I’ll make no apologies for being an AI skeptic. I maintain that position today, despite being an active subscriber of Google’s Gemini AI. I really like it for guided learning.

Regardless, the primary problem coming for AI is very similar to what caused some issues during the early days of the internet. Back then, the internet was new and companies were scrambling to build a bunch of high throughput fiber optic cable across the US.

Unfortunately, the “last mile” problem still existed, and despite you maybe living only a mile away from one of the fiber hubs, you still had a 50 year old phone line trying to send that fiber optic internet to your house.

womp, womp

This created a weird problem during the early days of high speed internet - when, initially, the internet backbones had effectively too much capacity compared to their actual use. “Dark” fiber lines actually existed at this time, meaning that they were either not transmitting internet traffic due to lack of the need for it, or they were simply under utilized.

Imagine having spent a bunch of money to build those lines, only to realize that they weren’t being used as you had hoped. How did that look on your balance sheets?

Fast forward to today.

Add to our story an oblivious billionarie, Kevin Leary, of Shark Tank notoriety. In particular, let’s talk about his actions this past week where he successfully strongarmed a set of local county board members in Utah to approve his “Stratos” data center.

This is a 40,000 acre data center that’s going to use up to 9 gigawatts of power.

That’s a data center that’s 8 miles long. It’s bigger than the entirety of Bryce Canyon national park.

And 9 gigawatts of power is 3 times as much power as the entire state of Utah currently consumes!

Besides this effort being terribly mishandled optically, it also showcases a huge problem that Wall Street has yet to come to grips with.

The demands for AI model training and serving is simply insatiable and do not align with real world constraints.

And this isn’t because I’m getting grumpier with time and hate progress in the tech space. Part of this is due to the sheer amount of compute that’s required to train new models these days, a number that seems to only increase over time.

Let’s explain.

Imagine with me that this thing is built and it starts consuming 9 gigawatts of power, and let’s ignore other issues like pollution and its water usage in the desert region that is Utah.

Given that Utah taxpayers are already vying for their own usage of 3 gigawatts of power (in the form of their electric bills), a few things will necessarily have to happen:

  • more power plants will have to be built, along with more infrastructure to transmit more power to more places.
  • prices for electricity with have to go up to effectively throttle what is available amongst all interested parties.

In the first case, you can’t exactly stand up a bunch of new power plants and infrastructure overnight. You need to truck in concrete, do environmental reviews, hire engineers, design plans, and a lot more - all of which take time and (currently) require a lot of approvals. Are these power plants going to be run via coal or gas? Who has to live next to them?

What if Utah simply buys electricity from a neighboring state?

Same story is going to apply here. Electricity requires infrastructure, especially over long distances. Given normal economic supply and demand, neighboring states may find their own electricity infrastructure strained to supply Utah’s needs, which typically means their prices will go up too.

All because I want to occasionally generate a picture of a banana or create a fun borderline-copyright-infringed avatar for my profile somewhere.

I digress.

In the second case from above, you’re effectively taunting the public with class warfare. Basically, you’re enacting Hunger Games rules.

There’s already a growing anti-billionaire sentiment in this country, and giving special tax deals to them while footing Joe the taxpayer with the bill is not going to be an easy sell. Even in the conservative state of Utah.

There’s a third thing that may happen here.

This is why I mentioned the fiber-optic story earlier.

What if the data center is built but Utah can’t supply it with enough power?

Does this mean that the data center is going to sit half used or unused entirely? What kind of environmental impact is that going to additionally add here?

What will happen to all the money that various investors put into this project? Are they going to disappear to zero via a plummeting stock price? Is there another taxpayer funded bailout in the works?

The takeaway?

Do we really want our public infrastructure, paid for with public taxpayer funds, to go to sweetheart deals with billionaires and/or multinational companies? The same ones will like to “socialize the costs, privatize the profits”?

If the past is any indication of the future, all the dollars they receive from the actions made via these data centers aren’t exactly going to be shared by the public who helped facilitate the building of them.

Not to mention that companies like these will abandon these locations when they are no longer profitable to maintain them.

So I guess the real question comes down to this - with the limited resources of our existing power grid, do we feel like the best usage of it should be to power the data centers at the heart of a few tech companies pumping out new AI models?

The same models that are becoming increasingly expensive, by the way, per subscription. Let’s not forget that!

I think you already know my opinion on the matter.