Meta's Capex Bet: One Number Decides Whether It's Cheap or Expensive
In March we said Meta was fairly valued at $623 and warned that capex was the single biggest swing factor in the call. The stock is now around $583, down about 26% from its peak, and that one assumption has become the whole story: our model says about $659 if Meta's capex normalizes, and about $331 if it stays elevated.
Three months ago we ran the numbers on Meta Platforms and landed on a deliberately dull verdict: fairly valued. Our DCF put intrinsic value near $609 against a market price of about $623, close enough that neither buying nor selling looked like an obvious move. We did attach one warning, though, and it is worth repeating now:
"The single most important variable in Meta's valuation is capital expenditure normalization... If CapEx remains elevated, whether because AI infrastructure requires perpetual reinvestment, competitive dynamics force continued spending, or the investments fail to generate adequate returns, the Phase 2 FCF expansion may not materialize. In that scenario, Meta's intrinsic value could be materially lower than $609."
That caveat has since aged into the main event. Meta trades around $583 today, about 6% below where we wrote and about 26% off its August 2025 high of $790 (it fell as low as $543 in late June before recovering). Over the same months the hyperscalers announced yet another leg up in AI spending, with Meta among the most aggressive of the bunch, and we changed how our model handles precisely this kind of outlay. It is the same revision we walked through in our recent Microsoft piece, where the call was overvalued at $490 and the stock promptly fell about 35%. So Meta is due a fresh look.
What Meta is actually spending
The worry we raised in March was not abstract. Meta's capital budget has gone close to vertical, and the receipts sit in the cash-flow statement:
Table 1. Meta capital spending and cash flow, FY2020-FY2025.
| Fiscal year | Capex | Capex / revenue | Operating cash flow | Free cash flow |
|---|---|---|---|---|
| FY2020 | $15.1B | 17.6% | $38.7B | $23.6B |
| FY2021 | $18.6B | 15.7% | $57.7B | $39.1B |
| FY2022 | $31.4B | 27.0% | $50.5B | $19.0B |
| FY2023 | $27.3B | 20.2% | $71.1B | $43.8B |
| FY2024 | $37.3B | 22.6% | $91.3B | $54.1B |
| FY2025 | $69.7B | 34.7% | $115.8B | $46.1B |
The tension lives in the last two columns. Operating cash flow has nearly tripled since FY2020, to $115.8 billion, yet free cash flow in FY2025 ($46.1B) actually slipped below FY2024 ($54.1B), even though operating cash flow grew about 27% over the year. Capex ate the gap. It jumped 87% in a single year to $69.7 billion, hit 34.7% of revenue, and absorbed roughly 60 cents of every dollar of operating cash flow, a share Meta has reached only once before.
That last happened in 2022, and it is worth remembering how it went. A weak ad market and Meta's first big data-center and Reality Labs push drove free cash flow down to $19 billion and the stock under $90. Three years later it was back near $800. This time it is the AI build-out doing the spending, but the pattern is the same, just bigger.
Meta is far from alone here. The four biggest hyperscalers (Microsoft, Alphabet, Amazon and Meta) have together guided to something like $725 billion of capital spending across 2026, up around 77% on the prior year.
Why we stopped assuming the spending lasts forever
When we valued Meta in March, the model did what most DCFs quietly do: it carried today's capital intensity straight through the forecast and into the terminal value. The scale of the AI build-out pushed us to revisit that, and we concluded it was not a reasonable assumption.
Assuming capex stays this high forever does not hold up. Heavy spending on new capacity is what a fast-growing company does; once growth slows to a steady long-run rate, it no longer needs to build at that pace, and capex drifts back toward the cost of simply replacing worn-out equipment, which is close to the depreciation line. So we made two changes:
- Across the explicit forecast, capex now glides down from Meta's recent (elevated) rate toward maintenance, rather than staying pinned at the build-out level. For Meta that means falling from about 28% of revenue (its normalized recent rate) toward roughly 11% (maintenance, close to depreciation) over the projection.
- In the terminal value, capex is set to maintenance, the only level a perpetuity can actually fund.
It is a truer picture than freezing one figure for eternity, and because it no longer treats the AI spend as permanent, it raises the fair value. Which is exactly why this single lever earns so much scrutiny: it carries most of the answer.
The fork, in numbers
Freeze every other input and move only the long-run capex assumption, and Meta's intrinsic value fans out as follows (price about $583):
Table 2. Meta intrinsic value under different long-run capex assumptions.
| What you assume long-run capex does | Fair value | vs price |
|---|---|---|
| Glides down to maintenance (our standard model) | ~$659 | +13% |
| Stays elevated for ~5 years, then normalizes | ~$623 | +7% |
| Stays elevated permanently (~28% of revenue) | ~$331 | -43% |
| Stays at the FY2025 pace (~35%) permanently | ~$160 | -73% |
The near-term assumption barely shifts the result: keep capex elevated for five years and only then let it fade ($623) and you still land within about $36 of the standard model ($659), because a DCF holds most of its value in the terminal period. The terminal assumption is the whole game, and for Meta the range it opens is unusually violent. Believe the build-out fades and Meta looks roughly 13% cheap at $659. Believe capital intensity is permanently stuck near its 28% norm and fair value collapses to about $331; at the full FY2025 pace of ~35% it falls to around $160.
Why is Meta's spread so much wider than Microsoft's? Because its spending runs so far ahead of its depreciation. Microsoft's elevated rate sits near 20% of revenue; Meta's normalized rate is closer to 28%, with a 34.7% spot reading. The further today's outlay runs above steady-state maintenance, the more brutal a "stays elevated" world becomes.
Because that one input does so much, every DCF page on the platform now carries a "Hold Capex Elevated" toggle. Leave it off (the default) and capex glides to maintenance: Meta is worth about $659. Switch it on and the recent rate holds across the forecast and the terminal: the figure drops to roughly $331. You can flip it on Meta's valuation page and watch the valuation move.
Where that leaves us at $583
A few honest points to finish on.
We never made a dramatic call on Meta the way we did on Microsoft. In March we said fairly valued and flagged capex as the swing factor, and that has aged reasonably: the stock is down about 6% from our write-up and about 26% from its peak, with capex now the center of the debate rather than a side note.
The market has already done much of this work. At $583 Meta trades at about 21 times earnings, modest for a company still growing revenue near 18% a year. Work back from the price and it assumes capex settling near 17% of revenue over the long run, above the ~11% needed just to maintain the asset base but well below the ~35% Meta spent last year. The market isn't treating the build-out as permanent, or as finished. It has landed in between, and most of the gap between today's price and our full "normalizes" value has already closed.
The rest is a call on capex. Fade back toward maintenance and the model says $659, about 13% above today. Stay elevated at 28-35% of revenue and fair value drops to $331 or worse. That range is far wider than Microsoft's, which already trades near its "stays elevated" value. Meta has more room on both sides, and the same one number decides which way it breaks.
We are not pushing you either way. The model lays both futures out in the open, and the "Hold Capex Elevated" switch lets you price the one you believe. Meta is a candidate for your own research, not a recommendation.