AI revolution comes with a high power cost

The rise of artificial intelligence (AI) is sending electricity demand soaring across the United States as data centers work overtime to power chatbots, image generators, and video tools, according to recent reports by BloombergNEF and MIT Technology Review.

According to BloombergNEF, US data-center power needs could more than double by 2035, jumping from nearly 35 gigawatts in 2024 to 78 gigawatts. By then, data centers could consume 8.6% of all US electricity, up from 3.5% today. A handful of tech giants, including Amazon Web Services, Google, Meta, and Microsoft, control nearly half of this capacity, giving them major influence over regional energy systems.

AI's thirst for computing power is driving the growth. MIT Technology Review reports that training a model like OpenAI's GPT-4 used roughly 50 gigawatt-hours of electricity, enough to power San Francisco for three days. While training consumes huge energy, generating responses for users now accounts for 80 to 90% of AI's electricity use. Simple text queries can consume just over 100 joules, but creating short videos can require more than 3 million joules, equivalent to riding 38 miles on an e-bike.

Data centers powering AI often rely on dirtier electricity than the US average. MIT Technology Review found that AI data centers' carbon intensity is 48% higher, with natural gas and coal still dominating despite corporate pledges to expand nuclear power. BloombergNEF notes that building new AI-ready data centers can take seven years, which moderates near-term energy growth.

Looking ahead, AI-specific electricity use could reach 165 to 326 terawatt-hours annually by 2028, enough to power roughly 22% of US households, says MIT Technology Review. Transparency around energy consumption remains limited, making it difficult for regulators and researchers to fully understand the environmental impact of AI expansion.

As AI becomes a daily part of life, from automated agents to multimedia creation, its demands on the power grid are set to grow, raising urgent questions about sustainable energy planning and infrastructure for the AI-driven decade ahead.