Is this an AI bubble?
A bubble is not high prices. It is prices that require a story nobody can defend with numbers. Ask what the price assumes, then judge the assumption.
Replace a vibe with four checks you can actually run.
How Is this an AI bubble? works, in one picture
The same argument as the text, as a chain. Each step is what makes the next one possible.
One investment is the fund
Most go to zero and that is not a failure of selection, it is the shape of the asset class. It is why a venture investor has no use for a company that will merely do quite well.
- 1
Every real technology gets a bubble, and the technology still wins
Railways were a bubble and railways transformed the world. The internet was a bubble and the internet ate the economy. The bubble and the revolution are not alternatives; historically they arrive together, because a genuinely enormous idea attracts more capital than it can absorb in the short run.
So 'AI is transformative' and 'AI stocks are wildly overpriced' can both be true, and the argument between the two camps is mostly people talking past each other.
- 2
Check what the price already assumes
This is the only check that really matters. Take today's price, run it backwards through a discounted cash flow, and solve for the growth rate that would justify it. Now you have a single, checkable claim: the market thinks this compounds at X% for a decade.
Then ask whether that number is possible. Not desirable, possible. If the implied growth would make the company larger than its entire addressable market, the price is not a forecast. It is a hope.
Nobody ever calls a top by looking at a P/E. They call it by noticing the P/E requires something arithmetically impossible.
- 3
Separate the sellers from the buyers
The companies selling AI infrastructure are booking enormous, real, cash revenue. That part is not in question. The companies buying it are converting cash into depreciating assets on the expectation of a future return, which is a bet.
In every technology build-out, the equipment sellers get paid regardless of whether the buyers ever earn a return. Ask which side of that trade you are on.
- 4
Watch the return on capital, not the revenue
A company earning 5% on capital that costs it 8% destroys value by growing. The tell for whether the AI spending is rational is not how big the numbers get, it is what happens to return on invested capital as those data centres start depreciating in earnest.
That is a slow signal. It is also the one that actually settles the argument.
You can name the growth rate today's price implies, and say whether it is arithmetically possible.
Read next
A bubble is not high prices. It is prices that require a story nobody can defend with numbers. Ask what the price assumes, then judge the assumption.
The near-monopolies and the commodities, side by side, because they look identical from outside and they are not.