How to value an AI company
When there are no earnings, you are not valuing a company. You are pricing a story about margins that do not exist yet. Make the story explicit.
Value a business whose profits are entirely in the future without fooling yourself.
How to value an AI company works, in one picture
The same argument as the text, as a chain. Each step is what makes the next one possible.
The only five moats there are
If you cannot name which of these a company has, it probably does not have one. It is merely doing well, which is a different and far more temporary condition.
- 1
The P/E does not exist, so people reach for P/S. Be careful
Price to sales ignores whether the revenue is profitable at all. A pound of software revenue at 85% gross margin is worth many times a pound of hardware revenue at 20%, and P/S cannot tell them apart. It is a fallback, not a valuation.
- 2
Value it on the margin it will have, and write that number down
The honest method is to forecast the business at maturity: what revenue, at what operating margin, in what year. Then discount it back. Every one of those inputs is a guess, which is exactly why you should state them out loud rather than hide them inside a multiple.
The discipline is that a stated assumption can be argued with. A P/S ratio of 30 cannot.
If your mature margin assumption is higher than any company in history has sustained, that is not a bold thesis, that is an error.
- 3
Treat capital intensity as the central question
Software historically needed almost no capital, which is why it earned such spectacular returns. AI needs data centres, power, and chips that depreciate fast. That is a fundamentally different business model wearing software's clothes.
Free cash flow, not revenue, is where this shows up. Watch the gap between the two widen.
- 4
Ask who captures the value
A technology can be transformative and still make nobody money, if the benefit competes straight through to the customer. Airlines transformed travel and destroyed a century of shareholder capital.
The question is not whether AI is valuable. It is whether the value is defensible, and that is a question about moats: switching costs, proprietary data, scale. If a model can be replicated for a fraction of the cost six months later, there is no moat, however good it is.
You can state the mature revenue, mature margin and year your valuation assumes.
Read next
When there are no earnings, you are not valuing a company. You are pricing a story about margins that do not exist yet. Make the story explicit.
The near-monopolies and the commodities, side by side, because they look identical from outside and they are not.