AI is useful. I use it.
So when people say "AI bubble", I do not think the right answer is "AI is fake". That feels too simple. The product is real. The usage is real. The demos are real enough.
The part that looks weird to me is the money.
A normal software business has a nice story: build the product, sell it many times, and margins should get better. AI is not that clean. Every prompt has a cost behind it. Every better model needs training. More users mean more GPUs, more power, more cooling, and more data centers.
OpenAI is the clearest example. It reportedly made about $13B in revenue in 2025, but spent about $34B. I know revenue of that size is impressive. But spending that much more than revenue tells me the business model is still not fully proven.
Big Tech can survive this longer because they already print money from ads, cloud, Windows, ecommerce, and all the old boring businesses. But even for them, the scale is wild. Alphabet raised 2026 capex guidance to $180B-$190B. Meta expects $125B-$145B. Microsoft expects about $190B. Amazon said its trailing free cash flow dropped sharply because property and equipment purchases went up, mainly from AI.
These are not normal "let's hire more engineers" bets. These are "let's build the factories first and hope the demand catches up" bets.
Maybe it works. I am not saying it cannot. Microsoft says demand still exceeds capacity. Many companies are actually using AI. I also use it for real work, so I do not want to pretend this is just hype on X.
But usage and profit are different things. If a company sells a very expensive thing too cheaply, people will use it. The question is what happens when the bill becomes visible. Can prices go up? Will companies still pay? Can model costs fall fast enough? Will all these data centers stay full?
The data center part is what makes this feel more serious than a normal software hype cycle. It needs land, electricity, water, chips, and cooling. Axios wrote about political anxiety around data centers and energy prices. The Guardian reported that many planned US data centers are in drought-hit areas. So the cost is not only in spreadsheets. People living near these projects may feel it too.
My view is simple: AI is useful, but the infrastructure race looks bubbly.
The bubble is not "AI will disappear". The bubble is "too much money is being spent at the same time based on the same optimistic assumption."
If demand catches up, the winners will look very smart. If not, there will be a lot of expensive buildings and GPUs sitting around, depreciating.
So I am not anti-AI. I am skeptical of the spending curve. Useful technology can still have a messy investment cycle around it. We saw that with railroads, telecom, internet, crypto, and now probably AI.
The product may be real. The math still has to become real too.