#9 of 30 · Model hub & inference platform
Hugging Face
Built the AI Energy Score; hosts a million models on other people's clouds.
In plain words
How scores workHugging Face scores 59 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 · Runs the AI Energy Score leaderboard (166 models, 10 tasks) and the CodeCarbon tracker; the Hub's traffic skews to small models.
- Transparency · Model cards carry a carbon-emissions field; the BLOOM study disclosed ≈50 tCO₂e for a full life cycle; measurement tools are open source.
- Conduct · Built and open-sourced the tools the field uses to measure AI energy, and publishes comparisons that name proprietary models.
What hurts
- Efficiency · No figure for the energy of its own hosted inference.
- Transparency · No corporate inventory or report.
What nobody knows
- Per-prompt water disclosed
- Annual environmental report
- Location-based Scope 2 reported
- Water-use effectiveness reported
- Hardware reuse or e-waste reported
Hugging Face is where open-weight models live: the Hub, Inference Endpoints and Spaces run on AWS, Google Cloud and Azure. Its climate lead Sasha Luccioni co-authored the BLOOM life-cycle study (about 50 tCO₂e for training and running a 176B model), built the CodeCarbon tracker and the AI Energy Score with Salesforce, Cohere and Carnegie Mellon — the benchmark this index leans on. The company publishes no corporate footprint of its own, which is small, and its hosted traffic inherits its clouds' claims.
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
Most Hub inference runs small and mid-size open models: the measured sub-10B median is 0.042 Wh and 70B-class models are about 0.25 Wh. Uncertainty roughly ±3×.
On-site cooling only, so comparable with Google and OpenAI's figures.
How we got this number
0.1 Wh per prompt × 0.15 L/kWh (AWS's disclosed fleet WUE (primary host)). Counting the water evaporated to generate the electricity adds ≈0.18 mL.
Location-based: what the grid actually emitted. Renewable certificates would bring the market-based figure near zero.
How we got this number
0.1 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
7 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
No inventory exists; Hub downloads and hosted inference have grown roughly by half each year.
How we got this number
No disclosure; AWS's fleet figure applies to the capacity it rents there, Google's 1.15 to the rest.
How we got this number
AWS, Google and Microsoft all match 100% of electricity annually; Hugging Face 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
Hugging Face serves from AWS, Google Cloud and Azure rather than its own fleet; a few thousand GPU-equivalents at any time. ±3×.
Megawatts the AI fleet pulls. One megawatt runs roughly 750 US homes.
How we got this number
5,000 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
7 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
5,000 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
5,000 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
7 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
7 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
- Runs the AI Energy Score leaderboard (166 models, 10 tasks) and the CodeCarbon tracker; the Hub's traffic skews to small models. Hugging Face, Salesforce, Cohere, Carnegie MellonMasters of Scale
- No figure for the energy of its own hosted inference. Masters of Scale
- No data of its own; AWS (0.15 L/kWh) and Google (1.15) fleet WUEs apply to hosted capacity. AmazonGoogle (arXiv 2508.15734)
- Model cards carry a carbon-emissions field; the BLOOM study disclosed ≈50 tCO₂e for a full life cycle; measurement tools are open source. Luccioni, Viguier, Ligozat (arXiv 2211.02001)Hugging Face, Salesforce, Cohere, Carnegie Mellon
- No corporate inventory or report. Masters of Scale
- No corporate climate or water targets; commitments are to measurement tooling. Hugging Face, Salesforce, Cohere, Carnegie Mellon
- Built and open-sourced the tools the field uses to measure AI energy, and publishes comparisons that name proprietary models. Hugging Face, Salesforce, Cohere, Carnegie MellonLuccioni, Viguier, Ligozat (arXiv 2211.02001)
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
- New York HQNew York, USA
- Paris officeFrance
- AWS us-east-1Virginia, USA
- Google Cloud us-central1Iowa, USA
Timeline
Incidents and developments involving this company
