AI Data Centers Set to Double Electricity Use by 2030 as Researchers Race to Cut Energy Consumption
Summary
AI data centers are projected to more than double electricity consumption to 945 terawatt-hours by 2030, adding 220 million tons of carbon emissions, while MIT researchers develop breakthrough solutions to cut GPU power by 70% and optimize renewable energy integration.
Key Points
- Data centers powering generative AI are projected to more than double electricity consumption by 2030 to 945 terawatt-hours, with 60% of increased demand met by fossil fuels adding 220 million tons of carbon emissions
- MIT researchers are developing solutions including reducing GPU power consumption by 70% with minimal performance impact, stopping AI training early to save half the electricity used for final accuracy improvements, and scheduling workloads when renewable energy is available
- Scientists are creating AI-based tools to accelerate renewable energy grid connections, optimize solar and wind predictions, and develop frameworks like the Net Climate Impact Score to help organizations measure AI projects' environmental costs versus benefits