Meta Platforms (META)

Family of Apps is an advertising machine with $226.0bn of revenue at a 46.9% operating margin, 3,600 million daily users and a wide moat (network effects, data and computing scale). Revenue is accelerating —+27.6% over the trailing twelve months, +28% in the second quarter— because artificial intelligence improves both time spent and the price of ads. The quality of the advertising business is not in question: the return on installed capital remains at 28.3%. What is in question is how much of that return is being reinvested in building the infrastructure of the next decade, and at what rate it pays off.

Meta is valued on a single multiple over owner earnings, not by sum of the parts: Family of Apps carries the value and Reality Labs is treated separately, as an option and a cost. The metric is NOPAT —operating income after tax—, which expenses stock-based compensation ($25.1bn, ~11% of revenue) and which reported free cash flow adds back.

The base case no longer projects margin expansion but the opposite: operating margin falls from 38.1% to ~34% in 2026 —the level implied by the company's own expense guidance— and recovers only partially to 38% by year 5, because depreciation of artificial intelligence capital expenditure is a cost that does not go away. On that basis the five-year value is ~$1,159 per share, a total return of ~+16% annually from the current ~$565. At today's price it trades at 20× owner earnings, a multiple that compresses toward 9× over five years.

The verdict is Very undervalued. There is a margin of safety: the market's perception is meaningfully worse than reality. At ~$565 the expected annual return is around ~+16%, which has to be read against the method's scale: 4% covers inflation, 10% matches the average return of equities and 15% defines a great investment. The base assumes NOPAT grows from $72.2bn to $153.0bn by year 5, which demands two things at once: that revenue sustains an orderly deceleration from +21.5% to +12%, and that the operating margin recovers from the 34% guided for 2026 to 38%.

The margin of safety narrowed versus the previous reading, and not because of the price but because of the business: the same revenue path now produces considerably less profit. It is a quality business bought at a reasonable price, not a deep-value opportunity.

The central risk is not advertising growth —which is accelerating— but capital allocation, and this quarter put numbers on it. Capital expenditure is guided to $130-145bn for 2026, invested capital grew $77.8bn over twelve months and NOPAT only $6.8bn: an incremental return of 8.8%, below the 10% bar. Part of it is a matter of timing, because data centers under construction do not yet bill, but it is the measurement to follow quarter by quarter, and if it does not recover the investment is malinvestment.

Three signals accompany it and are best read together: buybacks stopped entirely (zero in the first half of 2026, against $26.3bn in the prior year), the company issued $24.9bn of debt, and second-quarter free cash flow was $784 million. Management is choosing to build rather than return capital. The second front remains Reality Labs: the loss has stabilized at ~$19.1bn a year, neither converging nor blowing out.