Microsoft, Amazon, and Google have collectively pumped out 119 million metric tonnes of CO₂ equivalent in the past year - about a third of France's total emissions - and that's a 17% jump from last year. The culprit? Datacentres, which are being built at a frenzied pace to power AI tools that may or may not be worth the climate cost.

In the previous year, the trio emitted roughly 101 million tonnes, matching Czechia's 2024 output. But then came the AI boom, and suddenly everyone needed cloud storage for their chatbot conversations. Cecilia Rikap, an economics professor at University College London, calls the companies' green cloud claims "a marketing strategy" and warns that governments should be wary when these same firms pitch AI as a climate solution.

The emissions increases were documented in the companies' annual sustainability reports. Microsoft's emissions rose 25% to 20 million tonnes, driven by "the expansion of our datacentre infrastructure." Google's were up 18%, with supply chain activities fueling the growth - though it claims its AI systems helped reduce emissions elsewhere by 41 million tonnes. Amazon reported a 16% overall increase and a 20% rise in supply chain emissions, but still framed it as "making progress" toward net zero by 2040.

The global push to build AI infrastructure is costing tech giants an estimated $765 billion this year, with datacentres sprouting up from Norway to North Tyneside. This marks a sharp reversal from years of declining emissions. Prior to this year, Microsoft's emissions had flatlined at 16 million tonnes. All three companies still pledge net zero: Google and Microsoft by 2030, Amazon by 2040.

"The increases in total carbon emissions are strongly correlated with [the companies'] AI investment," said Shaolei Ren, a professor at UC Riverside. He also noted Microsoft's report hinted at a shortage of carbon credits to offset their pollution - because apparently, there aren't enough virtual trees to go around.

JLL, a property consultancy, expects about 1,200 datacentres to be built globally by 2030, with demand overwhelmingly driven by AI. The Uptime Institute estimates that big datacentre projects announced last year would consume 1.3% of the world's electricity, nearly doubling current demand - mostly from US projects. So, you know, progress.