The world is well aware of the environmental issues we are already facing, which is why there have already been measures in place such as alternative power sources and electric vehicles. However, some harmful emissions are unavoidable, such as the growing carbon footprint brought forth by AI data centers.
Google's Carbon Footprint Grows
Every time you submit a prompt or ask a question to an AI tool, you are effectively contributing to the need for powerful data centers that heavily consume energy. With the AI race continuing to rage on, more companies are establishing data centers to support consumer use.
With Google being one of the biggest AI developers in the market right now, the tech giant's carbon footprint has significantly grown since the AI boom. Despite aiming to cut its carbon emissions in half by 2030, it has instead grown to 48%, as reported by The Verge.
In 2023, Google produced 14.3 million metric tons of pollution, which equates to a 13% year-over-year increase. Data centers are known for their heavy use of electricity, and it has become the biggest source of the search engine giant's additional emissions.
With the growing development of Gemini AI and its functions, Google is almost releasing enough CO2 in the environment to rival the emission of 38 gas-fueled power plants in one year. The company itself has already admitted to the struggle of cutting down on greenhouse gas emissions.
"As we further integrate AI into our products, reducing emissions may be challenging due to increasing energy demands from the greater intensity of AI compute, and the emissions associated with the expected increases in our technical infrastructure investment," Google stated.
Unfortunately, companies have yet to find a way to cut down on electricity usage for AI operations, but the company assured that it is already trying to make AI models, hardware, and data centers more energy-efficient.
Some are Worse Than Others
Evidently, Google's carbon footprint is already bad enough, but it is not the only AI company processing AI requests right now. Hugging Face conducted an experiment to determine how much carbon is emitted by its own large language model, BLOOM.
Over a period of 18 days, Hugging Face found that training its LLM led to 25 metric tons of carbon dioxide emissions, but that has doubled when the company included the emissions produced by manufacturing the computer infrastructure needed for training.
That is equivalent to 60 flights between London and New York, as per MIT, and BLOOM is mostly powered by nuclear energy which does not produce carbon emissions. Companies like OpenAI and Meta, however, still heavily rely on fossil fuels.
In the article's estimation, OpenAI is emitting more than 500 metric tons of CO2, and this was back in November 2022 when the company was still training a less powerful AI, the GPT-3. Meta has released limited data so it's harder to estimate its carbon footprint.
This raises a larger concern considering that AI is already being integrated into everyday operations, and that has been the case even before the AI boom. Google and Microsoft is already adding it to their products, and social media companies like Meta continue to release AI features as well.
