Google has released its Environmental Report 2024, a more than 80-page document detailing all of the massive company’s efforts to apply technology to environmental problems and mitigate its own contributions. But it completely sidesteps the question of how much energy AI uses – perhaps because the answer is “much more than we would like to say.”
You can read the full report here (PDF), and honestly, there’s a lot of interesting stuff in there. It’s easy to forget how many plates a company the size of Google keeps spinning, and there’s some truly remarkable work in there.
For example, one is being worked on water replenishment program, where it hopes to offset the water used in its facilities and operations, ultimately creating a net positive outcome. This is done by identifying and funding watershed restoration, irrigation management and other work in that area, with dozens of such projects around the world being at least partially funded by Google. In this way, 18% of water consumption has been replenished (by whatever definition of that word is used here) and this is improving every year.
The company is also taking great care to leverage the potential climate benefits of AI, such as optimizing irrigation systems, creating more fuel-efficient routes for cars and boats, and predicting floods. We’ve already highlighted a few in our AI coverage, and they can be quite useful in many areas. Google doesn’t have to do these things, and many big companies don’t either. So credit where credit is due.
But then we get to the part of ‘Responsibly managing AI resource consumption.’ Here Google, hitherto so certain of every statistic and estimate, suddenly spreads its hands and shrugs. How much energy does AI use? Can somebody Real be sure?
Still, it must be bad, because the first thing the company does is downplay the entire data center energy market by saying that this is only 1.3% of global energy consumption, and that the amount of energy Google uses is only 10% of that. – so only 0.1%. According to the report, % of the world’s energy is used to power its servers. A little something!
Notably, in 2021 it decided it wanted to reach net-zero emissions by 2030, though the company admits there’s a lot of “uncertainty,” as it likes to call it, about how that will actually happen. Especially since emissions have increased every year since 2020.
Our total greenhouse gas emissions will be 2023 [greenhouse gas] emissions amounted to 14.3 million tCO2e, represent an increase of 13% year-on-year and an increase of 48% compared to our target base year of 2019. This result was mainly due to the increase in data center energy consumption and supply chain emissions. As we further integrate AI into our products, reducing emissions may be a challenge due to the increasing energy demand resulting from the greater intensity of AI computing, and the emissions associated with the expected increase in our investments in technical infrastructure.
(Emphasis mine in this and the quote below.)
Yet the growth of AI is lost among the above-mentioned uncertainties. Google has the following excuse for not being specific about the contribution of AI workloads to overall data center energy bills:
Predicting the future environmental impacts of AI is complex and evolving, and our historical trends likely do not fully reflect the future trajectory of AI. As we deeply integrate AI into our product portfolio, the distinction between AI and other workloads will not be meaningful. So we focus on data center-wide metrics because they include the total resource consumption (and therefore environmental impact) of AI.
“Complex and evolving”; “the trends are probably not fully aligned”; “the distinction … will have no meaning”: This is the kind of language used when someone knows something but really would rather not tell you.
Does anyone really believe that Google doesn’t know, down to the cent, how much AI training and inference has added to the energy cost? Isn’t being able to break down these numbers precisely part of the company’s core competency in cloud computing and data center management? It has all these other claims about how efficient its custom AI server units are, how it does all this work to reduce the energy required to train an AI model by 100x, and so on.
I have no doubt that there are many great green efforts happening at Google, and you can read all about them in the report. But it is important to highlight what it seemingly refuses: the enormous and growing energy costs of AI systems. The company may not be the main cause of global warming, but despite its potential, Google doesn’t appear to be net positive yet.
Google has every reason to downplay and gloss over these numbers, which can hardly be a good thing even in its limited, highly efficient state. We’ll be sure to ask Google to get more specific before we find out if they get any worse in the 2025 report.