Two years ago, Latitude Media reporter Maeve Allsup wrote that the artificial intelligence boom had a metrics problem. Hyperscalers, nearly two years into making investments in AI, had started acknowledging in their sustainability reporting that the technology would complicate their decarbonization efforts.
However, Maeve wrote at the time, “despite the cumulative hundreds of pages of reporting, the picture of AI’s energy and emissions footprint remains fuzzy and incomplete.” Everyone uses different reporting approaches, and many obfuscated how much energy consumption — and emissions — came specifically from data centers.
Today, the companies are still relying on different metrics and levels of disclosure, but their messaging is coalescing around a single pitch: improving efficiency, even as emissions and energy use soars.
This week, both Google and Amazon released reports that are their clearest yet about the dramatic impact of the AI boom on their sustainability efforts. And for starters, the numbers aren’t good.
According to the latest edition of Google’s environmental report, the company’s total emissions rose 18% in just a year, to roughly 14.5 million tons of carbon dioxide equivalent. However, that is an adjusted number, dubbed “ambition-based emissions” as opposed to the absolute emissions, which came in at about 19 million tons of CO2e. (The difference has to do primarily with the company’s accounting under the Science Based Targets initiative, which doesn’t include certain Scope 3 emissions.)
That adjusted “ambition-based” number represents an emissions increase of 18% over last year, and an 81% increase over its 2019 base year. Google said that much of the increase is due to Scope 3, or “supply chain” emissions: meaning all the physical infrastructure Google relies on, like servers, chips, and the equipment inside data centers.
In other words, AI is causing Google’s emissions to rise dramatically — which is exactly as the company predicted in its 2024 edition of this report that covered the boom’s early days.
Amazon, in its own version published a day later, reported that it emitted about 81 million tons of CO2e in 2025, a 16% increase over 2024. That total is significantly higher because the retail and logistics company has a much bigger physical footprint than Google, which deals primarily in software.
While both companies were explicit about the fact that the rise in emissions was related to data centers (as well as, for Amazon, its use of fuel for deliveries), neither broke out emissions or energy use numbers for their cloud computing business units specifically.
New efficiency numbers
Both couched their emissions spikes in the frame of efficiency — a move climate writer and analyst Ketan Joshi dubbed “efficiency-washing.”
Google went so far as to change its metric for reporting its carbon intensity. Where in previous years the company reported intensity just for its operations (i.e. it excluded Scope 3 emissions), it now bases the number on its total ambition-based footprint: that 14.5 million tons of CO2e number. It also restated previous years’ totals. The company’s energy intensity now looks higher, but shows a sharp decline from 620.2 tons of CO2e per gigawatt-hour down to 328.6 tons — never mind that the total energy use has soared so much as a result of the AI boom that the company had to change the unit from megawatt-hour to GWh to keep the numbers legible.
Google, for the first time, also expanded its reporting down to the level of individual inference (which is much less energy-intensive than training its models): The median Gemini text prompt uses 0.24 watt-hours and 0.26 milliliters of water, and emits 0.03 grams of CO2e. And the company introduced an entirely new metric, the per-chip “compute carbon intensity.”
The company’s fleet-wide power usage effectiveness, a metric of how efficiently a data center uses energy, was 1.09, which is slightly below where it has hovered since 2021.
However, these numbers are a distraction from the fundamental truth that even if the technology is getting more efficient, it’s still devouring more and more resources as the company expands its work. “The path to achieving our climate ambitions will not be linear — given our AI infrastructure buildout is currently accelerating faster than the grid is decarbonizing,” Google wrote.
Amazon, meanwhile, continues its habit of disclosing far less than its peers. The company does not publish an energy intensity metric at all, but rather shares its emissions intensity per dollar of revenue: 112.8 grams of CO2e per dollar, which is down 38% since 2019. Again, the company’s total revenue has ballooned in those years.
It shares its energy use primarily through its PUE of 1.14, which the company says is better than the public cloud industry average (though higher than Google’s).
This mirrors how the company talks about its water use. Last month, Brandon Oyer, AWS’ head of energy and water for the Americas, told Latitude Media that “a water efficiency metric is the best metric to focus on,” while total water volumes are “just big numbers.”
Amazon did not share more granular emissions or energy data per query, like Google has begun to do. Incidentally, Google’s report is more than twice the length of Amazon’s.
