Meta Q3 2026: A $9.9 Billion Asterisk, a $25 Billion Warning, and an Ad Machine Still Running at +26%

中文版 Chinese version

This is an outside read of Meta’s Q3 2026 earnings release and earnings call (October 21, 2026) — public numbers only. Figures are as reported that day and will be revised in filings. Not financial advice.

The print in one paragraph

Meta reported Q3 2026 revenue of $64.60 billion, up 26% year over year (26% in constant currency too), beating the ~$63.3 billion consensus. Underneath that clean headline sits a messy income statement: a $9.89 billion legal charge cut GAAP operating income 37% to $12.99 billion and halved the operating margin to 20%. Excluding the charge, operating income would have been $22.88 billion (+11% YoY) and diluted EPS $7.40, versus the reported $3.83. Capital expenditures hit $34.91 billion in the quarter, pushing free cash flow to negative $2.12 billion in what was, by revenue, the biggest quarter in company history. The stock rose ~4% on the release, touched +6%, then gave back most of it during the call, closing the after-hours session up about 2.2% (~$712.68 vs a $697.52 close) — a shrug, relative to the ±12% move options had priced in.

Two banks had pre-called this shape almost exactly. Bank of America expected the ~$10 billion charge and printed a $3.79 EPS estimate; reported GAAP EPS came in at $3.83. Bernstein feared EBIT of $11.34 billion against a $19.02 billion “clean” consensus; reported operating income of $12.99 billion landed between the two — the miss was in the legal line, not the business.

The scoreboard

Every year-over-year comparison in this table needs a footnote. Last year’s EPS of $1.05 was crushed by an 87% effective tax rate (an $18.95 billion tax provision tied to U.S. tax-law changes); this year’s reported EPS is crushed by the legal charge instead. The honest comparison is the middle column against history: an ad business growing 26% at a ~35% operating margin — still below last year’s 40%, because R&D ($20.05B, up from $15.14B) and cost of revenue ($13.37B, up from $9.21B) are compounding faster than revenue. The margin story is no longer only about one-offs; the AI cost base is in the run rate now.

The $9.89 billion charge — and the $25 billion one behind it

The charge comes from Meta’s August 2026 agreement with 52 state attorneys general resolving the teen-mental-health litigation, which provides for payments of up to ~$18.59 billion. Meta recognized $9.89 billion of it in Q3, inside general and administrative expenses — which is also why Family of Apps operating income fell 31% to $17.12 billion despite 26% revenue growth: the charge lives in the apps segment. CFO Susan Li noted the quarter’s 24% tax rate came in above guidance largely because part of the legal expense is non-deductible; excluding the charge, the rate would have been 17%.

Treat this as a recurring line, not a one-off, until proven otherwise — because on the same day as earnings, a New Mexico judge entered a $25 billion civil penalty against Meta. A Santa Fe jury had found in September that 26 of 29 public statements (2010–2021) about privacy, data practices, and content moderation were misleading under the state’s Unfair Practices Act — 43.9 million counted violations, a theoretical $219.5 billion statutory maximum, reduced by Judge Francis Mathew to $25 billion, with 15% annual interest until paid. For scale: five times the FTC’s record $5 billion penalty from 2019. Meta says it will appeal and calls the suit “a tax on speech”; the judgment post-dates Q3, so none of it is in these numbers. Add the youth-related trials Susan Li flagged as scheduled for the coming months, and the correct mental model is: Meta now has a legal-expense run rate, not a legal accident.

The ad engine is not the problem

Advertising revenue rose 26% to $63.06 billion, on impressions +10% and average price per ad +14% — both volume and price, in every region. Family daily active people hit 3.64 billion (+3%). Management’s AI receipts, with numbers attached:

• Generative retrieval on Facebook Feed: ~+2% conversions in targeted segments; Reels watch time +1% from the new recommendation architecture.
• GEM ranking model expansion: +1.7% off-site conversions, +9.6% app-ad conversions. Andromeda improvements: +1.4% Facebook conversions.
• Content understanding moved onto Meta’s own Muse models: unit processing costs down more than 80% across Instagram, Threads, Facebook, and ads.
• Nearly 10 million advertisers used at least one generative-AI creative tool in Q3; partnership ads grew >97% to a >$15 billion run rate.
• Family of Apps “other” revenue — WhatsApp paid messaging, subscriptions — rose 59% to $1.1 billion; Meta One passed 15 million subscriptions and trials.

This is the part of the call that justifies the stock’s first reaction: the core machine is not merely intact, it is visibly converting AI spend into ad yield right now, in measurable single-digit increments that compound.

Muse: the new line item

The quarter’s strategic headline was the Muse personal agent and the Muse Wonder frontier model. From the call:

• More than 10 million people in the U.S. and Canada set up Muse in its first 1.5 months — with no global launch yet. Zuckerberg contrasted the ramp with Threads’ start (“more than twice the level”).
• About 25% of people who set it up use it on any given day.
• Shopping is the top use case: over 25% of active users, average purchase size above $200 (including travel). Zuckerberg volunteered that Stripe had to raise credit limits in week one and Plaid had never seen connector activity like it.
• Monetization is planned to start with subscriptions, then transaction fees, paid access to advanced models and tokens, and enterprise (model API, Muse Code, business agents). Management explicitly said Muse will be monetized earlier than Meta’s consumer surfaces usually are — because of its compute costs.
• The constraint is supply, not demand: Meta says it is holding back global rollout until compute efficiency improves, targets >10x efficiency gains over the next year, and is building capacity intended to serve more than 1 billion people.

That last point is also the risk the stock faded on: Susan Li stated plainly that expanding Muse’s user base “will require significant investment, and we expect total serving costs to rise.” A free, compute-hungry agent used daily by a quarter of its users is a cost curve before it is a revenue curve. The next several quarters will be about whether the >10x efficiency claim lands faster than the user count.

The bill: capex, cash flow, and a balance sheet crossing

• Q3 capex: $34.91 billion (including finance-lease principal), up ~80% YoY. Operating cash flow was $32.79 billion (+9%), so free cash flow was −$2.12 billion, versus +$10.63 billion a year ago.
• Full-year 2026 capex guidance was narrowed upward to $135–145 billion (from $130–145B); full-year expense guidance rose to $175–179 billion, with the increase essentially equal to the legal charge.
• Q4 revenue guidance: $73–76 billion (with an ~1% FX tailwind), midpoint above the ~$73.6 billion consensus. Q4 tax rate guided to 15–17%.
• Management declined to give 2027 revenue guidance, said it will “continue investing aggressively,” and pushed quantitative 2027 color to the Q4 call — while still expecting ex-charge 2026 operating income to exceed 2025’s.
• The balance sheet quietly crossed a symbolic line: cash and marketable securities of $82.45 billion now sit against roughly $83.7 billion of debt, and net property and equipment has swollen to $253.22 billion from $176.40 billion at the end of 2025. Headcount, meanwhile, fell to 66,775 — down 15% year over year after the May 2026 reduction. Fewer people, vastly more machines.

Reality Labs: smaller losses, clearer verdict

Reality Labs revenue fell 6% to $443 million (Quest softness, partly offset by AI glasses growth) and the operating loss narrowed to $4.13 billion, better than the ~$5.1 billion loss analysts expected. Full-year 2026 RL losses are now guided to $18–19 billion (below 2025), and 2027 losses to around $15 billion, as investment is reallocated to AI. Zuckerberg added that new VR devices will come only with major technological breakthroughs, while Muse arrives on Meta’s glasses next month. Read together: the metaverse hardware bet is being converted, in public, into a glasses-plus-agent bet — and the losses now have a declining glide path attached, which is new.

The quiet big news: Class C stock

Buried in the release is the most structurally important item for long-term holders. The board approved a new non-voting Class C common stock, to be distributed as a stock dividend of one Class C share for every Class A or Class B share, effective January 19, 2027, with trading expected in late January 2027. Economically it resembles a 2-for-1 split; mechanically it creates a permanent equity currency — usable for capital raising, employee compensation, and acquisitions — that never dilutes voting control. Announced on the same day the company guided to $135–145 billion of annual capex and negative quarterly free cash flow, the intent isn’t subtle: the AI build-out is now big enough that Meta wants a funding instrument beyond cash flow and debt, and it wants it without touching the governance structure. How the Class C trades relative to Class A (the discount, if any) will be one of January’s cleaner market signals.

Bottom line

Strip the asterisks and Q3 2026 is two reports stapled together. Report one: the best advertising business in the world grew 26% at a ~35% underlying margin, guided Q4 above consensus, and produced a list of AI-driven conversion gains with receipts. Report two: that same company spent $34.9 billion on capex in 90 days, burned cash doing it, took a $9.9 billion legal charge with a $25 billion judgment waiting in the wings, launched a compute-hungry agent it deliberately hasn’t rolled out globally yet, and created a new share class to help pay for all of it.

The market’s verdict — up 6% at the peak, +2% by the end of the evening — was precisely “both reports are true.” What resolves the tension isn’t next quarter’s ad number; it’s three things to watch into the Q4 print: the 2027 capex number (promised in January), the first Muse monetization evidence versus its serving costs, and whether legal charges finally stop arriving in $10–25 billion denominations.

Sources: Meta Q3 2026 earnings release and earnings call, October 21, 2026 (figures and management commentary as reported by TradingKey’s release summary and call transcript) · InvestingLive — results vs consensus · NY Post — after-hours reaction and Class C · Morningstar/Dow Jones — after-hours price · Reuters via DevDiscourse — New Mexico $25B judgment · PYMNTS — Meta response to New Mexico ruling