Skip to content

#19 of 30 · AI lab

Anthropic (Claude)

A safety-first brand with no environmental report — and gigawatts on order.

San Francisco, California·anthropic.com ·Verified Oct 16, 2026
DLittle data
Index score
Little data

In plain words

How scores work

Anthropic scores 42 out of 100 — laggard. Little or no disclosure, fossil-backed capacity, or active disputes with regulators and communities. Confidence in this score is little data (D).

What helps

  • Nothing on record.

What hurts

  • Efficiency · No per-prompt, per-model or fleet energy data; Claude's frontier models are closed-weight so independent benchmarks cannot measure them.
  • Carbon · The $50bn Fluidstack build-out was announced without naming the sites' electricity source.
  • Transparency · No environmental report, no model-level disclosures, no data-centre locations beyond what partners publish.

What nobody knows

  • Per-prompt energy disclosed
  • Per-prompt water disclosed
  • Training emissions in model cards
  • Annual environmental report
  • Location-based Scope 2 reported
  • +3 more

Claude is trained and served on AWS Trainium (Project Rainier, ≈500,000 chips) and Google Cloud TPUs, with up to one million TPUs and 'well over a gigawatt' arriving in 2026, up to 1 GW on Azure, and $50 billion of its own data centres in Texas and New York with Fluidstack. Anthropic has published no environmental disclosure of any kind — no per-prompt figures, no training emissions, no annual inventory, no power source for the new sites — so its score borrows from its hosts' clean-energy claims and is capped by the absence of data.

Products
Claude · Claude Code · Claude API
Hosting
AWS (Trainium2, Project Rainier) and Google Cloud TPUs today; Azure (up to 1 GW) and own Fluidstack-built sites in Texas and New York from 2026.

Key figures

Per-prompt and fleet metrics. Hover a title for what it means; open “How we got this number” for the working.

How to read the labels:Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.Measured independentlyMeasured independently. Measured or estimated by researchers or analysts with no stake in the result.ecogpt.co 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.DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.Not disclosedNot disclosed. The company publishes nothing on this and no defensible estimate exists.— hover any label.
Energy per promptElectricity used to answer one typical request. Under 0.3 Wh is good. A 10 W LED bulb uses 0.3 Wh in about two minutes.
0.30
Wh / typical prompt
≈ 25.0 smartphone charges per 1,000 prompts
ecogpt.co 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.Epoch AI

ecogpt.co estimate; no company figure.

How we got this number

Claude's frontier models are closed and unmeasured. They are comparable in class to GPT-4o, for which Epoch independently estimates about 0.3 Wh for a typical query. Uncertainty roughly ±3×.

Water per promptWater evaporated cooling the servers for one request. Scope matters: on-site cooling only, or also the water used to generate the electricity (which adds ≈1.8 mL per Wh).
0.05
mL / typical prompt
≈ 0.09 500 mL bottles per 1,000 prompts
ecogpt.co 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.Epoch AIAmazon

On-site cooling only, so comparable with Google and OpenAI's figures.

How we got this number

0.3 Wh per prompt × 0.15 L/kWh (AWS's disclosed fleet WUE, where Claude is mostly served). Counting the water evaporated to generate the electricity adds ≈0.54 mL.

Carbon per promptGreenhouse gas from one request on the grid where it actually runs. Below 0.1 g is clean. A kilometre in a petrol car is about 170 g.
0.117
gCO₂e / typical prompt
≈ 0.69 km driven (petrol car) per 1,000 prompts
ecogpt.co 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.Epoch AI

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).

Annual emissionsThe company's whole yearly footprint, in million tonnes of CO₂e. Clouds and device makers dwarf pure labs because they also count shops, logistics or products. Compare the trend, not the size.
1.23
MtCO₂e / year (AI compute only)
ecogpt.co 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.AmazonAnthropic

Location-based estimate for the AI fleet only; not comparable with a full corporate inventory.

How we got this number

450 MW × 8,760 h × 80% utilisation × 390 gCO₂/kWh (US grid). Compute only — nothing else the company does is counted.

Emissions trendDirection of travel against last year or the company's own baseline. Negative is good. Positive means the AI build-out is outrunning clean energy.
+200
% (2025 → 2026 contracted capacity)
ecogpt.co 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.AnthropicPure AI
How we got this number

From ≈450 MW on Rainier to well over 1 GW of TPUs plus up to 1 GW on Azure during 2026 — roughly a tripling of compute, and of the emissions behind it.

Water per kWh (WUE)Litres of water consumed per kilowatt-hour of server electricity, across the fleet. 0.15 is excellent, 0.3 is good, 1.8 is the US data-centre average.
0.15
L/kWh (inherited from AWS)
ecogpt.co 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.Amazon
How we got this number

Anthropic publishes no water data; AWS's 0.15 L/kWh applies to the Rainier capacity that serves most Claude traffic.

Clean electricityShare of electricity matched with renewable or carbon-free sources. Annual matching counts certificates bought anywhere; hourly matching on the same grid is far stricter.
100
% annual renewable matching (inherited from hosts)
ecogpt.co 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.AmazonGoogle
How we got this number

AWS and Google both match 100% of electricity annually; Anthropic has no procurement of its own and the Fluidstack sites' supply is undisclosed.

Hardware & materials

The chips, power, copper and future e-waste behind the fleet — the part sustainability reports leave out

Factors we use
AI chips in service
500,000
Trainium2 chips (Project Rainier, AWS)
≈ 6,944 racks of 72 GPUs
Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.AmazonAnthropicPure AI

GPUs, TPUs or custom accelerators the company runs. Analyst figures are in H100-equivalents, company figures in physical chips.

How we got this number

Plus up to one million Google TPUs ('well over a gigawatt' in 2026) and up to 1 GW of Azure GPUs contracted in late 2025.

Power draw
450
MW (2025)
≈ 337,500 US homes' worth of power
ecogpt.co 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.AmazonAnthropicPure AIIndustryWeek

Megawatts the AI fleet pulls. One megawatt runs roughly 750 US homes.

How we got this number

≈500,000 Trainium2 chips at about 0.9 kW each all-in. Contracts signed in late 2025 add well over 1 GW of TPUs and up to 1 GW on Azure during 2026, plus $50bn of own data centres with Fluidstack.

Electricity a year
3.15
TWh / year
≈ 300,000 US homes for a year
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.AmazonAnthropicPure AIIndustryWeekSemiAnalysis

Terawatt-hours per year. One TWh is a year of electricity for about 95,000 US homes.

How we got this number

450 MW × 8,760 h × 80% utilisation. Compute fleet only.

Carbon baked into the hardware
257
ktCO₂e embodied in chips and servers
≈ 55,870 petrol cars driven for a year
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.AmazonAnthropicPure AIarXiv 2607.05408NVIDIA developer blog

CO₂e emitted making the chips and servers before they answer a single prompt.

How we got this number

500,000 accelerators × 514 kg CO₂e each (chip plus share of the host server). Emitted in manufacturing, before the first prompt is served.

Future e-waste
8,500
tonnes of servers to retire within ~5 years
≈ 47 empty Boeing 747s by weight
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.AmazonAnthropicPure AINVIDIAThe Register

Tonnes of server hardware that will be retired in three to five years.

How we got this number

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.

Copper in the build
12,150
tonnes of copper in the build
≈ copper for 146,386 electric cars
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.AmazonAnthropicPure AIIndustryWeekBHPS&P Global Market Intelligence

Tonnes of copper in cabling, busbars and transformers.

How we got this number

450 MW × 27 t of copper per MW (cabling, busbars, transformers).

Fossil generators on site
173
backup diesel generators (implied)
≈ 450 MW of backup capacity
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.AmazonAnthropicPure AIIndustryWeekLatitude MediaLatitude Media / Floodlight

Diesel backup units or gas turbines at the data centres, permitted or observed.

How we got this number

450 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.

Hardware reused or recycled
—
No hardware reuse or e-waste figure published.
Not disclosedNot disclosed. The company publishes nothing on this and no defensible estimate exists.

Share of retired servers and parts given a second life.

Mineral depletion
—
Only Mistral has published a mineral-depletion figure.
Not disclosedNot disclosed. The company publishes nothing on this and no defensible estimate exists.

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

Energy efficiency — How much electricity one answer costs.weight 20% · confidence D
45
  • No per-prompt, per-model or fleet energy data; Claude's frontier models are closed-weight so independent benchmarks cannot measure them. Amazon
Carbon intensity — How clean that electricity is, and whether total emissions are rising.weight 25% · confidence C
52
  • Compute comes from AWS (100% renewable matching, net-zero 2040) and Google Cloud (24/7 CFE goal); Anthropic itself reports nothing. AmazonAmazon
  • The $50bn Fluidstack build-out was announced without naming the sites' electricity source. IndustryWeek
Water — How much water the data centres consume, and whether it is replaced.weight 15% · confidence C
60
  • Rainier sites in Indiana use outside-air cooling with AWS reporting 0.15 L/kWh; no Anthropic-specific data. Amazon
Transparency — How much the company actually publishes.weight 20% · confidence D
12
  • No environmental report, no model-level disclosures, no data-centre locations beyond what partners publish. AmazonIndustryWeek
Commitments — Whether it has dated, measurable climate and water targets.weight 10% · confidence C
25
  • No public climate, energy or water targets of its own. Amazon
Conduct — How it behaves: permits, lawsuits, gas turbines, community water.weight 10% · confidence C
60
  • No environmental enforcement actions on record; its CEO co-signed the 2023 extinction-risk statement that Fortune reports both labs have since softened. Fortune

Models

What this company ships, and what each prompt costs

ModelWeightsEnergy / promptWaterTraining CO₂e
Claude (Opus / Sonnet / Haiku)
Frontier (undisclosed)
Closed0.30 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.0.045 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.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

OperationalConstructionPlanned
  • Project Rainier (AWS)Indiana, USA · ≈500,000 Trainium2 chips.
  • Google Cloud TPU regionsMultiple · Up to 1M TPUs in 2026.
  • Fluidstack site, TexasTexas, USA · Part of a $50bn programme; power source undisclosed.
  • Fluidstack site, New YorkNew York, USA

Timeline

Incidents and developments involving this company

NotableTransparency·Nov 12, 2025

Anthropic commits $50 billion to US data centres — and 2 GW of TPUs and Azure — without naming a power source

In three weeks Anthropic announced up to one million Google TPUs ('well over a gigawatt' in 2026), $50 billion of Fluidstack-built data centres in Texas and New York, and up to 1 GW on Azure. None of the announcements said where the electricity would come from.

Anthropic