MARCH 2025
The AI Power Surge
Growth Scenarios for GenAI Datacenters Through 2030
By Karl Smith, Joseph Majkut, Cy McGeady, and Barath Harithas
Abstract
This paper examines the feasibility and implications of sustained growth in generative AI (genAI)
infrastructure through 2030, focusing on three potential constraints: nancing, hardware availability,
and electric power demands. Using a novel modeling approach, we analyze three growth scenarios
based on historical technological adoptions: an eight-or-more year “Second Industrial Revolution”
scenario, a ve-year “PC Explosion” scenario, and a two-year “Dot-Com Boom” scenario.
Our ndings indicate that cumulative capital spending in the most aggressive scenario could reach $2.35
trillion by 2030, with compute deployment reaching 7.0E+30 total oating-point operations (FLOP) and
power demands adding 83.7 gigawatts (GWs) to the U.S. grid. While nancing appears manageable with
U.S. capital markets, and graphics processing unit (GPU) production constraints seem surmountable
through technological advancement, the unprecedented scale of electrical power demands—equivalent
to adding a new state the size of Texas to the grid—poses signicant infrastructure and policy challenges.
Introduction
Generative articial intelligence (genAI) rapidly became the focus of technoloy sector investment
after the release of ChatGPT 3.5 in November of 2022. Since then, genAI has consistently captured the
attention and resources of industry leaders, solidifying its place as a transformative technoloy.
The projections for genAI investment are staggering. Sam Altman, CEO of the leading genAI provider
OpenAI, has reportedly discussed creating a $7 trillion fund for genAI investment before 2030. Such
an enormous number underscores industry leaders’ condence in the transformative potential of
genAI for the technoloy sector and the global economy. Nvidia, the leading producer of GPUs that
provide compute for genAI, saw its revenue grow from $5 billion in the last quarter of scal year 2021
to $35 billion in the nal quarter of FY 2024. This growth has been fueled by the increasing demand for
advanced computational hardware capable of handling the exponential increase in genAI workloads.
