The environmental cost of AI far exceeds urgent climate targets. This is evident in the 2026 environmental reports of Alibaba, Amazon, Apple, Google, Meta, Microsoft, and Tencent. These 7 companies alone emitted over 168 million mtCO2e in 2025 – the equivalent of almost half of Australia’s and about a third of Brazil’s annual emissions. Their water withdrawal stood at 9,510 million cubic litres, as much as the U.K., Estonia and Namibia’s annual withdrawals taken together.

The data often tell a different story than the narrative parts of these environmental reports. The narratives promote greenwashing and invite people to drink the Kool-Aid of future (AI) promises. Much of the actual data and impact is hidden in appendices, separate data sheets, adjustments to calculations, or percentages rather than absolute numbers. Once found, the stories the data tell point to (public) electricity grids failing, water resources depleting, and communities coming under high stress with climate adaption. And that’s where we have the data. Companies like OpenAI and Anthropic don’t even publish any of their assessments – if they even bother to conduct them.

You can read my overall assessment on Medium.

There are several additional stories related to my assessment that I would still love to tell but go beyond the space of this post. These include the implications of tech companies’ pivot to nuclear power and fusion bets; the backlash many new data centre projects are facing across communities, Texas for instance; the varying approaches to policy and advocacy these companies adopt and how they influence regulation; and the consequences of water depletion both for people as well as for our economies that are so tightly linked to AI investments. Hopefully, I can carve out time for these soon.

Feedback, as always, welcome. Full analysis.