The snapshot Meta took on October 21, after the US market close, has two sides that sit together uneasily. In the foreground is an advertising machine running at full tilt, with revenue of $64.601 billion in the quarter ended September 30, up 26% from $51.242 billion in the same period of 2025 — identical growth on a constant-currency basis. In the background, less brightly lit, is a company that now spends more on building data centers than it takes in, and that is getting ready to ask the market for money.

Meta’s third-quarter 2026 numbers

According to the company’s release, advertising revenue from its family of apps (Facebook, Instagram, Messenger, WhatsApp) reached $63.061 billion, up 26%, driven by a 10% increase in ad impressions served and, above all, by a 14% rise in the average price per ad.

Daily active people across the group’s apps averaged 3.64 billion in September, up 3% year over year, while other family-of-apps revenue — powered by paid messaging on WhatsApp and by subscriptions — grew 59% to $1.097 billion. Reality Labs, the headset and glasses division, took in $443 million, down 6%, and lost $4.129 billion.

Further down the income statement the picture gets more complicated. Operating income fell 37% to $12.988 billion and the operating margin was cut in half, from 40% to 20%, while net income jumped 264% to $9.86 billion, or $3.83 per share against $1.05 a year earlier. Two numbers that seem to describe two different companies and that, read together, call for some explanation.

A beat that depends on how you look at it

During the quarter Meta booked a $9.89 billion legal charge tied to the settlement reached in August with 52 US attorneys general, closing out the lawsuits over teen mental health and social media. Stripping out that item, operating income would have been $22.88 billion, up 11%, and earnings per share $7.40.

The 264% jump in net income is just as misleading, resting almost entirely on a base effect. In the third quarter of 2025 Meta paid an effective tax rate of 87%, inflated by an extraordinary tax charge, while this year the rate came in at 24% (it would have been 17% without the non-deductible portion of the legal settlement). In short, the comparison measures last year’s tax bill more than this year’s profitability.

The verdict on Wall Street’s expectations also changes depending on the thermometer. Some surveys put consensus at $3.57 per share, which Meta beat; others at $5.46, which it missed — a sign that not every analyst had folded the legal charge into their estimates. The most telling figure is a different one. Even excluding the settlement, the operating margin stopped at 35.4%, against 40.1% a year earlier. Nearly five points lost along the way that have nothing to do with the courts and a great deal to do with data centers.

The first problem: cash

The number that weighs most in this quarter is not in the income statement but in the cash flow statement. Against operating cash flow of $32.786 billion, up 9%, Meta spent $33.942 billion on property and equipment and another $964 million on finance lease payments, for a total of $34.906 billion — the largest quarterly outlay in its history.

The result is negative free cash flow of $2.12 billion, against a positive $10.625 billion in the third quarter of 2025. According to data compiled by StockTitan, it is the first time since 2012 that capital spending has exceeded the cash generated by the business. Over the first nine months, free cash flow came to $11.049 billion, 63% below the $29.508 billion of the same period last year.

And the worst, at least in terms of outlays, may still be ahead. Meta narrowed its 2026 capital expenditure range to $135–145 billion, raising the floor from the $130 billion it had indicated in July. Since it spent $85.8 billion in the first nine months, staying inside that range means investing between $49.2 billion and $59.2 billion in the fourth quarter alone, between 41% and 70% more than it spent from July to September.

With revenue expected between $73 billion and $76 billion, it is hard to imagine free cash flow turning positive before the end of the year. For 2027 the company gave no figures, limiting itself to a promise to keep investing «aggressively»; analyst consensus, according to surveys cited by Mitrade, sits at around $196 billion.

The balance sheet carries the marks of this race. Meta closed September with $82.45 billion in cash and marketable securities and, as chief financial officer Susan Li noted, $83.7 billion of debt. In other words, the group that for years was a vault full of cash has arrived at a net financial position that is essentially flat — slightly negative, in fact — and that is before counting off-balance-sheet structures such as the $29.5 billion vehicle created in 2025 to finance the Hyperion data center in Louisiana. Long-term debt rose to $80.92 billion from $58.74 billion at the end of 2025, after $24.91 billion of bond issuance during the year, and accounts payable more than doubled, from $8.89 billion to $19.1 billion.

Two further details complete the picture. The first concerns share buybacks, which absorbed $26.25 billion in the first nine months of 2025 and fell to zero in 2026, removing one of the stock’s historical supports. The second is the interest and other financial income line, which went from $1.128 billion to $13 million in the quarter — a hint of how the cost of debt is starting to eat into the returns on cash.

Costs are running faster than revenue

Excluding the legal charge, total costs grew 36% against a 26% increase in revenue. That is the textbook definition of negative operating leverage, and breaking out the line items shows where the pressure comes from. Cost of revenue rose 45%, research and development 32%, and depreciation and amortization 44%, to $7.154 billion.

Meta itself points to infrastructure costs, spending on third-party cloud and the cost of artificial intelligence tokens purchased from outside providers as the main drivers. A company that builds its own models and its own data centers also pays for someone else’s computing power while it waits for its own to come online.

What stands out is that all of this is happening with fewer people: at the end of September the headcount was 66,775, down 15% year over year after May’s cuts. The spending that is growing, then, is not salaries but machines, energy and chips. And there is a new chapter set to weigh in. Muse, the personal agent launched at the end of the summer, is by Li’s own admission very expensive to run, and its total serving costs are set to rise as adoption spreads. The promise is to improve its efficiency more than tenfold within a year; until that happens, every new user is a cost line before it is a revenue line.

The settlement with the attorneys general calls for payments of up to roughly $18.59 billion, but only $9.89 billion was booked in the quarter. How and when the remainder will be recorded is one of the questions the 10-Q — the quarterly filing US companies submit to the Securities and Exchange Commission (SEC) — will have to answer. There is also a cash dimension the income statement does not show.

Judging by the cash flow statement, much of the charge appears to have been accrued rather than paid, given that the other liabilities line contributed $7.439 billion to operating cash flow; the payments, when they come, will weigh on a cash position already under pressure. At the same time, restricted cash recorded under other assets rose to $12.535 billion from $1.612 billion a year earlier — another item to read carefully in the quarterly filing.

Meta itself hedges in its fourth-quarter outlook, flagging growing scrutiny of issues involving minors in several markets and a series of trials already scheduled in the United States in the coming months.

Class C shares: funding AI without giving up control

The news set to be debated longest, though, came at the tail end of the results. The board approved the creation of a new class of common stock, Class C, listed but with no voting rights, to be distributed starting January 19, 2027 to existing shareholders at a rate of one share for every Class A or Class B share held. For anyone holding the stock, the immediate effect will resemble a two-for-one split, with the Class A share price mechanically halving and the new Class C shares beginning to trade at the end of January.

The official rationale speaks of flexibility to raise capital, compensate employees and make acquisitions without diluting voting rights. The documentation filed with the SEC tells a richer and at times more awkward story. Mark Zuckerberg controls 60.6% of the votes against a 13.4% economic stake, and he informed the committee of independent directors that he would not support any capital increase capable of eroding his majority.

The committee had asked in exchange for sunset clauses on the super-voting shares — tied, for example, to his departure from the company or to a reduction in his stake — but the proposal was rejected. Class A shareholders, for their part, were not asked to vote: the transaction was approved by written consent from Zuckerberg’s holding entities, which hold 56.1% of the votes.

In return the founder signed an agreement locking up 80% of his new Class C shares, releasing them at a rate of 2.5% per quarter over eight years on condition that he remains at the helm of the group, and committed to maintaining his California residency. The remaining 20%, equal to 10% of his entire economic stake, will instead be sellable immediately.

The filing explicitly links that possibility to Proposition 40, the wealth tax on large fortunes that California votes on November 3 and that could cost Zuckerberg several billion dollars. Rounding out the architecture is the promise of further stock dividends in Class C shares, one for one for all shareholders, every time the stock doubles from the $580.60 level of September 10, 2026, with a compound annual return of at least 10%.

According to the same documentation, the long-term plans presented to the board anticipate the need to raise hundreds of billions of dollars over several years, starting as soon as this fall, and the committee’s advisor estimated that Meta could issue up to roughly $700 billion of capital before Zuckerberg falls below 50% of the votes.

On the conference call, Susan Li said there were no specific financing plans to announce, but the SEC filing states in black and white that the company expects to issue Class C shares over time. The committee itself acknowledges that the new shares will probably trade at a discount to the Class A stock, forcing the company to issue more of them for the same amount of capital raised, and that some institutional investors and indexes may not welcome them. This is not Menlo Park’s first attempt. In 2016 Facebook proposed a very similar Class C, only to withdraw it in September 2017 on the eve of the shareholder trial brought in Delaware.

The case for the quarter

It would be unfair, as well as unhelpful, to read this quarter only through the cracks. Fourth-quarter guidance, with revenue between $73 billion and $76 billion, sits above the $73.6 billion expected by the consensus compiled by Bloomberg and implies growth of between 22% and 27% against the $59.85 billion of late 2025. Cost estimates indicate that, net of the legal settlement, 2026 operating income will remain above the 2025 level, and Reality Labs losses should fall to $18–19 billion this year and to around $15 billion in 2027.

Artificial intelligence, moreover, is already working for the advertising business. New recommendation systems have pushed time spent on Instagram and on Facebook Reels to record highs, ad ranking models have increased conversions across several surfaces, and creator ad partnerships have passed an annualized run rate of $15 billion, nearly doubling in a year. Muse topped 10 million users across the United States and Canada in about six weeks, with roughly a quarter of them using it daily and usage heavily oriented toward shopping. Meta One, the unified subscription for consumers and businesses, has passed 15 million between subscriptions and free trials. These are numbers that justify management’s enthusiasm, provided one remembers that for now they generate more costs than revenue.

How Meta stock reacted

The market chose to see the glass half full. Meta shares closed the October 21 session at $697.52, up 0.47%, and in after-hours trading climbed as much as roughly 5% before settling at $713 at 10:05 p.m. New York time, about 2.2% above the close. The stock still remains far from the $779.82 high reached on September 24, and is up around 6% since the start of the year — a modest performance for a group growing at this pace.

The reaction marks a change of tone from recent months. After the April and July results, both accompanied by an increase in the capital spending range, the stock fell more than 6% and nearly 8% respectively in after-hours trading. This time the spending increase was more contained and revenue growth acted as an umbrella. It remains to be seen how long investors’ patience will last when Class C turns into an actual issuance.

Editor’s note

This article was originally published in Italian on money.it by Flavia Provenzani on October 22, 2026 as «Meta, i ricavi corrono ma la cassa si ferma. Tutte le zone d’ombra della trimestrale». It has been translated and adapted for an international audience by the Money.it International desk.