Google has published its 2024 Environmental Report, an 80+ page document that explains all the efforts the giant is making to apply technology to environmental issues and reduce its own environmental footprint. But it doesn't address the question of how much energy AI uses at all, probably because the answer is “a lot more than we'd like to say.”
You can read the full report here (PDF) and, frankly, it's full of interesting takeaways. It's easy to forget that a company as big as Google has multiple tasks on its hands, but there's some really noteworthy stuff here.
For example, the company is working on a water replenishment program to offset the water used by its facilities and operations, ultimately generating a net benefit. It does this by identifying and funding watershed restoration, irrigation management, and other work in that area, with dozens of similar projects around the world funded at least in part by Google. The company has replenished (whatever definition of the term) 18% of its water usage this way, and is making improvements every year.
The company has also been careful to preface potential benefits of AI for the climate, such as optimizing irrigation systems, creating more fuel-efficient routes for cars and ships, predicting floods, etc. Some of these we've already covered in our AI articles, but they could actually be very useful in many areas. Google doesn't have to do this, and neither do many big companies, so kudos for the work they've done.
But then we get to the section “Responsibly managing AI resource consumption,” where Google, who had been so confident in all their statistics and estimates, suddenly spreads their hands and shrugs. How much energy does AI consume? Can anyone really be sure?
But the report must be bad because it underestimates the entire data center energy market, stating that it represents just 1.3% of global energy use, and that Google uses just 10% of that. In other words, according to the report, just 0.1% of the world's energy powers Google's servers, which is negligible.
Notably, in 2021 the company committed to achieving net-zero emissions by 2030, though the company acknowledged there was “considerable uncertainty” about how that would actually happen — especially since the company's emissions have increased every year since 2020.
In 2023, our total greenhouse gas emissions [greenhouse gas] Emissions were 14.3 million tCO2e, up 13% year-on-year and 48% compared to the target base year of 2019. This result was primarily driven by increased data center energy consumption and supply chain emissions. As AI is further integrated into products, reducing emissions may be challenged by increased energy demand from greater AI computing intensity, as well as emissions from expected increases in technology infrastructure investment.
(Emphasis here and in the following quotes is mine.)
Image credit: Google
But the growth of AI gets forgotten amid the aforementioned uncertainty. Google offers this excuse for why it hasn't specified how much AI workloads contribute to its overall data center energy bill:
Predicting the future environmental impact of AI is complex and evolving, and historical trends may not fully capture AI's future trajectory. As AI becomes more deeply integrated across our product portfolio, distinguishing between AI and other workloads becomes less meaningful. That's why we're focusing on data center-wide metrics that include AI's overall resource consumption (and therefore environmental impact).
“It’s complex and changing,” “trends don’t fully capture it,” “the distinction is… meaningless”: These are the kinds of phrases used when someone knows something but really, really doesn’t want to tell you.
Does anyone really think that Google doesn't know every penny of energy costs incurred by AI training and inference? The ability to break down these numbers so precisely is the company's core strength in cloud computing and data center management. The company has made various statements about the efficiency of its custom AI server units and how it is making every effort to reduce the energy required to train AI models by a factor of 100.
There's no doubt that Google does a lot of great environmental work, the details of which you can read about in the report. But it's important to highlight something the company noticeably refuses to address: the huge and growing energy costs of its AI systems. The company may not be a major driver of global warming, but despite that potential, Google doesn't appear to be making a net profit yet.
Google has every incentive to downplay and be vague about these numbers. They are not good, even in their reduced and highly efficient state. I will certainly be pushing Google to provide more concrete numbers before we find out if they're worse off in their 2025 report.