#8 of 30 · Chipmaker
NVIDIA
Runs its own operations on 100% renewables; its chips are the industry's footprint.
In plain words
How scores workNVIDIA 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 · Publishes embodied carbon intensity per exaflop (0.66 g for HGX H100 vs 0.50 g for HGX B200) and claims up to 25× better inference efficiency per generation.
- Carbon · 100% renewable electricity for offices and data centres under operational control in FY2025 (from 76%); SBTi-validated 50% Scope 1+2 cut by FY2030.
- Transparency · Full sustainability report with limited assurance on the renewable share, product carbon footprints per GPU and Scope 3 by category.
What hurts
- Efficiency · Efficiency gains are per unit of compute; absolute power keeps climbing, with a GB200 NVL72 rack drawing around 120 kW.
- Carbon · Scope 3 jumped from 3.5 to 6.9 MtCO₂e in a year, mostly purchased goods — chip and system manufacturing.
- Carbon · The electricity consumed by GPUs after sale is the real footprint: de Vries projected 85–134 TWh a year from NVIDIA servers by 2027.
What nobody knows
- Per-prompt energy disclosed
- Per-prompt water disclosed
- Water-use effectiveness reported
NVIDIA makes the accelerators behind almost every company in this index. Its own offices and data centres reached 100% renewable electricity in FY2025 and it publishes a product carbon footprint for the H100 (about 164 kg CO₂e per GPU), but Scope 3 emissions doubled to 6.9 MtCO₂e in a year as Blackwell shipments ramped, and the lifetime electricity its sold GPUs consume — which de Vries projected at 85–134 TWh a year by 2027 — dwarfs anything on its own books.
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
NVIDIA's Nemotron models are frontier-class and served through NIM on H100 and B200 systems; Epoch's 0.3 Wh for a comparable GPT-4o query is the anchor. Uncertainty roughly ±3×.
On-site cooling only, so comparable with Google and OpenAI's figures.
How we got this number
0.3 Wh per prompt × 0.3 L/kWh (estimated WUE for NVIDIA's own and partner-hosted clusters). Counting the water evaporated to generate the electricity adds ≈0.54 mL.
Location-based: what the grid actually emitted. Renewable certificates would bring the market-based figure near zero.
How we got this number
0.3 Wh × 390 gCO₂/kWh (US grid, location-based). NVIDIA's own sites are 100% renewable-matched, so its market-based figure is near zero.
Scope 1: 13 kt. Scope 2 (market-based): 0.
Driven by purchased goods as Blackwell ramped.
How we got this number
No disclosure. NVIDIA's own clusters are liquid-cooled and DGX Cloud runs in partner hyperscale sites that report 0.15–0.30 L/kWh.
Mix of on-site renewables, utility tariffs, certificates and PPAs.
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
Eos alone is 4,608 H100s and NVIDIA leases DGX Cloud capacity from partners; its own fleet is small next to what it ships — analysts estimated 1–1.5 million Hopper GPUs a quarter at the 2024 peak.
Megawatts the AI fleet pulls. One megawatt runs roughly 750 US homes.
How we got this number
0.82 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
821,200 MWh, up more than 200,000 MWh in a year.
CO₂e emitted making the chips and servers before they answer a single prompt.
How we got this number
50,000 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
50,000 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
94 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
94 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.
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
- Publishes embodied carbon intensity per exaflop (0.66 g for HGX H100 vs 0.50 g for HGX B200) and claims up to 25× better inference efficiency per generation. NVIDIA developer blog
- Efficiency gains are per unit of compute; absolute power keeps climbing, with a GB200 NVL72 rack drawing around 120 kW. The Register
- 100% renewable electricity for offices and data centres under operational control in FY2025 (from 76%); SBTi-validated 50% Scope 1+2 cut by FY2030. Data Center DynamicsNVIDIA
- Scope 3 jumped from 3.5 to 6.9 MtCO₂e in a year, mostly purchased goods — chip and system manufacturing. Data Center Dynamics
- The electricity consumed by GPUs after sale is the real footprint: de Vries projected 85–134 TWh a year from NVIDIA servers by 2027. Cell Press via EurekAlert
- No fleet WUE; its own data-centre water use is small. Upstream, TSMC — which fabricates every NVIDIA data-centre GPU — consumed 129 million m³ of water in 2024. TSMC
- Full sustainability report with limited assurance on the renewable share, product carbon footprints per GPU and Scope 3 by category. NVIDIAData Center Dynamics
- No absolute figure for the lifetime energy of sold products, only an intensity target. Data Center Dynamics
- SBTi: 50% absolute Scope 1+2 cut by FY2030; 75% cut in Scope 3 intensity of sold GPUs per PFLOP by FY2030; suppliers covering 80% of supply-chain emissions engaged. Data Center DynamicsNVIDIA
- No absolute Scope 3 target, which is where 99% of its emissions sit. Data Center Dynamics
- No environmental enforcement actions; publishes product carbon footprints that let customers count embodied emissions. NVIDIA developer blog
Models
What this company ships, and what each prompt costs
| Model | Weights | Energy / prompt | Water | Training CO₂e |
|---|---|---|---|---|
| Nemotron (Ultra / Super) 49B–253B (Llama-derived) | Open | 0.30 Wh 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. | 0.090 mL 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. | 5,000 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
- Santa Clara (Eos / Selene)California, USA · Research clusters; 4,608 H100s in Eos.
- TSMC fabsHsinchu / Tainan, Taiwan · Where every data-centre GPU is made; 129 Mm³ of water in 2024.
- TSMC Fab 21Arizona, USA · Blackwell wafers from late 2025.
Timeline
Incidents and developments involving this company
NVIDIA pays a reported $20 billion to license Groq's inference chips and hire its founders
NVIDIA confirmed a non-exclusive licensing deal for Groq's LPU inference technology, with Groq's CEO and senior team joining NVIDIA; CNBC put the price at $20 billion. Groq had claimed a tenth of the energy per token of GPUs without publishing a measurement, and keeps operating independently.
NVIDIA, GroqNVIDIA hits 100% renewable electricity as its Scope 3 emissions double to 6.9 Mt
NVIDIA's FY2025 sustainability report shows its own offices and data centres fully matched with renewables and Scope 2 at zero — while emissions from making its chips jumped from 3.5 to 6.9 MtCO₂e as Blackwell ramped. The electricity its sold GPUs will consume is not counted in absolute terms.
NVIDIAGenerative AI could produce 2.5 million tonnes of e-waste a year by 2030 — a thousandfold increase
A Nature Computational Science study by Peng Wang and colleagues projects 1.2–5 million tonnes of accumulated e-waste from generative AI by 2030, using the NVIDIA DGX H100 server as its benchmark. Circular measures — longer server lives, component reuse — could cut it by up to 86%.
NVIDIA