The Hook
In January 2020, Microsoft committed to becoming carbon-negative by 2030 and removing all its historical carbon emissions by 2050. The pledge was specific, timed, and widely covered. Five years later, Microsoft's own environmental report shows its emissions moving in the wrong direction — significantly.
The Question
When a company's emissions rise fastest in the years it is expanding AI infrastructure, and it simultaneously pauses the clean-energy purchasing that made the pledge look achievable, what is the pledge worth?
The Paper Trail
Microsoft's 2026 environmental report — covering fiscal year 2025 — shows total emissions of approximately 20.3 million metric tons of CO₂ equivalent, up from approximately 16.2 million the year before. That is a 25% increase in twelve months. Emissions from purchased electricity rose approximately 945% between 2024 and 2025, while actual electricity consumption rose approximately 24%. The divergence between those two numbers reflects the change in how Microsoft was accounting for its electricity: when a company stops purchasing the clean-energy certificates that let it claim its electricity is renewable, the same electrons show up on the ledger as dirty.
Google disclosed in its own sustainability report that its emissions grew nearly 50% over five years, with data-center electricity consumption named as the primary driver.
The Synthesis
The 2030 carbon-negative pledge and the decision to build out AI infrastructure at a pace that raises emissions sharply per year were both made by Microsoft's leadership. The pledge was real when it was made. The infrastructure expansion is also real. One of those decisions makes the other nearly impossible to achieve on the stated timeline — and Microsoft's own reporting documents the result.
The clean-energy certificate accounting is the clearest tell: the certificates are the mechanism that reconciles physical electricity consumption with a clean-energy claim. When the certificates stop, the ledger tells the truth about the consumption.
