Enterprise RFPs are starting to ask which model powers your AI stack, where it's hosted, and what it emits. Here's the carbon disclosure answer sheet that keeps that from costing you a scoring point.
by Ayush Gupta's AI
The problem
A procurement team at a bank, a healthcare system, or a large CPG company is now running a weighted scorecard on vendor RFPs that includes a line nobody used to ask about: what AI models power this engagement, where are they hosted, and what's the emissions estimate behind the work. It shows up folded in next to cost, security, and accuracy — one more weighted criterion, not a standalone sustainability audit. Most agencies can answer the cost and accuracy questions in their sleep. Ask which model generates the client's ad copy, whether it's a low-carbon region, or what the per-token inference footprint looks like, and the honest answer is nobody has ever been asked to find out. That's not a values problem. It's a scorecard problem — a blank or vague answer on one line item in a weighted rubric is exactly the kind of thing that quietly moves a bid from first choice to second.
The fix
Build a one-page AI carbon and energy disclosure answer sheet — which models, which hosting regions, what emissions method backs the estimate — before a procurement scorecard asks for it, so the question costs an email forward instead of a week of scrambling mid-pitch.
The Playbook
List every AI model actually touching client deliverables
Not the one model in the pitch deck — every model actually in the production path: the LLM drafting copy, the image generator producing creative, the transcription tool on client calls, the model embedded inside a third-party tool the agency resells. Procurement questions ask about the real stack, not the headline tool.
Pull what each vendor actually publishes on hosting and energy
Most major model providers now publish something — a trust center page, a sustainability report, or a stated data center region and power usage effectiveness figure. It won't be perfectly standardized across vendors, and that's fine. The scorecard question is usually 'can this vendor produce a method-backed estimate,' not 'is the number identical to a competitor's.'
I'm compiling an AI carbon and energy disclosure sheet for enterprise RFP responses. For each AI tool below, summarize what the vendor publicly discloses about:
1. Hosting region(s) and whether the local grid is disclosed as low-carbon
2. Any stated Power Usage Effectiveness (PUE) figure for their data centers
3. Any per-model or per-token energy/emissions estimate, and the method behind it
4. Whether they publish a sustainability or trust center page I should cite directly
Tools: [LIST YOUR AI TOOLS]
Where a vendor discloses nothing on one of these points, say so plainly instead of guessing — I need to know what's a real disclosed fact versus a gap I still have to flag.Turn the gaps into a position, not a blank
Some vendors won't have a clean answer on every line, and pretending otherwise is worse than saying so. Where the honest answer is 'not independently disclosed, best available estimate is X,' write that sentence down once so the account team doesn't improvise a weaker version of it live on a call.
Draft the standing answer sheet
One page: tools in use, hosting regions, whatever emissions or PUE figures the vendors disclose, and the agency's own stated method for choosing lower-footprint options where accuracy allows. This becomes the thing that gets attached to an RFP response instead of written from scratch under deadline.
Using the vendor disclosure summary above, draft a one-page AI carbon and energy disclosure sheet for enterprise RFP responses. Structure it as:
1. AI tools used in delivery, and what each is used for
2. Hosting region and any disclosed low-carbon grid commitment, per tool
3. Any published PUE or per-token emissions estimate, with the source cited
4. Where no figure exists, one honest sentence stating that and what we use as a proxy
5. A short statement of how we choose between models when carbon footprint is one factor among cost, quality, and client requirements
Keep it factual and sourced. No marketing language, no claims we can't back with the vendor's own published disclosure.Refresh it quarterly, not per-RFP
Model providers update their disclosures and hosting footprints faster than agencies update their boilerplate. Put a recurring reminder on the same cadence as other vendor-questionnaire refreshes, so the sheet attached to a bid next quarter isn't quietly citing a year-old PUE figure.
What changes
A sourced, one-page answer ready before a procurement scorecard asks for it, instead of a scramble that turns one weighted line item into the reason a bid loses on points nobody on the account team saw coming.
A procurement lead reading through a vendor RFP response hits a line most agencies have never been asked about before: which AI models power this engagement, where are they hosted, and what's the estimated energy or carbon footprint behind the work. It's not framed as a sustainability audit. It's one weighted line on a scorecard that also covers cost, security, and accuracy — and a blank or vague answer on that one line is enough to move a bid from first choice to second without anyone saying the word "sustainability" out loud on the call.
This is now a structural feature of enterprise procurement, not a fringe ask. Carbon and data-center disclosure questions have become a standing line item in vendor templates across financial services, healthcare, manufacturing, and large tech buyers, and the shift for AI vendors specifically is getting more granular: which models, which hosting region, what Power Usage Effectiveness figure for the data center, what per-token inference footprint. None of that is a question most agencies have ever had to answer, because nobody asked it eighteen months ago.
Why this lands differently than the security questionnaire did
Agencies already learned to handle the security and data-handling questionnaire — most have a reference document for that by now. This is the same shape of problem wearing a different question: not "is our data safe with you," but "what's actually running under the hood, and can you prove it with something a vendor published, not something you're guessing." The honest starting answer at most agencies is that nobody has ever checked which region a model is hosted in, let alone what its disclosed emissions profile looks like.
The fix isn't a sustainability initiative
Nobody needs to rebuild the agency's AI stack around carbon optimization to answer this well. The fix is smaller and more boring: list every model actually touching client work, pull whatever each vendor already discloses publicly about hosting and energy, write down the honest answer where a vendor discloses nothing at all, and keep the whole thing on one page that gets attached to a bid instead of assembled under deadline pressure for the first time.
Bottom line
This question is still rare enough that most competing agencies will also get caught without an answer. That's the actual opportunity — not winning on carbon virtue, but being the one bid that answers a scorecard line everyone else leaves blank. A half-day to build the sheet and a quarterly refresh is cheap insurance against losing points on a question that didn't exist on last year's RFP template and now quietly does.