capacity into account and therefore recommend a safety factor of at least 2. Using the electricity mix for the region in question, we can then calculate the emissions.
The turnover of AI companies is also public information. If you specify how much you pay for which AI services, we can calculate on a pro rata basis how much CO₂ your AI usage generates.

Common questions about the calculation method

Why doesn’t the form ask about prompts or token usage?
You might think that someone who enters lots of prompts every day causes more emissions than someone who only uses generative AI occasionally. But that isn’t necessarily the case.

This is because the computational effort depends not only on how many queries you make but also on how complex the responses are. A query might be resolved with a short answer to a yes-or-no question. Or it might require a lengthy text analysis, a video or a complex programming task.

AI companies measure the scope of these responses using tokens. A token is a small unit of text, such as a word or a character. It is therefore not just the input or query that counts, but above all the response to it. This can range from a handful to many thousands of tokens. The more tokens there are, the more computing power is required, the more energy is consumed and the more emissions are generated.

The number of tokens can therefore serve as a useful basis for estimating emissions from AI models running on your hardware. However, our form focuses on the use of models running in the providers’ data centres. How many tokens are actually consumed by individual users is usually very difficult or even impossible to determine.

Furthermore, the method described above takes into account more than just the direct emissions caused by the processing of AI queries. It also includes additional emissions arising from the provision and use of the AI infrastructure. It therefore makes more sense for our estimate to consider factors other than individual token consumption.
Is it fair that the emissions generated by free accounts are counted towards the share of emissions attributable to paying accounts?
AI is currently funded through investment, by paying users and (to a small extent at OpenAI) through advertising.

When calculating the emissions generated by paid AI subscriptions, we also take into account the emissions generated by free usage that are not funded by advertising. These emissions arise solely because there is a paying customer base.

To provide you with an estimate of your AI emissions, we take into account not only the narrowly defined, direct emissions resulting from the processing of AI queries but also the emissions caused by what made large language models (LLMs) possible in the first place. This includes model training, as well as the marketing activities of AI companies. To draw an analogy: if we were to calculate the carbon footprint of a can of Red Bull, it would be too simplistic to disregard the supply chain through which the can reaches the shop where you buy it and to ignore the massive marketing campaigns that make Red Bull the brand it is. The same applies to ChatGPT.

Whilst an AI service based exclusively on a paid model would be possible, a free one would be difficult to implement without paying users (OpenAI’s advertising-funded model has yet to prove itself). Without revenue and without new investment, an AI company would soon go bankrupt. It seems, therefore, that AI companies consider the free service necessary to attract customers and investment capital. So if we are to assign responsibility for emissions to end users, it should be the paying customers.

Furthermore, as already mentioned, training generates more emissions than processing AI queries. Even if AI companies were to eliminate free AI usage entirely, emissions from training would only cease once the money stopped coming in.

Common questions about donating

What are the benefits of supporting climate protection through a nonprofit?
ForTomorrow is an official German nonprofit company. That means that you can deduct your donations from income tax in Germany. Moreover, that ensures that your donations are really spent on climate protection measures. We use all of our revenue for our climate protection measures. We can't pay out profits.
How can I deduct my donations from tax?
You will automatically receive a donation receipt in next year's first quarter. If you pay income tax in Germany, you can deduct your donations with that receipt. If you pay taxes in another country, it depends on if your government recognizes donations to German non-profits as deductible.
Do I get a certificate for my donation?
You you will immediately get a certificate for your donation on the confirmation page. You can also get a ForTomorrow-badge for your website to show your engagement with reliable climate protection in Europe.

Calculate AI emissions

Calculate your AI emissions

Would you like to calculate your personal or your company’s AI usage?

Anthropic

Anthropic

Anthropic

Anthropic

Anthropic

Google

Google

Google

Google

Open AI

Open AI

Open AI

Open AI

Mistral

Mistral

Mistral

Mistral

Choose your safety factor

Donate €0 per month to ForTomorrow

Donate €0 per month and take responsibility for one tonne of CO₂ per year

Donate €0 per month and take responsibility for 0 tonnes of CO₂ per year

Donate €? per month to ForTomorrow

Donate €? per month and take responsibility for one tonne of CO₂ per year

Donate €? per month and take responsibility for ? tonnes of CO₂ per year

Julian Ströh

“Cookies”

Your privacy preferences

Calculate AI emissions

Calculate your AI emissions

Would you like to calculate your personal or your company’s AI usage?

Anthropic

Anthropic

Anthropic

Anthropic

Anthropic

Google

Google

Google

Google

Open AI

Open AI

Open AI

Open AI

Mistral

Mistral

Mistral

Mistral

Choose your safety factor

Donate €0 per month to ForTomorrow

Donate €0 per month and take responsibility for one tonne of CO2 per year

Donate €0 per month and take responsibility for 0 tonnes of CO2 per year

Donate €0 per month and take responsibility for one tonne of CO₂ per year

Donate €0 per month and take responsibility for 0 tonnes of CO₂ per year