This is what people are missing.
AI doesn’t have to stop working for the bubble to break.
The economics just have to stop getting better fast enough.
AI infrastructure spending is exploding while the cost of producing AI output keeps falling.
Token prices have fallen roughly 70% per year as efficiency improves, according to Man Group. That means companies need enormous increases in usage just to keep revenue growing.
Meanwhile, the spending keeps getting bigger.
Big Tech is expected to spend more than $740 billion on AI this year.
And now we’re seeing the problem in earnings.
Marvell grew revenue 37%, raised its FY2027 outlook to about $12 billion, and still got hammered because the market wanted the AI revenue sooner.
Its potential Google deal could be worth $120 billion, but the major revenue contribution isn’t expected until FY2029.
That’s the issue.
The money is being spent now.
The promised payoff keeps getting pushed into the future.
And if AI gets cheaper faster than AI usage grows, companies can build more capacity while making less money per unit of computing.
That’s a terrible setup for a capital spending boom.
The bubble doesn’t need ChatGPT to disappear.
It just needs investors to realize that $1 of AI revenue isn’t necessarily worth $1 of new infrastructure spending.
Once that happens, the companies selling the picks and shovels have a problem too.
Because eventually somebody has to prove all this spending can earn a return.