For three years, Meta Platforms has told a clean story: the advertising machine pays for the artificial-intelligence build, and the build eventually pays for itself. On Wednesday afternoon, that arithmetic briefly came apart in public.
The company reported third-quarter revenue of $64.6 billion, up 26% from a year earlier and ahead of the $63.3 billion analysts had expected. Net income rose 264% to $9.86 billion, or $3.83 a diluted share. Then came the rest of the page.
A $9.9 billion legal charge lifted total costs 68% and pushed operating income down 37%, to $13.0 billion. Free cash flow was negative $2.1 billion for the quarter. For a business that has spent most of its existence producing cash faster than it could deploy it, the sign in front of that number is the story.
The charge has a courtroom attached. On Wednesday, a New Mexico state judge ordered Meta to pay $25 billion after a jury found the company liable for deceiving users about privacy, the largest penalty ever imposed on the Facebook parent. Jurors concluded that Meta had committed roughly 43 million violations of the state’s consumer protection law, in a case that grew out of the Cambridge Analytica data scandal.
“Facebook promised users that they controlled their data,” Judge Francis J. Matthew wrote in the judgment. “This was false.”
Meta has said it will appeal the September verdict. New Mexico Attorney General Raúl Torrez told reporters the money would be tied up during that process and that, under the state constitution, penalties of this kind flow to education spending. “We may have paid for the construction of schools for the next 100 years,” he said. The state had sought $35 billion to $40 billion; Meta argued that anything above $3.4 billion would violate due process. The judge landed between the two figures, closer to the state.
Meta shares rose about 3% in after-hours trading. Investors, for now, are treating the penalty as a legal event rather than an operating one, in part because the appeal will take years and in part because the company settled a separate multistate case in August for $18 billion without visible damage to its multiple.
The more consequential signal in the release was about capital structure. Alongside the results, Meta’s board approved a new class of publicly listed, non-voting Class C common stock, to be distributed as a dividend of one Class C share for each existing Class A or B share. The charter amendment is set to take effect Jan. 19, 2027, and the company describes the mechanics as working like a two-for-one split for Class A holders.
The new shares are a currency. Meta said it intends to use them for acquisitions, employee compensation and raising capital, according to people familiar with the matter. Because they carry no votes, they let the company issue equity without diluting the control Mark Zuckerberg holds through supervoting B shares. The initial distribution itself raises no cash, which is the point: it creates a pool of stock to spend later.
That is an implicit acknowledgment of what the next phase of the build requires. Meta raised its full-year expense forecast to $175 billion to $179 billion, a figure that includes the legal charge, and narrowed capital spending guidance to between $135 billion and $145 billion. Fourth-quarter revenue guidance of $73 billion to $76 billion came in above what analysts had modeled, which is why the shares held their gain.
The clearest evidence that all this spending has a customer is Muse, the company’s consumer agent. Zuckerberg said more than 10 million people in the U.S. and Canada have set up a Muse agent since its September launch, and that about 25% of them use it on any given day. Shopping is the top activity. Meta said more than a quarter of those users have made a purchase through the agent, with average order values above $200.
If those numbers hold, Muse is less a chatbot than a storefront with a memory, and the advertising business has a new surface to sell against. That is the version of the story Wall Street wants: agents that generate commerce rather than merely consume compute.
Not every line cooperated. Reality Labs, the unit that houses Meta’s glasses and virtual-reality work, lost $4.13 billion in the quarter. Headcount fell 12% from the prior quarter, to 66,000, a reduction steep enough to suggest that the company is already paying for infrastructure with payroll as well as with debt.
Across the largest technology companies, the pattern of this earnings season is similar. AI infrastructure is now funded partly from operating cash flow and partly from capital markets, and the gap between the two is where investor nerves live. When free cash flow turns negative while depreciation climbs, the burden of proof migrates from the product roadmap to the income statement.
Zuckerberg’s bet is that agents like Muse convert into commerce and ad demand quickly enough to cover the depreciation curve. The Class C shares suggest management is hedging that bet: if appetite for AI spending cools, Meta will want a cheaper way to pay for things than cash, and a way to compensate employees in stock that cannot vote.
The open question is not whether Meta can afford the build. It is whether the build can be financed without asking shareholders to accept a lower vote and a lower multiple in the same quarter.
