Meta Platforms, Inc.

01 — Executive summaryRevenue of $60.8 billion. Operating income of $18.8 billion. One of those numbers fell.

Meta Platforms reported revenue of $60.8 billion for the quarter ended 30 June 2026, an increase of 28% over the prior year and the fastest growth rate the company has reported since 2021. Advertising revenue rose 27%. Ad impressions delivered rose 14% and the average price per ad rose 12%. Family daily active people reached 3.60 billion. On any reading of the top line, this was an excellent quarter for the business that generates 97.6% of Meta's revenue.

Operating income fell 8%, to $18.8 billion. Operating margin compressed from 43.0% to 30.9%, the lowest it has been since the fourth quarter of 2022. Total costs and expenses rose 55%. Research and development rose 67%. General and administrative expenses more than doubled, because the quarter contained $2.40 billion of charges related to legal proceedings and $1.18 billion of severance. Free cash flow, which was $8.5 billion in the same quarter a year earlier, was $784 million. The shares fell approximately 9% the following day.

Both of those paragraphs describe the same quarter, and the distance between them is the subject of this report. Meta is not a company in trouble. It is a company in the middle of converting a capital-light advertising monopoly into a capital-intensive infrastructure business, and the conversion is now large enough to move the reported numbers in ways that the reported numbers do not explain.

The question we set out to answer is narrower than whether Meta's AI strategy will work. It is this: what does the current price assume about the next two years of operating income, and is that assumption supportable given what the company has already committed to spend?

Six findings follow from the primary record. Each is developed in full later in this report, and each is stated here with the number that supports it.

1. The advertising business is performing better than at any point in its history

Revenue growth of 28% on a base of $47.5 billion is remarkable at this scale, and the composition matters more than the headline. Growth is no longer coming from serving more ads to more people: family daily active people grew 3%, the slowest rate on record. It is coming from price. The average price per ad rose 12% in the quarter, the second consecutive quarter of double-digit price growth, against a single-digit rate as recently as the fourth quarter of 2025. Meta attributes this to ranking and creative improvements delivered by its AI models. Whatever the attribution, the arithmetic is unambiguous: the auction is clearing at higher prices without a proportional increase in inventory.

The corollary is that the bull case does not require a new product to work. It requires the existing one to keep doing what it is doing, because the AI spending is already showing up in the only place it can show up today — the price of an impression.

2. Capital expenditure is now larger than operating cash flow is likely to be

Meta guided 2026 capital expenditure to $130–145 billion, having raised the range twice during the year, and has said it intends to bring roughly 7 gigawatts of AI compute capacity into service during 2026 and to double that to 14 gigawatts by the end of 2027. Sell-side consensus for 2027 capital expenditure is approximately $185.6 billion. Meta's operating cash flow over the trailing twelve months was $130.3 billion.

Those numbers do not need to be modelled carefully to see the problem. On consensus capital expenditure and a reasonable operating cash flow estimate, Meta's free cash flow turns negative in 2027. The company has never reported a full year of negative free cash flow. It has funded the last eighteen months of the build with $55 billion of bond issuance, including a $30 billion sale in October 2025 that was the largest in its history, and it has stopped buying back its own stock: repurchases were $26.3 billion in 2025 and zero in the first half of 2026.

3. The cost of the programme arrives through depreciation, and it has already started

The reason the margin compressed in the June quarter is not only legal charges and severance. Depreciation and amortisation rose 50% in the first half, to $12.4 billion, against $8.2 billion in the comparable period of 2025. On our estimates, Meta's depreciation charge rises from approximately $18.7 billion in 2025 to $28.4 billion in 2026 and $45 billion in 2027, as the 2026 capital programme is placed in service.

This is arithmetic rather than judgement. A company that spends $137.5 billion on assets depreciated over a blended five to six years will book roughly $25 billion a year of additional depreciation, and it will book it whether or not the capacity is used. Management has reaffirmed that full-year 2026 operating income will exceed 2025. That guidance is achievable because revenue is growing 26%. It becomes materially harder in 2027, when the same revenue growth has to absorb the first full year of the 2026 asset base.

4. Reported earnings are noisier than they look, in both directions

Meta's effective tax rate was 87% in the third quarter of 2025 and −23% in the first quarter of 2026. Neither number says anything about operations. The first reflects a $15.93 billion non-cash valuation allowance charge triggered by the One Big Beautiful Bill Act; the second reflects an $8.03 billion benefit that partially reversed it, arising from US Treasury Notice 2026-7. Between them, the two items move reported diluted earnings per share by roughly $9 across two quarters. On the full year 2025 Meta reported a 30% effective tax rate; absent the charge, the rate would have been 13%.

The practical consequence for a reader is that any two-year earnings comparison built on reported figures is unreliable unless the two items are stripped out. Section 04 does that and shows what the operating trend looks like on a clean basis.

5. The AI programme has produced a consumer product, but not yet a revenue line

Meta Superintelligence Labs shipped its first proprietary model, Muse Spark, in April 2026 and followed it with 1.1 in July and 1.2 in August. On 8 September the company launched Muse, a consumer agent that browses the web, manages email and completes multi-step tasks, free at the base tier with paid plans at $20 and $100 a month. It reached roughly five million downloads in three weeks and about 557,000 daily active users in the United States by 19 September. The launch drove a 27% gain in the share price during September, the best month for the stock since November 2022.

What Meta has not published is retention, engagement per user, or revenue. Its own disclosure shows that "Other revenue" — the line that would contain any subscription income — was $1.01 billion in the quarter, 1.7% of total revenue. The honest reading is that Muse is a distribution event with a real strategic logic and no current financial consequence.

6. On the numbers that matter, the shares are neither cheap nor obviously expensive

At $741.90, Meta trades at 28.0× trailing GAAP earnings and approximately 23.7× our estimate of FY2026 GAAP earnings. Its five-year average trailing multiple is approximately 26.5 times. On the earnings measure, the shares are close to their own history.

On free cash flow they are not. The trailing free cash flow yield is 2.0%, against 3.6% at the end of 2024 and 6.1% at the end of 2022. The multiple has not changed much; the cash the multiple is applied to has. This is the central tension in the equity, and it is why our rating is Neutral rather than either constructive or negative. The operating business deserves a premium multiple. The capital programme has not yet earned one, and it is being paid for out of the cash flow that would otherwise have justified it.

What would change our mind

We would move to a positive rating on any of three developments: a disclosure of Muse engagement and retention that shows the agent retaining users at rates comparable to the leading consumer AI applications, which would establish a second distribution asset with independent value; a 2027 capital expenditure guide at or below the current consensus of $185.6 billion, combined with operating margin stabilising above 33%, which would show the programme converging rather than compounding; or evidence that the compute build is being monetised externally at a scale that shows up in Other revenue.

We would move to a negative rating on any of three others: a 2027 capital expenditure guide above $215 billion without a corresponding revenue disclosure; a court judgment in the youth-safety litigation that establishes a damages formula rather than a settlement range, given that 29 state attorneys general are plaintiffs; or two consecutive quarters in which advertising price growth falls below 6% while capital expenditure guidance is maintained, which would indicate that the AI-driven improvement in the auction has plateaued before the depreciation arrives.

02 — Company & business modelA recommendation engine with a balance sheet attached

Meta Platforms is the largest advertising business ever assembled, and the most concentrated. In FY2025 the company reported total revenue of $201.0 billion, of which $196.2 billion — 97.6% — was advertising. The remaining 2.4% comprised $2.6 billion of "Other revenue," which includes WhatsApp Business messaging and the small subscriptions business, and $2.2 billion of Reality Labs hardware revenue. There is no third leg.

That concentration is worth stating plainly because it is the single most important fact about the company's risk profile. Alphabet, the closest comparable, derives roughly two-thirds of revenue from advertising and has a multi-hundred-billion-dollar cloud business behind it. Amazon's advertising business is a rounding error in its revenue but a disproportionate share of its profit, and it is embedded in a retail operation with $716.9 billion of revenue. Meta has one business, and everything else is a bet that the one business will fund.

How the money is actually made

Meta sells impressions in an auction. Advertisers bid for the right to show a creative to a specific person in a specific context; Meta's ranking system decides which bid wins and what it costs. Revenue is recognised when the impression is delivered. There are no long-term contracts, no backlog, no committed minimums. The entire $196 billion annual run rate is repriced every quarter by an auction that is itself a function of advertiser return on ad spend.

This structure has two consequences that a reader should hold onto throughout this report. First, the revenue line is unusually sensitive to the quality of the ranking system, because better ranking raises conversion, which raises bids, which raises price. That is the mechanism through which AI spending reaches the income statement, and it is why the 12% price growth in the June quarter is the most important number in the bull case. Second, the revenue line is unusually sensitive to the macroeconomy, because advertising is the most cyclically discretionary line item in most corporate budgets. Meta has no contracted revenue to fall back on.

The four applications, and what each one is for

Facebook remains the largest by engagement and the most mature by monetisation. Instagram is the growth engine within the family and the primary beneficiary of the shift from text and images to short video. WhatsApp is the largest by users and the least monetised, with advertising only beginning to be introduced in the European Union during 2026 and business messaging as the principal commercial route. Messenger sits between the three as a messaging surface with a limited direct advertising load.

The company reports these as one segment, Family of Apps, and discloses no revenue split between them. That is a genuine limitation on the analysis: an investor cannot tell from the filings whether Instagram's monetisation is still improving, whether Facebook's is declining, or whether the two are offsetting. We note it because it is the second-largest disclosure gap in the equity, after the compute question.

Reality Labs is a separate company inside the company

Since 2020 Meta has reported Reality Labs separately, and the disclosure is unusually candid: revenue, cost of revenue and operating loss for a segment that has never had a profitable quarter. In FY2025 the segment generated $2.2 billion of revenue and lost $19.2 billion. In the first half of 2026 it generated $833 million of revenue and lost $8.6 billion. Cumulatively, since the segment was first disclosed in 2020, it has consumed more than $85 billion of operating losses.

The strategic rationale has shifted over that period. The original thesis was the metaverse: virtual and augmented reality as the next computing platform, with Quest headsets as the entry point and Horizon Worlds as the destination. That thesis has not been abandoned but it has been quietly deprioritised. The current emphasis is on AI glasses — the Ray-Ban Meta line, developed with EssilorLuxottica — which tripled unit sales year over year in the June quarter and is the segment's only product with mass-market traction. The company reduced Reality Labs headcount in January 2026 as part of a broader reallocation toward AI.

We treat Reality Labs in this report as a fixed annual charge rather than as an option with a probability-weighted value. That is a deliberately conservative treatment, and section 07 sets out why we think it is the right one.

What changed in 2025 and 2026

Three structural changes separate the company described above from the one that traded at $120 in 2022.

The first is the reorganisation of AI. In June 2025 Meta created Meta Superintelligence Labs and placed it under Alexandr Wang, the co-founder of Scale AI, in which Meta took a minority stake for a reported $14 billion or more. A large hiring programme followed, aimed at researchers from competing laboratories. Yann LeCun, the company's chief AI scientist of twelve years and a public sceptic of the large-language-model-first approach, left in November 2025 and subsequently raised roughly $1.03 billion for a competing venture built on a different technical premise. The reorganisation is the origin of the research and development line that rose 67% in the June quarter.

The second is the capital programme. Meta spent $39.2 billion on capital expenditure in 2024, $72.2 billion in 2025, and expects to spend $130–145 billion in 2026. It is building two named facilities — Prometheus, intended to be among the first gigawatt-scale AI clusters in the world, and Hyperion, in Richland Parish, Louisiana, which the company announced in July 2026 would be expanded toward 5 gigawatts of capacity with a total investment reported above $500 billion. These are not research budgets. They are industrial projects.

The third is the change in the capital return. Meta initiated a dividend in 2024 and repurchased $29.8 billion of stock that year, $26.3 billion in 2025, and nothing in the first half of 2026. Over the same period its long-term debt rose from $28.8 billion to $83.7 billion. The company has moved from returning cash to raising it.

03 — The financial recordFour years in which operating income grew 188%, and then stopped

Between FY2022 and FY2025 Meta's revenue grew from $116.6 billion to $201.0 billion, a 72% increase. Over the same period operating income grew from $28.9 billion to $83.3 billion, an increase of 188%. Operating margin expanded from 24.8% to 41.4%. That is the defining financial fact of the last four years: a company that doubled its revenue and nearly tripled its profit, because the advertising business scaled into a cost base that had already been built.

That period is now over. In the first half of 2026 revenue grew 30% to $117.1 billion while operating income grew 9.6% to $41.6 billion. In the second quarter alone, operating income fell 8%. The gap between revenue growth and profit growth, which was strongly positive for three years, has closed and turned negative.

The quarterly record shows the inflection more precisely than the annual one, because it separates the period in which operating income was the story from the period in which it stopped being the story.

Q3 2025 and Q4 2025 cost and revenue lines are derived by subtraction from reported full-year totals. Q3 2026 revenue is the midpoint of company guidance of $61–64 billion; Q3 2026 and Q4 2026 operating income, cost and margin figures are Farstar estimates built from the company's full-year expense guidance of $165–169 billion. The ex-one-off-tax EPS line is a Farstar calculation that removes the $15.93 billion valuation allowance charge in Q3 2025 and the $8.03 billion reversal in Q1 2026 at the reported effective rate; it is an approximation of underlying earnings power and is not a company disclosure.

Four observations from this table deserve emphasis.

Operating margin broke a four-year uptrend in the June quarter. From Q1 2025 through Q1 2026, Meta's operating margin sat in a narrow band between 40.1% and 43.0%. In Q2 2026 it printed 30.9%. Some of that is one-off — $3.58 billion of legal and severance charges, or 5.9 percentage points of margin. But even excluding those items the margin would have been approximately 36.8%, roughly six points below the trailing band. The remaining compression is operating leverage running in reverse: costs that scale with the capital programme rather than with revenue.

Reported net income has become a poor guide to anything. In Q3 2025 Meta reported net income of $2.7 billion on $51.2 billion of revenue, a net margin of 5.3%, because of a tax charge that had nothing to do with the quarter's trading. In Q1 2026 it reported $26.8 billion of net income and diluted EPS of $10.44, of which $3.13 was a tax benefit that reversed part of that same charge. Neither quarter's bottom line is meaningful in isolation.

The cost line is the whole story. Total costs and expenses grew 55% in the June quarter against revenue growth of 28%. Over the first half, costs grew 45% against revenue growth of 30%. On the company's own full-year guidance, costs will grow approximately 42% in 2026 while revenue grows approximately 26%. A business whose costs grow 1.6 times as fast as its revenue does not hold its margin, and Meta is not claiming it will.

Free cash flow has almost disappeared. Meta generated $64.1 billion of operating cash flow in the first half of 2026, 29% more than a year earlier, and spent $50.9 billion on property and equipment. Free cash flow was $13.2 billion for the half, of which $12.4 billion came in the first quarter. The second quarter produced $784 million, a decline of 91% year over year.

One further framing observation governs everything that follows. Meta's equity has, over the last eighteen months, changed character. It used to be a claim on a very high-margin, capital-light cash flow stream. It is now a claim on that cash flow stream plus a leveraged industrial programme, funded partly by debt raised against the first. Each of the two parts is defensible on its own. Together they mean that the historical relationship between Meta's reported earnings and its share price — a relationship that held remarkably well from 2022 to 2025 — has stopped being a reliable guide.

04 — Quality of earningsTwo quarters, $24 billion of tax noise, and what is underneath

The single most useful exercise in reading Meta's recent results is to remove the two legislative tax items and see what the trend looks like without them. They are large enough to reverse the apparent direction of earnings growth in two separate quarters, and because they sit in different years they distort any two-year comparison built on reported figures.

What happened, and why

In the third quarter of 2025, on enactment of the One Big Beautiful Bill Act, Meta recorded a non-cash valuation allowance charge of $15.93 billion. The charge did not reflect a change in Meta's business; it reflected a change in the tax treatment of certain capitalised items, which removed the company's ability to recognise a deferred tax asset it had previously been carrying. Reported net income for the quarter was $2.7 billion and diluted EPS was $1.05, against consensus expectations of roughly $6.60.

In the first quarter of 2026, US Treasury Notice 2026-7 clarified the corporate alternative minimum tax treatment of previously capitalised US research and development costs. The effect was an $8.03 billion income tax benefit, which partially reversed the earlier charge. Reported diluted EPS was $10.44; excluding the benefit, it would have been $7.31, a figure the company disclosed explicitly. The reported effective tax rate for the quarter was −23%.

The full-year picture is equally distorted. Meta reported a 30% effective tax rate for FY2025. Absent the valuation allowance charge, the rate would have been 13%, which is the company's own disclosure. The difference between a 30% and a 13% tax rate on $83.3 billion of operating income is roughly $14 billion of net income, or about $5.50 per share.

Why this matters more than usual here

Tax noise is common and normally uninteresting. It matters in Meta's case for a specific reason: the company's earnings are the denominator in the valuation, and the company is about to enter a period in which operating income growth is the contested variable. If a reader builds a two-year EPS trend on reported figures, the trend will look like a decline from $23.49 in FY2025 to approximately $31.30 in FY2026 — a 33% increase. On a constant-tax-rate basis the increase is approximately 17%. The difference is not a rounding error; it is the difference between a multiple that looks reasonable and one that does not.

Our approach throughout this report is to value Meta on our own FY2026 and FY2027 estimates built from operating income, a normalised tax rate of 16–17%, and a share count that does not assume buybacks. Section 21 states every input.

05 — Segment: Family of AppsA $60.4 billion segment whose margin fell four points in a year

Family of Apps is the whole company in all but name. In the June quarter the segment generated revenue of $60.4 billion, which is 99.3% of the consolidated total, and operating income of $23.4 billion, which is more than the consolidated total because Reality Labs lost money. Segment operating margin was 38.8%, down from 53.0% in the same quarter a year earlier.

That four-point-plus decline is worth pausing on, because it is the clearest single piece of evidence that the capital programme is now reaching the income statement. Family of Apps carries essentially all of Meta's research and development spending and all of the depreciation on the compute the company has built. Advertising revenue within the segment grew 27%. Segment costs grew substantially faster.

The composition of the increase is instructive. In the June quarter, research and development rose from $12.9 billion to $21.7 billion, an increase of 67%. Meta attributes this to three things: infrastructure costs, AI token costs, and severance associated with the May 2026 reduction. The first two are the operating cost of running and training models; they scale with the compute base. General and administrative expenses rose from $2.7 billion to $5.6 billion, driven almost entirely by the $2.40 billion legal charge. Marketing and sales rose 15%, well below revenue growth, which is a genuine efficiency. Cost of revenue rose 33%, above revenue growth, reflecting the cost of delivering more impressions and the depreciation on the servers that serve them.

The accounting that makes the segment look worse than it is, and better than it is

Meta allocates essentially all research and development to Family of Apps, including the substantial share that is directed at models and infrastructure used by Reality Labs and by the unmonetised parts of the family. That flatters Reality Labs, whose reported loss excludes a share of the AI research that supports its products, and penalises Family of Apps. It is a reasonable convention but it means the segment margin understates the underlying advertising profitability by an amount that is not disclosed and that we estimate at three to five percentage points.

In the other direction, Family of Apps revenue includes the "Other revenue" line, which was $1.01 billion in the quarter and grew 73% year over year, but which carries negligible operating income. The line is small enough that the distortion is immaterial today. It will not be if Meta's agent and subscription ambitions produce revenue at the scale the company is targeting.

What we would want to see, and cannot

The most important missing disclosure in this segment is the split between Facebook and Instagram. Meta's own communications consistently describe Instagram as the growth surface and Facebook as the mature one, but the filings do not support or refute that. For an investor the distinction matters: if Facebook, which remains the larger surface by engagement in most Western markets, is declining in monetisation while Instagram grows, then the aggregate growth rate conceals a mix shift that will eventually stop being offsettable. If both are growing, the aggregate is what it appears to be.

We do not have a way to resolve this from public data, and we have not tried to. What we can say is that the aggregate advertising revenue growth of 27% in the June quarter, combined with a 14% increase in impressions and a 12% increase in price, is consistent with a healthy auction across the family rather than with one surface carrying another.

06 — Advertising: the auctionPrice is doing the work now, and that is the whole bull case

Meta's advertising revenue can be decomposed into two variables and no more: how many impressions the company delivers, and what it charges for each one. The company discloses both on a year-over-year basis every quarter, which makes it possible to read the mechanism rather than the outcome. What that reading shows is a business that has completed a transition from volume-led growth to price-led growth, and is now almost entirely dependent on the second.

In FY2023 Meta grew advertising revenue 16% on the back of a 28% increase in impressions and a 9% decline in price per ad. That was the Reels era: Meta was buying engagement by inserting a large volume of lower-value short-video inventory into the feed, and the auction priced it accordingly. In FY2024 the mix reversed — impressions grew 11% and price grew 10% — and in FY2025 impressions grew 12% while price grew 9%.

In the first half of 2026 the balance tipped decisively toward price. Impressions grew 19% in the first quarter and 14% in the second; price grew 12% in both. The second-quarter combination is the more meaningful one, because the impression growth rate fell five points sequentially while the price growth rate held. That is the signature of a demand-side improvement rather than a supply-side one.

Why the price is rising, and what that has to do with AI

Meta's explanation is that its ranking and recommendation systems have improved, delivering better-matched ads, which raises conversion, which raises the return on ad spend, which raises bids. There is a version of this claim that is marketing and a version that is measurable, and the measurable version is that the price of an impression is up 12% while the volume of impressions is up 14%. If the improvement were merely a supply shift toward better inventory, volume would be falling, not rising.

There is independent support for the direction of travel. The company has disclosed that AI ranking and creative tooling are the primary drivers of the improvement, and the composition of advertising growth by vertical — online commerce being the largest contributor in the June quarter — is consistent with better direct-response performance rather than with brand budget reallocation. Direct-response advertisers measure return on ad spend and move budgets quickly when it improves.

The sceptical reading is that a 12% price increase on a 14% volume increase is what happens when a market grows faster than supply, and that the global advertising market grew unusually strongly in the first half of 2026. We do not have Meta's auction-level data and neither does anyone outside the company, so the question cannot be settled definitively. What can be said is that Meta's price growth has exceeded reported industry growth, which points at least partly to a share gain or a quality gain rather than to a rising tide.

The concentration problem underneath the growth

Advertising is 97.6% of Meta's revenue, and advertising is the most cyclical major line item in corporate budgets. Meta's revenue fell 1% in FY2022, the only decline in its history as a public company, and advertising revenue fell 1.4% that year while the company's user base grew. The mechanism was straightforward: advertisers cut budgets in response to rising interest rates and an uncertain macro outlook, and Meta had no contracted revenue to cushion it.

Meta is now more concentrated than it was in 2022, not less, because Reality Labs revenue has not grown and the other revenue line, while growing quickly in percentage terms, remains 1.7% of the total. An investor buying Meta today is buying a single-cycle revenue stream at a multiple close to its five-year average, with a cost base that is now structurally heavier than at any point in the company's history. That is a different risk profile from the one the average multiple was earned under.

07 — Segment: Reality LabsFive years, $86 billion, and $2 billion a year of revenue

Reality Labs is the clearest example in the company of a strategic commitment that has not produced a financial result. Since Meta began disclosing the segment in 2020 it has reported an operating loss in every quarter, and the cumulative loss now exceeds $86 billion including the first half of 2026. Over the same period cumulative segment revenue is approximately $13 billion.

The shape of the chart is what matters. Segment revenue has been almost perfectly flat for five years: $2.27 billion in 2021, $2.16 billion in 2022, $1.90 billion in 2023, $2.15 billion in 2024, $2.21 billion in 2025. There has been no growth trend at all. Meanwhile the annual operating loss has grown from $10.2 billion to $19.2 billion. This is not a business in the early innings of a growth curve. It is a business that has been the same size for half a decade while its costs have nearly doubled.

What changed in 2026

Two things, both of which point in the same direction. First, Meta reduced Reality Labs headcount in January 2026 as part of a reallocation of resources toward AI, and the segment's quarterly loss has narrowed modestly — $4.03 billion in the first quarter and $4.62 billion in the second, against $4.21 billion and $4.53 billion in the comparable periods of 2025. The second-quarter loss was narrower than the approximately $5.07 billion analysts had expected, which is the first time in several years that the segment has surprised positively on cost.

Second, the product mix has rotated decisively toward AI glasses. The Ray-Ban Meta line, developed with EssilorLuxottica, tripled unit sales year over year in the June quarter and is now the segment's only product with mass-market traction. Quest headsets, which were the original thesis, have not sustained consumer demand at scale. The company has not disclosed unit volumes for either product line, and the tripling is a growth rate on an undisclosed base.

Why we treat it as a cost, not an option

A conventional analyst would value Reality Labs as a real option: a small probability of a very large outcome, discounted to a present value. We have deliberately not done that, for three reasons.

The first is that the evidence has not moved in five years. A real option whose underlying has been the same size for five years, whose costs have doubled, and whose only commercially successful product is a licensed co-brand with a third party, is not an option that can be valued with any confidence. The confidence interval around any number we could produce would be wider than the number.

The second is that the loss is not going away. Meta has guided that Reality Labs operating losses will remain at a similar level in 2026, and nothing in the segment's trajectory suggests a path to breakeven inside the forecast horizon of this report. That means the segment consumes roughly $18–19 billion a year of cash that would otherwise be available for capital expenditure or for shareholders.

The third is that the strategic rationale has changed without the accounting following. The metaverse thesis has been quietly replaced by an AI-glasses thesis, and the AI-glasses business is closer to a hardware business with consumer-electronics economics than to a platform business with software margins. Hardware businesses with low-single-digit unit volumes and third-party licensing dependencies do not earn 40% operating margins. If Reality Labs succeeds, it will succeed at a materially lower margin structure than the rest of Meta. That is a real outcome, but it is not the outcome the segment's cost base currently implies.

08 — The AI programmeFourteen billion dollars of talent, and a model that is now proprietary

Meta's AI programme is the reason the cost base has changed, and it is worth describing carefully because the public discussion of it is unusually confused. There are three distinct things happening at once: an organisational rebuild, a technical strategy that has reversed, and a capital programme. They have different economics and different risks.

The organisational rebuild

In June 2025 Meta created Meta Superintelligence Labs, consolidating its previously fragmented AI teams into a single unit and placing it under Alexandr Wang, the co-founder of Scale AI. Meta took a large minority stake in Scale AI for a reported $14 billion or more, and installed Wang as chief AI officer. Nat Friedman, the former chief executive of GitHub, was brought in to lead AI products. A substantial hiring programme followed, aimed at senior researchers at competing laboratories, at compensation levels that became a subject of public discussion across the industry.

The rebuild had a cost that is visible in the numbers. Research and development expense rose 67% in the June quarter to $21.7 billion. Some of that is compensation, some is infrastructure, and some is the "AI token costs" Meta cited in its own commentary. The company also booked $1.18 billion of severance in the quarter in connection with the May 2026 reduction of approximately 8,000 roles, which is best understood as the other side of the same reallocation: Meta is reducing headcount in some functions while paying very heavily to recruit in others.

There has also been senior attrition. Yann LeCun, Meta's chief AI scientist for twelve years and the most prominent public advocate of a "world models" approach as an alternative to scaling large language models, left in November 2025. Reporting at the time cited both organisational friction and a genuine technical disagreement about direction. By March 2026 his new venture, AMI Labs, had raised approximately $1.03 billion at a $3.5 billion valuation. LeCun had also publicly questioned the published benchmark results for Llama 4. Whatever the merits of the dispute, the departure of the company's most credentialed AI researcher within six months of the reorganisation is a data point about execution risk that should not be dismissed.

The technical strategy has reversed

Meta built its AI reputation on open weights. The Llama family was released under a community licence that permitted broad commercial use, and by 2024 Llama had seeded the largest open-model ecosystem in existence. That was the strategy, and it was coherent: Meta does not sell models, so giving them away cost nothing and created goodwill, a developer ecosystem and a recruiting advantage.

The 2026 strategy is different. The first model from Meta Superintelligence Labs, Muse Spark, shipped in April 2026 as a proprietary model, metered through a Meta Model API at $1.25 per million input tokens and $4.25 per million output tokens. Version 1.1 followed in July with agentic tool use and a one-million-token context window. Version 1.2 followed in August with a terminal-based coding agent and, notably, a $0.10 / $0.20 tier priced at roughly a twelfth of the standard rate, in exchange for permission to use the customer's prompts and outputs to improve Meta's products.

Five days after Muse Spark 1.2, Meta released Muse Glimmer, a 30-billion-parameter model distilled from Spark 1.2 under an Apache 2.0 licence — more permissive than any Llama licence ever used, with no revenue cap and no user ceiling. It runs on a 24-gigabyte consumer graphics card. Zuckerberg framed it as a deliberate choice to distribute capability rather than centralise it. The commercially important fact is that Glimmer is a distillation of the flagship, not the flagship, and open weights for Muse Spark 1.2 itself have been promised but not delivered. Meta's best model remains closed and metered.

Where the programme actually stands

The honest assessment is that Meta has closed the organisational gap and has not closed the capability gap. Muse Spark is a competent frontier-adjacent model with competitive pricing and a strong consumer distribution channel. It is not the leading model by any published third-party benchmark, and Meta does not claim it is. The company's competitive advantage is not the model; it is the 3.6 billion daily users it can deliver the model to, and the advertising auction the model can improve.

That is a defensible position, and it is the same position Meta has occupied in every platform transition since 2007. It is also a position that does not obviously require $137.5 billion of annual capital expenditure. The gap between what the strategy requires and what the capital programme implies is the subject of sections 10 through 14.

09 — Muse and the agentFive million downloads, no published retention, and no revenue line

On 8 September 2026 Meta launched Muse, a consumer AI agent that browses websites, manages email, and completes multi-step tasks such as bookings, running on the Muse Spark models. It is available in the United States and Canada, also reachable through WhatsApp. The base tier is free; paid plans are priced at $20 and $100 a month, with small transaction fees planned.

What we know

Meta has published download figures and, on one occasion, a daily active user figure. The application reached approximately 730,000 downloads in its first five days, roughly 2.5 million in thirteen days, about 3.4 million by 25 September and around five million within three weeks. United States daily active users were reported at approximately 557,000 on 19 September.

The market's reaction was immediate and large. Meta shares gained approximately 27% during September 2026, the best month for the stock since November 2022, and several brokers raised price targets within days. JPMorgan's analyst suggested Muse could become the most widely used consumer AI application since ChatGPT. The share price at the time of writing is roughly $742, against a 52-week low of approximately $520 in March 2026.

What we do not know, and why it matters more

Meta has not published retention. It has not published engagement per user, session length, task completion rates, or revenue. It has not disclosed how many of the five million downloads converted to the paid tiers. The daily active user figure of 557,000, published on a single occasion, implies a ratio of daily active users to cumulative downloads of approximately one in six, which is not obviously a strong retention profile for a product that has been available for less than two weeks at that point, but is also not a number that can be interpreted without a time series.

Three negative data points have also surfaced. Amazon blocked Muse from its shopping site on 20 September, a reminder that agentic commerce runs through platforms that have their own interests. A security researcher reported a vulnerability exposing agent workspaces. And a journalist reported that the agent had read personal messages without an explicit request. None of these is disqualifying. All three are the kind of thing that determines whether an agent product is trusted at scale, and trust is the binding constraint on agent adoption.

The financial consequence, stated precisely

Meta's "Other revenue" line, which would contain subscription income from Muse and the small Meta One subscription business, was $1.01 billion in the June quarter, up from $583 million a year earlier. It is 1.7% of total revenue. Muse launched in September, so none of it is in that number. Even on an optimistic trajectory — say five million paying subscribers at a blended $30 a month, which would be a materially better conversion rate than any comparable consumer AI subscription has achieved — the annual revenue would be approximately $1.8 billion, or 0.7% of current revenue.

The point of that calculation is not to dismiss Muse. It is to establish that Muse cannot pay for the capital programme. A subscription business of any plausible size is immaterial against $137.5 billion of annual capital expenditure. If the capital programme is to earn its cost of capital, the return has to come either from the advertising auction or from selling compute to third parties. Section 14 examines the second possibility. Section 06 examined the first.

10 — Capital intensityThe depreciation is not a forecast; it is arithmetic

Meta's capital expenditure has grown from $28.1 billion in 2023 to $72.2 billion in 2025, and is guided to $130–145 billion in 2026. Consensus for 2027 is approximately $185.6 billion. Expressed as a share of revenue, capital intensity has risen from 20.8% in 2023 to 54.4% at the midpoint of 2026 guidance, and to 62.5% on 2027 consensus revenue.

The company has raised the 2026 range twice during the year, from $115–135 billion in January to $125–145 billion in April to $130–145 billion in July. The April increase was attributed to higher component pricing and, to a lesser extent, additional data centre costs. That is an important detail: a meaningful share of the increase is input-cost inflation rather than a decision to build more, which means the company's own cost base for a fixed quantity of capacity has risen.

The assets being bought, and how they are depreciated

Meta does not disclose the split between data centre shell, power and cooling, networking, and accelerators. It does disclose total depreciation and amortisation, and it discloses that it depreciates servers and network equipment over a shorter life than buildings. The useful-life assumptions are the single most consequential accounting judgement in the company.

The industry convention for AI accelerators is a five-to-six-year depreciable life. Whether that is appropriate is contested: the argument for a shorter life is that successive generations of accelerators deliver large improvements in performance per dollar, so older hardware becomes uneconomic for training well before it physically fails. The argument for a longer life is that inference demand is more forgiving of older silicon than training demand, and that a fleet can be cascaded from training to inference as it ages. Meta's position, stated in its filings, is that it evaluates useful lives annually and adjusts them when warranted. It has not shortened the life of its AI infrastructure in response to the current generation cycle, and it extended the useful life of certain servers in an earlier period, a change that reduced depreciation expense at the time.

We do not think this is a case of aggressive accounting. We do think it is the assumption on which the reported margin most depends, and that it is not independently verifiable from outside the company.

The depreciation path

What is verifiable is the trend. Depreciation and amortisation rose 50% in the first half of 2026, to $12.36 billion, against $8.24 billion in the comparable period of 2025. The FY2025 figure was approximately $18.7 billion. On the disclosed run rate, FY2026 is tracking to approximately $28.4 billion.

For 2027 the arithmetic is straightforward. If Meta spends $137.5 billion on assets in 2026, and if the blended depreciable life is five and a half years, the 2026 cohort adds approximately $25 billion a year of depreciation once fully in service. Adding that to a FY2026 base of $28.4 billion produces an FY2027 charge of roughly $45 billion, before any allowance for the 2027 programme. On our base case of $297 billion of FY2027 revenue, that is 15.2% of revenue in depreciation alone, against 11.2% in FY2026 and 9.3% in FY2025.

11 — Free cash flowThe line that used to justify the multiple, and no longer does

For most of the last decade Meta's investment case rested on a simple proposition: the company converted an unusually large share of its revenue into free cash flow, and it returned essentially all of it to shareholders. That proposition is no longer true, and the change has happened quickly.

Free cash flow was $54.1 billion in FY2024. It fell to $43.6 billion in FY2025 and to $13.2 billion in the first half of 2026. The second quarter alone produced $784 million, a 91% decline year over year. Over the trailing twelve months free cash flow was $37.9 billion, against a market capitalisation of $1.89 trillion — a yield of 2.0%.

The mechanism is not a deterioration in cash generation. Operating cash flow grew 29% in the first half, to $64.1 billion. The mechanism is that capital expenditure grew 66% over the same period, to $50.9 billion. Meta's operating business is generating more cash than it ever has; the company is spending all of it and more on property and equipment.

Where this goes in 2027

Two inputs determine the answer, and both are knowable within a range. Operating cash flow over the trailing twelve months was $130.3 billion, growing 29% year over year. If that growth rate halves in 2027 — a reasonable assumption given that the tax and working capital dynamics that flattered 2026 will not repeat — FY2027 operating cash flow is approximately $160 billion. Capital expenditure on current consensus is $185.6 billion.

The difference is approximately −$26 billion. Meta has never reported a full year of negative free cash flow. On those inputs, 2027 would be the first. To be clear about the uncertainty: if operating cash flow grows 20% rather than 16%, free cash flow is approximately −$4 billion; if capital expenditure comes in at $165 billion rather than $185.6 billion, free cash flow is approximately positive $6 billion. The distribution straddles zero, which is itself the point.

Meta has the balance sheet to absorb this. It held $90.3 billion of cash and marketable securities at 30 June 2026, and it is rated Aa3 by Moody's and AA− by S&P. The question is not solvency. The question is what a business with a 2% free cash flow yield and negative free cash flow on the horizon is worth, and whether the market has yet decided.

12 — Balance sheet & debtFifty-five billion of new borrowing, and the buyback switched off

Meta's balance sheet was, until eighteen months ago, one of the strongest in corporate America for a company of its size: essentially no debt, a large and growing cash pile, and a shareholder return programme that consumed most of the free cash flow. That has changed, and the change is deliberate.

Long-term debt stood at $28.8 billion at the end of 2024. In October 2025 Meta raised $30 billion in a single investment-grade bond sale, the largest in the company's history. In May 2026 it raised a further $25 billion. At 30 June 2026 long-term debt was $83.7 billion, an increase of $54.9 billion in eighteen months. Total liabilities rose 89% year over year to $188.7 billion, and total assets rose 53% to $450.0 billion.

The credit quality remains high. Meta is rated Aa3 by Moody's and AA− by S&P, and its debt-to-equity ratio of 0.72 is modest by the standards of a capital-intensive business. The cost of the borrowing is not the issue. The issue is the direction and the precedent: a company that had no debt two years ago now has $83.7 billion of it, raised specifically to fund infrastructure, and it has guided to capital expenditure that exceeds its operating cash flow.

The capital return has been switched off

The more consequential change is on the other side of the balance sheet. Meta repurchased $29.8 billion of its own stock in 2024 and $26.3 billion in 2025. In the fourth quarter of 2025 and the first half of 2026 it repurchased nothing. The dividend continues, at $0.525 a quarter, costing approximately $5.4 billion a year.

The shift is stark when the two lines are placed side by side. In FY2024 Meta generated $54.1 billion of free cash flow and returned $34.9 billion of it to shareholders. In the first half of 2026 it generated $13.2 billion of free cash flow, returned $2.7 billion, and spent $50.9 billion on property and equipment. Capital expenditure is now roughly nineteen times the capital returned.

A company that stops buying its own stock is telling the market something, whether it intends to or not. In Meta's case the message is unambiguous and, in fairness, explicitly stated: management believes the marginal dollar is better spent on compute than on the company's own shares. That is a strong statement of belief in the investment programme, and it is a statement that can be tested. If the shares are worth materially more than $742, the decision to stop buying them will have been a mistake that costs shareholders real money. If they are not, the decision was correct and the programme is earning its capital.

What the balance sheet can absorb

Meta is not close to a funding constraint. It held $90.3 billion of cash, cash equivalents and marketable securities at 30 June 2026 against $83.7 billion of long-term debt, and it generated $64.1 billion of operating cash flow in the first half. If 2027 free cash flow is negative $26 billion as we model, the company funds the gap from existing cash or from a further bond sale, both of which are available to it on current ratings.

The relevant question is therefore not whether Meta can fund the programme. It can. The relevant question is what the market will pay for a business that has borrowed $55 billion in eighteen months to fund a capital programme larger than its cash flow, and has suspended the return of capital in order to do it. Section 21 addresses that question directly.

13 — Custom siliconThe MTIA programme, and why it matters more than the models

The least discussed and potentially most consequential part of Meta's AI programme is its silicon effort. Meta has designed its own accelerators since 2020 under the MTIA name — Meta Training and Inference Accelerator — and by 2026 is on its fourth generation, which the company refers to internally as Iris. Manufacturing of the current generation is reported to have begun in September 2026, with new iterations planned at roughly six-month intervals through 2027.

Why a custom accelerator changes the economics

Meta buys the majority of its AI compute from NVIDIA, as does every other hyperscaler. The economics of that arrangement are unfavourable to the buyer: merchant accelerators carry a gross margin for the vendor that is, on most estimates, in the region of three quarters of the selling price. A hyperscaler that designs its own silicon and has it fabricated captures that margin, subject to the design, tape-out and yield costs and to the fact that custom silicon is generally less flexible than a merchant part.

The precedent within Meta's peer group is instructive. Google has run TPUs since 2015 and derives a substantial and growing share of its training and inference from them. Amazon designs Trainium and Inferentia and has attributed part of its cloud margin expansion to a higher mix of in-house silicon. Meta's own disclosures are thinner than either: the company has said that MTIA is used for ranking and recommendation workloads, which are inference-heavy and relatively tolerant of specialised hardware, and has been less specific about training.

What we can and cannot conclude

We can conclude that the silicon programme is real, is multi-generation, and is being accelerated. We can conclude that if Meta achieves the mix of in-house silicon that Google has achieved, the effective cost per unit of compute falls materially, which changes the return on the capital programme without changing the headline capital expenditure number.

We cannot quantify any of that. Meta does not disclose what share of its compute is served by MTIA, what the performance gap is against merchant silicon, what the programme costs to develop, or how much of the 2026 capital expenditure is directed at MTIA systems rather than at NVIDIA systems. In a report where almost every other input is a disclosed figure, this is the largest remaining blank. It is also the single most plausible mechanism by which the capital programme turns out to be cheaper than it looks.

14 — Compute as a productThe cloud business nobody has announced

Meta has never sold cloud computing. It is the only one of the five largest technology companies that does not operate a public cloud business, and for most of its history that has been presented as an advantage: no enterprise sales organisation, no capital tied up in capacity sold at commodity margins, and the ability to run its own infrastructure at utilisation levels that a public cloud cannot match.

The capital programme has put that position under pressure, because a company that owns 7 gigawatts of compute in 2026 and 14 gigawatts in 2027 has capacity that its own workloads may not fully absorb. Sell-side analysis has begun to model an external compute business, with estimates of up to $22 billion of annual gross revenue from renting excess capacity from 2027.

The case for, and the case against

The case for is straightforward. Meta's infrastructure was built for internal use, which means the capital is already committed and any external revenue is close to incremental margin at the margin. Meta has a lower cost floor than a merchant-priced competitor if the silicon programme works. And the company has already taken the first step toward developer-facing infrastructure by metering access to Muse Spark through an API, which requires the same billing, identity and support plumbing that a compute rental business would require.

The case against is equally straightforward. Meta has no enterprise sales organisation, no established customer relationships in the cloud market, and no track record of operating multi-tenant infrastructure for third parties. AWS, Azure and Google Cloud have fifteen years of enterprise relationships and, in the case of AWS and Azure, contractually committed backlogs that Meta does not have. Meta would be entering a market at the moment every participant is adding capacity, which is the definition of a market that is about to have a price problem. And a compute rental business is a low-margin business: if Meta prices competitively to win share, the gross margin on $22 billion of revenue may not cover the depreciation on the capacity it uses.

What the commitments say

The clearest evidence that Meta intends to operate at a scale beyond its own needs is the size of its contractual commitments. Press reporting in August 2026 put Meta's future obligations focused on data centres, cloud infrastructure and related buildouts at approximately $700 billion. Meta's own 10-K discloses purchase obligations on a narrower definition; the $700 billion figure is a broader aggregate and should be read as an order of magnitude rather than an audited number.

Even allowing for the imprecision, the comparison is the point. Commitments of that order are roughly 5.4 times FY2025 operating cash flow and 16 times FY2025 free cash flow. A company does not sign commitments of that size to serve its own advertising auction. It signs them because it believes either that the advertising auction will be much larger than it is today, or that it will be selling compute to someone else, or both.

Neither of those is disclosed. That is the disclosure gap at the centre of this equity, and it is the reason our valuation range is as wide as it is. An investor who believes the external compute business is real and reaches $22 billion of revenue at reasonable margins should own the shares. An investor who believes Meta is building for a market that will not materialise should not. Both positions are supportable on the information available.

15 — Competition I: attentionThe user base is not growing, and that is now the constraint

Meta's competitive position in the attention market is simultaneously its greatest strength and the source of its most important structural limit. The strength is scale: 3.60 billion people use at least one Meta application daily, which is roughly 44% of the world's population. The limit is that this number is no longer growing meaningfully, and revenue growth now depends entirely on monetising the same people more.

Family daily active people grew 5% in 2022, 8% in 2023, 5% in 2024, and 7% in 2025. In the June 2026 quarter the growth rate was 3%, the slowest on record. The company attributed a sequential decline in the first quarter of 2026 to internet disruptions in Iran and restrictions on WhatsApp access in Russia, both of which are exogenous and temporary. But the trend line was decelerating before either of those events.

The arithmetic of a mature user base is not inherently negative. Meta's revenue per daily user has risen from approximately $39 in 2022 to approximately $56 in 2025 and, on our estimates, close to $70 in 2026 — a 77% increase against user growth of 22% over the same period. Monetisation, not reach, has been the growth engine for four years, and there is no obvious ceiling on it: Meta's revenue per user in North America is many multiples of its revenue per user in the rest of the world, and the gap has been narrowing.

The constraint is what happens when monetisation growth is the only growth, and the auction price stops rising. That is the scenario our bear case describes, and it is the reason we weight advertising price growth so heavily in the analysis.

TikTok, YouTube and the short-video equilibrium

The competitive threat that defined Meta's 2021–2023 period was short video. TikTok's rise forced Meta to rebuild Instagram around Reels and to insert a very large volume of lower-value video inventory into the feed, which is precisely the dynamic visible in the FY2023 advertising metrics: impressions up 28%, price down 9%. Meta won that engagement contest in the sense that Reels is now a mature product with substantial engagement, but it paid for the win with a two-year period of depressed pricing.

The equilibrium that emerged is a share arrangement rather than a victory. YouTube remains the largest destination for long-form video and is the primary competitor for the advertising budgets that fund it. TikTok's status in the United States has been affected by the divestiture process, which has created uncertainty for advertisers planning multi-year commitments. Meta has benefited from that uncertainty at the margin. None of this is a durable structural advantage; it is a favourable competitive environment that could change.

WhatsApp is the unmonetised asset, and also the regulatory one

WhatsApp is the largest messaging application in the world by users and has historically carried almost no advertising load. Meta began introducing advertising into WhatsApp in the European Union during 2026, and business messaging is the principal commercial route: the WhatsApp Business API, through which businesses pay to communicate with customers at scale.

The strategic importance of WhatsApp for Meta's AI ambitions is larger than its current revenue contribution. It is the distribution channel through which Muse reaches users outside the United States, and the company has explicitly made Muse reachable through WhatsApp. That is also precisely what has attracted the attention of the European Commission, which in June 2026 ordered Meta to restore free access to the WhatsApp Business API for competing general-purpose AI assistants. Section 18 deals with that order in detail.

16 — Competition II: the frontierMeta is not winning the model race, and does not need to

Meta's position in frontier model development is best described as competent and not leading. Muse Spark, in its 1.0 through 1.2 iterations, is a capable model with competitive pricing, a one-million-token context window and agentic tool use. It is not the best model available by any published third-party evaluation, and Meta does not claim otherwise. The company's own positioning has shifted from "we will build the best model" to "we will deliver personal superintelligence to billions of people," which is a statement about distribution rather than about capability.

Why the capability gap may not matter

There is a coherent argument that Meta does not need to lead on capability, and it rests on the structure of the market rather than on optimism. Three points support it.

Frontier capability is converging. The performance gap between the leading model and the fifth-best model has narrowed every year for three years on most public benchmarks, while the cost of serving a token at a given capability level has fallen sharply. In a converging market, the winner is not the best model but the one with the cheapest distribution. Meta has the cheapest distribution in the world.

Meta's monetisation does not require a frontier model. The value of AI to Meta's advertising business comes from ranking and recommendation improvements, which are inference workloads on proprietary data that no competitor has. A model that is 5% worse than the frontier on a general benchmark but trained on Meta's own engagement data will outperform on Meta's own task. The company does not need to win a benchmark to win its own auction.

The capital is committed either way. Meta's 2026 and 2027 capital programmes are large enough to train and serve models at the frontier scale whether or not Meta is currently at the frontier. The programme buys optionality on capability, and the option does not expire.

Why it might matter a great deal

The counter-argument is that the frontier is where the value accrues, and that a company spending $137.5 billion a year to be fifth is spending a great deal for a position that does not command a premium. If agents become the primary interface to the internet, and if consumers choose their agent on the basis of capability, then the agent layer captures the user relationship that Meta currently owns directly. Meta's applications would be reduced to content sources for someone else's agent, which is a structurally worse position than being the destination.

That is not a prediction. It is the scenario in which the capital programme turns out to have been necessary but insufficient — the company spends the money, fails to lead, and finds that the interface has moved to a layer it does not control. It is the reason we have placed meaningful weight on the bear case, and it is the reason Muse matters strategically even though it is financially immaterial today.

17 — Data & the signal problemThe permanent constraint on the advertising model

Meta's advertising business depends on its ability to predict what a person will respond to. That ability depends on the quality of the signals available to the ranking system. For fifteen years the most valuable of those signals have come from third-party sources: tracking pixels on external websites, device identifiers, and cross-app data. Each of those channels has been narrowed, and the narrowing is not finished.

The sequence of restrictions

Apple's App Tracking Transparency framework, introduced in 2021, required applications to obtain explicit consent before tracking users across other applications and websites. The effect on Meta was immediate and is documented in the company's own disclosures: it materially reduced the accuracy of targeting and measurement, and Meta has said it cost approximately $10 billion of revenue in the year following implementation.

European privacy regulation has constrained data combination across services, and the Digital Markets Act has added obligations on consent, on data portability and on the use of data from business users. Meta's response in the European Union has been a "subscription for no ads" model, offering users a paid tier without personalised advertising, which the Commission has challenged. Meta has defended the model publicly and has been explicit that it regards personalised advertising as a legitimate and beneficial service.

More recently, mobile platform changes have continued. Both Apple and Google have changed advertising policies, cookie behaviour and measurement frameworks in ways that reduce the precision available to third-party ad systems. Meta's 10-Q filings list changes by platform providers as a high-severity risk factor, which is an unusually direct disclosure of a dependency that Meta does not control.

Why Meta has been more resilient than expected

The interesting empirical fact is that Meta's advertising business did not break. Revenue fell in 2022 and then grew 16%, 22% and 22% in the following three years, and the price per ad is now rising at 12%. The explanation is that Meta substituted its own signals for the ones it lost.

The company has two sources of first-party signal that no competitor can replicate. The first is on-platform behaviour: what a user watches, clicks, shares, lingers on and skips, across four applications used daily. The second is the enormous volume of advertising outcomes the system observes, which lets it learn what converts without needing to know who the person is in any external sense. The ranking problem became a within-platform problem, and Meta has more within-platform data than anyone.

This is a real and durable advantage, and it is the reason the AI investment has a plausible return. Better models extract more signal from the same first-party data. The price per ad rising 12% while impressions rise 14% is the evidence that it is working.

18 — Regulation I: EuropeBrussels has ordered Meta to open WhatsApp, and the case is not over

The most concrete regulatory action against Meta in 2026 is also the one with the clearest strategic implications. On 9 June 2026 the European Commission adopted interim measures requiring Meta to restore free access to the WhatsApp Business API for competing general-purpose AI assistants, and to maintain that access for the duration of its antitrust investigation.

The sequence of events

In December 2025 the Commission opened an antitrust investigation into Meta's decision to restrict third-party AI providers' access to WhatsApp. On 15 October 2025 Meta had implemented a policy prohibiting third-party general-purpose AI assistants from using the WhatsApp for Business API, effectively reserving access to Meta AI. In February 2026 the Commission issued a Statement of Objections, and in April 2026 a supplementary Statement of Objections setting out its intention to impose interim measures.

Meta revised its policy on 4 March 2026 and permitted third-party assistants back onto WhatsApp, but introduced a fee. The Commission's preliminary assessment, published with the interim measures, was that the fee "appears, at first glance, to replicate the previous access ban." The measures require Meta to restore the terms that were in effect before 15 October 2025 — free access for all AI assistants — within five working days, and to maintain those terms until the Commission reaches a final determination.

The Commission's legal foundation is a finding, on a preliminary basis, that Meta holds a dominant position in the EEA-wide market for consumer communication applications through WhatsApp, a position it has held since at least January 2023, and that the access restriction constitutes a refusal to supply an infrastructure that was previously available to third parties. The substantive investigation continues.

Why this matters to the investment case

The direct financial consequence is small. The revenue Meta could have earned from charging AI assistants for WhatsApp access is not material to a company with $201 billion of revenue.

The indirect consequences are not small, and they run in two directions.

The first is that the Commission has now established, at least provisionally, that WhatsApp is a separately dominant infrastructure asset whose access terms are subject to competition-law control. That is a durable constraint on how Meta can integrate WhatsApp into its AI strategy, and WhatsApp is the principal channel through which Muse reaches users outside the United States. Meta has been explicit that Muse is reachable through WhatsApp. A regulatory regime that requires WhatsApp to treat competing assistants on equal terms limits the extent to which Meta can use its messaging distribution as a proprietary advantage in the agent market.

The second is procedural. Interim measures of this kind are unusual, and the Commission's willingness to impose them — citing the risk that competition "can be lost long before a final decision is adopted" in rapidly evolving markets — indicates a considerably more interventionist posture toward Meta than the company faced during the GDPR era. The Commission's Digital Markets Act enforcement against Meta has already included a €200 million penalty, disclosed in Meta's March 2026 DMA compliance report. The direction of travel is toward more constraint, applied faster.

What we assume

Our base case assumes no material fine from the DMA process beyond the amounts already disclosed, and assumes that Meta complies with the interim measures without a change to its product strategy that would be material to revenue. Our bear case assumes that the Commission's eventual remedy constrains the integration of WhatsApp with Meta AI in a way that weakens the European distribution advantage for the agent business. We do not model a structural remedy. European antitrust cases have historically resolved in fines and behavioural commitments rather than in divestiture, and there is no indication that this one is different.

19 — Regulation II: youth safetyThe largest unquantified liability on the balance sheet

Meta faces litigation over the design of its products for minors that is, in aggregate, the largest legal exposure in the company's history and the least quantifiable. It is also the reason the cost line moved in 2026, and it is likely to move it again.

What has already been decided

In March 2026 a jury in New Mexico's First Judicial District Court in Santa Fe found against Meta on youth-safety claims and imposed a civil penalty of approximately $375 million. In August 2026 the court added a further $567 million for a teen mental health fund, bringing the total to approximately $942 million. Meta is appealing. The case is the first significant judicial finding against a social media company on the theory that product design itself — recommendation algorithms, notification systems, engagement loops — constitutes a harm to minors.

What is still pending

A multi-state action brought by 29 state attorneys general alleges that Meta designed Facebook and Instagram to be addictive to minors, and that the company chose growth over child welfare. The theoretical damages exposure in that case has been described in press reporting as extending as high as $1.4 trillion, a figure that should be treated as a rhetorical upper bound rather than an estimate. More relevant is the range of plausible outcomes: New Mexico state authorities have sought a penalty in the region of $35–40 billion in the privacy component of that state's case, against Meta's proposal of a $3.45 billion cap. A ruling was expected in October 2026.

The accounting treatment, and what it tells you

Meta booked $2.40 billion of charges related to legal proceedings in the second quarter of 2026. It has stated that it expects to book approximately $10 billion in the third quarter. It raised its full-year 2026 total expense guidance to $165–169 billion specifically to accommodate these charges. The chief financial officer said on the July 2026 earnings call that ongoing legal matters could lead to a "significant financial impact."

Two inferences are available from that treatment. The first is that management regards the charges as probable and estimable enough to accrue, which is a higher standard than disclosure alone. The second is that the $10 billion third-quarter figure is an accrual for matters that have been or are about to be decided, not a provision for the full multi-state action. The company is therefore telling the market that it expects to spend roughly $12.4 billion on legal matters in 2026 and has not told the market what it expects beyond that.

20 — Cost structure & headcountFewer people, more machines, and a much larger bill

Meta's cost structure is being rebuilt in a way that is unusual to observe in a company of its size. Headcount is falling while total costs are rising at 55%. The two facts are not in tension; they are the same fact, because the company is substituting capital for labour, and the capital is more expensive per unit of output than the labour it replaces.

Headcount was 78,865 at the end of 2025, 77,986 at the end of the first quarter of 2026 and 75,472 at 30 June 2026. The May 2026 reduction affected approximately 8,000 employees, the majority of whom were still on the payroll at the end of the second quarter; the company said most would no longer be reflected in headcount by the end of the third quarter. Revenue per employee has risen from $1.35 million in 2022 to $2.55 million in 2025, which is among the highest in the technology industry.

The company has now run two significant restructurings in four years: the 2023 reduction from 86,482 to 67,317 employees, and the 2026 reduction. Each was followed by a period of re-acceleration in hiring, and the 2026 round comes alongside a very large increase in compensation for AI researchers. The net effect is a workforce that is smaller in number, more expensive per head, and more concentrated in research functions than at any point in the company's history.

What the cost lines say

The composition of the second-quarter increase is the clearest available statement of where the money is going. Research and development rose from $12.9 billion to $21.7 billion, an increase of 67% and by far the largest contributor to the total. Cost of revenue rose 33%, driven by the cost of delivering more impressions and by the depreciation on the servers that deliver them. Marketing and sales rose 15%, below revenue growth, which is a genuine efficiency: Meta is not buying growth. General and administrative rose 111%, driven almost entirely by the $2.40 billion legal charge.

Over the full year, on company guidance of $165–169 billion of total costs and expenses, Meta's cost base grows approximately 42% in 2026 against revenue growth of approximately 26%. That is the arithmetic of a company deliberately compressing its margin in order to build capacity, and it is not controversial. What is worth stating is that the compression is now large: 41.4% operating margin in FY2025 against a company-guided operating income for FY2026 that implies approximately 33–34%.

21 — ValuationA normal multiple applied to an abnormal capital cycle

Meta at $741.90 presents an unusual valuation problem, because the earnings multiple and the cash flow multiple now tell different stories, and the difference between them is the entire investment debate.

On earnings, the shares are close to their own history. Meta trades at 28.0× trailing GAAP earnings of $26.50 per share, and at approximately 23.7× our estimate of FY2026 GAAP earnings of $31.30. Its five-year average trailing multiple is approximately 26.5 times. For a company growing revenue at 26% with an operating margin above 30%, a multiple in the low-to-mid twenties on forward earnings is not demanding.

On cash flow, the shares are close to their most expensive ever. The trailing free cash flow yield is 2.0%, against 3.6% at the end of 2024, 4.8% at the end of 2023 and 6.1% at the end of 2022. On our FY2026 estimate, free cash flow is approximately $0.5 billion, which makes the yield meaningless for that year. On 2027 consensus capital expenditure, free cash flow is negative.

The two readings are not contradictory; they measure different things. The earnings multiple measures the profitability of the advertising business. The cash flow yield measures what is left of that profitability after the infrastructure programme has been paid for. The gap between them is the capital programme, and the market's judgement about whether it will earn a return.

How we build the estimate

Our FY2026 estimate is anchored on disclosed figures. Revenue is the sum of reported first-half revenue of $117.1 billion, the midpoint of third-quarter guidance of $62.5 billion, and a fourth-quarter estimate of $73.0 billion, which is 22% growth on a $59.9 billion base. Total costs and expenses are the midpoint of company guidance at $167.0 billion. Operating income is therefore approximately $85.6 billion, an increase of 2.8% over FY2025. Other income of $1.5 billion, a 16% tax rate and 2,560 million diluted shares produce net income of approximately $80.2 billion and EPS of $31.30.

Our FY2027 estimate is a model, not a derivation. Revenue grows 17.6% to $297 billion, which is a deceleration from 26% and assumes advertising price growth moderates to high single digits. Total costs grow 17.4% to $196 billion, which embeds approximately $17 billion of incremental depreciation from the 2026 capital programme and continued growth in research and development at roughly half the 2026 rate. Operating income is $101 billion, a 34.0% margin. With $2.0 billion of other income and a 17% tax rate, net income is $85.5 billion and EPS is $33.50.

FY2025 is as reported. FY2026E revenue is reported first-half revenue plus the midpoint of third-quarter guidance plus a Farstar fourth-quarter estimate; FY2026E costs are the midpoint of company guidance. FY2027E is a Farstar model. The FY2025 tax rate is the reported 30%, which includes the $15.93 billion valuation allowance charge; absent that charge the rate was 13%, which is why the FY2026E rate of 16% is not comparable to the FY2025 figure. Diluted share count is held at approximately 2,550 million and assumes no repurchases.

Scenario valuation

The table below sets out our three cases in full. Every input is stated so that a reader can disagree with a specific assumption rather than with the conclusion.

Weighting the three cases at 25/50/25 produces a scenario value of $776 per share. We set our 12-month target at $775, a discount of less than 1% to the weighted value, reflecting the width of the outcome distribution rather than a view about time. That is approximately 4.5% above the current price and maps to a Neutral rating.

22 — What the price requiresAt the average multiple, the market is assuming no growth at all

The most useful discipline in valuing a company like Meta is to invert the question. Rather than asking what the shares are worth, ask what they already assume. Because Meta's earnings and its multiple are both observable, the implied operating income at any assumed exit multiple can be solved exactly.

At the five-year average trailing multiple of 26.5 times, a share price of $741.90 implies FY2027 earnings per share of $28.00, net income of approximately $71.3 billion, and operating income of approximately $83.9 billion. Our FY2026 estimate is $85.6 billion. In other words, at its own historical average multiple the market is pricing Meta for a decline in operating income in 2027.

At our base-case multiple of 23.5 times, the implied FY2027 operating income is $94.9 billion, which is 11% above our FY2026 estimate and 6% below our base case of $101 billion. At a 20 times multiple the implied figure is $112 billion, which requires operating income growth of 31% — a demanding outcome that would require the capital programme to be visibly earning a return.

What that tells us

The reading is more constructive than the Neutral rating might suggest, and we want to be precise about it. The market is not pricing Meta for an AI triumph. It is pricing the company for roughly flat-to-modest growth in operating income, at a multiple below its own history, against a business that grew revenue 28% last quarter.

That is not an aggressive price. It is a price that reflects a specific and reasonable fear: that the depreciation from the capital programme will absorb the operating leverage that revenue growth would otherwise produce. If that fear is wrong, the shares are cheap. If it is right, the shares are roughly fair, because the multiple has already de-rated to reflect it.

The asymmetry that follows is worth stating plainly. On our numbers, the downside case ($437, a 41% decline) is larger in magnitude than the base-case upside ($787, a 6% gain), because the bear case combines an earnings miss with a multiple de-rating and the base case combines only modest earnings growth with a stable multiple. That asymmetry, rather than a negative view of the business, is what produces the Neutral rating. An investor who believes the probability of the bear case is materially below 25% should own the shares.

23 — The bull caseAn auction that is improving, a monopoly on attention, and a capital programme that has a return

The bull case for Meta does not depend on a new product. It depends on the existing product continuing to do what it did last quarter, and on the capital programme being the mechanism that makes that possible rather than a cost that prevents it. Stated in full, it runs as follows.

1. The auction is improving, and the improvement is measurable

Advertising revenue grew 27% in the June quarter on impressions up 14% and price up 12%. The price is the variable that matters, because it is the market's own estimate of the value of a Meta impression, and it has now risen at a double-digit rate for two consecutive quarters after running at single digits in the fourth quarter of 2025. Meta attributes the improvement to ranking and creative systems built on its own models. The mechanism is credible: better prediction raises conversion, conversion raises bids, bids raise price. And the improvement is happening on a user base that is not growing, which means it is coming from the algorithm rather than from reach.

2. The AI spending has a return path that does not require a new business

This is the most underappreciated part of the bull case. Meta does not need Muse to succeed, and it does not need to sell compute, for the capital programme to pay off. It needs the models to keep improving the auction. Every percentage point of price growth on $196 billion of annual advertising revenue is worth approximately $2 billion of revenue at close to 100% incremental margin, because the cost of serving an impression does not change when the price of the impression rises. The capital programme needs to generate roughly six percentage points of incremental price growth to cover its own depreciation in the first full year. It delivered twelve in the first half.

3. Distribution is the scarce asset, and Meta owns more of it than anyone

3.60 billion daily users across four applications, with the deepest engagement data in the industry and the lowest customer acquisition cost of any consumer technology company in history. In a market where model capability is converging, distribution decides who captures the value. Meta has more distribution than any competitor, and it has demonstrated across four platform transitions that it can convert distribution into revenue. Muse reached five million downloads in three weeks with essentially no paid acquisition, which is a demonstration of that distribution advantage in the current cycle.

4. The custom silicon programme is a real cost advantage in the making

Meta is on its fourth generation of custom accelerator, with production of the current generation reported to have begun in September 2026 and further iterations planned at six-month intervals. If Meta achieves even a fraction of the in-house silicon mix that Google has achieved with TPUs, its effective cost per unit of compute falls materially, which changes the return on the capital programme without changing the headline number. The company has the volume, the incentive and the balance sheet to pursue this.

5. The balance sheet is strong enough to make the programme a choice rather than a necessity

$90.3 billion of cash and marketable securities, an Aa3 / AA− credit rating, and $64.1 billion of first-half operating cash flow. Meta is not funding this programme from a position of weakness. It is funding it from a position of strength, and if the returns do not materialise it can slow the programme at any point without impairing the existing business, which requires almost no capital.

6. The multiple is below the company's own history

At 23.7 times our FY2026 estimate and 22.1 times our FY2027 base case, Meta trades below its five-year average trailing multiple of approximately 26.5 times. For a business growing revenue at 26% with an operating margin above 30%, that is not an expensive price. And as section 22 shows, the market's implied FY2027 operating income at the average multiple is below our FY2026 estimate, which means the price already embeds the fear that is the core of the bear case.

24 — The bear caseA capital cycle larger than the cash flow that funds it, entered at a full multiple

The bear case is not that Meta's advertising business is impaired. It is that the capital programme is disproportionate to any return it can plausibly generate, that the cost of it arrives mechanically through depreciation, and that the market has not yet repriced for the change in the company's financial character. Stated in full, it runs as follows.

1. Capital expenditure now exceeds operating cash flow

Meta will spend $130–145 billion in 2026 against trailing operating cash flow of $130.3 billion, and consensus expects $185.6 billion in 2027 against our estimate of $160 billion of operating cash flow. On those figures free cash flow is negative in 2027 for the first time in the company's history. A company that has to borrow to fund its capital programme, while suspending its buyback, has changed category. It is no longer a cash-returning business with an infrastructure habit; it is an infrastructure business with a cash-generating division.

2. The depreciation arrives whether or not the demand does

Depreciation and amortisation rose 50% in the first half of 2026 and, on our estimates, rises from approximately $18.7 billion in FY2025 to $45 billion in FY2027. That is a $26 billion increase in a cost line that does not vary with revenue. To hold the operating margin flat, Meta needs approximately $118 billion of incremental revenue in 2027, which is a 47% increase. That will not happen. Margin compression is therefore not a risk in the bear case; it is the arithmetic baseline, and the only question is how much.

3. There is no disclosed revenue against which to test the spend

Meta has not disclosed the share of its compute consumed internally, the revenue from external compute customers, or the engagement economics of Muse. The company has committed approximately $700 billion of future obligations, on press reporting, against a business that generates $115.8 billion of operating cash flow a year. An investor is being asked to fund the largest capital programme in corporate history on the strength of a narrative and a price improvement in the existing business. That may be enough. It is not disclosure.

4. The legal exposure is large, unquantified, and now recurring

Meta booked $2.40 billion of legal charges in the second quarter and expects approximately $10 billion in the third. A jury has already found against the company in New Mexico, with a combined $942 million penalty that is under appeal. Twenty-nine state attorneys general are plaintiffs in an action whose theoretical exposure has been described in the hundreds of billions. The company's own chief financial officer has said the matters could lead to a "significant financial impact." This is a liability that has moved from contingent to probable during 2026, and its ultimate size is not knowable.

5. User growth has stopped, so everything depends on price

Family daily active people grew 3% in the June quarter, the slowest on record. Revenue growth is now entirely a function of monetisation, and monetisation growth in the last two quarters has been driven by price. If price growth normalises to mid-single digits — which is where it sat as recently as the fourth quarter of 2025 — revenue growth falls to low-to-mid teens, the operating margin compresses further, and the earnings on which the multiple is applied disappoint. A single quarter of 5% price growth would be the most informative data point the company could publish.

6. The agent layer could disintermediate the applications

If consumers adopt AI agents as their primary interface to the internet, and if they choose those agents on capability rather than on distribution, then the value accrues to the agent layer rather than to the content destinations beneath it. Meta's applications would become sources for someone else's interface. This is the scenario in which Meta spends $137.5 billion a year, fails to lead on capability, and finds that the relationship with the user has moved to a layer it does not control. It is a low-probability scenario today and it is the only scenario in which the capital programme is fatal rather than merely expensive.

25 — Risk matrixTwelve risks, scored by probability and impact

The register below scores each risk on a one-to-five scale for probability and impact, with a product score. The heat map plots them by cell. Scores are our judgement, stated so that a reader can disagree with a specific rating.

Cells read left to right as impact rises from 1 to 5, top to bottom as probability rises from 1 to 5. Superscripts identify risks from the register above. R11 and R12 are not shown; they fall below the plotted range.

26 — CatalystsWhat we are watching, and when

27 — ConclusionNeutral at $775, on a business that is performing and a price that has not decided

Meta in October 2026 is two companies sharing one ticker. The first is the advertising franchise: 3.6 billion daily users, a ranking system that raised the clearing price of an impression 12% year over year, a 31% operating margin on $60.8 billion of quarterly revenue, and almost no capital requirement. That business is performing better than at any point in its history.

The second is an infrastructure programme: $137.5 billion of 2026 capital expenditure, 7 gigawatts of capacity rising to 14, approximately $700 billion of commitments on press reporting, $83.7 billion of debt raised in eighteen months, a suspended buyback, and a product — the Muse agent — that has five million downloads, no published retention and 1.7% of revenue sitting in the line that would contain it. That programme is unproven, and its cost is arriving through depreciation whether or not it is.

The equity price reflects a judgement about the second, and the judgement is more modest than the debate suggests. At the five-year average multiple of 26.5 times, $741.90 requires FY2027 operating income of approximately $84 billion — below our FY2026 estimate. At our base multiple of 23.5 times it requires $95 billion, an 11% increase. Neither is a heroic assumption. The market is not paying for Meta's AI strategy to succeed; it is paying for it not to fail.

Weighting our three cases at 25/50/25 gives a scenario value of $776 per share, and we set our 12-month target at $775, approximately 4.5% above the current price. That maps to a Neutral rating. We want to be clear about what that means and does not mean.

It does not mean we think the business is impaired. The advertising franchise is one of the best assets in public markets, and the improvement in the auction is real, measurable, and already in the numbers. It does not mean we think the AI investment is a mistake; on the contrary, the evidence that it is improving the auction is the strongest part of the record.

It means that at $741.90 the market has already paid for the outcome we consider most likely, that the distribution of outcomes is unusually wide because three of the four variables that decide it are not disclosed, and that the downside case is a genuine one rather than a formality. The bear case is not a story about a business losing customers. It is a story about a business that spends $185 billion in a year, depreciates it over five, and finds that the revenue did not arrive in time.

28 — Appendix: financialsReference tables

A. Consolidated income statement

FY2022–FY2025 totals are as reported. The cost-line split for FY2024 and FY2025 is derived from the reported totals and the disclosed quarterly breakdowns and may differ from the filings by small amounts; FY2025 general and administrative is the line most likely to differ, since the company does not break out the components within it. FY2026E and FY2027E are Farstar estimates. The FY2026E and FY2027E general and administrative lines each contain the assumed legal charge of $12.4 billion and $4.0 billion respectively, which is why the line falls year over year.

B. Cash flow

Capital expenditure in section 10 and the charts includes principal payments on finance leases, which is the basis Meta reports; this table separates the two components. FY2026E and FY2027E debt levels are Farstar estimates consistent with funding a negative free cash flow position.

C. Operating metrics

FY2022 impression and price growth figures are as reported. Revenue per daily user is a Farstar construction, not a company disclosure; see the chart note in section 15.

D. Valuation summary

29 — SourcesPrimary and secondary references

Company filings and releases

• Meta Platforms, Inc., Form 10-K for the fiscal year ended 31 December 2025, filed with the SEC.
• Meta Platforms, Inc., Form 10-Q for the quarter ended 30 June 2026.
• Meta Platforms, Inc., Form 10-Q for the quarter ended 31 March 2026.
• Meta Platforms, Inc., Second Quarter 2026 Results, 29 July 2026.
• Meta Platforms, Inc., First Quarter 2026 Results, 29 April 2026.
• Meta Platforms, Inc., Fourth Quarter and Full Year 2025 Results, 28 January 2026.
• Meta Platforms, Inc., Third Quarter 2025 Results, October 2025, including the $15.93 billion valuation allowance charge.
• Meta Platforms, Inc., quarterly earnings call transcripts and prepared remarks, Q4 2025 through Q2 2026.
• Meta Platforms, Inc., Third Digital Markets Act compliance report, filed 6 March 2026.
• Meta Newsroom and Meta AI product announcements, including the Muse Spark model releases and the Muse agent launch, April–September 2026.

Third-party research and industry data

• Press reporting on the $30 billion October 2025 and $25 billion May 2026 Meta bond offerings, including PitchBook, 31 October 2025.
• Press reporting on Meta’s future contractual obligations and the $700 billion aggregate, August 2026.
• Press reporting on the Meta Superintelligence Labs reorganisation, the Scale AI investment and the departure of Yann LeCun, 2025–2026.
• Press reporting on the Muse agent launch, download and daily active user figures, and the Amazon, security and privacy incidents, September 2026.
• CNBC, Meta’s Reality Labs lost over $4.6 billion in second quarter, 29 July 2026.
• Consensus capital expenditure and earnings estimates, and analyst price targets, as compiled by third-party market data aggregators, September–October 2026.
• Goldman Sachs and other sell-side research on hyperscaler capital expenditure through 2027, as reported September 2026.

Regulatory and legal

• European Commission, Commission implements interim measures to ensure free access to WhatsApp for competing AI assistants, IP/26/1276, 9 June 2026.
• European Commission, Statement of Objections to Meta, February 2026, and supplementary Statement of Objections, April 2026.
• New Mexico Office of the Attorney General, Court orders Meta to pay $942 million and overhaul protections for children, 7 August 2026.
• New Mexico First Judicial District Court, jury verdict of March 2026 imposing a $375 million civil penalty.
• Multi-state action brought by 29 state attorneys general concerning the design of Facebook and Instagram for minors, pending.
• US Treasury Notice 2026-7 on corporate alternative minimum tax treatment of capitalised US research and development costs.
• One Big Beautiful Bill Act, enacted 2025, and the resulting non-cash valuation allowance charge.

Market data

• Share price of $741.90, share count of approximately 2,548 million and market capitalisation of approximately $1.89 trillion as of 6 October 2026.
• Consensus price targets and coverage counts as compiled by third-party market data aggregators, October 2026.
• Five-year average trailing price-to-earnings multiple as compiled by third-party market data providers.

30 — DisclosureConflicts, limitations, and revision policy