One Load, Two Footprints: What Hyperscaler Scope 2 Figures Depend On

Two numbers for one load

Scope 2 covers the emissions from electricity a company buys. The GHG Protocol has companies report it on two bases, because the two views are not interchangeable⁵. The location-based figure applies the average emissions of the grids where power was actually consumed to metered use. The market-based figure instead reflects the contracts the company holds, chiefly energy attribute certificates and power purchase agreements, which it matches against its consumption.

For a planner, the practical consequence is that a provider's emissions figure is not one number. A planner reading a provider's emissions figure is therefore choosing a basis before reading any physical fact.

Three operators, both bases

The table sets out what the operators' own documents and one secondary report show. Fiscal periods differ (Microsoft's year ends 30 June; Meta's and Google's are calendar years), and the bases differ, so the rows are not summed or compared as totals.

All four rows are operator-reported figures. None is an independent measurement.

Dividing the reported totals by the reported electricity gives implied intensities. These are derived by this brief, not published by the operators. Microsoft's FY2025 location-based figure implies about 0.325 tCO2e per MWh and Meta's 2024 figure about 0.324, close to each other, as one would expect of grid-average factors. The market-based figures diverge: about 0.073 tCO2e per MWh for Microsoft in FY2025, against about 0.00007 for Meta in 2024. Meta's location-based total is roughly 4,400 times its market-based total. Meta states that it procures and retires one certificate for every MWh of electricity used².

Microsoft's tenfold move

Microsoft's own two years show how far the market-based figure can travel while the physical load does not. Between FY2024 and FY2025, electricity rose 24% and location-based Scope 2 rose 21%. Market-based Scope 2 rose from 259,090 to 2,707,428 tCO2e, a factor of 10.4. In FY2024 the fact sheet's matched renewable electricity equalled total electricity; in FY2025 it was 34,534,392 of 37,026,353 MWh, about 93%¹.

The fact sheet also records that Microsoft discontinued purchasing spot certificates in February 2025, and that these are excluded from the FY2025 matched figure¹. The coincidence with the jump is plain, but the fact sheet does not itemise what produced the market-based increase, so the link is an inference and not a statement by Microsoft.

Market-based Scope 2 is an outcome of purchasing policy. It can move by an order of magnitude in a year in which the physical load moved by a quarter.

Google and Amazon: what could and could not be read

For Google, The Planet Brief reports that the company's 2026 Environmental Report, published on 30 June 2026 and covering 2025, shows market-based Scope 2 of 2.815 million tCO2e, down 3% while electricity rose 37%, and location-based Scope 2 that "rose to 15.15 million tonnes"⁵. Google's own report could not be read as text in this work, so these figures are secondary. The Planet Brief also notes that Google matched 100% of its electricity with renewable purchases on a global annual basis, which does not mean every facility ran on carbon-free power every hour⁵. A 2025 commentary in Internet Policy Review gives a 2024 Google market-based figure of 3,059,100 tCO2e⁶; set against 2.815 million, that implies a fall larger than 3%. The two secondary sources are not reconciled here, so the year-on-year change should be taken from Google's report.

For Amazon, the carbon methodology states that Amazon "reports emissions according to GHGp's market-based method" for Scope 2 and Scope 3 as applicable³. No location-based Scope 2 figure appeared in the documents read, and the report's data tables could not be reached, so this brief does not claim that none is published. Amazon's total footprint rose from 69.55 to 80.85 million tCO2e between 2024 and 2025, but that total is calculated on a market-based method including environmental attribute credits and is not a Scope 2 figure⁴.

The rule is being rewritten

The GHG Protocol consulted from 20 October 2025 to 31 January 2026 on counting certificates in the market-based method only if they are matched to hourly consumption and connected to deliverable grid regions⁷. Its standards board has since leaned towards two market-based methods, one with hourly matching, deliverability and incrementality, the other keeping the 2015 requirements, with a consolidated consultation slated for the second quarter of 2027 and publication for the fourth quarter of 2028⁷. None of this is final. Figures reported under the present rules should therefore be read as figures under a method that is itself under revision.

What is not established

Nothing here is independently metered. Assurance differs by operator: a Deloitte review engagement for Microsoft's FY2025 information¹, and limited assurance from EY for select Meta metrics². Clarity AI's June 2026 analysis finds, at sector level only, that data-centre-dependent software firms report Scope 2 emissions 76% below their grid-based consumption, using a median ratio over a consistent panel; it gives no per-company figures⁸. No certificate prices or procurement costs appear in any source read, so the cost of moving a market-based figure is unknown. The brief also cannot say what produced the residual market-based emissions at Microsoft or why Meta's residual is near zero.

What a planner should take from this

Treat a market-based Scope 2 figure as a statement about contracts and a location-based figure as a statement about grids. Before using any provider's number, ask for both for the same period, the share of consumption matched and by which instrument, and the matching period and grid boundary. Use the location-based figure to estimate the physical grid burden of a workload. Use the market-based figure only where your own reporting framework allows contract-based claims, and expect that allowance to tighten. If a provider supplies only a market-based figure, as Amazon's methodology describes, that figure cannot be reconciled to the grid load it is meant to represent.