Written by Paul Wiseman for Industrial Info Resources (Sugar Land, Texas)--Debates about the place of artificial intelligence (AI) in society--both in utility and in energy demand--have come to the forefront in the last five years. During that time, AI purveyors have begun to realize that the general power grid cannot possibly grow fast enough to meet their skyrocketing demands. So they have scrambled to find power off-grid, from geothermal, field-harvested natural gas and other grid-independent sources.
Trump Touts Stargate JV to Invest in AI Infrastructure
On January 21, President Donald Trump's second day of his second term, he announced a joint venture (JV) among three top data firms, OpenAI (San Francisco, California), Oracle (NYSE:ORCL) (Austin, Texas) and SoftBank (Tokyo, Japan). The JV, called Stargate, is poised to invest between $100 billion and $500 billion in AI and associated power structure, with a particular focus on the industry's growth in Texas.
The Big Picture
Many experts believe AI-based data centers will demand 2% of all global power in 2025, totaling about 536 terawatt-hours (TWh). It is expected to expand to 3-4% of global power demand by 2030. While this is all newsworthy, the concept of humanlike machines and computer brains stretches back more than a century.
In this series we will start with some history of humanlike machines and brains and an overview of the power challenges. Next will be a deeper dive into how its power demands have skyrocketed in the last five years, and how that is expected to further strain an already maxed-out grid struggling with electric vehicles, all-electric homes, Power-to-X, intermittent renewable power supply and other energy transition demands. Then there will be a deeper dive into how AI firms are getting off-grid with field-supplied natural gas, etc. Finally, we will look at the cost-versus-benefit of AI and machine learning (ML) versus the ways their electricity demands challenge the energy transition.
I, Robot
The first known use of the word robot came from a 1921 science fiction play by Czech playwright Karel Čapek. In the play, entitled "Rossum's Universal Robots," the machines were referred to as "artificial people." The first robot was actually built in 1929 by Japanese professor Makoto Nishimura.
Twenty years later, in 1949, the book Giant Brains, Computers that Think came from computer scientist Edmund Callis Berkley.
But it was a conference at Dartmouth in 1955 led by a math professor at that school, John McCarthy, which gave AI its name and also laid the groundwork for its development.
Man vs. Machine
Computers have made news for competing against humans. In 1952, Arthur Samuel created the first computer to play a game on its own (checkers). Samuel is also noted for coining the term "machine learning" in 1959. Then there was IBM's Watson, which in February 2011 defeated the two top all-time Jeopardy! winners. Now AI is making headlines for much more significant things. It manages massive amounts of data, looks for trends and outliers, helps solve complicated problems, does some basic writing, answers questions and much more.
Stretching the Grid
All that intelligence requires a large amount of electrons even for tasks that used to be simple for computers. For instance, Goldman Sachs Research has estimated that a ChatGPT query uses 10 times as much power as a Google search.
The vast majority of this growth in energy use will be centered on North America and Europe. And much of the new power must come from fossil fuels, including natural gas and some coal. Both these regions are already struggling to reduce greenhouse gas (GHG) emissions and to prepare their grids for rises in electric vehicle (EV) charging requirements and other energy transitions that strive to replace fossil fuel use with clean-generated electricity.
On top of the rising electricity demand from data centers between 2025 and 2030, EVs are expected to add another 2.5% to 4.6% of total energy demand in the U.S. While personal EV charging can be largely done during off-peak hours, commercial vehicles may need to be charged during both peak and off-peak hours.
Power-to-X, or using cleanly-generated electricity from electrolysis to create green hydrogen (separating hydrogen from oxygen in water molecules), is used in hard-to-decarbonize industries such as steelmaking, manufacturing and other factory-oriented processes, is another area of significant power demand growth.
A significant purpose of Power-to-X is to use water to produce hydrogen instead of using natural gas, which today accounts for more than 90% of hydrogen produced. Separating hydrogen from water releases oxygen as a byproduct, whereas releasing it from natural gas releases the greenhouse gas carbon dioxide. The natural gas process is much less expensive, so ongoing Power-to-X research is focused on leveling the playing field regarding production costs.
Hydrogen hubs, which use Department of Energy (DOE) funding to research more-efficient ways to create hydrogen from water, moved forward in 2024, and their connection to the grid will likely grow as well.
Clean, but Intermittent
Another challenge for the grid involves the intermittency of the two most popular clean energy sources--wind and solar. While AI data centers churn ones and zeroes 24/7, solar panels bed down at sunset and wind turbines are unpredictable. Increasing amounts of battery storage can help level the supply, but there is a long way to go before that system can supply baseline power.
AI demands 24/7 reliability, so its only real option is in natural gas. Geothermal is reliable, but it is still mostly in development stages, so its current supply options involve the cost and commitment of building specifically for a data center--which some giants like Google Incorporated (Mountainview, California) have the wherewithal to do, but as a general solution, geothermal is still a few years away.
There is also some renewed interest in nuclear power. For instance, Microsoft Corporation (NASDAQ:MSFT) (Redmond, Washington) has signed an agreement with Constellation Energy Corporation (NYSE:CEG) (Baltimore, Maryland) to reopen the Three Mile Island nuclear plant, and Amazon (NASDAQ:AMZN) (Seattle, Washington) is looking to buy nuclear power from the Susquehanna nuclear power station in Pennsylvania, although the Federal Energy Regulatory Commission (FERC) has currently rejected the interconnection request.
Amazon is also looking at small modular reactors for use on a single location.
Next: How AI's Power Demands have Risen in the Last Five Years
Industrial Info Resources (IIR) is the leading provider of industrial market intelligence. Since 1983, IIR has provided comprehensive research, news and analysis on the industrial process, manufacturing and energy related industries. IIR's Global Market Intelligence (GMI) platform helps companies identify and pursue trends across multiple markets with access to real, qualified and validated plant and project opportunities. Across the world, IIR is tracking more than 200,000 current and future projects worth $17.8 trillion (USD).
