Exponential Growth in AI Could Lead to Massive Increases in Energy Use
The growing use of artificial intelligence (AI) in applications like ChatGPT could lead to exponential increases in energy consumption and carbon emissions if left unchecked, according to a new analysis. In a paper published in Joule, researcher Alex de Vries warns that adding generative AI to Google Search alone could increase its energy use more than tenfold, potentially rivaling the annual electricity consumption of entire countries.
De Vries suggests that while AI is currently responsible for 10-15% of Google’s energy use, this could dramatically rise as AI models grow in size and complexity. He estimates that running 100,000 Nvidia AI servers at full capacity could require up to 8.9 TWh of electricity per year. Further growth to 1.5 million servers by 2027 could consume up to 134 TWh annually – comparable to the current energy demands of Bitcoin mining.
As AI Proliferates, Efficiency Improvements May Not Keep Pace
While Google states that its AI energy efficiency is improving faster than compute demands are rising, de Vries argues efficiency gains may not be able to offset rapid AI adoption across the tech industry. He advises companies to carefully evaluate their AI needs rather than blindly expanding its use, in order to avoid unnecessary energy expenditure and emissions.
De Vries notes it’s still early to quantify AI’s full environmental footprint. But increased awareness now could help avoid runaway energy consumption as AI becomes ubiquitous across search, social media, customer service and other applications. With climate change an urgent crisis, AI developers and tech companies have a responsibility to prioritize efficiency and renewable energy sources.
The path forward likely involves a mix of optimizing AI systems, thoughtful deployment, powering data centers sustainably, and exploring alternative compute architectures. But inaction could lead to AI becoming a major polluter just as it promises to revolutionize industries. With planning and foresight, its benefits can be realized without prohibitive environmental costs.