Microsoft's AI Emissions Rose 25%. Google's Fell 2%. The Gap Isn't Mostly Physical.

Microsoft and Google each published a full-year environmental disclosure this summer, nine days apart, and both describe the same underlying story: aggressive AI datacenter expansion pushing electricity demand sharply higher. The headline numbers, though, point in opposite directions. Microsoft's own report says its total emissions rose 25% year over year. Google's own report says its Scope 1 and Scope 2 emissions fell 2% - despite a 37% jump in electricity demand.

Read side by side, that looks like a scorecard: one company scaling AI responsibly, one not. It isn't. Both companies' own disclosures show the gap is substantially a function of which clean-energy accounting choices each one made this year, not primarily a difference in how carbon-intensive their AI buildouts physically are. A compliance or procurement team using either number as a like-for-like measure of "whose AI infrastructure is greener" is comparing two things that were measured differently.

What Microsoft's Own Report Says Moved the Number

Microsoft's sustainability disclosure, published 9 July 2026 and covering fiscal year 2025, states that total emissions across Scopes 1, 2 and 3 - measured on a market-based accounting approach - increased 25% year over year. Scope 2 emissions specifically grew to represent 13% of Microsoft's total footprint, up from "nearly 2%" the year before.

Microsoft's own explanation for the rise names two causes, not one: the expansion of its datacenter infrastructure, and a deliberate decision to pause its use of "non-additional, unbundled renewable energy certificates" in favor of investments that bring net new power onto grids. The company frames the second cause as a policy choice with a near-term cost: moving away from certificates that don't add new clean generation to the grid, toward the kind of long-term power-purchase deals that do, at the price of a worse-looking number this year.

What Google's Own Report Says It Kept Doing

Google's environmental report, published 30 June 2026, states that electricity demand rose 37% year over year - a scale of growth in the same range as what's driving Microsoft's datacenter buildout. But Google's Scope 1 and Scope 2 emissions, on the same market-based accounting basis Microsoft uses, fell 2% over the same period.

Google's own explanation: it matched 100% of its electricity consumption with renewable energy purchases for the ninth consecutive year, and signed agreements for more than 12 gigawatts of net-new clean energy. Google's report does not describe pausing any category of certificate purchase the way Microsoft's does. It also states plainly, in its own words, that "our AI infrastructure buildout is accelerating faster than the grid is decarbonizing" - the same underlying tension Microsoft's report describes, acknowledged rather than resolved.

Why the Percentage Isn't the Comparison It Looks Like

Everything in this section is inference - neither company's report states it directly, and no rule requires either one to. Read together, the 25%-versus-negative-2% gap looks like it is telling you as much about each company's certificate-accounting choices this year as about the physical emissions of its AI buildout. Microsoft's number moved in part because it stopped counting a certificate category it describes as "non-additional" - one that doesn't add new clean generation to the grid. Google's number held in part because it kept counting new clean-energy agreements at scale. Neither report states what its own figure would look like under the other company's accounting choice, so that direct comparison isn't something either disclosure actually supports. What their own numbers do support is a narrower point: a market-based emissions total is only as comparable across companies as their underlying certificate-accounting choices are, and this year those choices diverged in exactly the direction that would produce this exact split.*

Neither company describes this as a compliance failure, and nothing in either report suggests one. Both figures were produced under the same general market-based framework both companies name in their own disclosures; the difference sits in a discretionary choice within that framework, not outside it.

Why This Matters More as AI Procurement Scales

For a technology leader or CXO citing a cloud vendor's emissions trajectory in a board deck, an RFP, or an internal AI-vendor selection scorecard, the practical consequence is specific: a year-over-year percentage change in a vendor's self-reported emissions is not, on its own, evidence of relative AI-infrastructure efficiency. It reflects that vendor's own accounting choices for that year as much as its physical footprint. An organization further along in AI adoption - running multi-cloud AI workloads, publishing its own Scope 3 figures that inherit vendor emissions, or answering to a board on AI's environmental footprint - is the one most exposed to citing this kind of comparison without the caveat. An organization earlier in AI adoption, still evaluating vendors, has the easier fix: ask what changed in the accounting before asking what changed in the number.

What a Well-Run AI Operation Checks Before Citing a Number

The operational implication isn't to distrust either company's disclosure. It's to read the methodology section before the headline figure - specifically, whether the reporting period's total reflects a market-based or location-based calculation, and whether the certificate mix changed year over year. Microsoft's report answers both questions about itself directly. Google's confirms its figures are market-based but does not say whether its own certificate mix changed this year. Neither answer is available from the headline percentage alone.

Source note: Figures and quotations are drawn from Microsoft's 2026 Environmental Sustainability Report announcement, "Responsibly building the AI future" (9 July 2026, covering fiscal year 2025) and Google's 2026 Environmental Report (30 June 2026).