When most people think of artificial intelligence (AI), they picture futuristic robots or clever software assistants — not power plants. But the reality of AI’s physical footprint is grounded in something far more prosaic: electricity. The need for power is the topic du jour at every Fortune 250 Business and IT planning session and not surprising that the entire power supply chain is up and to the right for the foreseeable future. Everything from copper (and gold and silver and lithium) and the equipment needed to mine it, the manufacturers that take raw materials and turn it into wire and connectors (and a thousand other distribution parts), and even the installers of these components into freshly minted data center structures (like Quanta, Turner and Holder). Yup, its the POWER that is driving nearly everything else for AI.
Artificial intelligence workloads — especially large language models, generative AI, and other advanced machine-learning systems — have driven massive growth in data center energy consumption. Ask an IT professional about power consumption by data centers prior to COVID and they would tell you that Data Centers consume “under 2%” of all power in the USA. Fast forward five years and data centers in the United States now account for roughly 6-7 % of total U.S. electricity use. IDC recently published a chart with contribution by Goldman Sachs that puts the topic in perspective:
Source: IDC, Goldman Sachs Global Investment Research, Department of Energy, Aterio
That share has climbed sharply with the explosion in AI deployments across industries. Five years ago, data center electricity — particularly from AI-oriented computing — was a much smaller slice of the grid’s total. Although exact AI-specific percentages are harder to pin down, estimates suggest the AI portion of data center energy use has roughly doubled during that period as generative models, real-time inference services, and cloud-based AI engines have proliferated. Now the scary part is what happens for the NEXT five years. Go ahead, mentally extend the line on the chart above that mimics the slope you see. Scary, right?
The trend reflects a broader shift: expanding AI buildouts that include more GPUs per farm, each requiring at least as much as the previous generation. Training a single large model can consume as much energy as hundreds of thousands of homes over its lifecycle, and operating them at scale — serving billions of daily inferences — adds further demand. The chart above makes sense and is not sensationalism, its reality!
Looking ahead just a few years, power projections paint an even more dramatic picture. Some forecasts anticipate U.S. data center power use could reach 12-15 % of the nation’s total electricity consumption by the end of this decade as AI and cloud computing continue their rapid expansion. And while we will find ways of reducing the footprint of AI, the power consumption will only increase.
What does this mean for the grid, utilities, and climate goals? The doubling of AI’s energy share in just a few years underscores the urgency of modernizing infrastructure, integrating more renewable energy sources, and improving efficiency at every layer of the AI stack. Whether powering the next breakthrough in medicine or enabling real-time language translation, the electricity behind AI shows that even the coolest digital innovations in AI have very physical ‘old school’ style demands.
Leave a Reply