The Trillion-Dollar Question: Valuing AI Companies
Standard valuation assumes a predictable revenue stream. Most AI companies are still working out what it is they sell.
The valuations placed on leading artificial intelligence companies have moved beyond the range where conventional analysis is much help. Revenue multiples are high, but not the highest ever seen; the difficulty is that the denominators are unstable in both directions.
Costs that rise with revenue, customers who are also competitors
- Revenue is growing fast but the cost of serving it is also growing, unlike classic software.
- Gross margins depend on future hardware prices and model efficiency, both of which are uncertain.
- Customer concentration is high, and several major customers are also competitors.
- The largest deals are structured with compute commitments that look like revenue and cost simultaneously.
The chip maker invests, the lab buys chips, revenue rises
The most discussed feature of the current market is the number of transactions in which an investor funds a company that then spends the money with the investor’s other holdings. A chip maker invests in a lab, the lab buys chips, the chip maker reports revenue and its valuation rises, which supports the investment.
This is not fraud and it is not unprecedented — vendor financing has a long history — but it does mean that reported demand is partly a function of capital flows rather than end-user consumption.
Three narrower questions worth asking
The honest position is that nobody can value these businesses with confidence, because the terminal state of the industry is unknown. The reasonable questions are narrower: how much of revenue comes from parties who are not also investors, what the cash cost of serving a customer is, and how long the current capital environment is assumed to persist.
Depreciation decides the margin
Software businesses are valued on the assumption that serving an additional customer costs almost nothing. Model providers do not have that property, and reported gross margins vary widely depending on whether they include depreciation on the hardware that produces the inference.
Two companies can report very different margins while running comparable operations, because one accounts for compute as a cost of revenue and the other capitalises the facilities and depreciates them. Analysts comparing headline figures are often comparing accounting policies.
What usage looks like once the novelty wears off
The question that matters most is whether usage persists when it is no longer novel. Subscription churn, the ratio of pilot spending to production spending, and the share of revenue from non-investor customers are the most informative available indicators.
None of them are consistently disclosed. That is itself a signal: companies confident in the durability of their demand tend to publish the numbers that demonstrate it.
Image credit and licence details for every photograph on this site are listed on the credits page. This article is editorial content; it carries no sponsored material.