TL;DR — Meta’s capital spending guidance nearly doubled to $130 billion to $145 billion for the year, on top of an already-record base. Advertising revenue is still the only line item paying for it.
Meta spent $72.22 billion on capital expenditures in 2025. This year, the company is guiding to $130 billion to $145 billion — nearly double, in a single year, on top of a base that was already the largest infrastructure budget in the company’s history. The products that spending is supposed to eventually justify — Llama, Meta AI, the “Business AIs” running inside WhatsApp and Messenger — don’t have a revenue line in the earnings release. Business AIs crossed 10 million weekly conversations last quarter. Disclosed revenue from those conversations: none.
The Business Actually Paying for This
Strip away the AI narrative and Meta is still, overwhelmingly, an advertising company. Revenue hit $60.8 billion in the second quarter of 2026, up 28% year over year, and nearly all of it comes from ads running across Facebook, Instagram, and the rest of the Family of Apps. That engine is healthy. It’s also the only thing funding a capex program that’s about to consume, in twelve months, more cash than the company spent across the two years before it combined. Every dollar going into a GPU cluster or a new data center is a dollar the ad business generated first.
Where “Not Yet Monetized” Actually Shows Up
Meta’s own language about its AI bet is telling. The company describes Meta Superintelligence Labs and its Muse model family as “strategic optionality” — a phrase that means, stripped of the branding, that even Meta isn’t claiming a near-term revenue case for this spending. That’s a meaningfully different posture than what we’ve seen elsewhere this earnings season. Microsoft’s $678 billion commercial backlog is signed, contracted revenue with a weighted average conversion timeline attached to it. TSMC’s advanced packaging capacity is sold out because paying customers have already committed to buy it. Meta’s capex, by contrast, is aimed largely at products the company hasn’t monetized yet, serving users who haven’t been asked to pay for them, built on a timeline nobody outside the company has visibility into.
The Bet Is Also Squeezing the Numbers That Already Work
The capex isn’t the only pressure on Meta’s financials this quarter. Total expenses rose 55% year over year to $42 billion, part of that a $2.4 billion legal charge and $1.18 billion in severance from a May headcount reduction, but stock-based compensation was already running hot before either of those hit — up 52% to $13.69 billion, the fastest growth of any major tech company we’ve tracked this season. Operating income fell 8% to $18.78 billion even as revenue grew 28%, and operating margin compressed from 43% to 31%. Three separate cost pressures — capex, compensation, and one-time charges — are landing on the same free cash flow line in the same two quarters.
What “Optionality” Costs When It Doesn’t Convert
The asymmetry worth understanding is depreciation, not just cash outlay. Data centers and GPU clusters get depreciated over roughly five to six years, which means this year’s $145 billion doesn’t just hit the cash flow statement once — it shows up as a drag on reported earnings for the rest of the decade, whether or not Llama, Meta AI, or Business AIs ever generate a dollar of revenue on their own. Compare that to Alphabet, which at least has a $240 billion contracted backlog converting to revenue on a known schedule, or TSMC, whose packaging lines are booked by customers who’ve already agreed to pay. Meta is running the capex intensity of a company with a backlog, without actually having one. If the AI products convert to real revenue on a reasonable timeline, the bet pays for itself. If they don’t, the depreciation bill still arrives regardless, funded entirely by an ad business that has to keep growing fast enough to cover a spending commitment made on faith.
We’ve been tracking how this earnings season is testing the gap between committed revenue and spending on faith — in Alphabet’s decision to zero out its buyback to fund its own AI infrastructure, and in TSMC’s sold-out packaging capacity, which is backed by actual customer commitments rather than internal optionality. Meta’s version of this trade is the least contracted of the three, and the most dependent on the ad business simply continuing to work.
Sources: Meta Platforms Q2 2026 earnings release and Form 10-Q; Meta Fourth Quarter and Full Year 2025 Results. Third Pole Markets holds no position in Meta Platforms as of publication. This is not investment advice — see our About page for our full disclosure policy.
The updated capex guidance is detailed in Meta’s investor relations materials and its 10-K filings on SEC EDGAR.
