Meta collected $60.8 billion in revenue and kept $784 million of it

Here is the deal. At 4:30 p.m. Eastern on July 29, Meta's second-quarter press release hit the wire, and the top line looked like a victory lap. Revenue of $60.8 billion, up 28% year over year, clearing the roughly $60.2 billion the Street had penciled in. Ad impressions grew 14% while the average price per ad rose 12% — volume and price pushing in the same direction at the same time, which is not a thing that happens often in a mature ad business. Family daily active people reached 3.60 billion, and Instagram crossed 2 billion daily users.

Then the stock fell more than 9% in after-hours trading. The next day, July 30, shares dipped toward the $529 range intraday before settling near $539, leaving Meta roughly 20% underwater for the year. On that exact same trading day, Microsoft ripped higher by more than 15% — its largest single-day market-value gain on record. Two mega-cap AI spenders, same week, opposite verdicts.

What the market read was not line one. It was every line underneath it. Total costs and expenses came in at $42.0 billion, up 55%. Income from operations fell 8% to $18.8 billion, and the operating margin compressed from 43% a year ago to 31% — twelve full points. Net income dropped 14% to $15.8 billion. Diluted EPS of $6.18 landed well below a consensus sitting in the low-to-mid $7 range. A quarter where revenue grows 28% and profit shrinks is genuinely rare in Meta's history.

And then the number that did the real damage: free cash flow of $784 million. A year earlier that same line read $8.5 billion. Operating cash flow for the quarter was $31.9 billion, and capital expenditures including principal payments on finance leases were $31.1 billion. Meta converted 98% of the cash it generated straight into data centers and servers. On top of that, management narrowed 2026 capex guidance from $125–145 billion to $130–145 billion — raising the floor by $5 billion while leaving the ceiling untouched. Raising a floor is not a forecast. It is a commitment.

The four names standing behind this income statement

Start with Mark Zuckerberg, who opened the call with a three-layer argument. AI is accelerating the core advertising business right now; it is the foundation for a next generation of personal agents; and it opens genuinely new revenue lines selling APIs, business agents, and eventually raw compute to enterprises. He framed Meta as "the only major company focused on distributing superintelligence widely rather than centralizing it, so that everyone can use it in the ways that matter to them." As a narrative for justifying an enormous spending program, it is tightly constructed. Whether it survives contact with a spreadsheet is the open question.

Then there is CFO Susan Li, who is the person actually explaining the arithmetic. She broke the 55% expense increase into four buckets: employee compensation, infrastructure costs, legal-related expenses, and third-party AI token costs. On top of those sat two one-time items — a $2.4 billion charge tied to legal proceedings and roughly $1.18 billion in severance from the May 2026 headcount reduction. Strip both out, Li argued, and operating income would have grown 9% instead of falling 8%. That is the company's defensive line, and it is not an unreasonable one.

Third is Meta Superintelligence Labs, a little over a year old, which shipped Muse Spark 1.1 and Muse Image within the past month. Zuckerberg said daily users of the Meta AI assistant grew 60% after Muse Spark was integrated. The cost of that organization is where it gets uncomfortable. Q2 R&D expense was $21.7 billion, up 67% year over year, moving from 27% of revenue to 36%. Stock-based compensation inside R&D alone was $6.8 billion, up 66% from $4.1 billion. Li attributed the increase to "technical talent hired over the past year, particularly in AI." Meta does not break out Superintelligence Labs payroll as a separate line — but total headcount actually fell 1% to 75,472 while total stock comp rose 57%. That combination answers the question without anyone having to disclose it.

The fourth name is the unexpected one: BlackRock. On July 28, the day before earnings, Meta announced a roughly $14 billion data center campus in El Paso, Texas, co-developed with BlackRock-managed funds. The ownership split is the interesting part — BlackRock's side takes 80%, Meta keeps 20%. Meta contributes land and construction-in-progress assets worth about $2.3 billion in kind, BlackRock funds roughly $4.9 billion in cash, and Meta receives about $1 billion back in a one-time true-up for the ownership adjustment. One gigawatt, targeted for 2028. Li described it on the call as "an example of partnerships that complement our approach." Translated: Meta has started experimenting with pushing capital expenditure off its own balance sheet.

The $80.3 billion that has not started depreciating yet

The single most important number in this quarter is not on the income statement. It is buried in the property and equipment footnote of the quarterly 10-Q filing. As of June 30, Meta's gross property and equipment stood at $292.9 billion — and $80.3 billion of that is construction in progress. At the end of 2025 that figure was $50.5 billion. It grew nearly $30 billion in six months.

Construction in progress does not depreciate, because it is not switched on yet. Which means the depreciation currently eating Meta's margins comes only from assets already running, while $80.3 billion sits in the queue waiting its turn. That is the part worth watching.

The run-rate confirms it. Depreciation of property and equipment was $6.00 billion in Q2, up 40% from $4.28 billion a year earlier. First-half depreciation and amortization totaled $12.36 billion versus $8.24 billion, a 50% increase. Accumulated depreciation sits at $67.1 billion — about 23% of gross PP&E, meaning the fleet is still young. Servers and network assets alone carry a $119.7 billion gross balance, up $22 billion in six months. Servers depreciate over roughly five and a half years; buildings over far longer. So the $130–145 billion Meta spends in 2026 does not fully hit the P&L until 2027 and 2028. The margin compression you are looking at right now is the opening move, not the endgame.

Here is how the four hyperscalers that reported in the same window stack up. Definitions differ slightly — Meta includes finance lease principal, the others report purchases of property and equipment — so read direction rather than decimal-precise comparisons.

Company Report date Q2 2026 capex Full-year capex guidance Quarterly revenue (growth) Free cash flow
Meta July 29 $31.1B (incl. finance lease principal) 2026: $130–145B (floor raised from $125B) $60.8B (+28%) +$784M
Alphabet July 22 $44.9B 2026: $195–205B (raised from $180–190B) $119.8B (+24%) −$5.9B
Microsoft July 29 $35.8B (PP&E additions) FY2026 actual: $115.9B; further growth signaled for FY2027 $90.0B (+18%) Not disclosed quarterly
Amazon July 30 $54.2B 2026: ~$220B (raised from ~$200B) $200.6B (+20%) −$7.6B (trailing twelve months)

Notice what the table says: Meta spends the least of the four in absolute terms. Alphabet burned $44.9 billion in a single quarter. Amazon burned $54.2 billion. Meta's $31.1 billion is the smallest number in the column. So why did the market single Meta out for punishment?

Two reasons. First, Meta has no revenue line directly wired to the spending. Alphabet has Google Cloud, which grew 82% to $24.8 billion this quarter with a $514 billion backlog. Amazon has AWS, which grew 37% to $42.2 billion with a $496 billion backlog. Meta's equivalent disclosure was Zuckerberg saying he would "share more soon."

Second, Meta has to carry all of this on advertising alone. Of the $60.4 billion Family of Apps generated, $59.4 billion was ads. Reality Labs posted $431 million in revenue against a $4.6 billion operating loss. So the structure is: ad profits cover the Reality Labs deficit, and then AI infrastructure spending gets stacked on top of what remains. Compare that with Alphabet, which booked $112.1 billion in quarterly net income (mostly from a roughly $98 billion unrealized equity gain, admittedly), or Microsoft, which printed $40.6 billion in operating income. Meta's cushion is visibly thinner.

Taxes got worse too. The Q2 effective tax rate was 16%, up from 11% a year ago, and the company raised its expected rate for the remaining quarters from 13–16% to 15–17%. Full-year total expense guidance moved up to $165–169 billion, absorbing the $2.4 billion legal charge. Revenue was the only line that improved. Everything below it deteriorated simultaneously.

Who is actually collecting money in this arrangement

The most obvious winners sell silicon and memory. Add up what Meta, Alphabet, Microsoft, and Amazon have publicly committed to spending in 2026 and you clear $600 billion comfortably. Andy Jassy explicitly cited rising memory prices as a reason Amazon lifted its guidance from about $200 billion to about $220 billion. Microsoft flagged component cost inflation inside its own quarterly capex. For a hyperscaler that is a cost line. For the vendor on the other side of the invoice, it is pricing power, and it is showing up in guidance documents rather than in analyst speculation.

The second winner is infrastructure finance. El Paso is the template. If Meta holds 20% and BlackRock-managed funds hold 80%, a $14 billion campus does not land fully inside Meta's consolidated capex figure. Meta keeps operational control while sourcing capital from outside. Amazon's recent $25 billion bond issuance runs on the same logic from a different direction. AI data centers are becoming a genuine asset class for private credit and infrastructure funds — which is attractive on a yield basis and also worth watching carefully, because the risk does not disappear when it moves off a balance sheet. It just relocates to someone whose disclosures you read less often.

The third group is AI researchers themselves. Meta's total headcount is down year over year while stock-based compensation rose 57%. Of $7.6 billion in total Q2 stock comp, $6.8 billion sat in R&D. That is a company shrinking overall and redirecting the freed resources into a small population of specialists — and the pattern is legible directly from the financials. Meta cut roughly 8,000 roles in May, and Li noted most of those departures will not clear the headcount statistics until the end of Q3.

What about advertisers and users? Here Meta's case is genuinely strong. Zuckerberg said integrating LLMs into recommendation systems drove double-digit year-over-year growth in global time spent on Instagram, and lifted Facebook video time spent 9% globally and more than 10% in the US and Canada. The simultaneous 14% impression growth and 12% price increase is plausibly the direct output of that improvement. As evidence that AI investment is monetizing inside the core ad business, that is sufficient.

The problem is whether it is sufficient to justify $31.1 billion in a single quarter. Year-over-year ad revenue grew by roughly $12.8 billion. Quarterly capex was more than double that increment. The question the market is asking is not "does AI help advertising" — it clearly does. The question is "does this much help justify this much spend." Zuckerberg said Meta has received "multiple offers at a significant premium to what we paid" for compute, then added in the same breath that "we believe selling intelligence is much higher margin than selling compute." Which is another way of saying the company has not committed to either path yet.

2022's metaverse shock, and where the rhyme breaks down

The déjà vu is legitimate. On October 26, 2022, Meta reported Q3 and guided 2023 capex to $34–39 billion while warning that Reality Labs operating losses would grow significantly. The stock dropped 25% the next day. Reality Labs revenue that quarter was $285 million. Spending tens of billions to support a division generating less than $300 million a quarter was the thing investors could not stomach.

What happened next gets cited by both sides of today's argument. In early 2023 Zuckerberg declared the "year of efficiency," cut more than 20,000 jobs across two rounds, and lowered capex guidance to $34–37 billion. The stock roughly tripled over the course of that year. Bulls read that as proof Meta knows how to change course under pressure. Bears read it as proof the stock recovered because spending fell, not because it rose. Same event, opposite lesson, and both readings are defensible.

But 2026 differs from 2022 in three ways that matter. First, the character of the asset. A large share of 2022 Reality Labs spending was R&D and content — money that simply evaporated if demand never arrived. 2026 spending buys GPUs and data centers. Li made the point directly on the call: the long useful life of these assets is itself a source of flexibility. If AI demand disappoints, compute can be redeployed or sold. It is not money that vanishes.

Second, the revenue trend runs the opposite direction. In Q3 2022 Meta's revenue declined 4% year over year. Today it is up 28%. The costs are frightening either way, but the engine paying for them is running dramatically better than it was.

Third, Meta is not alone this time. In 2022 Meta was the only company shoveling cash into the metaverse while the rest of big tech was cutting. In 2026 Alphabet, Microsoft, and Amazon are all spending more than Meta in the same direction. Whether an entire industry making the identical bet simultaneously makes it safer or considerably more dangerous is something you only learn afterward.

Three rivals got graded the same week and got very different marks

Alphabet went first. On July 22 it reported $119.8 billion in revenue, up 24%, with Google Cloud up 82% to $24.8 billion. Alphabet also raised 2026 capex guidance from $180–190 billion to $195–205 billion, and quarterly free cash flow flipped to negative $5.9 billion. The stock fell around 3% after hours. Alphabet is spending far more than Meta and got a far gentler reaction — because the line where capex converts into revenue was visibly accelerating at 82%.

Microsoft was the more dramatic control group. It reported fiscal Q4 the same evening as Meta: revenue of $90.0 billion (+18%), operating income of $40.6 billion (+18%), net income of $35.8 billion (+31%), Azure growth accelerating to 43%, Microsoft Cloud revenue of $59.3 billion (+27%), and commercial remaining performance obligation up 84% to $678 billion. Fiscal 2026 capex landed at $115.9 billion, and the company signaled further growth ahead in fiscal 2027. Azure crossed $100 billion in annual revenue for the first time. The stock jumped more than 15% the next day, the largest one-day market-value increase in market history. Microsoft told investors it would spend more, and investors cheered.

The difference was one number: Azure at 43%. One company proved the conversion rate from capex to revenue with a metric; the other said it would show us soon. Amazon passed the same test with the same grammar. Its July 30 report put AWS at $42.2 billion, up 37% — the fastest growth since 2021 — with $16.6 billion in AWS operating income and a $496 billion backlog. Jassy raised full-year capex to about $220 billion while saying that even that figure will not fully meet 2026 demand, and that the same will likely hold in 2027. Amazon's trailing-twelve-month free cash flow is already negative $7.6 billion. It got away with it because the backlog is a shield.

So the score the market handed out this week was not about the size of the spend. It was about the visibility of the return. Meta's problem is not that it plans to spend $130 billion. It is that no disclosed metric tells you when and through which revenue line that $130 billion comes back. Of the five opportunities Zuckerberg listed — personal agents, a public API, business agents, developer tools, and direct compute sales — exactly zero are broken out as a reported revenue line.

There is a skeptical read worth holding onto. When analyst Brian Nowak asked about 2027 capex, Li declined, saying the company is "not providing a 2027 capex outlook at this point" and adding that infrastructure plans remain "very fluid." Read charitably, that means they genuinely do not know yet. Read defensively, it means they do not want to put a number in front of the market that the market cannot absorb. Either way, the fact stands: Meta has published no picture at all of what margins look like in 2027 and 2028, when that $80.3 billion of construction in progress starts depreciating.

What actually changes, depending on where you sit

If you are an investor. The story of this quarter is not the earnings decline — it is the quality of earnings. Back out the $2.4 billion legal charge and $1.18 billion in severance and operating income grew 9% rather than falling 8%. One-time items produced a large share of the drop, and that is a fair defense. But depreciation is not a one-time item. First-half D&A is already up 50%, and $80.3 billion of unswitched-on capacity is waiting in line. The line to watch next quarter is not revenue growth. It is the rate of change in depreciation, plus where Q3 revenue lands inside the $61–64 billion guide. That guidance midpoint came in below expectations, and it was part of what drove the selloff too.

If you buy infrastructure for a living. The meaningful development is that Meta has begun saying out loud that it might sell its own compute. Zuckerberg confirmed "multiple offers at a significant premium," opened Muse Spark 1.1 through a public API, and described expanding distribution via partner channels and coding agents. If you procure AI capacity, that is potentially one more supplier in your evaluation set. That said, no pricing and no SLA have been published, so putting Meta into an actual sourcing plan today would be premature.

If you write code. The concrete change is Muse Spark 1.1 shipping as a public API. Zuckerberg described it as strong on agentic and coding workloads with good efficiency in computer use, tool use, and multimodal understanding. With developer tools and business agents both on the stated roadmap, how quickly an ecosystem forms around this API over the next couple of quarters is the thing to track. But no benchmark numbers and no pricing appeared in the earnings materials, so taking the performance claims at face value right now is not warranted.

If you just use the apps. What you will actually feel is ranking changes and glasses. Meta launched its own-brand Meta Glasses with EssilorLuxottica — the first product with Muse Spark built in, able to understand and respond to what the wearer is looking at. The company said early sales exceeded expectations. The next hardware update comes at Connect on September 23. Elsewhere, WhatsApp set an all-time record of 30 million messages per second during the World Cup final, and Threads passed 500 million monthly active users.

If you buy ads. A 12% increase in average price per ad means your costs went up — that is what that sentence means. But impressions rose 14%, so available reach expanded, and if Meta's claim about LLM-driven ranking accuracy holds, better conversion may offset the higher unit price. This is the kind of shift you verify inside your own account, not from a press release.

🥄 Three Things You’re Probably Wondering

— So what does this mean for me? It matters even if you own zero shares of Meta. The four biggest spenders have collectively committed north of $600 billion this year, and that money is pushing up prices for memory, GPUs, and electricity. Amazon named memory costs as a direct reason for raising its guidance. That pressure eventually reaches you through cloud bills and consumer device prices.

— When does the depreciation bill actually land? Calling an exact date would be guessing, but the direction is calculable. The $80.3 billion in construction in progress as of June 30 has not started depreciating. As those facilities come online in sequence, servers begin expensing over roughly five and a half years and buildings over much longer. First-half depreciation is already up 50%, so 2027 likely sees a steeper increase. Since Meta has published neither a 2027 capex outlook nor a margin outlook, any specific figure beyond that is speculation.

— Could Meta pull another 2023-style pivot? You cannot rule it out, but the conditions are different. In 2023 the things being cut were headcount and Reality Labs projects, and cutting them flowed straight to profit almost immediately. The 2026 spending is data centers already under construction and servers already ordered — hitting the brakes takes time to slow the vehicle. When Li said near-term capacity is more valuable than long-term capacity, the natural read is that Meta has no intention of easing off through at least 2027.

Further Reading

Numbers and criteria are as of announcement and may change. Investment calls are yours to make!