Key facts
- The Nature Sustainability study covers 403 hyperscale data centers and is described as the first peer-reviewed figures for data center emissions since the AI boom began.
- Google, Microsoft and Amazon together emitted 19 million metric tonnes of carbon dioxide equivalent in the last fiscal year, per company figures.
- That combined figure equals about a third of France's carbon output, according to the report.
- Goldman Sachs said it has not changed its projection for $1.4 trillion in AI-related hardware spending by the end of 2027.
- KR Sridhar of Bloom Energy said he sees nothing but a huge opportunity ahead for the sector.
A study published in the journal Nature Sustainability in the week of November 19, 2026 found that 403 hyperscale data centers accounted for nearly 2% of total US electricity consumption between May 2024 and April 2025. The report describes the study as the first peer-reviewed figures for data center emissions since the start of the AI boom, a sector it says has operated with little transparency.
The study also found that most of the energy used to power those data centers came from fossil fuels. According to company figures cited in the report, Google and Microsoft each reported carbon emissions up 25% year over year from 2025 to 2026, while Amazon's rose 16%. Together the three emitted 19 million metric tonnes of carbon dioxide equivalent over the last fiscal year, about a third of France's output, the report said.
The report said a COP31 pledge to power and build AI infrastructure and hardware sustainably has been signed by most countries, with only a handful of small countries yet to sign on, as the UN climate conference wrapped up in Antalya, Türkiye.
It also said rising data center energy consumption is pushing up consumer prices, drawing attention from politicians in both parties, and that the prior month's election result is expected to bring more scrutiny of the sector. Tim McDonnell, climate and energy editor at Semafor, wrote in an opinion piece that the AI boom's momentum remained intact, and the report cited Goldman Sachs saying it had not changed its projection for $1.4 trillion in AI-related hardware spending by the end of 2027. KR Sridhar of Bloom Energy said he sees nothing but a huge opportunity ahead, and Mark Daly, head of data centers at BloombergNEF, reportedly told McDonnell that the midterms did not present anything concrete that would cause a revision of views.
