- Generative AI models consume massive electricity and water resources, contributing to rising global carbon emissions and straining municipal infrastructure.
- Data center energy use is expected to more than double by 2026, potentially ranking AI data centers among the top five global electricity consumers.
- The environmental impact extends beyond energy, including the carbon footprint of GPU manufacturing, toxic mining practices, and rapid hardware obsolescence.
The rapid rise of generative AI, hailed for revolutionizing industries from scientific research to business productivity, is coming under scrutiny for its significant environmental impact. A new examination by MIT researchers reveals that the massive computational demands of training and deploying advanced AI models are driving up electricity and water consumption at alarming rates, adding to global carbon emissions and straining natural resources.
Training large generative AI models like OpenAI’s GPT-4 requires vast amounts of electricity due to the billions of parameters involved. Once trained, these models continue to consume energy during deployment and fine-tuning. The energy doesn’t just power computations — it also fuels extensive data center infrastructure, often cooled by water systems that put additional pressure on municipal supplies and local ecosystems.
Data centers, central to the AI boom, are expanding rapidly to accommodate demand. Their power consumption has surged globally, climbing from 460 terawatt-hours in 2022 to projections nearing 1,050 terawatt-hours by 2026 — potentially ranking them among the top global electricity consumers. Despite advancements in hardware, generative AI workloads require much more energy than traditional computing, making it difficult for grid operators to keep up, especially when demand fluctuates during model training.
Environmental concerns also extend beyond direct energy use. The manufacture of AI hardware — particularly GPUs — has a substantial carbon footprint due to the complex fabrication processes and emissions from global transportation. Extracting raw materials for these components often involves environmentally harmful mining practices and toxic chemical use, further amplifying AI’s indirect environmental cost.
With generative AI’s fast-paced evolution and short product cycles, older models become obsolete quickly, wasting the energy invested in their development. Researchers stress the need for a holistic approach to evaluating generative AI’s true cost. Without thoughtful regulation and sustainable development strategies, the industry risks undermining its own long-term benefits by contributing to escalating environmental degradation.
