The Most Expensive Spell-Check in History
Torsten Sløk, the chief economist at Apollo Global Management, recently dropped a data bomb that basically says AI has the financial ROI of a pet rock fed with premium wagyu beef. We are currently witnessing a capital expenditure frenzy that makes the 1630s Dutch tulip craze look like a disciplined retirement savings plan. Companies are pouring hundreds of billions into Nvidia chips and massive data centers, all in hopes that a chatbot will eventually figure out how to do something more profitable than writing a haiku about a depressed toaster.
Imagine walking into a casino, putting $50,000 on a single hand of blackjack, winning, and then realizing the casino only pays out in expired Arby’s coupons. That is the current state of the AI enterprise market. We’ve built the most sophisticated statistical engines in human history, capable of simulating the birth of a star, and we’re using them to help middle managers write slightly less passive-aggressive emails. It is a mismatch of cosmic proportions.
The Efficiency Trap is a Circular Firing Squad
Here is the punchline that the venture capitalists don't want to tell you: even when the AI actually works, it doesn't make anyone richer. It just makes everything cheaper for the customer until the entire industry is starving. This is the "Efficiency Trap." If every law firm in the country uses AI to draft contracts in ten seconds instead of ten hours, they don't get to keep that extra ten hours of profit. They just get into a bidding war until a divorce settlement costs the price of a McDouble.
It’s a race to the bottom where the only winner is the guy selling the sneakers. In this case, the guy selling the sneakers is Jensen Huang, and his sneakers are $40,000 H100 chips that consume more power than a small European nation. Corporate America is currently a group of people standing in a circle, pointing guns at each other, and shouting "I'm going to be so efficient that none of us will have any money left!"
Photo by Mark Stebnicki on Pexels
Building a Cathedral for a Hamster
We are currently seeing a massive, unbacked infrastructure bubble that assumes we will one day find a use for all this compute. It's like building a 12-lane highway into the middle of the Sahara Desert because you're convinced that one day, camels will start driving Ferraris. Since 2023, the "Magnificent Seven" tech companies have signaled they'll spend upwards of $200 billion a year on AI infrastructure. For context, you could buy the entire New York Yankees franchise about 28 times for that amount, and at least the Yankees occasionally provide entertainment value between the disappointment.
This infrastructure is being built on the assumption that productivity gains will eventually translate into earnings. But as Sløk points out, the needle hasn't moved. We have all these shiny new toys, and yet the actual economic output is just... sitting there. It's like buying a $10,000 professional espresso machine and then realizing you still prefer the taste of dirt-cheap instant coffee. Except in this scenario, the espresso machine also has a tendency to hallucinate that your milk is actually gasoline.
The Great Deflationary Ghost Dance
What we’re really doing is creating a deflationary cycle for services that will eventually hollow out the very companies trying to profit from it. If AI makes content creation, coding, and legal research essentially free, then the market value of those things becomes zero. You can't run a trillion-dollar economy on things that cost zero dollars. It’s the Napsterization of everything, but this time with more existential dread and fewer Metallica lawsuits.
Photo by Kampus Production on Pexels
The tech giants are essentially subsidizing the world’s most expensive experiment in commodity creation. They are spending the GDP of a medium-sized country to ensure that in five years, you can get a custom-written screenplay about a sentient taco for the low, low price of absolutely nothing. It is a bold strategy. It is also, from a purely fiscal standpoint, absolutely hilarious.
What This Actually Means
We are currently in the "fake it 'til you make it" phase of a technological revolution, but the "making it" part requires the laws of economics to take a permanent vacation. For AI to pay off, it needs to do more than just make things faster; it needs to create new value that people are actually willing to pay for. Right now, it’s mostly just making existing things faster and worse, which is a great way to go broke at the speed of light.
Expect a massive correction when the CFOs of the world finally look at the bill for their AI initiatives and realize the only thing they’ve successfully automated is the process of losing money. The bubble won't pop because the tech is bad—the tech is actually quite impressive. The bubble will pop because $600 billion is a lot of money to spend on a digital intern who thinks there are three Rs in the word "strawberry."
In the end, we’ll be left with a lot of very fast computers, a lot of very cheap digital garbage, and a lot of very confused investors wondering where their yachts went. But hey, at least the AI will be able to write a very poetic apology letter for the impending recession. It'll be grammatically perfect and completely soulless.
Quick Answers
Is AI actually making companies more profitable right now?
Mostly no; while it saves time on specific tasks, the massive cost of the hardware and the resulting price competition usually eat all the gains.
Why is Apollo’s economist worried about a bubble?
Because companies are spending hundreds of billions on infrastructure (GPUs and data centers) without a clear path to generating hundreds of billions in new revenue.
Is the "Efficiency Trap" real?
Yes—if everyone gets 20% faster at the same time, the market just lowers prices by 20%, leaving everyone exactly where they started but with higher electricity bills.
