The Ghost in the Machine
I spent the morning looking at the Capex charts for the Big Four tech giants, and the numbers are genuinely hallucinatory. We are on track to see over $200 billion in AI-related infrastructure spending this year alone, a figure that makes the Apollo program look like a bake sale. Yet, when you look at the broader economy, the productivity growth is... fine? It’s normal. It’s boring. It is certainly not reflecting the arrival of a god-like intelligence that was supposed to automate every white-collar task by lunch.
Torsten Sløk, the chief economist at Apollo Global Management, pointed out something that has been rattling around in my head for days. He noted that AI currently has "no impact on the aggregate economy." That’s a staggering thing to say when Nvidia’s market cap is hovering around $3 trillion. We have built the world’s most expensive engine, but nobody has figured out how to hook it up to the wheels of actual commerce. I keep wondering if we are mismeasuring the output, or if we are simply enamored with a technology that is very good at doing things that don't actually generate new money.
There is a fundamental disconnect between a tool being impressive and a tool being profitable. If I use an LLM to write a memo in five minutes instead of an hour, I have gained 55 minutes of life. But if every one of my competitors also gains those 55 minutes, the value of a memo drops to zero. We are running faster just to stay in the same place, and the only people getting rich are the ones selling the sneakers.
The Efficiency Trap and the Race to Zero
This brings me to a concept that feels like a glitch in the capitalist matrix: the efficiency trap. In a perfectly competitive market, if a new technology makes a service 20% cheaper to produce, the price of that service eventually drops by 20%. The consumer wins, sure, but the corporate profit margin remains exactly where it started. We are seeing this happen in real-time with coding, legal research, and basic copywriting. The cost of production is cratering, which sounds great until you realize it’s creating a deflationary spiral for services.
If AI makes a junior analyst's work instantaneous, companies won't just pocket the extra cash; they will be forced to lower their fees to win the next contract. The efficiency is neutralized by the competition almost immediately. This leads to a bizarre scenario where we have massive capital expenditure (buying the AI) without any long-term increase in earnings power. It’s like a global arms race where everyone buys a nuclear deterrent, but nobody actually gains any territory. You just have less money in the bank and a lot more hardware in the basement.
Photo by DS stories on Pexels
I’m trying to find the historical parallel here. When the steam engine arrived, it didn't just make weaving faster; it enabled the railroad, which changed the physical geography of the planet. When electricity arrived, it allowed for the 24-hour factory. AI, so far, seems to be a "point solution"—it makes existing digital tasks slightly faster. But making a digital task faster doesn't necessarily create a new market. It just saturates the old ones faster than ever before.
The Unbacked Infrastructure Bubble
There is a specific kind of vertigo that comes from looking at the scale of the data center build-out. Microsoft, Google, and Meta are essentially betting the farm on the idea that if they build the intelligence, the use cases will come. But what if the use cases are inherently low-value? We are currently using $40,000 H100 GPUs to help people write slightly more professional-sounding emails. The math doesn't check out. You cannot sustain a trillion-dollar infrastructure on a diet of email drafts and AI-generated cat pictures.
We are building power plants and laying subsea cables for a revolution that hasn't found its "killer app" beyond the chatbot. The risk here isn't just a market correction; it’s a structural misalignment of resources. We are diverting massive amounts of energy and capital away from physical infrastructure—housing, transport, actual power grids—to build a digital brain that might just be a very fancy search engine. If the ROI doesn't materialize by 2026, the pivot away from AI will be violent.
I keep looking for the silver lining, the hidden sector where AI is quietly creating billions in new, non-cannibalized value. Is it drug discovery? Maybe, but the clinical trial pipeline is still a decade-long slog that AI can’t bypass. Is it materials science? Possibly. But those are long-tail bets. In the short term, we have a massive, unbacked bubble of silicon and hope, and the economists are starting to check the receipts.
What This Actually Means
What we are likely witnessing is the difference between "technological progress" and "economic growth." They are not the same thing. You can have a world that feels incredibly futuristic—where you can talk to your phone and it talks back—while the actual GDP growth remains stagnant because the technology is just rearranging the furniture rather than building a new house. The Apollo warning is a reminder that markets eventually care about cash flow, not just compute power.
If the efficiency trap holds true, the biggest winners of the AI era won't be the companies using it, or even the companies building it, but the consumers who get more for less. But that creates a secondary problem: if corporate profits don't rise to justify the $1 trillion in AI investment, the stock market, which is currently priced for a productivity miracle, is in for a very rude awakening. We are in the "finding out" phase of the greatest capital experiment in human history.
Ultimately, I’m left wondering if we’ve confused speed with value. Just because a machine can do something in a second doesn't mean that thing is worth doing, or that anyone will pay for it. We have optimized the process of thinking, but we haven't yet figured out what to think about that actually pays the bills. Until that changes, the AI revolution is just a very expensive hobby for the world’s largest corporations.
Quick Answers
Is the AI bubble going to burst soon?
It depends on the next two quarters of earnings; if Big Tech can't show that AI is actually driving revenue—not just saving costs—the investment will dry up fast.
Why isn't AI making us richer yet?
Because most current AI applications just make existing tasks faster, leading to price wars and competition that eat up any potential profit gains for companies.
What was the Apollo economist's main warning?
Torsten Sløk argued that AI has had zero measurable impact on U.S. productivity data so far, despite the massive amount of money being spent on it.
