← Back to setup comparison

ROBOT SETUP ADVISOR / EVIDENCE

CO₂ sources & limits

Checked 23 October 2026. Published data, modeled electricity use and unsupported manufacturing assumptions are separate inputs.

What the total means: a planning scenario combining estimated electricity emissions and generic manufacturing assumptions. It is not a verified life-cycle assessment. Finding a report for one product does not validate other hardware in the catalog.

1. Electricity · used in calculations

Ember Yearly Electricity Data, 2025, Total generation rows. Values below are transcribed from the global CSV without rounding in the calculations.

Areag CO₂e/kWhData year
France41.2242025
Germany334.1172025
United Kingdom218.0012025
EU210.5362025
United States384.4032025
China528.2782025
India670.9452025
Japan473.7352025
Norway28.0502025
Canada191.5872025

These are annual generation-mix factors on a lifecycle basis. They include upstream and power-generation infrastructure impacts. They do not describe your electricity supplier, time of use or imported electricity mix and are not corporate Scope 2 reporting factors.

Electricity emissions = modeled kWh × g CO₂e/kWh ÷ 1,000. We hold the factor constant for the selected years; this is not a forecast. Robot and GPU power and training duration remain estimates. Cloud calculations assume the US generation mix and PUE 1.2; neither is verified for the selected provider.

Download the exact source rows and provenance (JSON) · Ember methodology, pp. 15–17

Source CSV modified 19 October 2026; retrieved 23 October 2026. Credit: Ember, CC BY 4.0. The JSON records the original file’s SHA-256 for auditing later revisions.

2. Robot manufacturing · a research reference

Stuhlenmiller et al., Robotics (2021), Table 2, estimates 462.2 kg CO₂e for production of a UR5 system, including 149.3 kg for its control box. The authors model component materials using ecoinvent 3.7; it is a research estimate, not a manufacturer-certified footprint.

This is the older UR5, not the UR5e listed here. We have not applied it to UR5e, other arms or humanoids, or used it to justify the existing 12 kg CO₂e/kg mass factor. It is evidence for a product-specific modeling approach.

Open-access paper, p. 9

3. GPU manufacturing · a reviewed manufacturer report

NVIDIA HGX H100 Product Carbon Footprint, p. 3 (2025): 1,312 kg CO₂e from raw materials through manufacturing for one baseboard containing eight H100 SXM GPUs. The report was reviewed against ISO 14067 and excludes use and end-of-life.

This is not a complete server or a single PCIe H100, A100 or RTX card. It does not validate the advisor’s generic 160 kg/GPU or 2,500 kg/server factors. We list it as a reference, not as a substitute for unmatched products.

4. Computers · use the exact configuration

Dell’s product carbon footprint library provides reports for individual desktops, laptops and servers. Match the model and configuration, then use the appropriate manufacturing stage. Do not add a whole-life footprint to separately calculated operating electricity: that would count use twice.

The advisor’s generic laptop and desktop entries do not identify a matching Dell configuration. Its 300 kg/PC and 30 kg/embedded-board values therefore remain assumptions.

What is still missing

Already-owned hardware adds no new manufacturing in this incremental purchase scenario; that does not mean its historical footprint was zero. Shared cloud hardware manufacturing is not allocated here. Treat rankings as scenarios, especially when hardware assumptions dominate.