Recommended Free Tools
Short answer: probably, for a specific comparison. A Guardian analysis by Isabel O’Brien estimated that the location-based emissions of in-house data centers operated by Google, Microsoft, Meta and Apple were 662% higher (7.62 times) than the companies’ officially reported figures for 2020–2022. That is an estimate of electricity-related accounting for four companies’ own facilities—not a direct measurement of every data center, all environmental impacts, or AI workloads alone.
What the 662% figure actually compares
The Guardian investigation, published on 15 September 2024 and amended on 18 September, compared two ways of calculating Scope 2 electricity emissions:
- Market-based emissions: figures linked to electricity contracts and qualifying instruments such as energy attribute certificates or renewable-energy certificates (RECs).
- Location-based emissions: emissions calculated from the average carbon intensity of the electricity grid where a facility actually uses power.
The 662% estimate covers 2020–2022 and only the companies’ in-house data centers. It does not establish that the firms violated reporting rules or intentionally misled the public. The difference largely reflects which accounting method is emphasized and what facilities can be identified from public disclosures.
Why market-based and location-based numbers diverge
Location-based accounting follows the local grid
Location-based Scope 2 accounting applies the average emissions intensity of the grid serving a facility. A data center running on a carbon-intensive grid therefore records higher emissions even if its owner buys certificates associated with renewable generation elsewhere.
#1 Best Overall
Market-based accounting follows contractual choices
Market-based accounting reflects electricity a company has purposefully chosen through contracts and qualifying instruments. An energy attribute certificate conveys the environmental attributes of generated electricity; it is not the electricity itself. Certificates may be issued and retired under rules covering market boundaries, tracking and timing, but the associated generation need not be physically delivered to the data center at the same time.
The Greenhouse Gas Protocol Scope 2 Guidance calls for companies operating in markets with qualifying contractual data to report both methods. The two figures answer different questions: the location-based number describes the grid supplying the facility, while the market-based number describes the contractual procurement claim.
The company examples reported by the Guardian
The investigation found especially large gaps in 2022 figures for Meta and Microsoft. These are the values reported in the Guardian analysis, not independent measurements of every facility those companies use.
| Company and year | Market-based figure | Location-based figure | Boundary and qualification |
|---|---|---|---|
| Meta, 2022 | 273 metric tons CO₂e | More than 3.8 million metric tons CO₂e | In-house data-center Scope 2 figures reported by the Guardian |
| Microsoft, 2022 | 280,782 metric tons CO₂e | 6.1 million metric tons CO₂e | Data-center-related figures reported by the Guardian |
The investigation said Google and Apple did not publish directly comparable, data-center-specific location-based figures in the material it examined. For those companies, the Guardian inferred likely data-center differences from broader Scope 2 comparisons. Those values should therefore be read as estimates rather than company disclosures.
Rank #3
Where Amazon fits—and where it does not
Amazon was excluded from the headline 662% calculation because its data-center-specific emissions could not be isolated consistently. The Guardian separately described Amazon as the largest emitter among the five companies, but that statement is not part of the four-company 662% result.
The investigation also reported a separate 275% (3.75-times) aggregate difference for the five companies over 2020–2022. That calculation used Amazon’s official Scope 2 figures for 2020 and 2021 because location-based figures were unavailable. It is a different population and methodology from the in-house-data-center estimate.
What the estimate leaves out
- AI-only emissions: the underlying years, 2020–2022, predate much of the later generative-AI data-center expansion. The calculation is not an AI workload estimate or a forecast of current emissions.
- Third-party and colocation facilities: companies may rent capacity or use cloud providers, making it difficult to assign electricity emissions between operator and customer. The Guardian said Scope 3 treatment is uncertain and noted that Apple’s third-party cloud-contract emissions were absent from the cited totals.
- Other environmental impacts: the quantified comparison concerns electricity-related greenhouse-gas accounting. It does not measure water consumption, land use, hardware manufacturing, local pollution or the full lifecycle of servers and buildings.
- Complete company footprints: an in-house Scope 2 number cannot be compared directly with a full-company Scope 2 total or a Scope 3 total without matching the organizational and facility boundaries.
The Guardian amended its article on 18 September 2024 after Apple clarified that only part of its location-based Scope 3 figure had been audited. An earlier explanation that attributed Apple’s gap to data centers was withdrawn; it should not be repeated.
How to evaluate a data-center emissions claim
When a company, analyst or later study publishes a number, check these six attributes before comparing it with the 662% estimate:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Best Value
- Accounting method: is it location-based, market-based or both?
- Facility boundary: does it cover owned and operated sites only, or also colocation and cloud providers?
- Scope category: is the number Scope 2 electricity, Scope 3 purchased services, or a combination?
- Reporting period: which year or multi-year period does it cover?
- Evidence type: was the figure directly disclosed, audited or inferred by an analyst?
- Workload boundary: does it concern data centers generally, or specifically AI training and inference?
Keeping those axes aligned prevents a low market-based figure for a company’s owned facilities from being presented as if it were a measurement of all electricity used on its behalf.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the finding means for AI and Big Tech accountability
The investigation’s central lesson is about transparency: a market-based figure can look very small when a company retires certificates, while the physical grid serving its facilities remains substantially more carbon-intensive. Location-based reporting makes that local electricity burden visible; market-based reporting shows the company’s contractual procurement position.
Jay Dietrich, research director of sustainability at Uptime Institute, told the Guardian: “Location-based [accounting] gives an accurate picture of the emissions associated with the energy that’s actually being consumed to run the data center. And Uptime’s view is that it’s the right metric,”
That is Dietrich’s assessment, not a formal declaration by the Greenhouse Gas Protocol. The protocol’s guidance requires both methods where qualifying contractual data exists, allowing readers to see the grid impact and the procurement claim rather than treating either number as the entire story.
So, are data-center emissions really 662% higher?
According to the Guardian’s estimate, probably—but only within its stated boundaries: four companies, their in-house data centers, electricity-related Scope 2 emissions, and the 2020–2022 period. Because public disclosures were incomplete and some company-specific values had to be inferred, 662% should be treated as an analytical estimate, not a fully observed census. It also should not be presented as a current-year total or as proof that AI alone caused the gap.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
