An hour after Microsoft showed the revenue side of the AI transition, Meta (NASDAQ: META) showed the cost side.
Revenue beat. Profit missed badly. And both stem from the same place.
What Was Reported
| Metric | Result | Year ago | Change |
|---|---|---|---|
| Revenue | $60.80 billion | $47.52 billion | +28% |
| Costs and expenses | $42.03 billion | $27.08 billion | +55% |
| Income from operations | $18.78 billion | $20.44 billion | -8% |
| Operating margin | 31% | 43% | -12 points |
| Net income | $15.85 billion | $18.34 billion | -14% |
| Diluted EPS | $6.18 | $7.14 | -13% |
Consensus stood at $7.23 and $60.26 billion. Revenue came in above - at the top of the company's own $58-61 billion guide. Earnings - a miss of more than a dollar.
The Gap Is Explained in One Line: Expenses
Revenue rose 28%. Expenses rose 55%. When costs run at nearly double the pace of revenue, the margin gets crushed - and that is exactly what happened.
But inside that 55% are three entirely different components, and separating them matters:
Decomposing the expense surge
Component one - one-time: about $3.6 billion. $2.40 billion of charges related to legal proceedings, plus $1.18 billion of severance from the May 2026 headcount reduction. These do not repeat in the same form - and management itself raised the lower end of its expense guidance by exactly the size of the legal charge. Component two - the AI infrastructure. Data centers, chips, power, depreciation on past investment - the running cost of the machine being built. Component three - Reality Labs, which continues to consume about $4.6 billion a quarter. In per-share terms, the one-time items alone are worth roughly $1.40 pre-tax (about $1.10-1.20 after tax) - on the order of the entire miss. Excluding them, earnings would have been near or above consensus - which sharpens how much this miss is a story of charges, not of the business. (That calculation is our approximation from published figures - $3.58 billion of charges over roughly 2.56 billion fully diluted shares.)
And on the Other Side: the Ad Engine in the Best Shape of Its Life
This is the point easy to miss beneath the red headline.
| Metric | Change |
|---|---|
| Family daily active people | 3.60 billion, +3% |
| Ad impressions | +14% |
| Average price per ad | +12% |
That combination is the ideal for an advertising business. Growth that comes only from volume means price is eroding; growth only from price means inventory is exhausted. Here both rise at once - more ads, and each one worth more.
And why is that happening? This is precisely the return on the AI investment. Better targeting means the ad reaches the right person - and the advertiser pays more for it. At Microsoft the AI return shows up as a separate revenue line (Azure). At Meta it is embedded inside the price of an ad - less visible, but no less real.
Zuckerberg put it directly in the release: "AI is accelerating our core business today." The number that confirms it is the 12% on price.
The Breakdown: Apps Versus the Lab
| Segment | Revenue | Operating income (loss) |
|---|---|---|
| Family of Apps | $60.37 billion | +$23.39 billion |
| Reality Labs | $431 million | $(4.62) billion |
Family of Apps earned $23.4 billion - down from $25.0 billion a year ago, because the expenses are booked there. And Reality Labs continues to consume about $4.6 billion a quarter - an annual run rate of roughly $18 billion, with no change in trend.
The Capex and the Balance Sheet - the Numbers Behind the Story
| Metric | Value |
|---|---|
| Quarterly capital expenditures | $31.08 billion |
| 2026 capex guidance | $130-145 billion (narrowed from $125-145) |
| Operating cash flow | $31.86 billion |
| Cash and marketable securities | $90.26 billion |
| Long-term debt | $83.66 billion |
| Dividend payments in the quarter | $1.35 billion |
Two points worth pausing on:
First - the floor rose. Narrowing the range from $125-145 to $130-145 billion means the minimum went up by $5 billion. In a market that judges Meta first on its capex, that is the substantive statement in the outlook.
And second - the debt. $83.7 billion of long-term debt is a relatively new number in the Meta story - a company that was historically nearly debt-free. The buildout is no longer funded from cash flow alone. Operating cash flow ($31.9 billion) is almost identical to the quarter's capex ($31.1 billion) - meaning free cash flow is close to zero, and the dividend and buybacks are funded at the margin from the balance sheet.
Guidance
| Metric | Guidance |
|---|---|
| Q3 revenue | $61-64 billion |
| 2026 expenses | $165-169 billion (lower end raised by the legal charge) |
| 2026 capex | $130-145 billion |
| Tax rate | 15%-17% (raised from 13%-16%) |
And the central commitment held: management continues to expect 2026 operating income to exceed 2025
- despite everything.
The Bull Thesis
Whoever reads it positively will point to the ad engine: impressions +14% and price +12% are the most direct evidence that AI is improving the core product. Revenue at the top of guidance, and next quarter's guide ($61-64 billion) implies continued growth above 20%.
Beyond that: much of the miss is one-time - $3.6 billion of legal charges and severance that are not part of the fixed cost structure. And the commitment that 2026 operating income exceeds 2025 held.
The Bear Thesis
Whoever reads it critically will note first that the margin fell 12 points in a year - and not all of it is one-time. The AI infrastructure is a growing fixed cost, and the depreciation on it will only accumulate.
Second, free cash flow has evaporated. $31 billion of capex against $32 billion of operating cash flow - the company is building at the limit of its current generation, and the $83.7 billion of debt shows it has already crossed it.
Third, Reality Labs is unchanged - a $4.6 billion quarterly loss, year after year, with no visible inflection.
And fourth, unlike Microsoft, Meta has no separate AI revenue line. The return is embedded in ad price - and when it is implicit, the market prices the explicit cost and discounts the implicit return.
The debate in one line
The bulls see an ad engine growing in volume and price at once, revenue at the top of guidance, a miss mostly explained by one-time charges, and a profit commitment that held. The bears see a margin down 12 points, free cash flow near zero, $84 billion of debt built to fund the capex, and a reality lab losing $18 billion a year with no end in sight. Both sides are reading the same report.
My Angle
A personal opinion of Ilan Abramov - not advice, not a recommendation
What catches me in this report is the comparison to Microsoft - because the two tell the same story from opposite sides.
At Microsoft, AI shows up as a revenue line: Azure, 43%, visible to everyone. At Meta it shows up as a 12% rise in the price of an ad - embedded, nearly invisible, but just as real. Targeting powered by better models makes the advertiser pay more for the same impression. That is a return on investment in every sense - it just does not get its own line in the report.
And that asymmetry explains the reactions. The market sees Meta's cost explicitly - $31 billion a quarter, a floor that rose - and its return implicitly. At Microsoft it is the reverse. Same strategy, different transparency, different pricing.
And what I flag as the truly important point: the cash flow. $31.1 billion of capex against $31.9 billion of operating cash flow - Meta is spending on infrastructure almost every dollar it generates, and the $83.7 billion of debt says sometimes more. A company that was a free-cash-flow machine has temporarily become a construction company. Not necessarily bad - but it changes the risk profile, because there is no cushion anymore.
And what I will follow: the price per ad. That 12% is the only metric where Meta's return on investment is visible in real time. As long as price grows double-digit, the bet is working. The day it moderates to single digits while capex is still climbing - that will be the sign the machine has peaked, and the questions about the $130 billion will get much sharper.
Summary
Meta delivered the evening's split report. The strong side: revenue of $60.8 billion (+28%) at the top of guidance, ad impressions +14% and average price per ad +12% - the ad engine in the best shape of its life, and AI is the reason.
The weak side: expenses up 55%, an operating margin down from 43% to 31%, and EPS of $6.18 against a $7.23 estimate - a miss largely explained by $3.6 billion of one-time charges (legal and severance), and partly by the structural cost of the AI buildout. The annual capex floor was raised to $130 billion, long-term debt stands at $83.7 billion, and Reality Labs keeps losing $4.6 billion a quarter.
And the commitment that held: 2026 operating income above 2025.
The question for the investor is not whether AI works at Meta - the price per ad answers that. The question is how long the market will tolerate an explicit cost and an implicit return - and which comes first: the margin stabilizes, or the patience runs out.
Sources: Meta Platforms' official results announcement for the second quarter of 2026 (July 29, 2026), as filed with the U.S. Securities and Exchange Commission on Form 8-K, including revenue, expenses and the one-time items, operating and net income, operational metrics, capital expenditures, the balance sheet, the segment breakdown and the CFO's outlook commentary for the third quarter and full year; CEO Mark Zuckerberg's remarks from the release; analyst consensus ahead of the report. The per-share estimate of the one-time items' impact is an approximation based on published figures. Data accurate as of the time of writing. The chart is shown in real time via TradingView. Nothing herein constitutes a forecast, recommendation or advice - see the full disclaimer at the bottom of the page.