A discounted cash flow (DCF) model values a business using the present value of free cash flows that the business is expected to generate over a forecast period, and beyond that through a terminal value. Free cash flow to the firm (FCFF) is typically calculated as net operating profit after tax (NOPAT), plus depreciation and amortisation, less net increase in working capital and capital expenditure (capex).
The ongoing build-out of AI infrastructure has led companies competing in this area to sharply increase their annual capex outlay. This dynamic has been particularly visible among hyperscaler businesses such as Amazon and Alphabet. Whilst this isn’t necessarily a bad thing, it does create problems when attempting to value such businesses with a DCF.
The key challenge is that a business can remain profitable at the NOPAT level whilst generating little or negative FCFF where capital expenditure is sufficiently high. In this situation, a DCF becomes especially sensitive to assumptions about how long elevated investment persists and the returns it ultimately generates.
For AI-related capex spending, the key consideration is at what point investment translates into future FCFF generating capacity. This is particularly important for terminal value – usually the largest proportion of a DCF’s value. A perpetual terminal value cannot sensibly be based on indefinitely negative FCFF. The model must therefore assume that free cash flow generation normalises to a sustainable level in or before the terminal year. The further that normalisation is pushed into the future, the more valuation depends on uncertain long-term assumptions rather than cash flows visible today.
