---
title: "Microsoft Emissions Up 25%: How to Read the Report"
description: "Microsoft reports emissions up 25% in fiscal 2025 alongside 100% renewable electricity. How to read Scopes 1-3, two accounting methods and what to ask vendors."
url: "https://staging.khoi.co.uk/en/blog/ai-tang-phat-thai-microsoft-cam-ket"
lang: en-GB
author: "Lê Quốc Khôi"
date: 2026-11-03
updated: 2026-11-03
topics: ["Ethics & corporate governance"]
tags: ["Microsoft", "Carbon emissions", "Sustainability reporting", "ESG-linked pay", "AI infrastructure"]
---

# Carbon negative by 2030, emissions up 25%: reading Microsoft's report

## Key takeaways

- In fiscal year 2025 Microsoft's total emissions were 20.29 million tonnes CO2e, up 25%, mainly from datacentre expansion and the end of "non-additional, unbundled" renewable energy certificates; the carbon-negative-by-2030 pledge dates from 2020 (Microsoft Corporation, 2026, p. 30; Microsoft Corporation, 2026, p. 8).
- The same report says 100% of electricity use was matched with renewable energy (Microsoft Corporation, 2026, p. 6). Both statements hold because they measure different things: market-based Scope 2 rose from 0.26 to 2.71 million tonnes, location-based Scope 2 by about 21% (accompanying Data Fact Sheet, my arithmetic).
- Cohen and colleagues (2023) find emissions-specific pay metrics go with lower Scope 1 emissions, on mainly descriptive evidence. Li and colleagues (2024) study CEO cash pay and equity holdings, not ESG-linked pay (Cohen et al., 2023; Li et al., 2024).
- The four-controls map is a synthesis, not a tested finding; the post uses it as a set of questions [Strategic leadership through four control mechanisms](https://staging.khoi.co.uk/en/framework/leadership-to-controls-map).
- The mid-sized retailer in the post is hypothetical, my own invention.

On 9 July 2026 Microsoft published its 2026 Environmental Sustainability Report (Smith and Nakagawa, 2026). On one page it says the company matched 100% of its global electricity consumption with renewable energy, and that total emissions rose 25% (Microsoft Corporation, 2026, p. 6). If you are about to run AI and cloud workloads at a larger scale, [this case](https://staging.khoi.co.uk/en/case/microsoft-ai-emissions-fy25) is a lesson in reading such a report. I set out only what the company discloses, and I do not conclude that it greenwashes.

## What does the report say about the 25%, and how does Microsoft explain it?

Total emissions for fiscal year 2025 were 20.29 million tonnes CO2e, against 16.215 million the year before (Microsoft Corporation, 2026, p. 30). The carbon-negative-by-2030 goal was set in 2020 (Microsoft Corporation, 2026, p. 8). The accompanying Data Fact Sheet gives 12.88 million tonnes for 2020, so the 2025 total stands about 58% above the baseline (my arithmetic).

The company names two causes: expansion of its datacentre infrastructure, and stopping the use of non-additional, unbundled certificates in order to prioritise investments that bring net new power to grids (Microsoft Corporation, 2026, p. 30). In their foreword, Brad Smith and Melanie Nakagawa write that growth-related emissions pressure "was expected" (Smith and Nakagawa, 2026). I found no AI-only emissions figure in the pages I read, so this post does not separate out AI's share.

Scope 1 is direct emissions from the company's own activities; Scope 2 is indirect emissions from purchased electricity, steam, heat or cooling; Scope 3 is everything else in the value chain. Scope 2 made up 13% of the total, up from nearly 2% a year earlier, and Scope 3 remains the largest (Microsoft Corporation, 2026, p. 30).

## How can "100% renewable" and "emissions up 25%" both be true?

They measure different things. One compares electricity used with renewable supply as the company defines it. The other is emissions with Scope 2 calculated on a market basis, which counts purchase contracts and certificates (Microsoft Corporation, 2026, p. 30). The Fact Sheet says the location-based method uses average emission factors for the grids, and that from fiscal 2025 short-term certificates no longer count in the market-based method (Fact Sheet, p. 9). The report adds that project contracts supplied more than 90% of the renewable energy counted towards the 2025 target, with the rest "grid mix" (Microsoft Corporation, 2026, p. 64).

| Data Fact Sheet (Tables 1B, 2, 5) | Fiscal 2024 | Fiscal 2025 | Change (my arithmetic) |
|:--|:--|:--|:--|
| Electricity consumed (MWh) | 29.83 million | 37.03 million | +24% |
| Scope 2, location-based (tCO2e) | 9.96 million | 12.03 million | +21% |
| Scope 2, market-based (tCO2e) | 259,090 | 2,707,428 | about 10.4 times |
| Total, management's criteria (tCO2e) | 16.215 million | 20.29 million | +25% |

Market-based Scope 2 rose by 2.45 million tonnes, about 60% of the 4.075 million rise in the total. The two Scope 2 series part company exactly where the company says it changed its certificate policy. The report does not split the rise numerically between growth and the change in accounting; reading the two series side by side is mine.

## What else sits behind the 25%?

Two further choices. First, which total: the 25% uses the company's "management's criteria", which net sustainable fuel certificates and renewable electricity out of three Scope 3 categories; on the Greenhouse Gas Protocol category basis the same table goes from 16.693 to 21.121 million tonnes, about 26.5% (Tables 1A and 1B). Second, the denominator: intensity per million dollars of revenue fell from 91.3 in 2020 to 75.0 in 2025, although it rose from 68.1 in 2024 (Table 4, computed on the Table 1A total), while the pledge is stated in absolute terms.

## Does ESG-linked pay close the gap?

Neither study in the library answers yes or no. Cohen and colleagues (2023) use data on 4,395 listed firms in 21 countries, 2011 to 2020 (Cohen et al., 2023). When firms put emissions-specific metrics into pay, that goes with a fall in Scope 1 emissions in later periods; ESG-linked pay in general shows no statistically significant link with emissions, goes with rising ESG ratings (two of three rating sets) and shows no link with better financial performance (Cohen et al., 2023). The authors state their own limits: the evidence is mainly descriptive, rests on public pay disclosures that are sometimes thin, and depends on the quality of ESG ratings (Cohen et al., 2023). Microsoft's Scope 1 in 2025 was 170,887 tonnes, about 0.8% of its total (Fact Sheet; my arithmetic), so that result says nothing about where Microsoft's rise sits. I found no mention of executive pay in the report or the Fact Sheet.

Li and colleagues (2024) analyse 8,028 firm-year observations of Chinese A-share firms, 2013 to 2022 (Li et al., 2024, p. 8). Higher CEO cash pay goes with more greenwashing (1% significance level) and CEO equity with less (5% level), after controls (Li et al., 2024, p. 10). The limits: the study looks at CEO cash pay and equity holdings, not ESG-linked pay; equity is only about 26% of executive pay in China against 55% in the US, on 2018 figures the paper cites (Li et al., 2024, p. 2); and the design is observational, even though the authors run several checks for endogeneity (Li et al., 2024, p. 11).

So pay may be a lever, but what it does depends on the metric, how it is measured and where. Neither study shows that it narrows the gap at a group like Microsoft.

## What do the four controls show when applied to this case?

The map is a synthesis, not a tested finding. I use it as four questions, not four verdicts.

> **Figure:** [Strategic leadership through four control mechanisms](https://staging.khoi.co.uk/en/framework/leadership-to-controls-map) (redrawn from the site's library framework)

| Control | What Microsoft discloses | What I could not see |
|:--|:--|:--|
| Resource allocation | A $1bn Climate Innovation Fund, $815m allocated (Microsoft Corporation, 2026, p. 9); agreements for up to 40 GW of new renewable energy in 26 countries (Microsoft Corporation, 2026, p. 62) | An emissions budget beside the datacentre budget |
| Decision rights | Nothing seen | Who approves new capacity, and whether they may refuse |
| Escalation | The Vice Chair and President and the Chief Sustainability Officer present to a board committee (Microsoft Corporation, 2026, p. 63) | At what level a problem goes up |
| Review | Limited assurance by Deloitte on fiscal 2025 data; goals, progress against goals and earlier periods outside scope (Fact Sheet, pp. 2, 23) | Whether internal review ever changed a decision |

From outside I see only the disclosure layer; the documents do not say whether the machinery behind it works.

## What should you ask a vendor before scaling up AI and cloud use?

Suppose the board of a mid-sized Vietnamese retailer (my hypothesis) plans to move data analysis and an AI assistant onto a large provider's cloud. In my reading of the Scope 3 definition, the cloud services it buys fall into its own Scope 3, so the provider's report is an input to its numbers. Four questions follow from the points where Microsoft's report is easy to misread.

| Question | Why ask |
|:--|:--|
| Market-based, location-based or both? | At Microsoft, Scope 2 moved in very different ways under the two |
| What is inside "100% renewable"? | Contracted projects, grid mix and short-term certificates are three different things |
| Which categories are netted for certificates, and is the cut absolute or per unit of revenue? | Microsoft's two totals differ by about 0.8 million tonnes; intensity fell against 2020 while the absolute total rose |
| Is there a figure for the metro area where my workload runs? | The 2026 report adds datacentre energy and water use by metro area (Microsoft Corporation, 2026, p. 8) |

This suits you if your volume is large enough for the vendor to answer in writing; it does not suit you if you buy a small plan through a reseller, where all that is left is the public report. The report also says some of Microsoft's goals may depend on customers and partners changing their behaviour (Microsoft Corporation, 2026, p. 63). On the company's side, four jobs could be assigned, one per control: who owns the emissions budget of the AI project, who may move workloads to another provider, what rise in electricity use must reach the CEO, and who reviews it each quarter. That is an application of a synthesis, not tested on Vietnamese firms. For environmental claims on packaging, see [Sustainable marketing is not CSR](https://staging.khoi.co.uk/en/blog/marketing-ben-vung-khong-phai-csr).

## Going deeper: what did I read, and what could I not check?

For Microsoft I read Smith and Nakagawa's post in full, pages 5 to 11, 28 to 31, 55 and 62 to 64 of the 66-page report, and searched the full text by keyword. I also read the 2026 Environmental Data Fact Sheet (aka.ms/SustainabilityFactsheet2026): Tables 1 to 7 and pages 2, 9, 10 and 23. It is not yet in the site's source library, so I name it without a citation key. Every ratio is my own arithmetic.

For Cohen and colleagues I read the ECGI working paper (revised April 2023), not the journal version, because the publisher's page blocked automated access, so I give no journal page numbers. Li and colleagues I read in full. The four-controls map: the library text only. I have not read Microsoft's proxy statement, independent estimates of AI electricity use, or Vietnamese rules on emissions reporting.

This post does not show whether ESG-linked pay works, nor that the company will miss its 2030 goal. It shows one documented year in which commitment and result diverge.

## Frequently asked questions

### What does Microsoft's carbon-negative goal cover, and where does 2025 stand?

According to the Environmental Data Fact Sheet published with the report, the company plans to bring Scopes 1 and 2 (market-based) close to zero, cut Scope 3 by more than half and remove what remains, with 2020 as the baseline year (Fact Sheet, p. 9). On management's criteria, Scope 3 was 12.31 million tonnes in 2020 and 17.41 million in 2025, up about 41% (my arithmetic from Table 1B). The target date is 2030, so this is one point on the way, not a forecast. Read the verbs closely: page 9 of the report says the carbon-removal projects are expected to contribute more than 45 million tonnes over the next three decades, while page 62 says the company has contributed more than 45 million tonnes (Microsoft Corporation, 2026, p. 9; Microsoft Corporation, 2026, p. 62).

### What are unbundled renewable energy certificates, and why did Microsoft stop buying them?

The report says the company stopped using non-additional, unbundled certificates so that it can prioritise investments that bring net new power to grids, and admits that this raises reported emissions in the near term (Microsoft Corporation, 2026, p. 30). The wording is not uniform: page 6 says "pausing" (Microsoft Corporation, 2026, p. 6), page 30 says "stopping", and the Fact Sheet says "discontinued" (p. 9). The Fact Sheet defines the discontinued kind as certificates bought separately from the power, outside a utility contract and under short-term agreements; the programme ended in February 2025, and the company says this may temporarily move it out of a carbon-neutral position (Fact Sheet, pp. 9-10).

### What does 'ESG-linked pay' mean in the study by Cohen and colleagues?

A firm counts as having it if at least one ESG criterion is a key performance indicator in its executives' pay scheme (Cohen et al., 2023). The criteria range from carbon emissions and other environmental variables to safety, diversity and inclusion, employee satisfaction, culture, compliance and governance (Cohen et al., 2023). On that broad definition, the share of firms in the ISS database using ESG metrics for executives rose from 3% in 2010 to over 30% in 2021 (Cohen et al., 2023). A firm with only a workplace-safety metric counts, so the 30% figure does not mean that many firms tie pay to emissions.

### How do Li and colleagues measure 'greenwashing'?

As the gap between a firm's Bloomberg ESG disclosure score and its Sino-Securities (Huazheng) ESG performance score, both standardised within industry; a positive gap means more is disclosed than done (Li et al., 2024, p. 8). In my reading, that measure needs two raters' scores across many firms, so it cannot be applied to a single report such as Microsoft's. This post does not conclude that Microsoft greenwashes; it reads what the company itself discloses.

## References

- Cohen, S. et al. (2023) ‘Executive compensation tied to ESG performance: international evidence’, *Journal of Accounting Research*, 61(3), pp. 805–853. https://doi.org/10.1111/1475-679x.12481
- Li, K., Lin, T.Y. and Zhu, G. (2024) ‘The effect of CEO's compensation in driving corporate ESG greenwashing: evidence from China’, *PLOS ONE*, 19(10), article e0312247. https://doi.org/10.1371/journal.pone.0312247
- Microsoft Corporation (2026) *2026 Environmental Sustainability Report: Responsibly building the AI future*. Microsoft (Microsoft Environmental Sustainability Report). Available at: https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/msc/documents/presentations/CSR/2026-Microsoft-Environmental-Sustainability-Report-PDF.pdf (Accessed: 2 November 2026).
- Smith, B. and Nakagawa, M. (2026) *Responsibly building the AI future*. Microsoft On the Issues. Available at: https://blogs.microsoft.com/on-the-issues/2026/07/09/responsibly-building-the-ai-future/ (Accessed: 2 November 2026).

---

Canonical HTML version: https://staging.khoi.co.uk/en/blog/ai-tang-phat-thai-microsoft-cam-ket