#7 of 30 · AI lab + cloud
Microsoft (Copilot / Azure)
Best-in-class water engineering, emissions 58% above its own baseline.
In plain words
How scores workMicrosoft scores 63 out of 100 — middle of the pack. Some disclosure or clean-energy commitments, offset by rising emissions or blind spots. Confidence in this score is good confidence (B).
What helps
- Efficiency · Fleet PUE and hardware efficiency are reported at facility level, and the Phi family targets small-model deployment.
- Carbon · Stopped counting non-additional unbundled renewable certificates, which raised reported Scope 2 from 2% to 13% of the total — a transparency gain that makes the number look worse.
- Carbon · 20-year PPA to restart Three Mile Island Unit 1 for carbon-free baseload from 2028; 40 GW of renewables contracted, 19 GW online.
What hurts
- Efficiency · No per-prompt energy disclosure for Copilot; Azure serves OpenAI's models, whose only public figure (0.34 Wh) comes without methodology.
- Carbon · FY2025 emissions of 20.29 MtCO₂e, up from 16.21 Mt — a 25% rise and roughly 58% above the 2020 baseline; electricity up 24% to 37 TWh.
- Commitments · The company itself says sustainability solutions 'are not scaling fast enough to meet demand'.
What nobody knows
- Per-prompt energy disclosed
- Per-prompt water disclosed
- Training emissions in model cards
Microsoft hosts OpenAI's models and its own Copilot stack on Azure. Its FY2025 emissions reached 20.3 MtCO₂e — up 25% in a year and about 58% above the 2020 baseline it promised to go carbon-negative against by 2030 — while electricity use hit 37 TWh. On water and hardware it leads: fleet WUE of 0.30 L/kWh, zero-water cooling in every design since August 2024, 90.9% of retired servers reused or recycled, and the first year it replenished more than it withdrew.
Key figures
Per-prompt and fleet metrics. Hover a title for what it means; open “How we got this number” for the working.
ecogpt.co estimate; no company figure.
How we got this number
Copilot serves OpenAI's models from Azure. OpenAI's own figure for an average ChatGPT query is 0.34 Wh and Epoch's independent estimate is about 0.3 Wh; Microsoft publishes nothing of its own. Uncertainty roughly ±3×.
On-site cooling only, so comparable with Google and OpenAI's figures.
How we got this number
0.34 Wh per prompt × 0.3 L/kWh (Microsoft's disclosed fleet WUE). Counting the water evaporated to generate the electricity adds ≈0.61 mL.
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).
≈58% above the 2020 baseline.
Hourly matching target for 2030.
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
≈3.2M NVIDIA plus ≈0.3M AMD; Fairwater in Wisconsin alone adds hundreds of thousands of GB200s.
Megawatts the AI fleet pulls. One megawatt runs roughly 750 US homes.
How we got this number
37 TWh a year ÷ 8,760 hours. Average draw implied by annual electricity; peak capacity is higher.
Terawatt-hours per year. One TWh is a year of electricity for about 95,000 US homes.
How we got this number
Up 24% in a year; more than double 2022.
CO₂e emitted making the chips and servers before they answer a single prompt.
How we got this number
3,500,000 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
3,500,000 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
4,224 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
4,224 MW ÷ 2.6 MW per generator, the average size permitted at US data centres. Permit counts are not published fleet-wide; this is the number a fleet this size normally carries.
Share of retired servers and parts given a second life.
How we got this number
3.2 million components handled by six Circular Centers; target met a year early.
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
- No per-prompt energy disclosure for Copilot; Azure serves OpenAI's models, whose only public figure (0.34 Wh) comes without methodology. Sam Altman (blog)
- Fleet PUE and hardware efficiency are reported at facility level, and the Phi family targets small-model deployment. Microsoft
- FY2025 emissions of 20.29 MtCO₂e, up from 16.21 Mt — a 25% rise and roughly 58% above the 2020 baseline; electricity up 24% to 37 TWh. ESG DiveS&P Global Energy
- Stopped counting non-additional unbundled renewable certificates, which raised reported Scope 2 from 2% to 13% of the total — a transparency gain that makes the number look worse. ESG Dive
- 20-year PPA to restart Three Mile Island Unit 1 for carbon-free baseload from 2028; 40 GW of renewables contracted, 19 GW online. World Nuclear NewsS&P Global Energy
- Fleet-wide WUE of 0.30 L/kWh in FY2024, down 39% from 0.49 in 2021; every new design since August 2024 uses zero-water cooling, and Fairwater's closed loop is filled once. MicrosoftMicrosoftConverge Digest
- Replenished more water than it withdrew globally for the first time in FY2025. ESG Dive
- Carbon negative, water positive and zero waste by 2030; 100% of electricity matched by zero-carbon purchases 100% of the time by 2030. Data Center Dynamics
- The company itself says sustainability solutions 'are not scaling fast enough to meet demand'. ESG Dive
- Backed nuclear restarts and additional clean generation rather than on-site gas for new AI capacity. World Nuclear News
Models
What this company ships, and what each prompt costs
No model-level entries — this company hosts or supplies others rather than serving its own models.
Where the compute lives
Named sites from company disclosures and press reporting
- QuincyWashington, USA
- West Des MoinesIowa, USA · GPT-4 training site.
- Mount Pleasant (Fairwater)Wisconsin, USA · 315 acres; hundreds of thousands of GB200s; closed-loop water.
- GoodyearArizona, USA
- San AntonioTexas, USA
- DublinIreland
Timeline
Incidents and developments involving this company
Microsoft's emissions rise 25% to 20.3 Mt, 58% above its 2020 baseline
Data-centre expansion and a decision to stop counting non-additional renewable certificates pushed FY2025 emissions to 20.29 MtCO₂e. Microsoft still calls its 2030 carbon-negative goal achievable.
MicrosoftNebius swaps 400 MW of gas engines for 328 MW of gas fuel cells at Microsoft's Vineland site — permit pending
The New Jersey campus serving Nebius's $17.4–19.4 billion Microsoft contract is designed for up to 300 MW of IT load with only 5 MW from the grid. The original plan for 400 MW of Bergen gas engines became 328 MW of Bloom Energy fuel cells, which still burn natural gas; Cleanview found the generation unbuilt and its air permit pending.
Nebius, MicrosoftMicrosoft signs a 20-year deal to restart Three Mile Island Unit 1
Constellation will bring the ~835 MW reactor back online as the Crane Clean Energy Center, with Microsoft buying its output for data centres from 2028, pending NRC approval.
Microsoft