Meta (META): Free Cash Flow Sinks as AI Capex Nearly Doubles

Meta’s free cash flow fell to $784 million last quarter, down from $8.5 billion a year earlier, even as revenue rose 28% to $60.8 billion. Read only that pair of numbers and you would guess something in the business broke.

Nothing broke: the ad business is compounding as fast as it has in years, and capital expenditure of roughly $31 billion in the quarter, almost double the $16.5 billion Meta spent a year ago, is what actually consumed the cash, a bet that only pays off if the infrastructure it bought starts producing revenue nobody else can match.

I want to keep those two stories separate here, because collapsing them into one narrative is how people either panic over a number that is not a demand problem or wave away a real one.

Capex nearly doubled to $31 billion

Operating cash flow held up fine, close to $31.9 billion for the quarter, so the shortfall is not about advertiser demand slowing down. Capital expenditure, including the principal payments on finance leases, came to about $31.1 billion, according to Meta’s own second-quarter release and the numbers filed with the SEC. Subtract one from the other and free cash flow lands at $784 million, a decline of roughly 91% from the $8.5 billion Meta generated in the same quarter last year.

That is a fast pivot. A year ago Meta was still building AI infrastructure at a measured pace; this quarter it spent nearly all of what the ad business throws off on data centers and chips, leaving less than a billion dollars over.

Most of that money is going into data center construction, the custom silicon Meta designs for training and serving its own models, and the power contracts needed to run all of it, rather than into a single flashy purchase I could point to. None of those assets earn anything the day they’re built; they earn something once the models running on them are good enough that advertisers, or eventually other paying products, will pay more for the output. That lag between spending and payoff is the entire risk in this stock right now, more than any single quarter’s cash flow number.

This is not the first time Meta has spent aggressively into a bet before the payoff was proven. The Reality Labs build-out, and the cost-cutting Meta itself later branded a “year of efficiency,” is a precedent for how fast sentiment can flip if a payoff takes too long to show up. I am not calling the AI buildout a repeat of that; the ad-targeting return on AI spend is already visible in the numbers above, which Reality Labs never managed. But the market’s patience is not infinite, and Meta has tested it before.

I made a similar point about the rest of the AI supply chain absorbing this same wave of spending in a piece on the semiconductor cycle hiding behind AI capex. The dollars show up on this side of the ledger too, just as a subtraction instead of an order. Meta is both a beneficiary of that cycle, since its ad-ranking models run on this hardware, and a direct driver of it: a handful of companies spending at this pace is most of what is behind current chip demand in the first place.

The ad engine keeps compounding

Revenue per user keeps climbing, and I read the mechanism as fairly mechanical: Meta’s ad-ranking models keep getting better at matching a given ad to the person about to see it, so advertisers convert more per dollar and bid more for the same inventory. Quarterly revenue of $60.8 billion was up 28% from a year ago, and full-year 2025 revenue of $201.0 billion was up 22% from $164.5 billion in 2024. Gross margin held at 82.0%, barely different from 81.7% a year earlier, which tells you the core ad business is not getting less efficient to run; the spending pressure shows up further down the income statement, not in the cost of serving an ad.

Full-year 2025 net income was actually down 3%, to $60.5 billion from $62.4 billion in 2024, despite that 22% revenue growth. Net margin sits at 30% today. That gap between rising revenue and falling profit is the AI spending, arriving early and showing up as expense before it shows up as anything else.

I would put Meta next to Alphabet here, another advertising business funding an AI buildout out of a cash-generative core, though Alphabet’s search moat and Meta’s social-graph moat monetize differently enough that the comparison is more about the shape of the trade than the specific numbers.

Here is what I am tracking in one place, so the trade-off is visible without scrolling back through the numbers above.

Metric	Value	Context
Revenue (Q2 2026)	$60.8 billion	up 28% year over year
Free cash flow (Q2 2026)	$784 million	down from $8.5 billion a year earlier
Capital expenditure (Q2 2026)	~$31.1 billion	up from ~$16.5 billion a year earlier
Operating cash flow (Q2 2026)	~$31.9 billion	funds both capex and the free-cash-flow gap
Full-year 2025 net income	$60.5 billion	down 3% from $62.4 billion in 2024
Trailing / forward P/E	25.1 / 21.0	implies about 19% EPS growth priced in
Analyst rating	86% buy, avg target $758	14% upside from $665.75

A quant grade slipped, the target didn’t

43 analysts cover this stock, 86% of them rate it a buy, and the average price target of $758 implies 14% upside from the $665.75 Meta trades at as I write this. That is a fairly bullish sell-side picture for a stock whose free cash flow just fell 91% year over year.

The quantitative score I track moved the other direction, from a C to a D over the same period, for the reason a sell-side analyst would wave off: quant models weight near-term cash generation and margin trend heavily, and both just got worse on paper even though the underlying ad business did not. I read that split as sell-side patience for a story it believes in, set against a model that only sees the current quarter’s numbers.

At 25.1 times trailing earnings and 21.0 times forward earnings, the market is pricing implied EPS growth of about 19%, a reasonable ask if the ad business keeps compounding the way it has and the AI spend does not need to stay this high indefinitely. I wrote about what a similar multiple demands in a reverse-DCF post, and the same logic applies here: a 21.0 forward multiple is not cheap, but it is not priced for perfection either.

The stock fell -8.0% on the day Meta reported these results, 2026-07-29. That is within the company’s own range; the average earnings-day move over recent quarters runs 9.6% in either direction, so this was not an outsized reaction by Meta’s own standard, even if it felt large in the moment.

The market still isn’t pricing in fear

Short interest on Meta sits at just 1.1% of shares outstanding, which is about as close to zero as a mega-cap stock gets. Almost nobody is structurally betting against this story right now, capex spike included.

The stock trades 15.0% below its 52-week high of $784 and 28% above its 52-week low of $520, a wide range that reflects how much this name has moved on AI sentiment alone over the past year rather than any change in what the ad business actually sells. Price to sales today is 7.2, versus a five-year average of 7.3; price to book is 6.3 against a five-year average of 6.6. Both sit below their own historical norms, which tells me the market has not re-rated Meta upward for the AI story the way it has some peers. It is paying roughly the going rate for the ad business and getting the AI infrastructure bet close to free by comparison, an asymmetry I find more persuasive than the free-cash-flow headline on its own.

Meta also still pays a small dividend, $2.10 a year for a yield of 0.32%, and nothing in this spending cycle has touched that payout. It is a minor detail next to $31 billion in quarterly capex, but it is one more sign management is not treating the balance sheet as stretched.

The case against my caution

Here is where I could be wrong: if free cash flow climbs back above roughly $5 billion in the third quarter while capex guidance holds flat rather than rising again, that tells you the spending is plateauing relative to what the ad business throws off, and the market’s current patience will look prescient rather than complacent. A rebound like that, paired with WhatsApp business messaging or other newer lines starting to show up as real revenue rather than a footnote, would be a stronger AI story than ad-targeting improvements alone.

What I can’t verify from outside

The one thing I cannot check from outside the company is whether the depreciation schedule on all this new hardware matches how long the equipment actually stays useful. If GPUs and custom chips age out faster than their assigned depreciation life, today’s margins are flattering future ones, a pattern that has bitten other hyperscalers before when they later shortened useful-life assumptions and took the hit to reported earnings all at once.

The number that decides the next quarter

I am watching free cash flow for the third quarter against two thresholds. Under $2 billion, with capex guidance rising again, says the buildout still has no ceiling in sight, and skepticism deserves more weight. Back toward $5 billion or higher, without a matching jump in spending, says the peak is behind this cycle, at least for now, and an 86% buy rating starts to look like it saw this coming rather than got lucky.

Analysis and opinion only, not investment advice. Figures come from Meta’s second-quarter 2026 results on its investor site and its filings on SEC EDGAR; valuation multiples are approximate and were checked on September 23, 2026.