Your employees opened a chatbot dozens of times today. They drafted emails, summarized contracts, cleaned up spreadsheets, and generated code. Every one of those requests ran on a server in a data center somewhere, drawing power and producing emissions. Almost none of it shows up in your carbon inventory. That gap is the part of your AI carbon footprint most companies have not started counting, and it is growing faster than almost any other line in the value chain.
The energy behind every prompt is real
Data center electricity demand grew 17 percent in 2025, and consumption at AI-focused data centers surged 50 percent, according to the International Energy Agency. The agency expects total data center demand to roughly double by 2030. Model providers reported a threefold jump in active users over a single year. The work your staff offloads to AI tools is part of that curve, not separate from it.
A single short text prompt is cheap. The problem is volume and complexity. Reasoning models, long-document analysis, image and video generation, and agentic workflows that now chain multiple steps together can each consume hundreds of times the energy of a basic query. Multiply that by a workforce reaching for AI throughout the day, across a full year, and the AI carbon footprint stops being a rounding error.
Where employee AI use lands in your AI carbon footprint
The instinct is to check Scope 1 and Scope 2 first, and both come up empty. Scope 1 covers fuel you burn directly, which a cloud chatbot does not involve. Scope 2 covers the electricity you purchase for assets you operate, and you do not run the data center or pay its power bill. When you buy AI as a service, whether that is an enterprise ChatGPT seat, Microsoft Copilot, Gemini, or Claude, the provider owns the infrastructure, and you have no operational control over it. Under the GHG Protocol, this places the emissions in Scope 3.
The specific home is Scope 3 Category 1, Purchased Goods and Services. The same logic that puts your accounting software, your CRM, and your cloud storage in Category 1 applies to AI tools. You are paying for a service whose emissions occur in a vendor’s operations, upstream of your own. That makes employee AI use a measurable input to your AI carbon footprint, not an externality you can wave off.
One exception is worth naming. If your company self-hosts models on hardware you own and power yourself, the electricity becomes Scope 2, and the servers carry embodied emissions under Scope 3 Category 2, Capital Goods. Most organizations are not doing that. They are buying access to someone else’s models, which keeps the AI carbon footprint in Category 1.
The shadow AI problem
A large share of employee AI use never appears on a procurement statement. Staff sign up for free tiers with personal accounts, paste work into consumer tools, and run tasks through browser extensions that the company never licensed. This is shadow AI, and it is invisible to the spend-based methods most carbon accounting teams rely on. If you estimate Category 1 emissions from what you paid a vendor, every free-tier prompt your team sends is an emission you never capture. The AI carbon footprint you report ends up smaller than the one you cause.
How to report your AI carbon footprint
You do not need perfect data to start. The GHG Protocol Scope 3 guidance supports a tiered approach, and AI fits cleanly within it.
Begin with a spend-based estimate. Take annual spend on AI tools and licensed platforms, apply an emission factor for cloud software, and you have a defensible first number. It will be rough. It is also far better than a zero that pretends the activity does not exist.
Improve it with vendor data. The major AI and cloud providers publish sustainability reports and, increasingly, customer-level carbon dashboards. Where a provider offers energy or emissions tied to your usage, use them. That moves the AI carbon footprint from a spend proxy toward a real activity-based figure.
Then deal with shadow AI directly. Survey teams on which tools they use. Set a policy that routes work through licensed, measurable accounts instead of personal logins, so the data exists in the first place. The governance win and the carbon accounting win come from the same action. Document your methodology and your assumptions so an auditor can follow the logic, which matters more as the EU’s CSRD and the ISSB’s IFRS S2 push Scope 3 disclosure from voluntary to expected.
Why companies are overlooking it
The activity hides between departments. IT signs the software contracts, the sustainability team builds the carbon inventory, and the two rarely reconcile a chatbot subscription against an emissions category. Materiality assumptions do the rest. A single SaaS license looks small next to freight, manufacturing, or purchased hardware, so it gets waved through as immaterial without anyone checking the math on a year of company-wide usage. And because AI tools arrived fast and spread informally, no one built a line item for them. A category that did not exist in last year’s inventory is easy to leave out of this year’s.
The result is an AI carbon footprint that grows while reading as zero on paper. That is exactly the kind of gap regulators and assurance providers have begun probing, because it signals an inventory that tracks the familiar while missing the new. A footprint that ignores a fast-growing source is not conservative. It is wrong in a direction that will be hard to defend later.
The hardware side of the AI carbon footprint
AI does not only draw power in the cloud. It accelerates the refresh cycle on the ground. Demand for AI-capable laptops, GPUs, and on-premises inference hardware is accelerating device upgrades, and every upgrade creates retired equipment. Counting the energy of AI use while ignoring the hardware it displaces leaves half of the AI carbon footprint unmanaged.
This is where disposition matters. Responsible IT asset disposition services keep retired AI hardware in use through refurbishment and resale, recover materials through certified recycling, and document the avoided emissions for your inventory. The same ESG and carbon footprint reporting discipline that should capture your AI use also tracks what happens to devices when they leave service. Pairing measured AI use with environmental accountability on the hardware side produces a footprint that reflects the whole picture, not just the parts that were easy to count.
The companies that get ahead of this will not be the ones using the most AI. They will be the ones who can show, with documentation, where their AI carbon footprint sits, how they measured it, and what they are doing about both the prompts and the hardware behind them.
Frequently Asked Questions
Is employee AI use Scope 1, 2, or 3?
For tools you buy as a service, it is Scope 3. You do not operate the data center or buy its electricity, so the emissions sit upstream in your value chain. The specific category is Scope 3 Category 1, Purchased Goods and Services. Self-hosted models running on hardware you own are the exception, which shifts the power use into Scope 2.
Which Scope 3 category covers AI tools?
Category 1, Purchased Goods and Services. AI software purchased on a subscription plan falls under your other cloud and SaaS spend, where the emissions occur in the vendor’s operations rather than your own. Category 2, Capital Goods, applies only if you purchase and own AI servers and hardware outright instead of renting access.
How do I calculate the AI carbon footprint of our tools?
Start spend-based: multiply annual AI software spend by a cloud emission factor for a quick estimate. Refine it with vendor-provided usage data where available. Survey staff to capture shadow AI on free tiers, then document your method so the figure holds up under CSRD or IFRS S2 assurance. A rough number beats an assumed zero.
What is shadow AI, and why does it matter for reporting?
Shadow AI is employee use of AI tools that the company never licensed, often free tiers on personal accounts. It matters because spend-based carbon accounting only sees what you paid for. Unlicensed usage produces real emissions that never enter your inventory, thereby understating your AI carbon footprint and weakening governance.
Does AI use really matter next to manufacturing or freight
Per prompt, no. At the scale of a full workforce using AI every day over the year, it adds up quickly, with AI data center electricity use rising sharply each year. Treating it as automatically immaterial without measuring it is a mistake. Measure first, then decide whether it clears your materiality threshold.