#14 of 30 · Enterprise AI lab
Cohere
Co-author of the AI Energy Score, with no footprint report of its own.
In plain words
How scores workCohere scores 49 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 mostly estimates (C).
What helps
- Efficiency · Command A (111B dense) serves on two H100s at 156 tokens a second — a deliberately small footprint for a frontier-class model; DeepSeek-V3 needs eight GPUs for the same context length.
- Conduct · No enforcement actions; backs a public efficiency benchmark that lets customers compare models.
What hurts
- Efficiency · No per-prompt energy figure published, despite co-creating the benchmark that measures it.
- Carbon · No procurement or inventory of its own.
- Commitments · No public climate or water targets.
What nobody knows
- Per-prompt energy disclosed
- Per-prompt water disclosed
- Training emissions in model cards
- Annual environmental report
- Location-based Scope 2 reported
- +2 more
Cohere builds the Command family for enterprises, serves it on Oracle Cloud, AWS, Azure and Google Cloud, and is anchor tenant of a multibillion-dollar CoreWeave-built data centre in Canada backed by C$240 million of federal money. It co-created the AI Energy Score benchmark with Salesforce, Hugging Face and Carnegie Mellon, and Command A is engineered to serve on two GPUs — but Cohere discloses no energy, water or emissions figures and has no climate targets.
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
Command A is a 111B dense model that Cohere serves on two H100s; dense models of that size sit a little above Epoch's 0.3 Wh for a typical GPT-4o query. Uncertainty roughly ±3×.
On-site cooling only, so comparable with Google and OpenAI's figures.
How we got this number
0.4 Wh per prompt × 0.3 L/kWh (estimated WUE across OCI, Azure and CoreWeave hosts). Counting the water evaporated to generate the electricity adds ≈0.72 mL.
Location-based: what the grid actually emitted. Renewable certificates would bring the market-based figure near zero.
How we got this number
0.4 Wh × 390 gCO₂/kWh (US grid, location-based).
Location-based estimate for the AI fleet only; not comparable with a full corporate inventory.
How we got this number
26 MW × 8,760 h × 80% utilisation × 390 gCO₂/kWh (US grid). Compute only — nothing else the company does is counted.
How we got this number
The CoreWeave-built site roughly doubles Cohere's dedicated capacity when it opens; no inventory exists to measure against.
How we got this number
No disclosure. Azure reports 0.30 L/kWh and CoreWeave's dense liquid-cooled racks are similar or lower; OCI publishes nothing.
How we got this number
OCI public cloud and AWS match 100% annually; Canada's grid is about 80% non-fossil. Cohere has no procurement of its own.
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
Cohere publishes no fleet size. Anchor tenancy of a multibillion-dollar CoreWeave-built site in Canada (likely 50–100 MW when complete) plus leased capacity on Oracle, AWS, Azure and Google points to tens of thousands of GPU-equivalents. Uncertainty ±3×.
Megawatts the AI fleet pulls. One megawatt runs roughly 750 US homes.
How we got this number
20,000 accelerators × 1.3 kW each, all-in. Fleet power implied by the chip count.
Terawatt-hours per year. One TWh is a year of electricity for about 95,000 US homes.
How we got this number
26 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
20,000 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
20,000 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
26 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
26 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
- Command A (111B dense) serves on two H100s at 156 tokens a second — a deliberately small footprint for a frontier-class model; DeepSeek-V3 needs eight GPUs for the same context length. The Logic
- No per-prompt energy figure published, despite co-creating the benchmark that measures it. Hugging Face, Salesforce, Cohere, Carnegie Mellon
- Hosted on OCI (100% renewable-matched public cloud), AWS and Azure; the Canadian site sits on a grid that is roughly 80% non-fossil. OracleData Center Dynamics
- No procurement or inventory of its own. Data Center Dynamics
- Co-created the AI Energy Score and publishes efficiency ratings for its models under it, but has no environmental report and no training-emissions disclosure. Hugging Face, Salesforce, Cohere, Carnegie MellonSalesforce
- No public climate or water targets. Data Center Dynamics
- No enforcement actions; backs a public efficiency benchmark that lets customers compare models. Hugging Face, Salesforce, Cohere, Carnegie Mellon
Models
What this company ships, and what each prompt costs
| Model | Weights | Energy / prompt | Water | Training CO₂e |
|---|---|---|---|---|
| Command A 111B dense | Open | 0.40 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. | 0.120 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. | 2,000 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
- Toronto HQ / leased cloudOntario, Canada
- CoreWeave-built siteCanada (location undisclosed) · Multibillion-dollar; C$240M federal backing.
- Oracle Cloud regionsUSA
Timeline
Incidents and developments involving this company
