Technology companies have spent years insisting that AI would be part of the climate solution: optimizing grids, forecasting renewables, wringing waste out of everything. A new report backed by the United Nations does not dispute that AI can help. It simply lays out, in hard numbers, how much the same boom is now working against the very targets these firms have pledged to hit. The two stories no longer add up.
The Greening Digital Companies 2026 report, published by the International Telecommunication Union and the World Benchmarking Alliance, tracks the emissions and climate plans of 200 major tech companies. Its headline finding is blunt: emissions from four leading AI and cloud providers rose by as much as 239 percent between 2020 and 2024. Over the same stretch, 14 major telecom firms actually cut their emissions by 11 percent. The contrast points the finger squarely at compute. As the ITU put it, AI "is both a driver of efficiency and a growing challenge," and right now the challenge is winning.
The scale is easy to underestimate until you convert it to something physical. The 200 companies consumed close to 500 terawatt-hours of electricity in 2024, roughly 1.7 percent of global consumption. The ten biggest users alone drew 269 terawatt-hours, more than the entire nation of Australia. With Amazon declining to report its electricity data, China Mobile topped the list at 63 terawatt-hours, nearly double the 32 apiece consumed by Alphabet and Samsung, with Microsoft close behind at 30.
What makes this more than an accounting story is the widening gap between promises and progress. Tech firms remain among the world's largest buyers of renewable power, and 151 of the 200 have set near-term emission targets. But only 85 were judged on track based on actual results, and just 81 had comprehensive plans to get there. Setting a target has become table stakes; meeting one clearly has not. Only 25 companies reported sourcing 100 percent renewable electricity, and only Swisscom earned a perfect climate score.
The uncomfortable subtext is that the demand curve is still bending upward. The report expects power needs to keep climbing as AI, cloud and digital infrastructure expand, which is why separate analyses have warned that data-center growth could prop up demand for fossil fuels even in a decarbonizing world. A pledge made in 2020, before the generative AI surge, was built for a much smaller machine than the one these companies are now racing to build.
That is the real test buried in the 239 percent figure. It is not whether AI can, in theory, help cut emissions somewhere in the economy. It is whether the firms selling that promise can bring their own footprint under control fast enough to be credible. "Environmental sustainability must be built into how we design, power and scale" these systems, said ITU chief Doreen Bogdan-Martin. The report is a reminder that, for now, it mostly is not, and that a climate pledge is only as good as the electricity bill behind it.