META Q3 2026 — The Ad Engine Held; the Commitment Book Needs a Lender

The setup

The Q2 dive ended on one line: Meta is spending at hyperscaler scale with no cloud revenue and no contracted backlog behind the build. Q3 answered the way the filing usually answers a narrative: with the cash line. It is the first quarter in which the spending outran the cash. Capital expenditures, including principal payments on finance leases, were $34.91B against $32.79B of operating cash flow, so free cash flow was −$2.12B, down from $10.63B a year earlier and $784M in Q2.

Strip out the legal noise and the rest of the print is a strong advertising quarter. Leave it in and the income statement looks broken. Neither is the story. Revenue was $64.60B, up 26% on the year and 6.25% on the quarter. Income from operations was $12.99B, down 37%, because general and administrative expense carried a $9.89B charge for the August agreement with 52 state attorneys general over teen mental health. Excluding that charge, Meta says income from operations would have been $22.88B and diluted EPS $7.40 rather than $3.83. Net income was $9.86B against $2.71B a year ago, when a one-time $15.93B tax charge sat in the comparison.

Put the lead number in its series. On the cash-flow statements Meta has filed with the SEC, operating cash flow less capex and finance-lease principal was last negative in Q2 2012, the quarter it listed, at about −$173M. The low of the 2022 build was $154M in Q3 2022, and in the eight quarters before Q2 2026 the lowest figure was $8.55B. One quarter of capex, $34.91B, is now 89.0% of everything Meta spent on capex in 2024. What it is not: a sign that the ads business is weakening. Operating cash flow was $32.79B, the second-largest quarter in the series. The engine did not slow; the bill outgrew it.

Against the Street, revenue beat a $63.35B consensus and EPS excluding the charge of $7.40 compared with a $5.46 estimate, per StockStory on Yahoo Finance. META closed at $697.52 on October 21 and traded at $654.85 at 9:08 a.m. PT on October 22, down 6.12%.

The numbers

Read the G&A bar before the margin. G&A was $14.86B, 23.0% of revenue, against $3.51B a year earlier; the attorneys general charge alone is 15.3% of revenue. Take it out and operating margin was 35.4%, against 40.1% in Q3 2025. That gap is the real cost story of the quarter, and it is infrastructure: cost of revenue rose 45% and R&D 32%, which the 10-Q puts down to data centers, technical infrastructure, third-party cloud services, and “third-party AI token costs.” The tax rate was 24%, 17% excluding the charge, because part of the settlement is not expected to be deductible.

What’s actually driving it

The auction got richer while the inventory slowed. The 10-Q explains the price line: average price per ad rose because of “an increase in advertising demand, which we believe is mostly due to ongoing improvements to our ad performance from our ad targeting and measurement tools.” Decode it against the volume line. Advertising revenue is impressions times price, and both are disclosed. Impression growth went 19%, 14%, 10% across Q1, Q2 and Q3 2026, while price per ad went 12%, 12%, 14%. The headline held at 26% because the mix inverted: in Q1 volume did most of the work, in Q3 price did. Price per ad at +14% is the fastest rate in the seven-quarter series, which otherwise ran between 6% and 12%; the last +14% was Q4 2024. Impressions at +10% are the slowest since Q1 2025’s +5%. A stable top line is not evidence of a stable engine; read the ingredients, not the level. Management sells this as AI working, and the price line supports it: pricing is the half of the formula better ranking is supposed to move. What the narrative leaves out is that pricing is also the half a weaker advertiser budget hits first, and volume is no longer there to cushion it.

The rest of the Family of Apps line confirms demand rather than undercutting it. Family daily active people averaged 3.64B in September, up 3%, and average revenue per person was $17.72, up 22.5% from $14.46. FoA other revenue, mostly WhatsApp paid messaging and subscriptions, rose 59% to $1.10B. Revenue grew 33% in the United States and Canada, 18% in Europe, 16% in Asia-Pacific, and 34% in the rest of the world. Online commerce was the largest contributor to the increase. On the call, CFO Susan Li said partnership ads grew more than 97% and passed a $15B annual run rate, and Mark Zuckerberg said more than 10 million people in the US and Canada set up Muse personal agents in about six weeks, per the Motley Fool transcript.

The segment margin moved in the infrastructure, not the lawsuit. Family of Apps earned $17.12B from operations on $64.16B of revenue. The segment note lists legal-related costs only among FoA expenses, so the charge sits there; add back $9.89B and FoA earned $27.01B, a 42.1% margin, against 49.2% in Q3 2025. Q2, on the same treatment of its $2.40B legal charge, was 42.7%. So the step down happened a quarter ago and held. FoA costs excluding the charge rose 44.0%. Employee compensation inside FoA rose only 14.5%; the other-costs line, which holds infrastructure, cloud, AI tokens and the legal accruals, rose 64.6% excluding the charge. The Q3 2025 release called this in advance: “Employee compensation costs will be the second largest contributor to growth.” Decode the order. Headcount fell 15% on the year to 66,775 and 11.5% from Q2 as the May reduction of about 8,000 roles worked through, yet share-based compensation rose 32.5% to $7.36B. Fewer people, each paid more in stock. The release was right about the order: the first contributor is the machines.

Depreciation is the second bill, and capex writes it. Depreciation and amortization was $7.15B, up $798M on the quarter, the largest quarterly step in the seven-quarter series; the prior steps ran between $357M and $621M. It is 11.07% of revenue against 9.69% a year ago. Server and network depreciation alone was $5.34B, up 48.3% on the year. The pipeline behind it is larger than the line: construction in progress was $89.50B at September 30, 2.03 times the $44.02B of a year earlier and $9.15B more than in June, and none of it depreciates until it is placed in service. Capex ran 4.88 times D&A this quarter. Every one of those dollars becomes an income-statement cost over the asset’s life, which is why the carrying cost is senior to the shareholder applies to Meta even without a cloud segment: depreciation is paid out of the ad margin before equity sees a dollar.

The commitment book is the leading indicator, and it grew again. Cash capex records decisions already made; commitments show the ones still running. At September 30 the 10-Q carried $454.52B of non-cancelable contractual commitments, “mostly” third-party cloud capacity, servers and network infrastructure, energy service agreements, and Reality Labs hardware, plus $355.73B of data-center, energy, colocation and network leases signed but not yet commenced. Together that is $810.25B. The series: $80.51B at June 2025, $139.33B at September 2025, $234.82B at December, $420.55B at March, $628.30B at June. The book added $181.95B in Q3 after $207.75B in Q2 and $185.73B in Q1, so the pace eased slightly but did not stop. Non-cancelable commitments alone are 5.60 times their level a year ago. Against the balance sheet, the book is 9.83 times cash and marketable securities and 169.2% of total assets of $478.84B. This is the library rule that a one-period change is a phase, not a decision, until you have the series, in its most useful form: the capex line says Meta spent $34.91B; the commitment series says it has signed for more than five years of building at the Q3 pace, and the book is still growing.

The total is the headline; the timing is the problem. Of the $454.52B, $45.85B is due in the rest of 2026, which is the fourth quarter alone, and $121.41B in 2027. The 2027 figure was $47.65B at March 31 and $81.65B at June 30, so it grew 2.55 times in two quarters and $39.76B in Q3 alone. Meta’s operating cash flow over the last four quarters was $133.09B, so the contracts already due in 2027 equal 91.2% of a year’s operating cash, before any lease starts paying rent and before any capex Meta has not yet put under contract. The Q4 tranche of $45.85B is 1.40 times Q3’s operating cash flow. Most of it is capex in all but name; some, the third-party cloud, will land as operating expense. Either way, it is cash.

Meta is posting collateral to its own suppliers. The Q1 release explained the first capex raise of the year: the higher range “reflects our expectations for higher component pricing this year.” Decode it against the restricted-cash note. At March 31 Meta held $5.00B of money market funds in escrow for “a multi-year purchase agreement.” At June 30 and again at September 30 it was $10.80B, under “certain multi-year infrastructure purchase agreements,” to be released between 2028 and 2030 “upon satisfying the underlying purchase obligations.” Total restricted cash equivalents were $12.83B. A buyer of Meta’s size does not usually post cash security to a vendor; it does when the vendor is the scarce side of the trade. A commitment is an allocation claim before it is a cost: the escrow is what Meta paid to keep its place in the line, and the $10.80B is cash that sits inside the $82.45B headline but cannot fund anything else until 2028 at the earliest. Unused letters of credit, mostly for the same long-term infrastructure commitments, were $1.53B.

Part of the build sits off the balance sheet on purpose. In July Meta put $2.26B of Texas land and construction in progress into a new data-center venture, took back a one-time $1.04B distribution, and kept 20%. It now holds 20% of two ventures, Louisiana (formed October 2025, about $27B of development cost) and Texas (about $14B). It does not consolidate either. It leases the campuses back, starting in 2029 and 2028, and guarantees the residual value: the maximum residual value guarantee rose from $28.09B at December to $41.49B at September 30, and Meta’s total maximum exposure to loss in the two ventures rose from $45.95B to $68.40B. Those guarantees are recorded at zero because payment is “not probable.” On who ultimately pays for the capex, this structure answers: the venture partners pay up front, Meta pays rent over up to 20 years, and if the campuses are worth less than the guarantee threshold when a lease ends, Meta pays the difference.

The leases are long and the revenue is not contracted. The not-yet-commenced leases run “from greater than one year to 30 years,” commencing through 2036; the July 2026 batch alone added about $68B of data-center leases with terms of 18 to 20 years. Oracle’s version of this problem, in our Q1 FY27 dive, was sign a nineteen-year cost against a five-year book: $288B of leases against a backlog mostly consumed within five years. Meta’s version has no book at all. Its revenue is an auction cleared every day. That is a strength in demand terms, because nobody can cancel a contract that does not exist, and a weakness in duration terms, because nothing on the revenue side is locked to the cost side.

The legal bill is partly paid in a future the quarter does not show. The Q3 2025 release warned that “a number of youth-related trials are scheduled for 2026, and may ultimately result in a material loss.” Decode what arrived. The August agreement provides for payments of up to about $18.59B. About $12.7B is payable to participating states in annual installments over 10 years; about $5.3B is contingent and not accrued. Meta booked $9.89B in Q3 and $11.89B across nine months for the agreement. Legal-related accruals rose from $7.32B at December to $19.88B at September 30, $8.30B of it non-current. That is why the charge barely touched Q3 cash: the cash-flow statement credits $7,439M to the change in other liabilities, against −$736M a year earlier. Q3’s −$2.12B of free cash flow was struck before the first installment. The next school-district bellwether trial is set for February 2027. Press reports on October 21 said a New Mexico judge ordered Meta to pay $25B in penalties in a separate case, which Meta plans to appeal; the 10-Q does not quantify it.

The funding stack has turned over. Meta held $82.45B of cash and marketable securities at September 30, down $7.81B in the quarter. Debt was $83.67B: $80.92B long-term plus $2.75B due in August 2027. For the first time in this series, debt exceeds cash and securities, by $1.22B. The path is two bond deals in under a year: long-term debt was $28.83B at September 2025, Q4 2025 added $30.00B of face value, and May 2026 added $25.00B in six series. Interest expense was $977M, up 330%, against interest income of $934M; the cash pile no longer pays for the debt. Meta repurchased no stock in the first nine months of 2026, against $26.25B in the same months of 2025, but still spent $12.77B on taxes for net share settlement of employee awards and $4.06B on dividends.

The new share class is a financing instrument, and the release says so. The board approved non-voting Class C stock, effective January 19, 2027, to be issued as a one-for-one dividend on every Class A and Class B share. The first reason the special committee gives: “The scale of investment required to compete, sustain growth, expand infrastructure, and pursue new strategic initiatives amidst the AI transformation taking place in the industry is significant,” and Class C “will provide Meta the ability to finance its growth.” Decode it with the cash line. A company with $96.87B of nine-month operating cash flow does not need a new share class to pay a $1.36B quarterly dividend; it needs one to sell equity, pay for acquisitions, or grant stock without diluting the founder’s vote. The Founder Alignment Agreement makes the vote part explicit: 20% of the Class C shares the Zuckerberg group receives are freely saleable, the other 80% only alongside a matching reduction in voting power, and any future Class A buyback must be matched by “at least an equal number of shares of Class C common stock on an aggregate basis.” On the call, Susan Li said Meta does not have any specific financing plans to announce at this time, per the Motley Fool transcript. No plan announced. The instrument is built anyway, and companies do not build financing instruments they expect never to use.

Here is the bull case at full strength. The ads business is accelerating on price, which is the part AI ranking improves: price per ad +14%, revenue +26%, Q4 guided to $73B to $76B, up 21.9% to 26.9% on the year. Operating cash flow was $96.87B over nine months, up from $79.59B. The settlement is a one-off with a 10-year payment schedule. The commitment book buys capacity the ad engine already monetizes, Muse and the frontier models add new revenue lines, and Reality Labs losses are guided down to about $15B in 2027 from $18B to $19B. Meta still expects 2026 operating income, excluding the charge, above 2025’s $83.28B. On that reading, negative free cash flow is the bottom of a build cycle, the same shape as 2022, when free cash flow fell to $154M and then rose to $13.64B within four quarters. We land the other way, for one reason: in 2022 the build was a capex wave, and capex can be cut. This one is $454.52B of contracts that cannot be cancelled and $355.73B of leases that have been signed. Meta has fixed the spending side for years and left the revenue side to the auction. If the auction keeps pricing at +14%, the bull is right. The financing still has to come first.

The guide and the Q4 bar

Meta guided Q4 revenue to $73B–$76B, raised full-year total expenses to $175B–$179B (from $165B–$169B, the full amount of the charge), and narrowed 2026 capex to $135B–$145B from $130B–$145B. The capex range has moved every quarter of the year: $115B–$135B in January, $125B–$145B in April, $130B–$145B in July, $135B–$145B now. The low end has risen $20B in nine months; the high end has not moved since April. The Q3 2025 release set the frame a year ago: “capital expenditures dollar growth will be notably larger in 2026 than 2025.” Decode the size. 2025 capex was $72.22B, up $32.99B on 2024. The 2026 guide is up $62.78B to $72.78B. “Notably larger” turned out to mean roughly double the dollar growth.

Now compute what Q4 has to deliver. Meta spent $85.82B on capex in the first nine months, so the full-year guide leaves $49.18B to $59.18B for Q4. Assume capex lands inside the guide and operating cash flow follows the pattern of the series. For Q4 free cash flow to come in at or above Q3’s −$2.12B, operating cash flow would need to reach at least $47.06B at the low end of the capex range, 29.9% above the best quarter in the series ($36.21B, in Q4 2025). For free cash flow to be zero, it would need $49.18B, 35.8% above that best quarter. Neither is impossible for a business guiding Q4 revenue up as much as 26.9%, but neither has happened. Add the $45.85B of commitments due in Q4, the $1.36B dividend, and RSU tax withholding that was $4.06B in Q3, and the cash pile shrinks in Q4 unless Meta borrows or sells something.

The operating side of the guide is easier. Full-year expenses of $175B–$179B leave $47.92B to $51.92B for Q4; against $73B–$76B of revenue that is Q4 operating income of $21.08B to $28.08B. To keep 2026 operating income excluding the Q3 charge above 2025’s $83.28B, Q4 needs more than $18.75B. Every point of the guide clears that. Reality Labs has $5.23B to $6.23B of Q4 loss left in its $18B–$19B range, against $6.02B in Q4 2025. The Q4 tax rate is guided to 15%–17%. The P&L bar is low. The cash bar is the one that binds, and it is the one the market traded on.

How peers read this

Meta is the purest version of the hyperscaler funding question, because it has no cloud customers to pass the bill to. Against Oracle, which turned customer prepayments into its funding source, Meta has no deferred revenue to lean on: every dollar of the commitment book is paid out of ad cash or borrowed. Against Microsoft, Alphabet and Amazon, the difference is the same one the Q2 dive drew: their AI capex has a contracted external revenue line behind it; Meta’s has an auction. Against CoreWeave, the shape is closer than it looks. Meta is not levered the way a neocloud is, but it now carries net debt, posts escrow, guarantees its landlords’ residual values, and buys third-party cloud capacity, the product the neoclouds sell. Demand-long, balance-sheet-short was written for the sellers of compute; the biggest buyer is starting to carry the same profile in smaller proportion.

For the suppliers, the read is constructive. $121.41B of non-cancelable commitments due in 2027, much of it servers and network equipment, is demand already signed for NVIDIA, the memory makers (see Micron), and the power and data-center chain. The escrow tells you who has the leverage in those contracts: the seller. The risk to suppliers is concentration, not demand. If Meta’s financing window narrows, the first thing it can slow is uncontracted capex, which is the part of the 2027 plan not yet in the book.

Valuation reality check

At $654.85 on 2,556.24M Class A and Class B shares outstanding at October 16, market value is about $1.67T. Add $1.22B of net debt and enterprise value is about $1.68T. On the last four quarters, that is 22.3 times GAAP diluted EPS of $29.32 and 19.9 times $32.89 if the Q3 charge is added back. Enterprise value is 18.8 times trailing operating income of $89.27B excluding the charge. Market value is 66.6 times trailing free cash flow of $25.13B, and that multiple gets worse in Q4 under any capex inside the guide.

The earnings multiple says Meta is a reasonably priced ad compounder. The free-cash-flow multiple says it is an infrastructure company priced on earnings. Both are true, and the gap between them is the depreciation that has not yet hit the income statement: $89.50B of construction in progress and $355.73B of leases that will become right-of-use assets and lease expense over the next decade. The commitment book is 48.4% of the market value. A reader paying 19.9 times adjusted earnings is paying for the ad engine and assuming the build earns its cost of capital; the filing gives no revenue line to check that assumption against, which is the single largest difference between Meta and the cloud builders.

What we’d watch from here

• A new bond or Class C offering filed with the SEC before the Q4 release. That is the call. No filing by then, and the cash pile carried Q4 alone.
• Q4 free cash flow against Q3’s −$2.12B. At the low end of the capex guide, Q4 operating cash flow needs $47.06B to match it, 29.9% above the best quarter in the series.
• The commitment book in the 10-K against $810.25B, and the 2027 tranche against $121.41B. Growth below the $181.95B Q3 add says the signing phase is ending; growth above it says it is not.
• Price per ad against +14% and impressions against +10%. Price falling below impressions again would mean the auction is no longer doing the work.
• Depreciation and amortization against $7.15B, with construction in progress against $89.50B. A quarter in which D&A rises faster than revenue is the margin cost of the build arriving.
• Restricted cash against $12.83B and residual value guarantees against $41.49B. Either rising means more of the build is being secured with Meta’s credit rather than paid for.
• The 2027 capex and expense guide on the Q4 call, against 2026’s $135B–$145B, and the Reality Labs loss against about $15B.

The trade

Evidence-led, not a chase. The 6.12% move on the reaction day prices the free-cash-flow turn and the share-class news, not a change in the ad business. A holder who wants to keep the ad engine and is uneasy about the funding could overwrite part of a position into the Q4 print, when the 2027 capex guide and the first full quarter of the commitment schedule arrive together. A bond issue would be the event that tells you how much the market charges Meta for the build; we would watch the spread on any new long-dated notes rather than the size. We have not priced strikes here.

The risk sits in Meta’s own notes: $454.52B of commitments that cannot be cancelled, $41.49B of residual value guarantees recorded at zero, $5.3B of contingent settlement payments not yet accrued, and a youth-litigation docket with a bellwether trial in February 2027.

Bottom line

Q2 asked whether Meta could keep spending at hyperscaler scale without a cloud business behind it. Q3 answers in two halves: the ad engine is strong enough to pay for the income statement, and not strong enough to pay for the commitment book in cash. Free cash flow turned negative for the first time since 2012, debt now exceeds cash, $121.41B of non-cancelable contracts fall due in 2027, and the board has just created a share class whose stated first purpose is to finance growth. Our call: Meta raises new external capital, notes or shares filed with the SEC, before it reports Q4 2026. Wrong if no such offering is filed by the Q4 release.