American Enterprise Institute analysis examines data center electricity consumption as a critical constraint on AI infrastructure buildout.

Nvidia became the U.S.'s most valuable listed company on Tuesday, surpassing Microsoft, thanks to the demand for its artificial-intelligence chips. Nvidia's chips have been the workhorses of the AI boom, essential tools in the creation of sophisticated AI systems that have captured the public's imagination with their ability to produce cogent text, images and audio with minimal prompting. CEO Jensen Huang articulated the broader transformation, saying: "We are fundamentally changing how computing works and what computers can do," and asserting that "The next industrial revolution has begun" during a conference call with analysts in May. Computing in the blossoming AI era will be qualitatively different from the past five decades, with these changes showing up in forecasts of increased investment in data centers.

The technological foundation underlying this shift lies in the difference between traditional CPU and GPU architecture. Past computing relied on CPUs (central processing units) that perform calculations based on instructions, with innovations symbolized by Moore's Law allowing more processors to fit in smaller chips. In contrast, the GPUs (graphical processing units) that Nvidia pioneered in the 1990s feature a parallel architecture enabling many independent calculations at the same time and speed, originally designed for graphics in video games. Through the 2000s, Huang and his colleagues recognized that this massively parallel chip architecture is actually a general-purpose technology beyond graphics processing. Nvidia evolved from a GPU company to a machine learning company to an AI company, building a coding platform and developer ecosystem around these capabilities, which now drive both AI training and the ongoing inference that delivers answers to users.

The energy implications of this technological shift are substantial. A 2020 Science article by researcher Eric Masanet and co-authors found that since 2010, global data center workloads and compute instances had increased more than sixfold, while data center internet protocol traffic had increased by more than 10-fold. Data center storage capacity grew by an estimated factor of 25 over the same period. However, electricity use per computation of a typical volume server dropped by a factor of four largely owing to processor efficiency improvements and reductions in idle power, while watts per terabyte of installed storage dropped by an estimated factor of nine. Yet as researcher Alex deVries noted in his 2023 paper in Joule, AI electricity requirements could soon be as large as a country, making data center electricity demand and its implications for infrastructure systems a critical concern.

This new demand arrives within a specific historical context for U.S. electricity consumption. Over the past 60 years, electricity consumption has grown from approximately 700 billion kWh in 1960 to over 4 trillion kWh in recent years, representing an average annual growth rate of approximately 2.2 percent. However, the recent trend has been notably different: between 2008 and 2019, electricity consumption growth was relatively flat, with an overall average annual growth rate of about 0.3 percent, following a significant decrease in 2008 due to the financial crisis. The surge in data center electricity demand driven by AI now poses a fundamental challenge to the infrastructure systems that have experienced minimal growth requirements for over a decade.