Why are towns all over the United States convulsed about development plans for data centers? Dennis Talluto says size matters.

“I think the primary concern that's really driving the paranoia and and the noise level is basically the size and the scope of data centers,” says Talluto, founder and principal of LVL Strategies. “Twenty, 30 years ago, you might have, say, a 50,000- or 100,000-square-foot data center that might draw four megawatts of power. And now you've got half a million square feet for a data center that's drawing five hundred megawatts to a gigawatt just for one site. So I think the the size of the site, the size of the facilities, the acreage involved, that is what's really driving the noise level.”

Talluto spent 45 years working in tech and has consulted for data center developers. He joins the Overton Window Podcast to discuss what is driving the widespread opposition to new hyperscale data center construction.

“I think another aspect that’s driving this is the actual applications and the workloads,” he says. “In bygone days, if you were in banking, there were a lot of transactional computer operations: people swiping credit cards, tellers making deposits. Those are very transactional in nature. The same thing in healthcare, although healthcare had some large data models, and the same in automotives. There were a variety of applications that would be very transactional in nature.

“Artificial intelligence has changed that to what they call the large language models, which are basically very conversational. You can ask all types of questions, some requiring large amounts of computing power and storage in order to give you the answer.”

Talluto considers the powers of eminent domain and the relationship of state to local government in the data center struggle. He faults media for overstating the energy- and water-usage challenges but points out some new features of hyperscale centers.

“What is really driving the agenda today, what’s driving the media noise today, is not these hundred thousand-square-foot data centers,” he says. “There are literally hundreds and hundreds in the Detroit metropolitan area. Many of them are simply in the basements of the companies that run them. People drive by them every day and have no idea what’s in that building. It looks like offices above ground. Below ground, it could be 50,000 or 100,000 feet of raised floor running rooms full of computers.

“But the AI agenda is for these hyperscale sites that are so large and require so much energy that many of them need their own power generation right on site. Many of them require special cooling facilities built on site. All of that takes acreage, to build these things out. Aside from the application, a lot of it is driven by the physics of the technology.”

Talluto compares artificial intelligence today with Google in its 1990s instantiation as a highly effective search engine serving human curiosity.

“I remember when Google first came out and people were raving, ‘You gotta see this,’” he says. “The total knowledge that Google had at its disposal was whatever it could ingest into memory and then index. So it would create pointers for every little piece of information that it knew about and throw enough computing and memory power at it that Google would provide a response very, very quickly — within a second or so, depending on how you structure the search.

“Artificial intelligence is taking it to a whole ’nother level, where you’re using large language modeling, essentially human language. You can just speak to it if you don’t want to type to it, and it will recognize and filter your everyday vernacular in order to come to a digital understanding of what those bits and bytes meant, of what what you asked it to do.”

Talluto notes that some of the data center resistence is informed by concerns about the human ends of artificial intelligence and the rise of legal and de facto social credit systems here and abroad.

“These large language models are being are being fed by people typing in complete paragraphs or speaking everything that they want, and so eventually the AI model you’re working with knows quite a bit about you: your image and likeness, your desires, your feelings, how you feel about politics one way or the other, how you feel about the economy and banking, how you feel about automotives, healthcare, you name it,” Talluto says.

“Over time, you become the product that is minable: Who you are, how you think, how you feel. And the owners who are paying for that data center will sell that product either individually or collectively. So over time they may know that this particular geographic area that is considered very old school, very conservative. You hear the comments about the Midwest values versus East Coast or West Coast. All of those people individually and collectively are going to provide information. They're going to provide the data points that are eventually going to be mined and used by those that are paying for it.”

Listen to the full conversation on the Overton Window Podcast.

Permission to reprint this blog post in whole or in part is hereby granted, provided that the author (or authors) and the Mackinac Center for Public Policy are properly cited.