AI Data Centers Project to Account for 8% of Global Power Demand by 2030
Artificial intelligence (AI) data centers are expected to account for 8% of global power demand by 2030, according to projections from the International Energy Agency. The significant energy consumption is attributed to the processing demands of AI, particularly the training of large language models (LLMs) which rely on graphics processing units (GPUs). These GPUs generate substantial heat, necessitating intricate cooling systems that further increase energy and water usage.
A typical ChatGPT query consumes 0.32 ml of water, and the cumulative effect of billions of daily queries results in considerable water consumption. Data centers frequently use evaporative cooling, which leads to significant water loss. Alternatives such as liquid and immersion cooling, waste-heat recovery, and relocating data centers to less water-scarce regions are being explored to reduce the water footprint.
Experts suggest that while energy use cannot be entirely eliminated, it can be made cleaner. Requiring data centers to utilize non-polluting energy sources, including nuclear, solar, or hydro power, is proposed as a solution. Additionally, strategies like model compression and carbon-aware scheduling of AI systems could reduce computational load and energy draw. Regulation of AI use, potentially limiting model size or token count, is also considered a way to manage resource consumption.