AI and Electrical Power

The views expressed are those of the author and do not necessarily reflect the views of ASPA as an organization.

By Barbara Neuby
December 2, 2024

Every state and federal agency now uses various artificial intelligence (AI) services but few administrators seem to realize that the benefits of AI come with huge power demands and water costs. This article, the first in a series of articles, will examine administrative use of AI in states and federal agencies and review the problems of electricity and water demand needed to operate AI systems. Future articles will explore present and future power generation and water conservation efforts, ways in which administrators are coping with power demands and water restrictions and explore best practices that lead to AI sustainability.

What is AI?

AI is a set of technologies that substitute for human intelligence, normally required to find patterns, recognize connections and/or generate content. AI includes hardware, software and a lot of programming to train the first two. AI, or “expert systems,” have been with us for decades. It includes things like tax software that estimates what one owes, QuickBooks spreadsheets that enables businesses to categorize expenditures and revenues properly or the telephone ordering system that turns a “press 2” into a prescription refill. In the 1980s, machines began to learn through memory and pattern recognition. Now broadly in use, statistical packages popped into a hard drive can perform regression analyses and solve higher equations. By 2010, computers learned deeply. That is, they categorized information properly on their own without much human programming. With each iteration of development, microchip sophistication and power needs have grown. Today, in 2024, AI is in its fourth stage, called generative learning, or “GPT,” which stands for “generative-pre-trained.” At this stage, using large language models (LLM), GPT can create content from large data sets, remember those patterns and connections and even suggest new items of interest, all without human intervention.

AI Usage

States have used AI for many different purposes including but not limited to establishing digital services for their citizens, planning transportation routes, logging geographic info, managing general building maintenance and expanding administrative office capabilities. State information officers believe AI will improve digital service efficiency, lower costs and improve cybersecurity. According to the National Association of State Chief Information Officers (NASCIO), 27 states have created a working group or task force on AI and 17 have developed policies on AI use. Thirteen states have created repositories to track AI use among state offices. NASCIO also reports that state employees need training to use AI appropriately. Perhaps the primary theme that stands out in these states is the desire to prevent anti-discrimination or anti-bias in AI systems: States are working to train their system to prevent bias in content returned or generated by AI. The National Conference of State Legislatures (NCSL) notes that a few states have passed legislation regarding AI but state assembly staff use AI to write emails, transcribe hearing reports and assist with formulas.

Nearly every federal cabinet department and most agencies employ AI in some way even though the Governmental Accounting Office (GAO) notes that departments and agencies have no clear plan available regarding appropriate use or anti-bias techniques. Advanced computing, emergency response, environmental modeling, climate forecasting and materials research are some of the primary ways federal workers use AI.

Power Consumption

Although the benefits of AI seem remarkable, two often unspoken problems lurk below the surface. AI is power-hungry and uses huge quantities of water. Sam Altman’s Chat GPT is now in widespread global use in Microsoft’s Bing search engine and Google has employed its own version of AI and is building a 750,000 square foot data center in Mesa, Arizona. One AI search uses the amount of electricity it would take to run a light bulb for 20 minutes and equals the power consumption of 10 Google searches. AI operates out of data centers that need to be closely located due to limitations of fiber optics. According to the International Energy Agency (IEA), machine learning power consumption has doubled every year since 2010 and, with the average age of the U.S. power grid nodes at 40 years, (50 in the European Union), grid capacity is a de facto limit on how much further AI can go. Global data centers use as much power yearly as Italy uses. Goldman Sachs estimates that power demands for all uses will increase 160 percent by 2030 but notes that little is being done to address this issue head on.

Currently, AI power consumption is approximately three percent of all power used according to Goldman Sachs and the Energy Information Agency (EIA). The EIA projects slow growth while Goldman Sachs and Barclays project that AI power needs will grow to eight or nine percent, respectively, by 2030. The Energy Research Institute believes that data center electricity use will double by 2030 and agrees with the Barclay’s projection of AI power demand of nine percent. Morgan Stanley claims AI power use will increase five times over in three years but Wells Fargo goes further, projecting a 550 percent increase in demand. GS’ analysts note that the United States will need to spend $50 billion on new AI power generation alone in the United States. Europe will need $1 trillion, as it will acquire many more large data centers and is further behind in modernization.

Power Sources

At a time when sustainability and decarbonization is a global focus, most of the power generated and used comes from burning fossil fuels. Oil contributes 37-38 percent; natural gas, roughly 35; coal, 10; nuclear, eight; and renewables, roughly 10 percent. Power generation is mainly a private sector monopoly with little to no generation via government means. Water is both a private sector and government operation with government serving as regulator of both power and water.

The next article in this series will delve into the current effort to grow both power and water supplies needed to sustain AI growth.

---

Author: Barbara Neuby teaches at Kennesaw State University, north of Atlanta, where she researches advancing issues like the central bank digital currency, global financial problems, and the pros and cons of artificial intelligence. She believes these issues are connected and will play out in a new global financial system. Neuby can be reached at [email protected].

(No Ratings Yet)
Loading...