· 2 min read
illuminem summarizes for you the essential news of the day. Read the full piece on MIT Technology Review or enjoy below:
🗞️ Driving the news: US data center emissions have tripled since 2018, now accounting for 105 million metric tons of CO₂ annually, as energy-intensive AI models like OpenAI’s Sora push energy consumption to new heights
• Data centers use 4.59% of all U.S. energy, with their power sources being 48% more carbon-intensive than the national average
• AI's rapid evolution into multimodal models (e.g., video, music generation) is expected to drive emissions even higher
🔭 The context: While data centers support diverse digital functions, AI-specific workloads like training and pinging large models are growing disproportionately
• Many centers are located in coal-heavy regions, exacerbating emissions
• The shift from text-based AI to resource-heavy multimedia models increases energy demands exponentially, with no immediate regulatory pressure to reduce carbon footprints
🌍 Why it matters for the planet: The carbon intensity of AI threatens to undermine sustainability efforts, particularly as AI adoption expands across industries
• Without energy efficiency improvements or cleaner power sources, AI could significantly contribute to worsening climate impacts
⏭️ What's next: Researchers urge policymakers to regulate data center emissions, while Big Tech faces growing scrutiny over sustainability practices
• Innovations in energy-efficient chips and renewable-powered facilities will be critical to mitigating AI’s environmental toll
💬 One quote: “As we scale up to images and video, the data sizes increase exponentially… and emissions will soon jump.” – Gianluca Guidi, lead author of the Harvard study
📈 One stat: 95% of U.S. data centers operate in regions with electricity sources dirtier than the national average, driving carbon emissions upward
See here detailed sustainability performances of green tech companies like DeepSeek, OpenAI and Nvidia
Click for more news covering the latest on carbon
