Meta Platforms used a strong third quarter to show both sides of its AI bet on Oct. 21: the company reported $64.601 billion in revenue and rising ad prices, but it also spent $34.91 billion on capital expenditures in a single quarter, posted negative free cash flow of $2.12 billion, and approved a new non-voting share class that gives it another way to fund expansion without weakening Mark Zuckerberg’s voting control.
That combination matters more than a routine earnings beat. The practical question for investors, advertisers, suppliers and employees is whether Meta’s core ad machine is producing enough measurable AI value to justify a 2026 capital-spending range of $135 billion to $145 billion — and whether shareholders will have adequate visibility as the bill grows.
The ad engine is carrying the load
The picture in Meta’s earnings release is of a company whose existing business remains unusually powerful. Advertising revenue was $63.061 billion out of the $64.601 billion total. Across Facebook, Messenger, Instagram and WhatsApp, average daily active people reached 3.64 billion in September, up 3% from a year earlier. Ad impressions rose 10%, and average price per ad rose 14%.
Those are the numbers that make the AI strategy financially possible. Meta’s Family of Apps segment generated $64.158 billion in revenue and $17.117 billion in operating income in the quarter. Management also forecast fourth-quarter revenue of $73 billion to $76 billion, suggesting the underlying ad business still has momentum heading into year-end.
The quarter was not clean. Total costs and expenses jumped 68% to $51.613 billion, with a $9.89 billion charge tied to Meta’s agreement with U.S. state attorneys general over teen mental-health and social-media litigation doing much of the damage. Net income still rose to $9.860 billion from $2.709 billion a year earlier, and Meta said adjusted earnings were $7.40 per share excluding that charge. The important point is not that the legal hit can be wished away; it cannot. It is that even after absorbing it, Meta still looks like a company funding expansion from a very large operating base. Headcount was down 15% year over year to 66,775, underscoring that this phase of spending is flowing more into compute, facilities and networking than into payroll growth.
What AI has proved — and what it has not
Meta says its AI investments are already improving recommendations, engagement and monetization in advertising. The quarter offers some indirect support for that claim. Higher ad impressions, higher pricing and continued revenue growth are consistent with a healthier recommendation and monetization system, even if they do not isolate AI from every other driver of performance. Right now, the most credible return case for Meta’s AI spending is not a new standalone AI business. It is AI as a productivity and revenue layer inside the existing ad machine.
That matters because the evidence for the second part of Meta’s AI story is thinner. The company said more than 10 million people in the United States and Canada have created accounts for its Muse personal AI agent, and Zuckerberg said on the earnings call that about 25% use it on a given day. That shows real consumer interest. It does not show retention, paid conversion, profitable inference or standalone margins. Meta also said it expects to improve Muse’s cost efficiency by more than tenfold over the next year, but that is a company forecast, not a measured result.
The spending, by contrast, is already measurable. Meta reported $34.91 billion in quarterly capex, including principal payments on finance leases, and free cash flow turned negative. It also narrowed its 2026 capex outlook to $135 billion to $145 billion, from $130 billion to $145 billion. Meta has not broken out how much of that outlay is going to model training, inference, networking, real estate or ordinary platform capacity. That gap matters because AI infrastructure economics depend on utilization as much as ambition. If ad monetization keeps improving and inference costs fall sharply, the fixed-cost buildout can look smart very quickly. If ad demand weakens, component prices rise or compute sits underused, the same buildout can pressure cash generation just as fast.
Reality Labs remains a reminder that Meta can sustain expensive bets for a long time without proving near-term returns. The unit produced $443 million in revenue and a $4.129 billion operating loss in the quarter. Meta says those losses should narrow to roughly $15 billion in 2027 as investment shifts toward other priorities, primarily AI. That is less an end to speculative spending than a reallocation of it.
The new share class widens the runway — and the blind spot
Beginning Jan. 19, 2027, Meta plans to add a publicly listed, non-voting Class C common stock and distribute one Class C share for each existing Class A and Class B share. Meta says the stock dividend should function economically like a two-for-one stock split without changing voting power. More important than the optics of the split is the financing logic behind it: the new class gives Meta another instrument for capital raising, employee compensation, acquisitions and general corporate purposes while preserving Zuckerberg’s control.
For a company entering one of the most capital-hungry periods in its history, that flexibility is meaningful. So is the trade-off. Future Class C issuance could dilute outside shareholders economically while leaving them with no additional influence over strategy. Meta has not quantified how much capital it expects to raise through Class C shares, how quickly it may issue them, or what the per-share effect could be. Investors therefore have a new question to track alongside capex and AI adoption: what exactly the additional stock is funding.
That governance issue is separate from the legal charge, and separate again from whether Muse becomes a business of its own. But all three threads meet in the same place: accountability for a historically large AI buildout. For now, Meta has clearly proved that its advertising business is strong enough to bankroll a huge infrastructure push. It has not yet proved that Muse and Meta’s broader model-and-product portfolio can earn returns commensurate with the scale of the spending behind them. The next checkpoints are practical ones: whether ad pricing and impressions keep improving, whether AI lowers inference costs fast enough to restore cash generation, and whether Class C stock is used as a disciplined financing tool rather than a quiet source of dilution.
By
By
By
By
By
By
