News :  When Big Data Comes to Small Town: Data Center Blog Series Part 1

This blog is based substantially on research in a Sept. 29, 2025 unpublished paper by Peter Montague

Part I: Fabrications, Hallucinations, Scheming and Untruthfulness

Part II: The Hype in HyperScale

Part III: BYONCE is Possible 

Introduction

Way back when, about a year ago, we almost never heard a peep about data centers.  Now every day its data center this, data center that. To some they are a bogeyman to be opposed outright, to others an inevitable nuisance, to others a hopeful development, and to still others they’re OK as long as they’re powered by clean energy rather than polluting methane gas.

To the oil and gas industry, they are a pretext to build even more methane burning power plants, because, after all, that is what AI demands, and we can’t possibly deny AI what it “demands.”  And we couldn’t possibly meet that “demand” with “alternative” energy like solar and wind. Or could we?

• In this 3-part blog, we look at the hype in hyperscale data centers. In Part I, we look at Generative AI itself, which is what the new giant data centers are serving. In Part II we look at the numbers behind energy needs for new data centers, and why they will raise electricity bills.  In Part III we look at some of the real world impacts of data centers, especially around noise, water use, jobs and pollution, and whether those impacts can be avoided.

Part I: Fabrications, Hallucinations, Scheming and Untruthfulness

What the heck are data centers, anyway?

Data centers are huge windowless buildings stuffed with electronic equipment. They are essentially giant computers shared by millions of people. They run the internet as we know it. That’s where we get email and news. It’s where Facebook lives, along with WhatsApp, TikTok, Instagram, X, and Bluesky. Your computer and your hand-held device depend on data centers, as do many or most businesses. 

There are already more than 5,000 data centers in the US, 10 times as many as in China or Germany. The internet will not shut down if new ones aren’t built right away.  We’ll be fine.

Most of the new data center boom is for “hyperscale” (i.e., humongous) data centers to support generative artificial intelligence, or genAI. So, the first thing to consider is genAI itself, the main product and purpose of most new data centers.

There are five kinds of Artificial Intelligence (AI)

Predictive AI predicts future outcomes, for example weather, or the maintenance requirements of power plants, or how proteins will fold in three dimensions based on two-dimensional pictures of amino acids (for discovery of new drugs).

Computer vision to mimic human vision of the real world — detect and identify objects, recognize humans, interpret x-rays, recognize roadway hazards, identify the condition of crops, etc.

Physical AI refers to physical objects that interact with the real world, such as autonomous cars and trucks, robots, and drones. 

Agentic AI is a broad term encompassing “agents” designed to do specific tasks, such as virtual voice assistants on mobile devices (Siri, Alexa, Hey Google), systems that optimize the use of energy in buildings or the charging of batteries in electric vehicles, apps that detect spam and malware in emails and text messages, search engines (Google search, Perplexity.ai), fraud detectors in financial services, language translators, facial recognition to unlock devices, and much more. http://bit.ly/46nafWy

Generative AI refers to applications that generate new content, such as text, images, audio and video. This is the AI that has excited the public since late 2022. Examples include OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, Meta’s LlaMa, China’s Deepseek, and many more.

Despite mountains of hype, the future financial success of genAI remains very uncertain. For one thing, genAI software inevitably (and spontaneously) produces some erroneous answers. These were formerly known as “hallucinations,” but more recently renamed “fabrications.”  If you ask Claude to make a shopping list (not sure why you can’t do that yourself, but whatever) and it forgets something, at worst the broccoli will be bland. For anything more consequential it might not be a great idea to rely on a technology that hallucinates or fabricates.

In addition, genAI suffers from two other apparently unavoidable problems: scheming and untruthfulness. According to OpenAI, the company that brought you Chatgpt, scheming occurs when a genAI starts “pretending to be aligned [with the user’s wishes and intentions] while secretly pursuing some other agenda.” GenAI has an observed tendency to make stuff up and, when challenged, to provide untrue explanations of how it reached particular conclusions.  In other words, AI lies and is deeply weird.  Is it intelligent to count on such a cunning technology?

Perhaps, what with arms races and techno-fomo the United States must do its utmost to master AI, but it’s not at all clear how AI will benefit our people, nor that communities must make the sacrifice of hosting the massive hardware that underlies the software that undermines our brains.

In the AI article with the greatest title, “Artificial Intelligence Meets Natural Stupidity,” energy guru Amory Lovins argues that even without the problems of fabrications, scheming, and untruthfulness, relentless hype about genAI seems likely to cause the construction of unneeded AI data centers, in which case some or many projects will go bust.

Therefore, data center proposals should always include discussion of what happens if the data center fails to thrive – how would the community be affected? In Part II, “The Hype in Hyper Scale,” we will look at the implications for utility rates. (Hint: they will go up.)