What Are Scope 3 Emissions? A Simple Guide
On this page
Scope 3 emissions are the indirect greenhouse gas emissions that happen across a company value chain, outside of its own operations and its own energy purchases. They are usually the largest and hardest-to-measure part of a company carbon footprint, and for many organizations the software and cloud services they buy, including AI, fall squarely into this category.
This guide explains the three emission scopes in plain language, shows why the AI tools you pay for count as scope 3 emissions rather than your own, and outlines practical ways to account for that usage. The framework comes from the Greenhouse Gas Protocol, the most widely used standard for corporate carbon accounting.
The three scopes, briefly
Emissions accounting splits a company footprint into three buckets. Scope 1 covers direct emissions from sources the company owns or controls, such as fuel burned in company vehicles or a gas boiler in an office. These are the emissions you create on site.
Scope 2 covers indirect emissions from the energy you buy, mainly the electricity, steam, or heating your facilities consume. You did not burn the fuel yourself, but you caused the emissions by drawing power from the grid. Scope 3 covers everything else indirect: business travel, purchased goods and services, waste, and the cloud and AI vendors in your supply chain.
- Scope 1: direct emissions from owned or controlled sources
- Scope 2: indirect emissions from purchased electricity and heat
- Scope 3: all other indirect emissions across the value chain
Why buying AI is a scope 3 emission
When you send a request to a hosted AI model, the electricity is consumed in someone else data center, not yours. You do not own the servers and you did not buy the power directly, so the resulting emissions are not your scope 1 or scope 2. They land in scope 3 as part of your purchased services.
This mirrors how any cloud service is treated. The energy behind a model reply is real, as we detail in how much electricity AI uses, but from your books it shows up as an indirect, upstream emission tied to a vendor. That is exactly what scope 3 was designed to capture, and it is why AI adoption can quietly grow this part of a company footprint.
How to account for AI usage
Accounting for AI in scope 3 starts with data on how much you use. The cleanest input is token volume, because tokens map to energy. A rough per-token figure ranges from about 0.0008 Wh for a small model to around 0.0038 Wh for a frontier model and 0.0040 Wh for a reasoning model, before you apply data center overhead.
From energy you derive carbon by applying a grid factor, such as roughly 0.395 kg CO2 per kWh in the US, and a PUE multiplier of about 1.56 to account for cooling. Our carbon footprint calculator walks through this chain. The challenge for most companies is getting usage data from vendors at all, which is why tracking scope 3 AI emissions often depends on the provider surfacing it.
The vendor data problem
You cannot report what you cannot measure, and most AI providers do not hand customers a per-request emissions figure. That leaves buyers estimating from token counts and public grid averages, which is workable but imprecise. The gap between estimate and reality is where scope 3 reporting tends to break down.
A provider that measures and reports energy, carbon, and water per request removes most of that guesswork. Instead of modeling your AI footprint from assumptions, you receive the actual measured impact tied to your account. That kind of transparency, offered through a carbon-tracked AI API, turns a fuzzy scope 3 estimate into a documented figure you can stand behind.
Reducing your AI scope 3 footprint
Once you can see the number, you can act on it. Choosing more efficient models for routine tasks, trimming unnecessary requests, and consolidating on providers that run in cleaner grid regions all lower the underlying emissions. We collect practical steps in how to reduce your AI carbon footprint.
You can also address what remains through providers that offset their measured impact. If a vendor funds verified offsets beyond what your usage causes, the AI portion of your scope 3 can shrink rather than grow. For teams building this into a wider strategy, green AI for business covers how to fold it into reporting and procurement.
The headline: AI you buy from a hosted provider counts as a scope 3 emission, an indirect footprint you should measure from usage data and address through efficiency and offsetting.
FAQ
What are scope 3 emissions in simple terms?
Scope 3 emissions are all the indirect greenhouse gas emissions in a company value chain that are not from its own operations or purchased energy. They include business travel, purchased goods, and cloud or AI services. They are usually the largest share of a company total footprint.
Why is AI usage a scope 3 emission and not scope 2?
Scope 2 covers electricity you buy directly for your own facilities. When you use a hosted AI model, the electricity is consumed in the provider data center, not yours. That makes it an upstream, indirect emission from a purchased service, which is scope 3.
How do I calculate the emissions from AI I use?
Start with usage data such as tokens processed, convert tokens to energy using per-token estimates, then apply a grid carbon factor and a PUE multiplier for data center overhead. The result is your estimated AI carbon footprint. Vendor-provided measurements make this far more accurate.
Can measuring AI emissions actually reduce my scope 3?
Measuring itself does not reduce emissions, but it lets you act. Once you see the figure you can choose efficient models, cut waste, and pick providers in cleaner regions. Providers that offset beyond their measured impact can reduce the AI portion of your scope 3.