Market-based or location-based? Meta's two emissions totals, explained

ecogpt.coPublished Nov 2, 2026 · 2 min read

Meta reported 8.2 million tonnes of CO₂e for 2024 under one method and 15.6 million under another. Both are in the same report. Here is what the two methods count and why the index shows both.

Location-based Scope 2 uses the real average mix of the grid where the electricity was used.
Market-based Scope 2 lets a company subtract renewable certificates and power-purchase agreements it has bought, anywhere.
Meta's 2024 footprint is 15.6 Mt location-based and 8.2 Mt market-based; the gap is certificates and contracts.
Microsoft stopped counting non-additional certificates in 2025 and its reported Scope 2 rose from 2% to 13% of the total.

One company, two totals

Meta's 2025 environmental data index lists its 2024 greenhouse-gas emissions as 8.2 million tonnes of CO₂e on a market basis and 15.6 million tonnes on a location basis. Nothing is hidden. The two figures answer different questions about the same electricity.

Location-based: what the grid emitted

Scope 2 covers emissions from purchased electricity. The location-based method multiplies each megawatt-hour by the average carbon intensity of the grid where it was consumed. A data centre in Iowa gets Iowa's mix, coal and gas included. It is the physical picture: what actually came out of smokestacks to run the servers.

Market-based: what the company bought

The market-based method lets a company replace the grid average with the contracts it holds. A power-purchase agreement with a wind farm, or an unbundled renewable energy certificate, counts as zero-carbon electricity against the company's consumption, even if the wind farm is in another state and the data centre was running on gas that night. Meta has matched 100% of its electricity this way every year since 2020, which is how 15.6 million tonnes becomes 8.2.

Why the gap matters for AI

Both methods are legitimate under the GHG Protocol, and companies must report both. The trouble starts when one number travels without the other. Meta's Llama 4 model cards report location-based training emissions (1,354 tonnes for Scout, 645 for Maverick) and market-based emissions of zero. The zero is an accounting result, not an absence of combustion.

Microsoft offers the clearest demonstration of what the choice is worth. In its 2026 report it stopped counting "non-additional" unbundled certificates, the kind that do not cause new clean generation to be built. Reported Scope 2 rose from 2% to 13% of its total, and the headline footprint jumped to 20.3 million tonnes. The electricity had not changed. The accounting had.

The stricter standard: hourly matching

Google publishes a third figure that neither method captures: the share of its electricity matched with carbon-free sources in the same hour on the same grid. It was 66% in 2024. Google, Microsoft and others have set 2030 goals to reach 100% on that basis. Hourly matching is far harder than annual matching, because it cannot be met with daytime solar certificates bought to cover night-time load.

How to read the index

Company profiles show both totals where both are reported.
The clean-electricity figure on each profile says which kind of matching it is.
Estimated annual emissions for companies that publish nothing are location-based, so they are not comparable with a market-based corporate total.

If a company tells you its AI runs on 100% renewable energy, ask which method, and which hour.