#20 of 30 · AI lab
OpenAI (ChatGPT)
A single blog-post number, and gigawatts of gas-backed capacity on the way.
In plain words
How scores workOpenAI scores 40 out of 100 — laggard. Little or no disclosure, fossil-backed capacity, or active disputes with regulators and communities. Confidence in this score is mostly estimates (C).
What helps
- Carbon · Azure-hosted capacity benefits from Microsoft's renewable matching.
What hurts
- Carbon · Abilene runs on the Texas grid plus one of the largest on-site gas build-outs in the US (≈360 MW of behind-the-meter generation by January 2026), with 41 more turbines proposed.
- Carbon · At least three of the seven announced Stargate sites plan on-site natural-gas plants.
- Water · The 0.000085-gallon per-query claim has no stated scope.
What nobody knows
- Training emissions in model cards
- Annual environmental report
- Location-based Scope 2 reported
- Water-use effectiveness reported
- Third-party verified AI figures
- +1 more
The only official footprint figure for ChatGPT is a line in a June 2025 blog post — 0.34 Wh and 0.000085 gallons of water per average query — with no methodology. Independent estimates (Epoch AI, 0.3 Wh for a typical GPT-4o query) broadly agree but note long inputs cost 10–100× more. Meanwhile OpenAI passed a million GPUs in 2025 and the Stargate build-out leans on behind-the-meter gas at Abilene and at least two other sites, permitted with 62 diesel generators and up to 51 gas turbines under rules written for dry cleaners.
Key figures
Per-prompt and fleet metrics. Hover a title for what it means; open “How we got this number” for the working.
Blog-post figure without methodology; Epoch independently estimates ~0.3 Wh.
0.000085 US gallons; scope unstated.
Location-based: what the grid actually emitted. Renewable certificates would bring the market-based figure near zero.
How we got this number
0.34 Wh × 390 gCO₂/kWh (US grid, location-based). Most ChatGPT traffic is served from Azure in the US.
Location-based estimate for the AI fleet only; not comparable with a full corporate inventory.
How we got this number
1300 MW × 8,760 h × 80% utilisation × 420 gCO₂/kWh (Texas grid plus on-site gas). Compute only — nothing else the company does is counted.
How we got this number
From roughly half a million GPUs at the start of 2025 to 'well over a million' by year-end, per the CEO; Stargate adds 1.2 GW at Abilene alone.
How we got this number
Azure's fleet WUE is 0.30 L/kWh; Abilene's closed-loop cooling evaporates almost nothing. 0.3 is the blended assumption.
How we got this number
Roughly half of OpenAI's 2025 capacity is on Azure (100% annual renewable matching) and half on Stargate campuses drawing on ERCOT plus on-site gas (≈25% clean).
Hardware & materials
The chips, power, copper and future e-waste behind the fleet — the part sustainability reports leave out
GPUs, TPUs or custom accelerators the company runs. Analyst figures are in H100-equivalents, company figures in physical chips.
How we got this number
'Well over a million', per Sam Altman in July 2025; Abilene alone is planned for 450,000 GB200s.
Megawatts the AI fleet pulls. One megawatt runs roughly 750 US homes.
How we got this number
≈1,000,000 GPUs × 1.3 kW all-in; Abilene alone is sized at 1.2 GW.
Terawatt-hours per year. One TWh is a year of electricity for about 95,000 US homes.
How we got this number
1300 MW × 8,760 h × 80% utilisation. Compute fleet only.
CO₂e emitted making the chips and servers before they answer a single prompt.
How we got this number
1,000,000 accelerators × 514 kg CO₂e each (chip plus share of the host server). Emitted in manufacturing, before the first prompt is served.
Tonnes of server hardware that will be retired in three to five years.
How we got this number
1,000,000 accelerators × 17 kg of server hardware each. Tomorrow's e-waste unless reused; generative AI could produce 2.5 Mt of it a year by 2030.
Tonnes of copper in cabling, busbars and transformers.
How we got this number
1,300 MW × 27 t of copper per MW (cabling, busbars, transformers).
Diesel backup units or gas turbines at the data centres, permitted or observed.
How we got this number
41 more gas turbines and 18 more generators proposed for the expansion.
Share of retired servers and parts given a second life.
Kilograms of antimony-equivalent, the life-cycle unit for scarce minerals consumed.
Score breakdown
Each pillar is 0–100; weights are fixed across the index
- Sam Altman states 0.34 Wh and 0.000085 gallons of water per average query; no definition of 'average' and no methodology. Sam Altman (blog)
- Epoch AI independently estimates ~0.3 Wh for a typical GPT-4o query, rising to ~2.5 Wh for a 10k-token input and ~40 Wh at 100k tokens. Epoch AI
- Abilene runs on the Texas grid plus one of the largest on-site gas build-outs in the US (≈360 MW of behind-the-meter generation by January 2026), with 41 more turbines proposed. CleanviewEpoch AILatitude Media / Floodlight
- At least three of the seven announced Stargate sites plan on-site natural-gas plants. Epoch AI
- Azure-hosted capacity benefits from Microsoft's renewable matching. Data Center Dynamics
- Abilene uses closed-loop cooling that trades water for electricity; the mayor says annual use equals about half a day of the city's consumption. TechRepublic
- The 0.000085-gallon per-query claim has no stated scope. Sam Altman (blog)
- No environmental report, no model-level training emissions, no per-prompt methodology; fleet size known only from CEO posts and partners' filings. Sam Altman (blog)Epoch AIIntrol
- No public net-zero, carbon-free-energy or water targets. Epoch AI
- Chose gas-backed behind-the-meter power for speed at Stargate sites; Abilene's generators and turbines were authorised through Texas 'permit by rule' filings with no public comment. CleanviewThe Texas Tribune
- CEO softening earlier AI-risk warnings as an IPO approaches, per Fortune. Fortune
Models
What this company ships, and what each prompt costs
| Model | Weights | Energy / prompt | Water | Training CO₂e |
|---|---|---|---|---|
| ChatGPT (average query, company figure) Frontier (undisclosed) | Closed | 0.34 Wh DisclosedCompany figure. Published by the company itself. We take it as given and record the scope next to it. | 0.320 mL DisclosedCompany figure. Published by the company itself. We take it as given and record the scope next to it. | 15,000 t Estimateecogpt.co estimate. Our own estimate from published factors. The working is shown next to the number; treat it as an order of magnitude, not a measurement. |
| GPT-4o (typical query, independent estimate) Frontier (undisclosed) | Closed | 0.30 Wh IndependentMeasured independently. Measured or estimated by researchers or analysts with no stake in the result. | 0.090 mL DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor. | 10,000 t Estimateecogpt.co estimate. Our own estimate from published factors. The working is shown next to the number; treat it as an order of magnitude, not a measurement. |
Where the compute lives
Named sites from company disclosures and press reporting
- Stargate AbileneTexas, USA · ≈1.2 GW planned; 450,000 GB200s; on-site gas.
- Stargate Shackelford CountyTexas, USA
- Stargate Doña Ana CountyNew Mexico, USA
- Stargate LordstownOhio, USA
- Azure (Microsoft) regionsMultiple
Timeline
Incidents and developments involving this company
Stargate Abilene adds 41 gas turbines and 62 diesel generators under a permit written for dry cleaners
A Floodlight and Texas Tribune investigation found Stargate's Abilene campus was authorised to run 10 gas turbines and 62 backup diesel generators through Texas 'permits by rule' that need no environmental study, public notice or comment — and has applied to add 41 turbines and 18 generators more.
OpenAI, Oracle, CrusoeStargate Abilene pairs the Texas grid with ~360 MW of on-site gas
OpenAI's flagship Stargate campus went live in September 2025 with 200+ MW and is scaling toward 1.2 GW. Cleanview's tracking shows 29 gas turbines ordered and roughly 360 MW of behind-the-meter generation built by January 2026.
OpenAI, CrusoeOpenAI's only footprint figure: one sentence in a CEO blog post
Sam Altman wrote that an average ChatGPT query uses 0.34 Wh and 0.000085 gallons of water. No methodology, no definition of 'average', no third-party review.
OpenAI