Microsoft's Emissions Jump 25% as AI Datacenters Expand

The AI boom has a carbon bill, and it's coming due faster than expected

Microsoft's latest sustainability report shows carbon emissions rose 25% in 2025, a jump the company directly attributes to datacenter infrastructure expansion for AI workloads. Microsoft isn't alone: Google and Amazon both disclosed significant emissions increases in the same reporting period, suggesting an industry-wide pattern rather than a company-specific issue.

The numbers, and where they diverge

The scale of the increase depends on which source you read. The Verge reports Microsoft's 2025 emissions totaled 34 million metric tons (without select interventions), while The Guardian puts the increase at 25%, reaching 20 million mTCO₂e. Sources differ on whether these figures represent the same emissions scope — it's unclear if one reflects total emissions and the other a narrower measure like Scope 3 or datacenter-specific output.

A similar discrepancy appears with Google: The Verge cites a 25% spike in supply chain emissions specifically, while The Guardian reports an 18% increase in overall emissions. Sources differ on whether these are the same metric described differently or genuinely separate figures.

Amazon reported cleaner numbers by comparison — a 16% increase in overall emissions and a 20% increase in supply chain emissions.

Taken together, Microsoft, Amazon, and Google's collective carbon emissions hit 119 million mTCO₂e in the financial year ending March 2026, up from roughly 101 million mTCO₂e the previous year — a nearly 19% year-over-year increase across the three companies.

This isn't a one-year blip

Microsoft's 2024 sustainability report showed a similar emissions rise, indicating this is a multi-year trend rather than an isolated spike. The company's own report states that sustainability solutions are not scaling fast enough to meet AI demand — a notable admission given Microsoft's stated goal of being carbon negative by 2030 (a target Google shares, while Amazon targets net zero by 2040).

Microsoft did stop purchasing non-additional, unbundled renewable energy certificates in February, a move toward more rigorous carbon accounting — but it hasn't offset the broader upward trend.

Resource strain beyond carbon

Emissions aren't the only concern. Amazon disclosed in June that its data centers used 2.5 billion gallons of water in 2025. This comes as the industry gears up for a historic infrastructure build-out: the world's biggest tech companies are on track to spend $765 billion this year, mostly on AI datacenters. JLL expects roughly 1,200 new datacentres globally by 2030, and the Uptime Institute estimates that announced projects alone would consume 1.3% of global electricity usage.

Marketing vs. reality

Not everyone is convinced current sustainability framing matches operational reality. Cecilia Rikap has characterized claims by Microsoft, Amazon, and Google about their clouds being ecologically friendly as a marketing strategy. Shaolei Ren has noted that increases in total carbon emissions are strongly correlated with the companies' AI investment levels — suggesting the emissions trend will likely continue as long as AI spending does.

On the other side of the ledger, Google says its AI systems helped reduce emissions elsewhere by 41 million tonnes of CO2 last year, though the methodology behind that claim isn't detailed in available reporting.

Why founders should care

For early-stage founders, this report points to a few probable shifts worth tracking:

Rising demand for efficiency-focused infrastructure. If AI-driven emissions growth continues at this pace, startups building energy-efficient AI hardware, cooling systems, or datacenter optimization tools may find an expanding market — potentially amplified by the $765 billion in planned datacenter spending industry-wide.
Possible growth in carbon accounting and ESG compliance tools. The unexplained gaps between different emissions figures (Microsoft's 34M vs. 20M metric tons; Google's 25% vs. 18%) suggest reporting standards may still be inconsistent — a plausible opening for startups offering standardized, third-party emissions verification.
Regulatory attention could increase. With experts publicly questioning whether sustainability claims reflect real operational change, founders in climate-adjacent or ESG spaces should expect scrutiny of self-reported metrics to intensify, which could favor independent auditing services.
Resource constraints (water, electricity) may become operational risk factors. Companies building AI-dependent products should watch for potential regulatory or grid-capacity friction as datacenter growth continues at scale.

What's missing from the picture

Several gaps remain unresolved in current reporting: there's no breakdown of how much emissions growth is attributable specifically to AI workloads versus general datacenter expansion, no clarity on interim progress toward the 2030/2040 net zero targets, and no detail on how discontinued renewable energy certificates previously factored into reported totals. Founders evaluating this space should treat current figures as directionally informative rather than precisely comparable across companies.