Meta reports second-quarter results after the close on July 29, and the entire market conversation is about one number: capital expenditure. The company guided to $125 billion to $145 billion of AI infrastructure spending in 2026, raised from an earlier range, and after Alphabet's shares fell more than 7% a week ago on its own capex increase, investors are braced for Meta to do the same. The advertising business is not really in question. Revenue is expected near $60.2 billion, up around 27% year over year, with ad pricing power intact. The fear is entirely about the spending.
But the most revealing fact heading into the print is the opposite of what everyone is watching for. Meta has not been spending too much. It has been spending strikingly little against its own budget, and that underspend, not a potential increase, is the detail that actually tells you something about how the AI buildout is going.
The underspend nobody is talking about
Here is the number that reframes the quarter. Meta told investors in April it would spend $125 billion to $145 billion this year. In the first quarter, it actually spent $19.84 billion, against a Street estimate of roughly $27.6 billion, which is barely a sixth of the low end of its full-year budget in the first of four quarters.
Do the arithmetic and the situation is stark. If Meta intends to hit even the bottom of its guidance, it has to spend something like $105 billion in the remaining three quarters, a wall of capital that has to land fast. So the question the market is asking, will Meta raise its budget, is arguably the wrong one. The more interesting question is whether Meta can deploy the budget it already has, and what it means either way. That single quarter of underspending sets up a genuine fork, and both branches are informative.
Why the catch-up cuts two ways
Follow the two branches, because the stock's meaning depends on which one is unfolding.
Branch one: the spending arrives on schedule, in a rush. If Meta is simply back-loaded, Q2 capex could roughly double Q1 and the following quarters could be enormous, on the order of $40 billion or more each. That validates the guidance and shows the AI buildout is real and proceeding, but it also means a staggering bill lands right as investors are already anxious, and free cash flow, the actual cash the business throws off after this spending, gets compressed hard. In this branch the worry is not that Meta lied about the number. It is that the number, actually delivered, is frightening to look at, and the market may recoil from seeing the checks clear even though it knew the total.
Branch two: the spending keeps lagging. If Q2 comes in well below the roughly $35 to $40 billion a smooth path would imply, it suggests Meta cannot actually deploy capital as fast as it guided, whether because data-center construction, chip supply, or power availability are bottlenecks. That would ease the near-term cash-flow pressure, which sounds good, but it carries a quieter and more troubling message: if the most aggressive AI spender among the megacaps cannot spend what it planned, the urgency of the entire AI infrastructure race looks less like a stampede and more like a story running ahead of what the physical world can build. An underspend relative to guidance is not obviously reassuring. It can mean the buildout the whole sector is priced on is moving slower than claimed.
That is why the underspend is the tell. Whichever way the catch-up resolves, it says something the headline capex figure does not: either the money is about to hit in an alarming wave, or the company that most wants to spend it cannot, and both readings matter more than another guidance tweak.
The EPS problem hiding under the revenue growth
There is a second number that complicates the comfortable "the ads business pays for everything" narrative, and it deserves to be stated plainly. Revenue is growing close to 27%, which is excellent. But earnings-per-share growth has stalled near zero even as revenue keeps climbing, because the AI spending, the depreciation on all those data centers and servers, plus rising costs and interest, is eating the incremental revenue before it reaches the bottom line.
This is the real fulcrum of the Meta debate, and it is more precise than "is AI worth it." A company can grow revenue impressively and still deliver no growth in profit if its costs rise as fast as its sales, and that is close to what is happening. Operating margin has slipped from a peak around 48% to roughly 41%, the first clear sign the AI spend is touching profitability rather than being comfortably absorbed. The bull case requires believing this is temporary, that the spending builds assets which later generate profit, so the flat EPS is an investment phase, not a permanent condition. The bear case is that the spending is a rising, permanent cost of staying competitive in AI, in which case Meta has traded a highly profitable advertising machine for a still-large but structurally less profitable one. Wednesday's numbers will not settle that, but the trajectory of margins and the capex pace are the evidence that will eventually decide it.
The cloud pivot, and the reason Meta might want it
The forward-looking wild card is strategic, and it connects directly to the spending problem. There is persistent reporting that Meta may enter the cloud-computing business, selling raw access to its enormous AI infrastructure to outside customers, which would make it a fourth hyperscaler alongside Amazon, Microsoft, and Google.
The logic is straightforward and worth understanding, because it turns the capex problem into a potential capex solution. Meta is building a vast amount of computing capacity primarily for its own AI needs. If it has more than it can use, or wants to defray the enormous cost, selling the excess to other companies would generate revenue that offsets the spending, converting a pure cost center into something that partially pays for itself. A confirmation that Meta is becoming a compute seller would be a significant strategic development, arguably the most consequential Meta announcement since its metaverse pivot, because it would reposition a portion of the AI buildout from a bet that must pay off through Meta's own products into infrastructure that earns money by being rented out.
That would also change how to read the capex figure entirely. Spending $145 billion to power your own apps is a bet on those apps. Spending it to build a business selling compute to others is a different bet with a different, more direct revenue path. Which of those Meta is actually making is something management commentary on Wednesday could clarify, and it matters more to the long-term story than the quarter's earnings.
How to read the print
The clean way to approach Wednesday is to watch the capex line and its composition more closely than the revenue or EPS beat, because the advertising strength is already known and the spending is where the genuine uncertainty lives. The specific things that matter: whether Q2 capex jumped toward the $40 billion pace a catch-up implies or stayed suppressed, whether full-year guidance moved again, whether management confirms and details a cloud-selling business, and what any of it does to the free-cash-flow and margin trajectory.
The reassuring frame is that Meta is a dominant, highly profitable advertising company trading at a below-market multiple of roughly 18 to 22 times forward earnings, spending aggressively to win a technology that could define the next decade, and if the spending converts to durable engagement gains or a real compute business, today's discounted stock could look cheap. The cautionary frame is that this is a company whose profit growth has stalled under the weight of that spending, which has raised its capex guidance more than once, and which now has to deploy an enormous sum in a compressed window with no guarantee the returns arrive on any particular schedule. Both frames are defensible, and the quarter will nudge the balance rather than resolve it.
The sharpest single lens, though, is the one the market is mostly not using. Everyone is bracing for Meta to announce it will spend more. The more telling question is whether it can spend what it already promised, and how that answer reflects on the pace of the entire AI infrastructure boom. A smooth, massive catch-up says the buildout is real and the bill is coming. A continued lag says the most motivated builder in the industry is hitting the limits of how fast even unlimited money can turn into data centers. Neither is the story the headline capex number tells, and both are more important than whether the guidance range moved by another few billion dollars.