Skip to content

#4 of 30 · AI cloud

Nebius

District heating from Finnish GPUs; a gas-to-fuel-cell pivot in New Jersey.

Amsterdam, Netherlands·nebius.com ·Verified Oct 17, 2026
BGood confidence
Index score
Good confidence

In plain words

How scores work

Nebius scores 64 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 · Portfolio PUE of 1.25 in 2025 (industry average 1.54), with Finland at 1.15 in its best months and 102 GWh of electricity avoided through design.
  • Carbon · 100% renewable electricity in Iceland, France and the UK in 2025; Finland moving from 95% low-carbon to 100% renewable.
  • Water · 0.018 L/kWh at an owned liquid-cooled site versus a projected US average of 0.45–0.49; free cooling in Finland.

What hurts

  • Efficiency · No per-workload energy figures.
  • Carbon · Vineland is designed to run behind the meter on 328 MW of natural-gas fuel cells with just 5 MW from the grid; generation was not yet permitted or operating in early 2026.
  • Conduct · Chose behind-the-meter gas generation at Vineland for speed; permit still pending with NJ DEP.

What nobody knows

  • Per-prompt energy disclosed
  • Per-prompt water disclosed
  • Training emissions in model cards
  • Third-party verified AI figures

Nebius, the AI cloud spun out of Yandex, publishes a full sustainability report: portfolio PUE of 1.25 in 2025 against an industry 1.54, 19.5 GWh of waste heat exported to Mäntsälä's district heating, 100% renewable electricity in Iceland, France and the UK, and a liquid-cooled site at 0.018 L/kWh. Its $17.4–19.4 billion Microsoft deal is served from Vineland, New Jersey, where a plan for 400 MW of gas engines became 328 MW of Bloom fuel cells — still gas-fed — with the air permit pending and only 5 MW from the grid.

Products
Nebius AI Cloud · Token Factory (inference) · Managed Kubernetes
Hosting
Own data centre in Mäntsälä, Finland; colocation in Paris, Iceland and Kansas City; Vineland, New Jersey campus for Microsoft.

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 AIConverge Digest

ecogpt.co estimate; no company figure.

How we got this number

Nebius serves frontier training and inference for Microsoft and others; Epoch's 0.3 Wh for a typical frontier query is the representative workload. 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).
0.03
mL / typical prompt
≈ 0.06 500 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 AIConverge DigestNebius

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

How we got this number

0.3 Wh per prompt × 0.1 L/kWh (estimated fleet WUE (0.018 L/kWh measured at one liquid-cooled site)). 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 AIConverge Digest

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). Vineland, the largest site, will run on gas fuel cells; the Nordic and French sites are renewable.

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.
0.68
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.Epoch AINebius

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

How we got this number

250 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
% (Vineland 50 MW → 300 MW)
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 AICleanview
How we got this number

The Microsoft campus is ramping from 50 MW toward 300 MW of IT load on gas fuel cells, tripling the fleet's emissions even as European sites stay renewable.

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.
0.10
L/kWh (fleet)
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.Nebius
How we got this number

One owned liquid-cooled site measured 0.018 L/kWh; colocation sites in Paris and Kansas City are undisclosed, so the fleet figure is higher.

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
% renewable electricity in Iceland, France and the UK (2025)
Company figureCompany figure. Published by the company itself. We take it as given and record the scope next to it.Nebius

Finland 95% low-carbon; US sites on gas fuel cells and grid.

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
120,000
GPU-equivalents (order of magnitude)
≈ 1,667 racks of 72 GPUs
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 AINebius

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

How we got this number

Epoch puts the Vineland, New Jersey site at about 57,000 H100-equivalents on 50 MW; Finland, Paris, Iceland and Kansas City add a similar amount. ±2×.

Power draw
250
MW (2026)
≈ 187,500 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.Epoch AICleanview

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

How we got this number

Vineland at 50 MW of IT power heading for 300 MW; Mäntsälä, Paris, Iceland and Kansas City together around 200 MW.

Electricity a year
1.75
TWh / year
≈ 166,667 US homes for a year
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.Epoch AICleanviewSemiAnalysis

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

How we got this number

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

Carbon baked into the hardware
61.7
ktCO₂e embodied in chips and servers
≈ 13,413 petrol cars driven for a year
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.Epoch AINebiusarXiv 2607.05408NVIDIA developer blog

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

How we got this number

120,000 accelerators × 514 kg CO₂e each (chip plus share of the host server). Emitted in manufacturing, before the first prompt is served.

Future e-waste
2,040
tonnes of servers to retire within ~5 years
≈ 11 empty Boeing 747s by weight
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.Epoch AINebiusNVIDIAThe Register

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

How we got this number

120,000 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
6,750
tonnes of copper in the build
≈ copper for 81,325 electric cars
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.Epoch AICleanviewBHPS&P Global Market Intelligence

Tonnes of copper in cabling, busbars and transformers.

How we got this number

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

Fossil generators on site
96.0
backup diesel generators (implied)
≈ 250 MW of backup capacity
DerivedDerived. Arithmetic on a disclosed or independent input using a published conversion factor.Epoch AICleanviewLatitude MediaLatitude Media / Floodlight

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

How we got this number

250 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 B
68
  • Portfolio PUE of 1.25 in 2025 (industry average 1.54), with Finland at 1.15 in its best months and 102 GWh of electricity avoided through design. Nebius
  • No per-workload energy figures. Nebius
Carbon intensity — How clean that electricity is, and whether total emissions are rising.weight 25% · confidence B
55
  • 100% renewable electricity in Iceland, France and the UK in 2025; Finland moving from 95% low-carbon to 100% renewable. Nebius
  • Vineland is designed to run behind the meter on 328 MW of natural-gas fuel cells with just 5 MW from the grid; generation was not yet permitted or operating in early 2026. Cleanview
Water — How much water the data centres consume, and whether it is replaced.weight 15% · confidence B
80
  • 0.018 L/kWh at an owned liquid-cooled site versus a projected US average of 0.45–0.49; free cooling in Finland. Nebius
Transparency — How much the company actually publishes.weight 20% · confidence B
70
  • Annual sustainability report with PUE, WUE, heat export, renewable share and site list. Nebius
  • Avoided-water and avoided-electricity figures are design projections, flagged as such. Nebius
Commitments — Whether it has dated, measurable climate and water targets.weight 10% · confidence B
50
  • Commits to renewable and low-carbon electricity across sites; no dated net-zero or water target. Nebius
Conduct — How it behaves: permits, lawsuits, gas turbines, community water.weight 10% · confidence B
55
  • Exports waste heat to Mäntsälä homes, cutting their heating costs about 10%. Nebius
  • Chose behind-the-meter gas generation at Vineland for speed; permit still pending with NJ DEP. Cleanview

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

OperationalConstructionPlanned
  • Amsterdam HQNetherlands
  • MäntsäläFinland · Free cooling; 19.5 GWh of heat to district heating.
  • VinelandNew Jersey, USA · Microsoft; 300 MW planned on gas fuel cells.
  • Paris (Equinix)France · 100% renewable.
  • IcelandIceland · 100% renewable.
  • Kansas CityMissouri, USA

Timeline

Incidents and developments involving this company

NotableAir & power·Feb 1, 2026

Nebius 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, Microsoft