Every major tech company now has an "AI for climate" announcement. Google is using AI to optimize wind farm output. Microsoft has pledged to use AI to accelerate carbon capture. Amazon claims AI-powered logistics reduce delivery emissions. The press releases are polished. The commitments are sweeping. And according to a new research report, the math simply doesn't add up.
Researchers have concluded that technology companies' claims about AI as a climate solution are, in the language of the report, greenwashing. The core argument: the energy demands of AI infrastructure — the data centers, the GPU clusters, the cooling systems, the power transmission — far outstrip any efficiency gains AI might enable elsewhere. You cannot build a net climate-positive AI system by pointing to what the AI does while ignoring what the AI costs.
The numbers don't flatter the industry. Major cloud providers are currently in the midst of the largest data center expansion in history, driven almost entirely by AI demand. Several have quietly revised their carbon commitments downward, or restructured them in ways that make progress harder to evaluate. Microsoft saw its carbon emissions increase by 30% between 2020 and 2023 — precisely the period when it was dramatically expanding its AI infrastructure. Google's reported emissions rose similarly. The pledges to reach net zero by 2030 are increasingly being walked back or reframed around purchased offsets rather than actual reductions at the source.
The "AI for climate" framing relies on a specific accounting approach: you count the emissions that AI might theoretically prevent in other sectors, and you don't count the emissions from running the AI itself. That's not how climate accounting works. If AI-optimized logistics saves 10 metric tons of carbon in delivery routes but the AI system running the optimization emits 12 metric tons during training and inference, the net result is negative. The tech industry has not, in any consistent way, done that full-cycle accounting — and researchers are now calling that out with data.
I've watched startup pitches where founders claimed AI would solve climate change. I've watched the same people build infrastructure that consumes power at a scale that would have seemed impossible ten years ago. This report does the math, and the math is not flattering. The technology is genuinely capable of improving efficiency in specific applications. But the claim that AI will be a net climate positive — at scale, across the economy — is not a technology claim. It's a marketing claim. And researchers are now calling it what it is.
