INFOGRAPHICS. How artificial intelligence has increased greenhouse gas emissions from tech giants

Published


Reading time: 3 min

A Google data center in Hanau, Germany, on January 6, 2023. (ARNE DEDERT / DPA / AFP)

Google and Microsoft’s CO2 emissions have increased by more than 40% in recent years. This increase is largely due to the construction of new data centers essential to AI.

They swore it with their hand on their heart. To slow climate change, the biggest tech companies have made numerous promises in recent years to become at least carbon neutral by 2030. But, far from decreasing, the greenhouse gas emissions of some of the internet giants have since swelled, particularly because of the development of one of the most fashionable technologies in recent months: artificial intelligence.

This is the case of Google, whose greenhouse gas emissions increased by 13% between 2022 and 2023 and by 48% in five years, according to the American group’s annual report published on Tuesday, July 2. The company’s activities (which include its search engine, but also the Google Drive and YouTube platforms, and Pixel smartphones) led to the emission of approximately 17.16 million tons of CO2 equivalent in 2023.

How can this increase be explained? Google justifies it in particular by an increase in so-called “scope 2” emissions, which include the production of electricity purchased by the company to power its data centers. However, these computer data storage units are essential to the functioning of the vast majority of AIs. The latter, like ChatGPT, must analyze millions of examples of texts or images to be able to generate new ones on demand – a process that is also energy-intensive.

Google also points to an increase in “scope 3” emissions, which include the emissions of all actors participating in the production chain. This therefore includes greenhouse gases linked to the construction of the company’s buildings (including data centers) but also emissions linked to the manufacture of electronic components (the chips used in its data centers, the manufacture of its smartphones, etc.).

One specific activity is therefore particularly singled out: artificial intelligence. “As we integrate AI into our products, reducing emissions may prove difficult”the company concedes. And for good reason: it has thrown itself headlong into the race with other tech giants (like Microsoft or OpenAI) to integrate as much generative AI as possible into its services, from automatic email writing to photo editing to so-called “augmented” internet searches.

Google isn’t the only tech company seeing its emissions soar: one of its main competitors, Microsoft, is in the same boat. The company’s operations contributed 17.16 million tons of CO2 equivalent to the atmosphere in 2023, up from 12.22 in 2020 (the first year such measurements were taken), according to its environmental impact report published in May. That’s a 40% increase in four years.

Here too, the company points the finger “the construction of new data centers”. Microsoft, which has partnered with OpenAI (the creator of ChatGPT), also produces its own AI software and integrates it into most of its services. However, the company aims to be “carbon negative” by 2030, but this goal was unveiled in 2020, “before the explosion of artificial intelligence”justifies Brad Smit, the president of Microsoft, to Bloomberg.

The two tech giants defend themselves by highlighting their efforts to reduce the consequences of their activities. In particular, they say they are improving the operation of their data centers and increasingly sourcing renewable energy, and assure that progress in AI will make it possible to optimize energy consumption or find new solutions to global warming. What about other companies in the sector? It’s complicated to know, because these emissions are difficult to quantify and often surrounded by a thick veil of secrecy.

Apple claims its CO2 emissions have fallen in recent years (PDF)but the iPhone designer is still lagging behind its competitors in AI, and the many AI features it unveiled in June will not arrive on iOS until the fall in the United States. For their part, Amazon and Meta (Facebook’s parent company) have not published data on the subject since 2022, before the explosion of generative AI applications. And that’s without counting the emissions of the hundreds of start-ups that have emerged in recent months to ride the wave.


source site-33