2 September 2026 · Built from web sources only: Meta’s FY2025 Form 10-K and results release (28 Jan 2026), the Q2 2026 Form 10-Q, results release and earnings presentation (29 Jul 2026), the Q2 2026 earnings call, and independent industry sources including WPP Media, dentsu, EMARKETER, court filings and competitors’ own filings. Not a valuation and not a recommendation.
Meta is an attention auction. It gives four communication apps away to 3.60 billion people a day and sells the right to interrupt them — by auction, priced per impression — to millions of advertisers who have no comparably cheap way to reach those people.
In Q2 2026 each of Meta’s 3.60 billion daily active people generated $16.86 of Family of Apps revenue and $6.53 of Family of Apps operating profit. A year earlier the same person generated $13.65 of revenue and $7.23 of profit. Revenue per person rose 23%; profit per person fell 10%. That single divergence is the whole investment question.
Profit per person is computed from Meta’s disclosed segment operating income and average DAP; every other figure is as reported.
| What the business is | A global advertising network wrapped around four free communication apps. |
|---|---|
| Industry | Digital advertising — worldwide ad revenue of roughly $1.3 trillion in 2026, of which digital is about 69%. |
| How it makes money | It ranks an ad against a user, bills the advertiser when the ad is delivered, and keeps the gap between the auction clearing price and the compute cost of serving the impression. |
| Unit of economics | One daily active person: $16.86 revenue, $6.53 Family of Apps operating profit per quarter. |
| What protects it | Owned distribution at a scale nobody can rent, first-party behavioural signal from four apps, and auction depth. Not patents, and not — per the November 2025 antitrust ruling — market power. |
| What drives earnings | Impressions delivered (+14%), average price per ad (+12%), and the cost of the compute that ranks both (capex 51% of revenue). |
| What to watch | Depreciation catching up to capex; whether ad pricing holds once the 2026 event calendar passes; the engagement effect of the August 2026 teen-usage remedies. |
| Cycle exposure | Medium — but rising, because the cost base has just been converted from variable to fixed. |
The problem Meta solves is not the one its users have. Users want to talk to people they know and be entertained by people they do not; Meta gives them four apps that do this for free. The paying customer is an advertiser who wants to reach a specific person at a specific moment and cannot do it economically any other way. Meta sits between the two and is paid by the second to reach the first.
Trace one unit. A person opens Instagram. The ranking system assembles a feed and reserves slots for ads. For each slot an auction runs in the time the screen takes to draw: advertisers hold standing bids expressed as an outcome they want — a purchase, an install, a lead — and Meta’s models predict how likely this person is to deliver it. Highest expected value wins. The advertiser is billed on credit and pays within weeks: receivables were $21.75bn against $60.80bn of quarterly revenue, a collection cycle of about a month.
Two consequences follow. The revenue is transactional, not contracted — deferred revenue was only $1.16bn at 30 June 2026, so there is no backlog and no visibility beyond the quarter. And the marginal cost of one more impression is the compute needed to rank it, which is why the entire economic argument now runs through the infrastructure line.
Advertising was 97.6% of FY2025 revenue. Family of Apps produced $60.37bn of revenue and $23.39bn of operating income in Q2 2026; Reality Labs produced $431m of revenue and a $4.62bn loss. Inside Family of Apps, a second and much smaller line is compounding fast: other revenue passed $1bn in a quarter for the first time, up 73%, on WhatsApp paid messaging and subscriptions. That matters out of proportion to its size, because it is the first evidence that WhatsApp — a product that peaked at 30 million messages a second during the World Cup final, with almost no monetisation history — can be charged for.
Reality Labs has changed character without changing name. Revenue grew 16% on AI glasses while Quest headset sales fell, and management has said roughly 70% of the segment’s 2026 operating expenses go to wearables. It has lost about $92bn cumulatively since 2020 and guides to a 2026 loss similar to 2025’s $19.19bn. Read it as a fixed ~$19bn annual charge levied on the advertising business, currently buying an option on glasses rather than on virtual reality.
Advertising exists because attention is scarce, perishable and impossible to store. What is actually sold is a probability: the chance that showing this message to this person now changes what they buy. The chain runs from advertiser through agencies and demand-side platforms to whoever owns the surface, and the profit pool has migrated decisively to the surface owners, because they alone hold the identity data that makes the probability estimable.
Global ad revenue should reach roughly $1.3 trillion in 2026, growing 8.9% excluding US political spend — an upgrade from the 7.1% WPP Media forecast in December, with US growth revised up to 11.9%. Digital is about 69% of the total. Within digital, three companies take 62.3% of worldwide spend.
The structural event of 2026 is that Meta is forecast to pass Google. EMARKETER projected in April 2026 that Meta will reach $243.46bn of net worldwide ad revenue against Google’s $239.54bn — a 26.8% share against 26.4%, the first time in the history of digital advertising that Google has not been first. The mechanism is growth rate, not scale: Meta accelerating from 22.1% to 24.1% while Google holds at 11.9%. The quarter corroborates the direction.
| Company | Ad revenue, Q2 2026 | Growth YoY | 2026 capex guidance |
|---|---|---|---|
| Meta — advertising | $59.4bn | +27% | $130–145bn |
| Alphabet — Google Search & other | $63.3bn | +17% | $195–205bn |
| Amazon — advertising services | $19.8bn | +26% | $220bn |
Capex figures are each company’s own full-year 2026 guidance, not comparable segment spend. Meta is growing advertising fastest while committing the least capital of the three.
Three do. Owned distribution: Meta reaches 3.60 billion people daily on surfaces it controls, so it pays no traffic acquisition cost to a gatekeeper — a line Google carries and Meta does not. Auction depth: an auction with millions of bidders clears higher than one with thousands, and every advertiser added raises the price paid by every existing one at no incremental cost. First-party signal: four apps used as communication infrastructure generate behavioural data no third-party broker can replicate.
Two commonly cited barriers do not bind. Advertiser switching costs are low — campaigns move platforms in days, which is precisely why performance rather than lock-in determines share. And Meta does not hold market power in the legal sense: on 18 November 2025 Judge James Boasberg ruled that the FTC had failed to prove a current monopoly, widening the relevant market to include TikTok and YouTube. The ruling removed existential divestiture risk. It is worth noticing what Meta had to argue to win it.
The court’s decision today recognizes that Meta faces fierce competition.
Among named rivals, Alphabet is the largest and is defending search economics against AI answers; Amazon is the fastest-growing threat, sitting on purchase intent Meta must infer; retail media is the fastest-growing digital channel of all. TikTok — identified by the court as Meta’s fiercest rival — resolved its US ownership question on 22 January 2026, when ByteDance closed the transfer of its US operations into a majority-American joint venture with Oracle, Silver Lake and MGX. Meta spent two years benefiting from TikTok’s uncertainty. That benefit has ended.
The industry evidence points to a clear answer on what kind of company wins here: one that owns global-scale distribution it does not rent, generates its own first-party signal, runs a self-serve auction deep enough to price long-tail demand, and can fund continual improvement of its ranking models. Meta is the purest example of that type in the market. The open question is not whether it is the right kind of company, but whether the capital intensity now required to stay that way leaves an adequate return.
Revenue grew 28% in Q2 2026 to $60.80bn, and 30% across the first half to $117.11bn. Almost all of it decomposes into two published numbers: impressions delivered rose 14% and average price per ad rose 12%, compounding to the 27% advertising growth reported.
| Component | Contribution | Evidence |
|---|---|---|
| Reported total revenue growth | +28% | $60.80bn vs $47.52bn |
| Currency | ≈ +1pt | +27% on a constant-currency basis; $685m FX effect |
| Acquired revenue | nil | $474m of acquisitions of businesses and intangibles in all of H1 2026; goodwill fell over the period |
| Organic, constant-currency | ≈ +27% | Volume (+14% impressions) × price (+12%) |
Reported and organic growth are effectively identical here, which is unusual at this scale and worth stating plainly: none of the headline rate was bought.
Price rises when the models predict outcomes better, because advertisers bid to a target return and better prediction lifts what a given bid is worth. But price also rises when the market is strong, and 2026 is strong. Both forces are operating; no public source separates them.
Daily active people grew only 3%, so most impression growth came from engagement per person and from what Meta calls ad load optimisations. Engagement gains are durable; ad load has a floor set by user tolerance, and Meta discloses no headroom.
On a user-geography basis, revenue grew 32% in the US and Canada and 36% in Rest of World, against 19% in Asia-Pacific. Rest of World grows from a low revenue-per-person base, so it adds impressions faster than dollars; the US adds dollars.
WhatsApp paid messaging and subscriptions. At 1.7% of revenue this changes nothing in 2026. It is the main evidence for the claim that Meta owns surfaces it has not yet charged for.
AI glasses are selling faster than management expected, but 0.7% of revenue against a $4.62bn quarterly segment loss makes this a cost centre with an option attached, not a growth engine.
What is conspicuously absent is user growth. Meta has effectively stopped adding people and is extracting more from the people it has. That is a mature-network pattern, and it makes pricing the whole story: if average price per ad stops compounding at low double digits, there is no volume lever large enough to replace it.
Meta’s margin structure used to be simple. Serving an ad cost almost nothing once the network existed, so revenue growth dropped through to operating income and the company earned a 41% operating margin on $201bn of FY2025 revenue. That arithmetic is being rewritten in real time.
In Q2 2026 the operating margin was 31%, down from 43%, on revenue that grew 28%. Total expenses grew 55%. Two items were one-off — $2.40bn of legal charges and $1.18bn of severance for the May 2026 reduction of about 8,000 roles — and excluding both, Meta states operating income would have risen 9%. That is the honest read of the quarter. But 9% operating income growth on 28% revenue growth is still severe operating deleverage, and its cause is not one-off.
R&D was 36% of revenue against 27% a year earlier. The two fastest-growing cost lines are depreciation and compensation: D&A reached $6.36bn in the quarter (up 46%), with servers and network assets alone at $4.62bn against $3.12bn; share-based compensation reached $7.66bn (up 58%) on AI hiring. Meta also disclosed a new category on the Q2 call — third-party AI token costs — meaning it is buying inference from others while building its own.
| US$ billions | FY2023 | FY2024 | FY2025 | H1 2026 |
|---|---|---|---|---|
| Revenue | 134.9 | 164.5 | 201.0 | 117.1 |
| Operating margin | 35% | 42% | 41% | 36% |
| Reality Labs operating loss | (16.1) | (17.7) | (19.2) | (8.6) |
| Capex, incl. finance leases | 28.1 | 39.2 | 72.2 | 50.9 |
| Free cash flow | 43.0 | 52.1 | 43.6 | 13.2 |
| Share repurchases | 19.8 | 30.1 | 26.2 | nil |
Comparability: FY2025 net income and the effective tax rate are distorted by a one-time $15.93bn charge on enactment of the One Big Beautiful Bill Act — excluding it the FY2025 rate would have been 13% rather than 30%. H1 2026 is a six-month figure, not comparable in absolute terms to the annual columns, and contains an $8.03bn tax benefit booked in Q1 2026 under CAMT transitional relief. The H1 2026 margin includes the Q2 legal and severance charges.
Free cash flow was $784m in Q2 2026 against $8.55bn a year earlier. Operating cash flow was healthy and rose 25% to $31.86bn; the collapse is entirely on the investing side, where capex including finance leases hit $31.08bn — 51% of revenue, against 22% in Q2 2024. Guidance is $130–145bn for full-year 2026, narrowed upward from $125–145bn, with management citing higher component pricing.
A large part of that bill has not reached the income statement. Construction in progress stood at $80.35bn at 30 June 2026, up from $50.52bn at year-end, and assets under construction do not depreciate. Reported margin currently reflects a fraction of the capital already committed.
The commitments extend well past the balance sheet. Leases signed but not yet commenced totalled $278.99bn, running to 2036 with terms up to 30 years, with more added in July. Meta holds 20% of a Louisiana data-centre venture with roughly $27bn of estimated development cost, a $12.31bn lease commitment from 2029 and residual value guarantees with a threshold near $28bn; its stated maximum exposure to loss on that venture alone is $46.03bn. A further $10.80bn of cash sits in escrow under multi-year infrastructure purchase agreements, released between 2028 and 2030.
Where the cash went, ranked across FY2025 and H1 2026: capital expenditure ($72.22bn, then $50.92bn), buybacks ($26.25bn, then nothing), non-marketable equity investments ($18.33bn in FY2025), dividends (~$1.35bn a quarter, maintained). Debt went the other way — the face value of senior notes rose from $59.00bn to $84.00bn after a $25bn six-series issue in May 2026 at coupons of 4.55% to 6.45%. The behaviour is clearer than any strategy statement: a company that stops repurchasing its own equity and starts issuing 30-year paper is telling you it believes the return on the next data centre exceeds the return on its own shares.
On 26 August 2026, mid-trial in the multi-district youth-harm litigation before Judge Yvonne Gonzalez Rogers in the Northern District of California, Meta settled with a bipartisan coalition of state attorneys general. Reported headline values differ across accounts — between roughly $16.7bn and $18bn — with approximately $12bn payable regardless and the remainder conditional on YouTube and TikTok adopting comparable safety measures, spread over ten years. Meta expects to accrue about $10bn of legal expense in Q3 2026, states this was not contemplated in the July guidance, and says all other guidance ranges are unchanged. This post-dates every reported figure above.
Advertising is discretionary spend that can be cut within a quarter, which historically made Meta cyclical on the revenue line but resilient on the cost line, because its costs were mostly people and its infrastructure was sized to traffic it already had. That symmetry has gone. The revenue remains as cancellable as ever while the cost base has been converted into depreciation schedules, 30-year leases and residual value guarantees that do not care what advertisers do next quarter.
Where it sits in its cycle: on volume and price, at or near a high — revenue growth accelerated from 22% in FY2025 to 30% in H1 2026, price per ad has risen 12% in each of the last two quarters against 6% in Q4 2025, and the market itself is accelerating. On margin, well below peak: 31% against 48% in Q4 2024. On free cash flow, at a trough: $784m in the quarter against $52.10bn for all of FY2024. A record top line coinciding with margin and cash flow at multi-year lows is unusual, and it is a function of deliberate spending, not of demand.
The downside case does not need an advertising recession; it needs advertising growth merely to normalise. If global growth reverts from 8.9% toward the mid-single digits dentsu forecast in December, and price per ad decelerates with it, revenue growth could halve while depreciation from the 2025–26 capex programme steps up regardless. Meta has never operated with this cost structure in a weak ad market. There is no analogue in its own accounts.
| Region | Revenue | Share | Growth YoY |
|---|---|---|---|
| US & Canada | $26.8bn | 44% | +32% |
| Europe | $14.3bn | 24% | +24% |
| Asia-Pacific | $11.2bn | 18% | +19% |
| Rest of World | $8.5bn | 14% | +36% |
Europe includes Russia and Turkey; Rest of World includes Africa, Latin America and the Middle East. Nearly a quarter of revenue sits in the jurisdiction now degrading Meta’s targeting signal.
| Indicator | Where published | What it tells you |
|---|---|---|
| Ad impressions and average price per ad | Quarterly earnings release, investor.atmeta.com | Separates volume from pricing. Price decelerating while impressions accelerate means ad load, not demand. |
| Depreciation of servers and network assets | Property and equipment note, 10-Q / 10-K | How much of the capex bill has reached the income statement. $4.62bn in Q2 2026. |
| Construction in progress | Property and equipment note, 10-Q / 10-K | What has been paid for but is not yet depreciating. $80.3bn at 30 June 2026. |
| Family daily active people and ARPP | Quarterly earnings presentation | Whether the teen-usage remedies show up in engagement from Q4 2026. |
| Leases not yet commenced | Commitments and contingencies note, 10-Q / 10-K | The fixed obligation building off balance sheet. $279.0bn at 30 June 2026. |
| Global and US ad revenue growth | WPP Media This Year Next Year; dentsu; Magna | Distinguishes Meta share gain from a rising market. |
The teen-usage remedies reduce the input, not just the profit. The August 2026 settlement requires a two-hour cumulative daily limit for teens across Facebook and Instagram, disableable only with parental permission; blocked access between midnight and 6am; notifications muted during school hours; a non-algorithmic feed option; autoplay disabling; and like counts hidden by default. Each of those reduces sessions or reduces the ranking signal within a session. Impressions are 14 points of a 27-point ad growth rate, and this removes some of them by court order rather than by competition.
Depreciation is a bill already paid, arriving later. Unlike headcount, depreciation cannot be cut in response to weak revenue — the cash left years earlier. With $80.35bn in construction in progress and $130–145bn of 2026 capex, the depreciation line has years of programmed increases embedded in it. If revenue growth normalises before that schedule matures, margin compresses on both blades at once.
Off-balance-sheet obligations convert a flexible business into a fixed one. $278.99bn of leases not yet commenced, plus roughly $28bn of residual value guarantees on the Louisiana venture, are contractual regardless of ad demand. That venture is deliberately non-consolidated because Meta is not the primary beneficiary, yet Meta discloses maximum exposure to loss of $46.03bn on it. An investor reading only the consolidated balance sheet will understate the fixed cost of this business.
Single-product concentration compounds everything else. Advertising was 97.6% of FY2025 revenue. Meta is financing a multi-year infrastructure programme, servicing $84bn of debt and absorbing a $19bn annual Reality Labs loss from one revenue line, sold to one buyer type, on a spot basis with no backlog.
European signal degradation is a slow leak on a quarter of revenue. Following the €200m DMA fine in April 2025, Meta began rolling out its Less Personalized Ads option in January 2026 after the European Commission acknowledged the revised model. Every user who takes it moves from behavioural to contextual targeting, which lowers the predicted outcome rate and therefore what advertisers will bid.
The antitrust win is not final. The FTC has appealed the November 2025 ruling. A reversal would revive divestiture exposure over Instagram and WhatsApp — and Instagram is the surface carrying the current growth.
These are findings, not gaps in the research. Each would need resolving before forming a thesis.