Microsoft disclosed that its greenhouse gas emissions climbed to 21 million metric tons in fiscal year 2025, up roughly 27% from about 17 million metric tons the prior year, in a filing that lays bare the direct tension between the company's AI infrastructure buildout and its long-standing climate commitments. The spike was driven in large part by a tenfold surge in Scope 2 market-based emissions -- those tied to purchased electricity -- as Microsoft's AI data centers scaled far faster than the clean-power supply available to run them.
Microsoft Chief Sustainability Officer Melanie Nakagawa notably declined to directly reaffirm whether the company remains on track to meet its long-standing goal of becoming carbon negative by 2030, telling reporters that "many of the sustainability solutions are not scaling fast enough to keep pace with AI infrastructure growth." That's a materially different posture than the confident sustainability messaging Microsoft has maintained in prior years, and it's a rare instance of a hyperscaler publicly acknowledging its AI ambitions may be structurally incompatible with its own prior climate targets on the original timeline.
The most concrete evidence of that tradeoff is Microsoft's data-center power sourcing. The company is developing at least two data centers that will rely on natural gas for power, in Texas and West Virginia, including a deal disclosed in June pairing Microsoft with Chevron on a planned West Texas natural gas plant whose output would feed a dedicated data-center campus. Natural gas, while cleaner than coal, is still a fossil fuel -- meaning Microsoft is directly trading near-term compute availability for emissions the company had previously committed to eliminating.
“The most concrete evidence of that tradeoff is Microsoft's data-center power sourcing.”
Microsoft isn't alone in this reversal. Google and Amazon both reported comparable emissions surges over the same period -- 18% and 16% respectively -- as all three companies now acknowledge, to varying degrees, that AI infrastructure expansion is outpacing their own decarbonization efforts. That parallel pattern across the three largest cloud providers suggests the tension isn't a Microsoft-specific management failure, but a structural feature of the current AI buildout: clean power generation, transmission and storage simply aren't scaling at the pace hyperscalers need to keep every new data center on a zero-carbon power supply.
For founders and operators building AI-heavy products, the emissions data is a reminder that the compute layer underneath every AI application carries a real and growing environmental cost that enterprise customers -- particularly those with their own sustainability commitments -- are increasingly likely to scrutinize in vendor selection. For investors, the gap between hyperscalers' public climate commitments and their actual emissions trajectory is becoming a material disclosure risk, not just a reputational one, especially as more jurisdictions consider mandatory climate-risk reporting.
The bear case: none of this is likely to slow AI infrastructure spending in the near term, since compute availability is currently the binding constraint on AI product development for every major lab, and climate commitments have historically proven to be the easiest corporate promise to quietly push back when they conflict with core business priorities. What to watch next: whether Microsoft, Google or Amazon issue any formal revision to their 2030 climate targets, and whether increased natural-gas-powered data center capacity becomes the default solution the rest of the industry adopts to solve near-term power constraints.