Meta beats on revenue but stock plunges 9%. A massive tax charge obscures record spending - and reveals the mounting pressure behind the AI infrastructure race.

There is a particular brand of corporate optimism that sounds like this: record revenue, beating analyst expectations, AI driving growth across every metric. And then, buried in the fine print, a $15.9 billion one-time tax charge that wipes out your earnings per share and sends your stock tumbling 9% in after-hours trading.

Welcome to Meta's Q3 2025 earnings report - a document that simultaneously proves Mark Zuckerberg's AI gamble is working and raises uncomfortable questions about whether anyone, including Zuckerberg himself, knows where this ends.

Meta reported record quarterly revenue of $51.2 billion on Wednesday, beating analysts' expectations of $49.5 billion for the third quarter, but its diluted earnings per share of $1.05 were far below the FactSet analyst consensus estimate of $6.72. 

The reason? Meta said that the implementation of President Donald Trump's One Big Beautiful Bill Act resulted in a one-time, non-cash income tax charge of $15.93 billion

Without that tax hit, Meta said diluted earnings per share would have been $7.25 - comfortably above expectations and consistent with a company firing on all cylinders. But Wall Street doesn't trade on hypotheticals. It trades on reported numbers. And the reported numbers looked like a disaster.

More troubling than the tax charge, however, was what came next: Meta announced it's raising capital expenditure guidance again, signaling that 2026 will see spending increase at a rate that makes this year's record outlays look modest. The company raised its 2025 guidance for capital expenditures, which will now come in the range of $70 billion to $72 billion, up from a prior outlook of $66 to $72 billion.

Meta raised the low end of its total expenses for the year by $2 billion, saying expenses will come in between $116 billion to $118 billion, a figure that was previously $114 billion to $118 billion.

The market's reaction was swift and brutal: Meta shares dropped 9% after the company reported third-quarter earnings on Wednesday Prior to earnings, the stock had risen 28% in 2025. In a single evening, much of that confidence evaporated.

The question investors are now asking isn't whether Meta's AI investments are working. It's whether they're sustainable - and whether Zuckerberg knows when to stop spending.

The Revenue Story: AI Is Actually Working

Strip away the tax charge and the guidance anxieties, and Meta's Q3 results tell a remarkably positive story about a company that has successfully weaponized artificial intelligence to print money.

Revenue of $51.2 billion represents strong year-over-year growth, driven almost entirely by advertising - the Family of Apps segment (Facebook, Instagram, WhatsApp, Messenger, Threads) that contributed roughly 98% of total revenue last quarter. And AI is the engine behind that growth.

Meta has transformed how digital advertising works. Where advertisers once manually created campaigns, defined audiences, and optimized budgets, Meta's AI now does most of that work autonomously. The system identifies potential customers, generates ad creative variations, adjusts bidding strategies in real-time, and optimizes conversion rates - all without human intervention.

The result? Advertisers get better returns on investment, which makes them spend more, which generates more data, which makes the AI smarter, which delivers even better results. It's a flywheel, and it's spinning faster every quarter.

On the earnings call, Zuckerberg discussed "Vibes," a new AI-enabled content type: "Vibes is an example of a new content type enabled by AI, and I think that there are more opportunities to build many more novel types of content ahead".

This is the monetization thesis in action: AI doesn't just improve existing products; it creates entirely new formats that didn't exist before, each one a potential new revenue stream.

Meta expects fourth quarter 2025 total revenue to be in the range of $56-59 billion, reflecting an expectation for continued strong ad revenue growth, partially offset by lower year-over-year Reality Labs revenue in the fourth quarter.

That Q4 guidance - $57.5 billion at the midpoint - slightly exceeded Wall Street's expectations but wasn't enough to offset concerns about spending. The market wanted blowout guidance to justify the capital expenditure trajectory. It got solid-but-not-spectacular instead.

Reality Labs: The $70 Billion Question

Meta's Reality Labs hardware unit reported a third-quarter loss of $4.4 billion on $470 million in sales, bringing total losses since the division first began reporting numbers in Q4 2020 to more than $70 billion.

Read that number again: $70 billion in cumulative losses.

This is not a rounding error. This is not investment for the future with clear returns on the horizon. This is Zuckerberg betting the company's advertising profits on a vision of the future that may or may not materialize - and doing so at a scale that would bankrupt almost any other company on earth.

The only reason Meta can absorb these losses is that its core advertising business is so absurdly profitable that it generates enough cash to fund what is essentially a decade-long science experiment.

Meta finance chief Susan Li said on the earnings call that Reality Labs revenue in Q4 is expected to come in lower compared to the year prior, due to the company not releasing a new VR headset this year and because retailers bought inventory of Meta's previous headset for the holiday shopping season during the third quarter.

There is, however, a bright spot: Li noted that "We're still expecting significant year-over-year growth in AI glasses revenue in Q4 as we benefit from strong demand for the recent products that we've introduced, but that is more than offset by the headwinds to the Quest headsets"

Ray-Ban Meta smart glasses - the AI-enabled wearables that let users ask questions, take photos, and interact with AI assistants - have found genuine commercial traction. They're not selling in iPhone-like volumes, but they're selling. And more importantly, they represent a proof of concept: wearable AI that people actually want to use.

But even if smart glasses become a genuine product category, they won't offset $70 billion in losses anytime soon. The metaverse bet remains what it has always been: a long-term gamble that requires faith more than evidence.

The Capital Expenditure Arms Race

Here is where things get uncomfortable.

Meta's much-watched capital expenditures hit a record high of $19.37 billion in the third quarter, up from $17.01 billion in the second quarter. That's nearly $20 billion spent in three months on data centers, AI infrastructure, servers, and the physical backbone required to train and deploy large language models at scale.

For context: that's more than most countries spend on their entire defense budgets in a year.

The company raised its 2025 guidance for capital expenditures to $70 billion to $72 billion, up from a prior range of $66 to $72 billion But the truly alarming signal came when executives indicated that 2026 spending would be "notably larger" than 2025.

Analysts are now projecting 2026 capex at nearly $97 billion - a 41% increase over current estimates - as Meta races to build AI infrastructure faster than competitors can keep up.

Zuckerberg has made his philosophy clear: he would rather overspend on AI than risk falling behind. The calculation is that being first to "superintelligence" - AI systems that can reason, learn, and improve autonomously - is worth almost any price. Because the company that gets there first may be unbeatable.

But this creates a prisoner's dilemma across the entire tech industry. If Meta spends $100 billion on AI infrastructure, Google must match or exceed that to stay competitive. If Google spends more, Microsoft must respond. If Microsoft escalates, Amazon can't sit idle. And so the spending spirals upward, each company convinced that the alternative - falling behind - is existential.

The problem: no one knows when the returns will justify the investment. Or if they ever will.

The Tax Charge: A Trump Gift That Feels Like a Curse

Meta's $15.93 billion one-time, non-cash income tax charge stems from President Trump's One Big Beautiful Bill Act The law restructures corporate taxation in ways that create immediate accounting hits in exchange for future cash flow benefits.

The company said it expects the act to result in "a significant reduction" in its U.S. federal cash tax payments for the rest of 2025 and future years.

So this is, technically, good news: Meta will pay less in actual cash taxes going forward. But because of how corporate accounting works, the benefit shows up as a massive one-time expense now, obliterating reported earnings even though no actual cash left the building.

It's the kind of accounting quirk that sophisticated investors should understand and look through. But markets don't always behave rationally, especially when a stock has run up 28% year-to-date and investors are looking for reasons to take profits.

The tax charge gave them that reason. And once the selling started, concerns about rising capex and modest Q4 guidance piled on, turning a routine post-earnings dip into a 9% rout.

What Wall Street Is Actually Worried About

The stock's 9% decline isn't really about the tax charge. Investors know that's a one-time accounting event with no long-term impact.

What's driving the sell-off is a more fundamental concern: Meta is spending faster than it can articulate a clear return on investment timeline.

Investors are worried that Meta expects its capital expenditures to be 'notably larger' in 2026 versus 2025, raising concerns about the size of the investment Zuckerberg is planning to make.

There's also anxiety about whether the company is building too much, too fast. Zuckerberg has said he'd rather "overspend than underinvest" in AI, stating that some companies may be overbuilding - but all investors are making rational choices.

That "rational choices" framing is doing a lot of work. Because from the outside, spending $70+ billion this year and planning to spend $97 billion next year on AI infrastructure - while simultaneously losing $70 billion on Reality Labs over four years - doesn't look rational. It looks like a company whose CEO has decided that winning the AI race is worth bankrupting the advertising business if necessary.

The difference is that Meta's advertising business is so profitable it can afford this level of spending. For now.

The User Growth That No One Is Talking About

Buried beneath the capex anxiety and tax charge drama is a data point that would be the headline at almost any other company: daily active users reached 3.4 billion last quarter, while revenue surged 22% to $47.5 billion, with the company's operating margin hitting 43%.

3.4 billion daily active users. That's nearly half of humanity logging into a Meta property - Facebook, Instagram, WhatsApp, Messenger, or Threads - every single day.

This is the moat. This is why Meta can afford to burn tens of billions on speculative bets. Because the core business is a money-printing machine at a scale that has never existed in human history.

An operating margin of 43% means that for every dollar Meta takes in, it keeps 43 cents as profit after all operating expenses. That's extraordinary for a company of this size. And it's improving, not eroding, even as AI investments ramp up.

The advertising business isn't just healthy. It's thriving. AI is making ads more effective, which makes advertisers spend more, which generates more profit, which funds more AI research, which makes ads even more effective. The loop is self-reinforcing and accelerating.

The Vision: "Personal Superintelligence"

Zuckerberg has reframed Meta's AI strategy around what he calls "personal superintelligence" - AI systems that understand individual users so deeply they can anticipate needs, automate tasks, and act as personalized assistants across all Meta products.

This isn't Siri or Alexa, which respond to commands. This is AI that proactively suggests, creates, and executes on behalf of users without being asked. An AI that knows you well enough to book travel, draft emails, create content, manage schedules, and make recommendations - all while learning and improving from every interaction.

If Meta succeeds in building this, it would represent a fundamental shift in how humans interact with technology. Instead of apps and interfaces, you'd have an AI layer that sits between you and the digital world, handling complexity so you don't have to.

The business model implications are staggering. An AI that knows you that well becomes indispensable. And an indispensable product can command almost any price.

But building "personal superintelligence" requires infrastructure at a scale that has never been attempted. It requires data centers that can process trillions of interactions simultaneously. It requires AI models trained on datasets so large they don't yet exist. It requires investment measured not in billions, but in hundreds of billions.

This is what Zuckerberg is betting on. And this is why he's spending like there's no tomorrow.

The Uncomfortable Question

Here is the thing about arms races: they only end when someone runs out of ammunition or decides the prize isn't worth the cost.

Meta isn't running out of ammunition. Its advertising business generates enough cash to fund AI infrastructure spending indefinitely. Google, Microsoft, Amazon, and OpenAI are all similarly well-capitalized and equally committed to winning the AI race.

Which means the only way this arms race ends is if one of two things happens: either someone achieves a breakthrough so significant that competitors can't catch up, or everyone collectively realizes they're spending themselves into diminishing returns.

The first scenario - a decisive AI breakthrough - is possible but unpredictable. The second scenario - coordinated de-escalation - requires competitors to trust each other not to defect. In tech, that trust doesn't exist.

So the spending continues. And investors are left to wonder: when does this end? And what happens if the returns don't materialize?

The Bottom Line

Meta's Q3 earnings revealed a company operating in two parallel realities.

In one reality, Meta is a stunningly profitable advertising juggernaut with 3.4 billion daily users, 22% revenue growth, 43% operating margins, and AI technology that's genuinely improving its core business. This company is winning, and winning big.

In the other reality, Meta is spending $70+ billion a year on AI infrastructure with returns that are promised but not yet proven, losing $70 billion on a metaverse bet that shows no signs of profitability, and raising spending guidance for 2026 to levels that even optimistic analysts find concerning.

Both realities are true. And which one matters more depends entirely on whether Zuckerberg's bet on "personal superintelligence" pays off.

If it does - if Meta builds AI systems that become as indispensable as smartphones - the current spending will look like bargain-basement prices for the future of computing. The company will have secured a monopoly position in the next platform, and investors who sold on a 9% dip will be remembered as the people who panic-sold Amazon in 2001.

If it doesn't - if the AI breakthrough never comes, or comes too slowly, or gets replicated by competitors - Meta will have spent hundreds of billions building infrastructure for a future that never arrived. And the advertising business, no matter how profitable today, won't be able to subsidize that level of spending forever.

Wednesday's 9% stock decline suggests that investors are starting to worry the second scenario is more likely than the first. They're not abandoning ship. But they're no longer willing to give Zuckerberg unlimited rope.

The most revealing moment of the earnings call came when Zuckerberg discussed new AI-enabled content types like "Vibes" and said there were "more opportunities to build many more novel types of content ahead." It was optimistic. It was forward-looking. It was classic Zuckerberg.

It was also completely non-specific about when these opportunities would translate into revenue that justifies the spending.

And that, more than any tax charge or capex increase, is what investors are really concerned about: that Meta is building the future faster than the future is willing to arrive.

The bet Zuckerberg is making is historic in its ambition and terrifying in its scale. He's wagering that being first to superintelligent AI is worth spending whatever it takes - even if "whatever it takes" turns out to be more than any company in history has ever spent on a single technological bet.

We'll know in a few years whether he was right. In the meantime, Meta will keep spending, investors will keep worrying, and the rest of the tech industry will keep racing to catch up.

Because in the AI arms race, the only thing more expensive than overspending is falling behind.