Counting factories tells you almost nothing about emissions

Mapping every heavy-industrial site on Earth and then attaching an emissions estimate to each one produces a result that inverts the intuition almost exactly.

Analysis by Inzonex · published 2026-09-25 · counted from published open data · how we work

Key figures

71.1% of mapped industrial sites account for 1.6% of the estimated emissions.

5.9% of sites account for 57.5%.

Per site, a cement plant is estimated to emit about 61 times what a food and beverage plant does.

The inversion

IndustryAtlas maps 30,533 industrial facilities worldwide, 30,363 of which carry an emissions estimate, totalling about 10,918 Mt CO₂ a year.

Sort those facilities by how many there are and you get one answer. Sort them by what they emit and you get almost the opposite one.

Food and beverage, textiles71.1Oil and gas fields, steel5.9
Share of industrial sites (percent of mapped sites).
Food and beverage, textiles1.6Oil and gas fields, steel57.5
Share of estimated emissions (percent of estimated emissions).

Food and beverage and textiles are 71.1% of the sites and 1.6% of the emissions. Oil and gas fields and steel plants are 5.9% of the sites and 57.5% of the emissions.

Every sector, both ways

SectorSites% of sitesMt CO₂ a year% of emissions
Food & Beverage11,98739.3%144.41.3%
Textiles9,72131.8%31.80.3%
Cement Plant2,1607.1%1,589.614.6%
Lime Plant1,5615.1%426.03.9%
Steel Plant9533.1%2,316.121.2%
Chemical Plant8532.8%606.55.6%
Oil & Gas Field8472.8%3,966.036.3%
Refinery7382.4%975.48.9%
Metals Plant4181.4%37.90.3%
Other Manufacturing3981.3%17.40.2%
Pulp & Paper Mill3441.1%92.20.8%
Aluminium Smelter3081.0%413.03.8%

The extreme case sits at the top and bottom of that table. Textiles are 31.8% of the world's mapped industrial sites and 0.3% of the emissions attached to them. Oil and gas fields are 2.8% of the sites and 36.3%.

Why the count misleads

The reason is temperature and chemistry, not size or effort. A brewery, a dairy and a dye house all run on steam at a few hundred degrees, and steam at that temperature is a solvable problem: it can be recovered, insulated, electrified or produced by a heat pump.

A cement kiln runs above 1,400 °C and releases carbon dioxide from the limestone itself, before any fuel is burned. A blast furnace uses carbon as the chemical reducing agent, not merely as a heat source. Those emissions do not go away when the energy supply gets cleaner, because they were never only about energy.

Per site, the gap is roughly 61 to one between a cement plant and a food and beverage plant.

Why this matters for policy and AI

Two practical consequences follow, and they point in opposite directions.

Decarbonisation effort is badly served by site counts. A programme that reaches the most factories reaches food and textiles, which is where most of the employment and most of the practical energy saving sits. A programme that reaches most of the emissions has to reach a few thousand sites that are harder, costlier and concentrated in a handful of countries. These are different programmes and conflating them wastes both.

For the AI power argument, the industrial base is already there. New demand does not land on an empty network. It lands on one that is already serving 30,533 mapped industrial sites, many on the same substations. And the flexibility of that existing load is not evenly distributed either: a fermentation schedule or a continuous kiln cannot move when an AI campus arrives on the same feeder.

What these numbers are and are not

The emissions figures are Climate TRACE estimates. They are modelled from satellite and other observation, not metered at the stack, and should be read as estimates throughout. The facility counts are counts of mapped sites, which is a coverage figure: sectors with good public registries are mapped more completely than sectors without them.

Two consequences for reading the table. Textiles and food are numerous partly because they are well documented, so their share of sites is a statement about data as well as about the world. And an estimate attached to a single facility can be well out; the pattern across thousands of facilities is far more robust than any one row.

None of these numbers measure electricity. They measure carbon dioxide and site counts, which is why nothing here should be read as a claim about connected load.

Sources and method

  • Facility counts, sectors and locations from IndustryAtlas, counted on 2026-09-25.
  • Emissions estimates for 30,363 of the mapped facilities come from Climate TRACE, which models facility-level greenhouse gas emissions from satellite and other observational data. They are estimates, not metered measurements.
  • Where facilities carry regulator-reported emissions instead, those come from the US EPA Greenhouse Gas Reporting Program and the EU Emissions Trading System transaction log.
  • Counts are of mapped facilities and do not represent connected electrical load or consumption.