Microsoft’s push into AI, particularly with ChatGPT, is causing a massive spike in energy use and carbon emissions. Training advanced AI models like GPT-4 has consumed an enormous amount of electricity, equivalent to the yearly needs of over 1,000 U.S. households. As Microsoft’s data centers power up to support these AI breakthroughs, their electricity use has soared from 11 to 24 terawatt-hours in just four years, leading to a 42% rise in carbon emissions. Despite ambitious plans to become carbon-negative by 2030, the company’s AI boom is putting its green goals at serious risk, reflecting a broader struggle in the tech industry to balance innovation with sustainability.
The AI weapons race by Big Tech comes at a high energy cost. According to research, for instance, training OpenAI’s GPT-4 required up to 62,000 megawatt-hours or the energy consumption of 1,000 American families over the course of five to six years.
This is also evident when monitoring Microsoft’s power use (measured in terawatt-hours) and associated carbon emissions (measured in million metric tons of CO2 equivalent) during the previous four years, as Pallavi Rao of Visual Capitalist explains below.
The company’s 2024 Sustainability Report, which covers FY 2020–23, is the source of the data. The fiscal year of Microsoft begins on July 1st and ends on June 30th.
Microsoft’s Emissions Goals Challenged by AI Expansion
Microsoft’s electricity use has more than doubled from 11 TWh to 24 TWh in just four years. To put things in perspective, Jordan (population: 11 million) utilizes 20 TWh of electricity annually.
The surge in power use is coupled with a 42% rise in overall carbon emissions, suggesting a greater proportion of renewable energy sources.
These tendencies align with the usage of Microsoft Azure for AI model training and execution, with OpenAI’s ChatGPT being the most well-known example.
Microsoft actually invested “hundreds of millions of dollars” in building a supercomputer just for ChatGPT, which required connecting thousands of Nvidia GPUs.
AI model training takes a lot of processing power. Compared to data centers offering standard email or internet services, ones constructed to supply the aforementioned computing have higher power requirements.
Actually, between 2020 and 2023, the building of Microsoft’s data centers alone was responsible for 30% of the increase in emissions.
This rise in emissions follows Microsoft’s announcement that the company wants to achieve carbon neutrality by 2030. Google is currently mired in a similar maze. It also set a 2030 deadline for becoming carbon neutral. Rather, since 2019, its emissions have increased by 48%.
Last month, GreatGameInternational reported that the MPRP plans a 70-mile transmission line in Maryland, seizing land via eminent domain due to a fossil fuel plant ban. This has sparked a backlash over AI power demands and clean energy needs.
