Grywacheski: How AI is supercharging America’s demand for energy

Electricity is the lifeblood of our economy. It powers our factories, lights up our homes and businesses and provides the spark that moves our transportation system.

This year, according to the U.S. Energy Information Administration, the U.S. is expected to consume a record-high 4,200 billion kilowatt-hours of electricity. Most of our nation’s electricity (42%) is fueled by natural gas. This is followed by nuclear power (19%), renewables (22%) — such as wind, solar and hydropower — and coal (16%). All fossil fuels, which include natural gas and coal, account for 60% of our total electricity consumption.

But the amount of electricity America consumes is expected to quickly surge. According to global consulting firm ICF International, U.S. electricity demand is expected to grow by 25% by 2030 and 78% by 2050. For perspective, between 2005-20, America’s average annual increase in electricity use was just 0.1%. So, what is driving this sudden surge in demand?

Experts point to the rapid growth of data centers used to support AI technology. AI data centers are highly specialized facilities built to support complex AI workloads such as machine learning, image generators and large language models used to understand, generate and translate human languages. At the core of these AI data centers are hundreds of thousands of high-performance microchips — typically graphics processing units (GPUs) and AI accelerators — that work in parallel to optimize AI models and algorithms. All this high-performance computing, however, requires a tremendous amount of energy.

There are hundreds of AI data centers spread out across the U.S. Many of these use electricity in amounts comparable to entire cities. For example, Colorado-based Vantage Data Centers is currently developing one of the largest AI data centers ever to be built. Once completed, the $25 billion facility in Shackelford County, Texas, will utilize 1.4 gigawatts of electricity. Just one gigawatt of electricity is enough to power a city of roughly one million people.

The growing demand for electricity has already started to strain existing, and often aging, local electrical grids. To address this issue, AI hyperscalers such as Microsoft, Alphabet (Google), Amazon and Meta (Facebook) have started taking their future electricity needs into their own hands. Microsoft, Alphabet and Amazon have already partnered with energy companies to develop modular nuclear facilities. A modular nuclear reactor is a factory-built and standardized reactor that is much smaller in size and output than a traditional nuclear facility. Additional nuclear modules can be installed to scale output as needed as their electricity demand grows. Other hyperscalers, like Meta, have also signed large purchase agreements of electricity with energy companies to ensure a steady supply of electricity to their AI data centers.

AI has quickly become a powerful force. Thought still in its relative infancy, the world is already seeing tremendous advancements in science, engineering, research & development, medicine and finance, among many other fields. But all those advancements brought about by AI require a massive amount of technology. And for now, at least, all that technology is fueled by an equally massive amount of electricity.