Valuing AI Without Valuing the Story

The central idea

AI adoption creates investment value only through the cash flows an owner can ultimately receive, relative to the price paid.

In this essay +

1. Separate four claims that often travel together
2. Follow the reinvestment bill
3. Use a small model to expose a big assumption
4. Ask who keeps the productivity gain
5. Reverse the valuation question
6. Scope

An impressive AI product is evidence of a capability. It is not, by itself, evidence that a security is attractive at today’s price. Between the demonstration and an investor’s return sit adoption, competition, reinvestment, financing, and the expectations already embedded in the valuation.

The useful question is how a specific technical change reaches a specific cash flow. Asking that question does not dismiss technological progress. It gives the investment claim a structure that can be challenged.

Separate four claims that often travel together

The first claim is that a capability is improving. The second is that customers will adopt it. The third is that a particular supplier will capture durable economics. The fourth is that buying that supplier’s equity at a given price offers an adequate return.

Evidence for one does not automatically establish the next. A useful product can face intense price competition. A profitable business can be purchased at a price that assumes even better results. A supplier’s revenue can grow while the capital required to support that growth consumes most of its operating cash.

A research note should write these claims separately and attach evidence to each. Otherwise, the most exciting part of the technology story can end up carrying an unsupported valuation conclusion.

Follow the reinvestment bill

A simplified enterprise cash flow bridge is:

free cash flow to the firm =
  after-tax operating income
  + depreciation and amortization
  − capital expenditure
  − increase in non-cash working capital

The definition and adjustments must remain consistent with the business being valued. Lease treatment, acquisitions, and other claims may require additional work. This compact formula is a starting point, not a universal shortcut.

Damodaran’s introduction to valuation connects value with expected cash flows, growth, and risk. His discussion of the determinants of growth links sustained growth to reinvestment and returns on capital. These are general valuation principles rather than an estimate for any AI company.

For an infrastructure-heavy business, useful questions include what must be built, when it must be replaced, and whether utilization supports the expected returns. For an application business, customer acquisition, retention, inference expense, and distribution can be more central. “AI company” is too broad a label to supply a single operating model.

Use a small model to expose a big assumption

Assume a fictional mature business will generate $100 million of free cash flow to the firm next year, growing at 3% forever. Under a constant-growth perpetuity, enterprise value is:

enterprise value = next-year free cash flow / (discount rate − growth rate)

The discount rate must exceed the perpetual growth rate. Holding the cash flow assumption fixed gives:

Discount rate	Perpetual growth	Illustrative enterprise value
8%	3%	$2,000 million
10%	3%	About $1,429 million
12%	3%	About $1,111 million

The operating story did not change, yet the valuation changed substantially. A precise output can therefore rest on uncertain assumptions about risk and long-term economics.

A young, rapidly changing business usually needs an explicit forecast period before any stable-growth terminal value. Applying the perpetuity directly to an early growth phase can hide the transition that matters most.

Enterprise value is also not equity value. Moving to common equity requires reconciling cash, debt, and other relevant claims; a per-share estimate then needs an appropriate share count and treatment of dilution.

Ask who keeps the productivity gain

Imagine a software tool halves the labor required for a task. Customers might pay for that gain, competitors might force the supplier to pass much of it through in lower prices, or the gain might expand usage while leaving margins under pressure.

These are alternative mechanisms, not forecasts. Distinguishing them requires evidence about contracts, switching costs, distribution, differentiation, and customer behavior. A growing addressable market does not determine the split of economic surplus.

Similarly, lower inference cost can improve unit economics, but it can also stimulate additional usage or invite price cuts. The net cash-flow effect depends on the response of customers and competitors. A valuation should make that response an assumption rather than burying it inside a revenue growth rate.

Reverse the valuation question

Instead of starting with a target price, ask what operating outcomes would be needed to justify an observed enterprise value. Then describe the revenue, margin, reinvestment, and competitive conditions those outcomes require.

A reverse valuation does not reveal what the market literally believes. Different combinations of assumptions can produce the same value. Its purpose is to turn a vague enthusiasm into explicit requirements that can be compared with evidence.

Keep scenarios internally consistent. Rapid growth alongside negligible reinvestment and expanding margins may be possible for some businesses, but it is a package of claims that needs support. Label downside cases with mechanisms, not just lower spreadsheet inputs.

Technical progress is worth understanding on its own terms. Investment analysis begins when that understanding is translated into cash flows, uncertainty, and a price. The bridge between those concepts is where the real work lives.

Scope

All numbers describe an invented business. This essay offers a valuation framework, not a fair-value estimate, security recommendation, or assessment of current AI-sector prices.