On the way back from a funeral in Shreveport, I decided to take I-20 exit 45 at Delhi and drive north about half a mile.
There I witnessed what may be the largest one-location construction project in history—Meta’s Hyperion data center.
For miles, there was nothing but big-building construction underway, with cranes, brick and steel frames, fuel storage tanks, thousands of pieces of building equipment and dozens of buses to ferry construction workers from nearby towns. Mind blowing.
I remember once viewing the Kemper Light Power Plant, which was the biggest construction project in Mississippi's history. Hyperion makes Kemper look like a sharecropper’s shotgun shack.
How big is it? One way to measure is by electricity consumption. The Amazon data center projects in Madison are expected to consume one billion kilowatts. Hyperion is five billion kilowatts. Five times bigger.
The Hyperion site is 2,250 acres, 3.5 square miles and $50 billion. Amazon Madison is about $20 billion.
The enclosed space is 10 million square feet, equivalent to 174 football fields. And unlike a warehouse, this isn't inexpensive square footage. It's packed with extraordinarily expensive AI computing equipment, electrical infrastructure, cooling systems and networking.
ChatGPT describes Hyperion as “one of the largest single-site industrial construction projects ever undertaken anywhere in the world.”
Seeing this project with my own eyes in the middle of rural Louisiana blew my mind. Thirty miles away in Tallulah my company publishes the weekly Madison County Journal. Madison Parish is one of the lowest income counties in the country. It’s been a struggle to keep publishing.
Looking out across the vast expanse, I thought to myself, “It is impossible to compete with a company of this size.” Indeed, for small community newspapers, Facebook is our biggest competitor.
The stock market has been booming, in part because of the AI boom. We’re experiencing our fourth really good year in a row. That’s unusual. You have to wonder what will happen to burst the bubble. Last time it was Covid.
Some prognosticators warn about the AI bubble. U. S. companies have invested $750 billion in AI. That makes AI the biggest investment in a project or industry in history: more than the Manhattan Project ($30 billion), the Apollo moon landing ($309 billion), and the interstate highway system ($634 billion).
Meanwhile, there is worry about the rapid drop in the cost of “tokens”— the unit of measurement by which AI companies charge for their computing power. The cost of AI tokens has collapsed at a rate historically faster than almost any other input in computing history. Over the last three years, the price of a unit of AI intelligence has dropped by 99 percent.
What happens if AI becomes a commodity? AI companies are spending vast resources on creating the perfect AI platform. But what if 99 percent of AI requests don’t require perfection but just “good enough.” I don’t need a Ferrari AI to help me change my oil myself. A simple Honda AI will do just fine.
If AI becomes an easy-to-deliver cheap commodity, how do the AI companies service their $750 billion in debt? And what happens to the huge grid upgrades and new power plants funded by rank-and-file ratepayers such as Entergy customers? If AI profits don’t materialize as predicted, Entergy customers like me and you will end up paying the cost of these “stranded assets.”
Of course, preventing such disasters is why the state legislature created the Public Service Commission, which for decades developed a go slow, meticulous process for approving grid upgrades and new power plants for electricity monopolies like Entergy and Mississippi Power.
But in the rush to get into the data center game, the Mississippi legislature chucked all that out the window. The Mississippi Development Authority supplanted the Public Service Commission, and the Amazon project was approved in record time. The 300-page Senate Bill 2001 was approved in less than eight hours. To this day, the public has no idea how much Amazon pays for its Entergy electricity. It is a secret.
Haste makes waste.
It is ironic that just as Mississippi is bending over backwards to welcome Amazon, the state of Texas has imposed a data center moratorium. We seem to always be a little too late to the party.
There is a theory that offsets the fear of the rapidly declining AI token costs. It’s called the Jevons Paradox, which postulates that when a resource becomes incredibly cheap, people don’t spend less money, they just use vastly more of it. So even if the cost of AI declines by 99 percent, the usage increases by a factor of more than 100.
I have my own back-of-the-envelope calculations. Goldman Sachs estimates worldwide investment in AI at about one trillion dollars. Is that too much? Let’s go through the math:
There are eight billion people in the world. If 10 percent of the people in the world paid $10 a month for AI and the software had a 40 percent gross profit margin, that would be $40 billion a year in profits. At a 10 percent cost of capital, that could justify a $400 billion investment less than half what is actually invested.
But the numbers can rapidly change if you tweak them. If you apply the current 27 price-earnings ratio of U. S. public stocks (instead of a 10 percent cost of capital), the $40 billion in AI profits could justify the trillion-dollar investment that we’re seeing.
And what if 20 percent of the world ends up paying $20 a month for AI? Then the annual profits would be worth $160 billion, easily justifying the investment we are seeing.
And that’s just consumer spending on AI. The current ratio of consumer AI spending to business AI spending is four to one, meaning business spends four times more on AI than a consumer. That could mean $640 billion in AI profits. At a 27 price-earnings ratio, that could justify $17 trillion in worldwide AI capital investment—17 times more investment than we are currently seeing.
These are the calculations being made by our tech titans such a Mark Zuckerberg, Jeff Bezos, Elon Musk, Larry Page, Sergey Brin, Bill Gates, Larry Ellison, Dario Amodei, Sam Altman, and dozens of huge private equity funds such as Black Rock, the Vanguard Group, Fidelity Investments, dozens of sovereign nation investment funds and thousands of public pension funds around the world, not to mention billions of private investors.
There is about $600 trillion in worldwide capital available. Given that, a trillion dollar investment in a technology that could change the world seems puny. That’s what all these AI investors are counting on.
I use AI constantly all day long in everything I do. I used it to research this column. A few years ago, it would have taken days to do the research that I was able to do in a few minutes. Indeed. AI will change the world. It has the capacity to usher in unimaginable prosperity.
Our main challenge is to properly manage this power, so it is used for good and not evil.
