Meta Platforms
The owner of Facebook, Instagram and WhatsApp: it gives the apps away and sells advertisers the attention of 3.6 billion people a day.
The scores
Growth4/4 ↑
Revenue grew 27.7% over twelve months against a peer median of 18.35%, with advert volume up 14% and price per advert up 12% together.
Business quality3/4 →
The highest operating margin in the cohort, but a court found after trial that users treat rival apps as substitutes for its own.
Financial position2/4 ↓
Net cash about $1.5 billion, free cash flow negative, $454.5 billion of signed commitments, and about $60 billion of planned equity not yet raised.
Execution record3/4 →
Revenue guidance has been met and the ranking work produces measured gains, but earnings missed consensus in two of the last four quarters.
Per share capture2/4 ↓
About $95 billion of new shares and convertible stock is planned, roughly 5% of market value, and the buyback has gone to nil.
Potential3/4 →
The format it sells grew 32.6% while search grew 11%, but no world total exists to size the position and no new line is quantified.
Risk profile3/4 ↓
One revenue source, one regulatory theme with trials the company says may cost materially, and a capital programme with no 2027 figure.
Valuation3/4 ↓
At 23.1 times operating profit and 27.9 times earnings, above its own six-year normal, the price now pays for growth, though less than rivals get.
Evidence quality4/4 →
Filed and vendor figures agree exactly, a federal court supplies the moat evidence under adversarial test, and a trade survey measures the market.
If it works 1.5× · if it fails you keep about 100%.
The Verdict
We rate Meta attractive. It gives away Facebook, Instagram and WhatsApp and sells advertisers the attention of billions of people, and the business is solid: it is showing more adverts and charging more for each at the same time. The price is fair, not cheap: on profit it is still the cheapest of its rivals, but on the cash left after its data-centre spending it is among the dearest. As a bet it is safe, with little upside: the advertising business alone covers today's price, so the glasses and new AI products are free, but even both succeeding adds little. The view hinges on the data-centre spending earning a return.
The long caseIf glasses that see and hear for you become an everyday purchase, the company already selling most of them through the biggest eyewear maker sells the next device.
The evidence under this - 9 claims
The chief financial officer attributes part of the 12% price increase to 'improvements in macro conditions relative to Q2 of last year and currency tailwinds' rather than to performance alone, so not all of it is pricing power. The offsetting reading is the combination: raising price 12% while raising volume 14% is not something a business without pricing power can do in the same quarter, whatever share of the price move is cyclical, and she notes the price gain was partially offset by strong impression growth from lower-monetising surfaces - meaning mix worked against the reported figure rather than for it.
- 2026-10-21editedQ3 2026 results: advertising +26%, impressions +10%, price per ad +14%.
Free cash flow of $784 million on operating cash flow of $31.86 billion means capital spending consumed 97.5% of the cash the business produced in the quarter. The offsetting argument is that this is an accounting distinction rather than an economic one: the spending buys servers and data centres with multi-year lives, and a company that expensed the same amount would show identical cash flow and far lower profit. Which reading is right is precisely what the thesis is about, and neither the filings nor the call resolve it - the company has attached no revenue to the spending.
- 2026-10-21editedQ3 2026 results: free cash flow turned negative at $2.12B on $34.91B of capital spending.
A capital budget roughly doubling year over year with no 2027 figure offered is the single largest uncertainty in this file, and the refusal is not evasion - the company says planning is dynamic and that it is 'gearing our current infrastructure plans toward maximizing capacity in 2026 and 2027'. Two things cut the other way. The company still guides operating income above last year's while spending this, which bounds the damage to the income statement. And it states that 'the industry has historically underbuilt for the wave of AI adoption, making existing capacity, including our own, extremely valuable' - a claim about scarcity that, if true, makes the spending an option rather than a sunk cost.
- 2026-10-21editedQ3 2026 release: capex range narrowed to $135-145B, expenses raised for the charge, Q4 revenue guided $73-76B, no 2027 capex number.
- 2026-10-22editedQ3 2026 10-Q: a 2027 figure exists even though no 2027 capital spending number was guided.
The two multiples describe the same company and cannot both be right about it, which is the thesis. The case against reading the profit multiples as cheap is that operating profit is stated after depreciation on an asset base that is roughly doubling, so today's depreciation charge understates tomorrow's - meaning the 23.1 times figure is struck on a profit number the capital programme will itself reduce. That is the strongest argument against this thesis and nothing in the filings refutes it; what bounds it is the company's own guidance that 2026 operating income will still exceed 2025's. The second argument is new: no quarter has been reported since August, so the whole move from 17.5 to 23.1 times is the market paying more for the same profit, and the discount the thesis was written to exploit is now smaller than it was.
- 2026-09-27editedRe-struck at the 2026-09-25 close of $751.66 from the refreshed research pack, 34.5% above the $559.02 it was written at; Meta is no longer the dearest of the cohort on free cash flow.
Amazon moved the opposite way on the same asset class in the same month, shortening a subset of servers and networking equipment from six years to five effective January 1, 2025 and attributing it to 'the increased pace of technology development, particularly in the area of artificial intelligence and machine learning'. Both are the companies' own estimates and both are defensible, but they cannot both describe the same hardware: one firm judged AI infrastructure to be lasting longer at exactly the moment the other judged it to be obsoleting faster, and the lengthening is the one that lowers near-term depreciation.
The absence is the finding and it cuts both ways. Against the company: a $130-145 billion capital programme with no quantified revenue attached is the reason the market is refusing to capitalise it, and this thesis cannot argue that the market is wrong on evidence the company has not produced. For the company: the largest single use of the compute is improving the advertising business, and that return is already visible in the disclosed 8.3% increase in ad clicks and 15.7% uplift in conversions on Facebook - so 'no revenue attached' understates it, because some of the return arrives as advertising revenue rather than as a new line.
- 2026-10-21editedQ3 2026 release still attaches no revenue to the spending.
EssilorLuxottica gives no revenue for the line and says the glasses weighed on its own margin; units are not Meta's revenue, and Reality Labs' 2025 sales rose only 3% as Quest headsets fell (C37).
- Meta discloses no revenue for the assistant, the agents, the API or Meta One, so the figure has nothing filed under it (C10).
- Meta's own 2024 forecast, unsealed in court, put generative AI product revenue at $2B to $3B in 2025, and none has been disclosed since: its own forecasts have run ahead of what it reports.
- Gemini and ChatGPT already hold the assistant users Meta would charge (C14).
The company
Meta runs Facebook, Instagram, WhatsApp, Messenger and Threads, used by 3.60 billion people a day. It charges users nothing. Advertisers pay to put a message in front of a chosen audience, and increasingly pay for a result, such as an install or a purchase, rather than for being seen.
Advertising is 99.3% of revenue. The rest is Reality Labs, which sells headsets and AI glasses at a large loss, paid for by the apps.
The evidence under this - 3 claims
User growth has halved: 7% in December 2025 against 3% in June 2026. At 3.6 billion daily users out of roughly 5.5 billion people online, the measure is approaching saturation by arithmetic, so deceleration is expected rather than alarming - but it removes one of the two levers on revenue and leaves the whole growth case resting on monetisation per user. It also means the erosion test in the moat block matters more than it would for a company still adding users.
- 2026-10-21editedQ3 2026 release: users grew 3% to 3.64 billion.
How it makes money
Every advertisement is sold by auction, in real time, so no advertiser is committed beyond the next one. Meta earns more by showing more advertisements and by charging more for each, and it did both in the June quarter.
The evidence under this - 1 claim
The chief financial officer attributes part of the 12% price increase to 'improvements in macro conditions relative to Q2 of last year and currency tailwinds' rather than to performance alone, so not all of it is pricing power. The offsetting reading is the combination: raising price 12% while raising volume 14% is not something a business without pricing power can do in the same quarter, whatever share of the price move is cyclical, and she notes the price gain was partially offset by strong impression growth from lower-monetising surfaces - meaning mix worked against the reported figure rather than for it.
- 2026-10-21editedQ3 2026 results: advertising +26%, impressions +10%, price per ad +14%.
Where the money comes from, stream by stream
| Stream | Share of sales (Q2 2026) | How it earns it |
|---|---|---|
| Advertising | 97.6% | Auctioned ads in Facebook, Instagram, Messenger, WhatsApp and Threads feeds, stories and Reels, priced per impression and click. |
| Family of Apps other revenue | 1.7% | WhatsApp business messaging fees, Meta Verified and other subscriptions; up 73% from a small base. |
| Reality Labs | 0.7% | Quest headsets and Ray-Ban and Oakley AI glasses sold to consumers, at a large operating loss. |
Who runs it
Its founder, Zuckerberg, started Meta in 2004 and still runs it. Susan Li has been finance chief since 2022, and a new president, Dina Powell McCormick, joined in January 2026. Zuckerberg owns about an eighth of the company but, through shares with ten votes each, controls most of the vote, so outside shareholders cannot overrule him. He takes a $1 salary. The other officers' bonuses are set by the board's judgement, not a formula, and their shares vest for staying, not for results.
The evidence under this - 3 claims
- 2026-10-21editedThe Class C recapitalization bears on founder control and on how the build may be funded.
Nothing in the plan pays on revenue, margin, return on capital or the share price as a formula, so the record of the capital programme costs the officers nothing except through the value of the shares they hold; the new President was hired in January 2026 on $60 million of service-vested units (S20).
Why it wins, and who could beat it
Why customers come back
Scale nobody can assemble twice. More users produce more measured sales, more measured sales train a better prediction of who will buy, and better predictions return more for each advertising dollar. Advertisers have no loyalty; they follow the return.
Users can leave, and a federal court said so. What an advertiser cannot find elsewhere is this much measured attention in one place. The proof is price: Meta charged 12% more per advertisement while earning the highest operating margin of its rivals, 38.1%.
The evidence under this - 4 claims
The chief financial officer attributes part of the 12% price increase to 'improvements in macro conditions relative to Q2 of last year and currency tailwinds' rather than to performance alone, so not all of it is pricing power. The offsetting reading is the combination: raising price 12% while raising volume 14% is not something a business without pricing power can do in the same quarter, whatever share of the price move is cyclical, and she notes the price gain was partially offset by strong impression growth from lower-monetising surfaces - meaning mix worked against the reported figure rather than for it.
- 2026-10-21editedQ3 2026 results: advertising +26%, impressions +10%, price per ad +14%.
The Federal Trade Commission has appealed, so the holding is not final. And the ruling is legally good news for Meta - it ends a case seeking to unwind the Instagram and WhatsApp acquisitions. The reason it appears here as counter-evidence to the moat rather than as support is that Meta won by proving its own attention is contestable: the company argued, and a court accepted after hearing myriad witnesses and duelling experts, that users treat TikTok and YouTube as substitutes. That is the strongest independent evidence available on the durability question, it was produced under adversarial conditions, and it says the moat is narrower than a 3.6 billion user count implies.
Meta is not the best business in its own cohort on three of the four operating measures - Reddit earns a higher return on invested capital, a higher gross margin and much faster growth. The scale difference is why the comparison still favours Meta and it has to be stated rather than assumed: Reddit produces those ratios on a tiny fraction of the revenue, where one large advertiser moving budget moves the percentage. Against that, a cohort in which the subject loses three of four rankings is weaker evidence for a quality score than one in which it wins them, which is why business quality is scored 3 rather than 4. The one it wins - operating margin - is also the one least sensitive to scale.
- 2026-09-27editedRe-read from the refreshed research pack; the only figure that moved is the cohort median return on invested capital, 10.25% to 10.3%.
User growth has halved: 7% in December 2025 against 3% in June 2026. At 3.6 billion daily users out of roughly 5.5 billion people online, the measure is approaching saturation by arithmetic, so deceleration is expected rather than alarming - but it removes one of the two levers on revenue and leaves the whole growth case resting on monetisation per user. It also means the erosion test in the moat block matters more than it would for a company still adding users.
- 2026-10-21editedQ3 2026 release: users grew 3% to 3.64 billion.
The threat
TikTok and YouTube. In 2025 a federal judge ruled that they compete for the same hours Facebook and Instagram sell. Alphabet, YouTube's owner, is the only listed company selling advertising at Meta's scale, with $294.7B of Google advertising revenue against Meta's $201.0B.
The evidence under this - 2 claims
The Federal Trade Commission has appealed, so the holding is not final. And the ruling is legally good news for Meta - it ends a case seeking to unwind the Instagram and WhatsApp acquisitions. The reason it appears here as counter-evidence to the moat rather than as support is that Meta won by proving its own attention is contestable: the company argued, and a court accepted after hearing myriad witnesses and duelling experts, that users treat TikTok and YouTube as substitutes. That is the strongest independent evidence available on the durability question, it was produced under adversarial conditions, and it says the moat is narrower than a 3.6 billion user count implies.
The rivals: GOOGL, GOOGL, SNAP, RDDT, AMZN, PINS, and 2 more that are private, in-house or listed abroad.
Who else does this, rival by rival
| Ticker | Rival | Where it competes | How far along | Priced against it |
|---|---|---|---|---|
| Private | TikTok, operated by ByteDancePriced against it: Private | Family of Apps | On sale | Private |
| GOOGL | YouTube, operated by AlphabetPriced against it: Yes | Family of Apps | On sale | Yes |
| Google, through Gemini | Reality Labs and Meta AI beyond advertising | On sale | ||
| Private | OpenAI, through ChatGPTPriced against it: Private | Family of Apps | On sale | Private |
| SNAP | Snap, through SpectaclesPriced against it: No usable figure | Reality Labs and Meta AI beyond advertising | Announced | No usable figure |
| RDDT | RedditPriced against it: Yes | - | On sale | Yes |
| AMZN | Amazon.comPriced against it: Shown, not counted | - | On sale | Shown, not counted |
| PINS | PinterestPriced against it: No usable figure | - | On sale | No usable figure |
How big it could get
How big this could get
Americans spent a record $294.6B on internet advertising in 2025, up 13.9%. Social media, the kind Meta sells, grew fastest of the large formats: 32.6%, to $117.7B, overtaking search.
Meta's own advertising, $59.4B in the June 2026 quarter, is booked worldwide, so its share of this American market cannot be read off.
The evidence under this - 2 claims
The survey covers United States revenue only, which is the reason the prize does not compose and is a real limit on what this figure can support - Meta reports advertising revenue globally and roughly half of it is earned outside the United States. The figures are also self-reported by participating publishers and platforms to a trade association, not audited, so they are an industry measure rather than a statistical one. What they do establish independently of Meta is the relative growth of formats, which is the part the driver block uses.
The chief financial officer attributes part of the 12% price increase to 'improvements in macro conditions relative to Q2 of last year and currency tailwinds' rather than to performance alone, so not all of it is pricing power. The offsetting reading is the combination: raising price 12% while raising volume 14% is not something a business without pricing power can do in the same quarter, whatever share of the price move is cyclical, and she notes the price gain was partially offset by strong impression growth from lower-monetising surfaces - meaning mix worked against the reported figure rather than for it.
- 2026-10-21editedQ3 2026 results: advertising +26%, impressions +10%, price per ad +14%.
Where the growth is coming from, business by business
Where the growth is coming from
| Business | Share of revenue | Growth: last quarter | Last half-year | Last year | Share in 3 years at its half-year rate |
|---|---|---|---|---|---|
| Advertising | 98% | +27.5% | +30.1% | +27.5% | 96% |
| Other revenue | 2% | +72.7% | +73.1% | +65.3% | 4% |
| Reality Labs and Meta AI beyond advertising | 1% | +16.5% | +6.5% | +5.8% | 0% |
| Whole company | 100% | +28.0% | +30.4% | +27.7% | 100% |
On the filed split through 2026-06-30, if each business keeps its last half-year's growth, the company grows about 31% next year and 31% in year 3, against 27.7% over the last twelve months. A scenario, not a forecast: no business holds its rate.
The results
The numbers
Sales grew 28% over the twelve months to Sep 2026, faster than the 22% in the year to Dec 2025.
Operating margin is 33%, about where it was (35%) in the year to Dec 2023.
Running costs take 48¢ of each sales dollar, up from 41¢ in the year to Dec 2021.
Each dollar put into the business earns 26% a year after tax, about where it was (31%) in the year to Dec 2021.
It turned 12% of sales into free cash, $28.6B over the last twelve months, after spending 43% of sales on buildings and equipment.
It owes $1.22B more than the cash it holds, about 0.0 years of the cash its operations bring in.
The share count fell 10% in 4.8 years.
All the numbers
Filed figures from the company's 10-K and 10-Q, read 2026-10-22. Fiscal years end in December.
| Dec 2021 | Dec 2022 | Dec 2023 | Dec 2024 | Dec 2025 | 12 mo to Sep 2026 | |
|---|---|---|---|---|---|---|
| Size and growth | ||||||
| Sales | $118B | $117B | $135B | $165B | $201B | $242B |
| Sales growth | +37% | −1.1% | +16% | +22% | +22% | +28% |
| Profit | ||||||
| Gross margin | 81% | 78% | 81% | 82% | 82% | 81% |
| Operating margin | 40% | 25% | 35% | 42% | 41% | 33% |
| Net margin | 33% | 20% | 29% | 38% | 30% | 31% |
| What growth costs, per sales dollar | ||||||
| Research and development | 21% | 30% | 29% | 27% | 29% | 32% |
| Sales and marketing | 12% | 13% | 9.1% | 6.9% | 6.0% | 5.4% |
| General and admin | 8.3% | 10% | 8.5% | 5.9% | 6.0% | 11% |
| Pay in shares | 7.8% | 10% | 10% | 10% | 10% | - |
| Capital spending | 16% | 27% | 20% | 23% | 35% | 43% |
| Cash | ||||||
| Cash from operations | $57.7B | $50.5B | $71.1B | $91.3B | $116B | $133B |
| Free cash flow | $39.0B | $19.3B | $44.1B | $54.1B | $46.1B | $28.6B |
| Balance sheet | ||||||
| Cash and short-term investments | $48.0B | $40.7B | $65.4B | $77.8B | $81.6B | $82.4B |
| Debt | $0 | $9.92B | $18.4B | $28.8B | $58.7B | $83.7B |
| Net cash | $48.0B | $30.8B | $47.0B | $49.0B | $22.8B | −$1.22B |
| Returns and shares | ||||||
| Return on capital | 31% | 18% | 25% | 32% | 24% | 26% |
| Shares (diluted) | 2.86B | 2.70B | 2.63B | 2.61B | 2.57B | 2.57B |
What growth costs
Growth used to make Meta more profitable, because selling and admin costs barely rise with sales. That has turned: research spending, which now carries AI staff and computing, is rising faster than sales, and the money spent on data centres has doubled as a share of sales and will reach profits slowly as wear-and-tear charges. Profit margins peaked in 2024.
The evidence under this - 2 claims
- The trailing year carries $2.4 billion of legal charges and $1.2 billion of severance from the June quarter (C05), roughly one and a half points of margin, so part of the fall is not the cost of growth.
- Capital spending is not yet an expense: it reaches the income statement as depreciation over five to 5.5 years (C27), so the margin has absorbed only the start of it.
Free cash flow of $784 million on operating cash flow of $31.86 billion means capital spending consumed 97.5% of the cash the business produced in the quarter. The offsetting argument is that this is an accounting distinction rather than an economic one: the spending buys servers and data centres with multi-year lives, and a company that expensed the same amount would show identical cash flow and far lower profit. Which reading is right is precisely what the thesis is about, and neither the filings nor the call resolve it - the company has attached no revenue to the spending.
- 2026-10-21editedQ3 2026 results: free cash flow turned negative at $2.12B on $34.91B of capital spending.
The last report (29 Jul)
Quarter ended June 2026, filed figures.
Sales beat the top of guidance, up 26% to $64.6B, with ad prices up 14% and ads shown up 10%. Profit fell 37% because of a $9.89B charge for the deal with 52 states; without it, profit would have risen.
The cash is the worry: spending of $34.9B left free cash flow at minus $2.1B, and net cash shrank to about $1.5B. A New Mexico court also ordered Meta to pay $25B, which it will appeal.
What it did to the thesisThe advertising engine passed its test, but cash flow turned negative and legal costs stepped up. The view holds; conviction is now low because the cushion is thinner.
The quarter in detail
Sales growth from the previous quarter, filed figures.
Earnings per share $6.18 against $7.19 expected; $6.18 as filed, with no adjustments.
The evidence under this - 5 claims
The adjusted figure is management's and flatters the quarter, but the direction it corrects is real: a 12-point margin decline of which roughly 6 points is two identified one-off charges is a different fact from a 12-point decline that is all run-rate. Against that, the legal charges are not obviously one-off - the same release warns of youth-related trials scheduled this year that 'may ultimately result in a material loss' - so a company that books $2.4 billion of legal charges in a quarter and expects more is arguably showing its true cost structure rather than an exception.
- 2026-10-21editedQ3 2026 results: revenue above the top of guidance; operating profit up about 11% only when the $9.89B charge is excluded.
The chief financial officer attributes part of the 12% price increase to 'improvements in macro conditions relative to Q2 of last year and currency tailwinds' rather than to performance alone, so not all of it is pricing power. The offsetting reading is the combination: raising price 12% while raising volume 14% is not something a business without pricing power can do in the same quarter, whatever share of the price move is cyclical, and she notes the price gain was partially offset by strong impression growth from lower-monetising surfaces - meaning mix worked against the reported figure rather than for it.
- 2026-10-21editedQ3 2026 results: advertising +26%, impressions +10%, price per ad +14%.
Free cash flow of $784 million on operating cash flow of $31.86 billion means capital spending consumed 97.5% of the cash the business produced in the quarter. The offsetting argument is that this is an accounting distinction rather than an economic one: the spending buys servers and data centres with multi-year lives, and a company that expensed the same amount would show identical cash flow and far lower profit. Which reading is right is precisely what the thesis is about, and neither the filings nor the call resolve it - the company has attached no revenue to the spending.
- 2026-10-21editedQ3 2026 results: free cash flow turned negative at $2.12B on $34.91B of capital spending.
- Net cash of $6.6 billion on a company earning $83 billion of annual operating income is not a solvency question, and the chief financial officer frames the debt deliberately: 'we've also been evolving our capital structure to include a greater mix of debt to bring down our cost of capital'. The finding is the direction rather than the level. In six months net cash fell 71%, the buyback stopped, and the capital that used to retire shares now buys servers - which is a real transfer from the current shareholder to a future one, whatever the return on it turns out to be.
- Net cash of $6.6 billion is also not the whole commitment. The 2025 10-K reports $45.95 billion of maximum exposure to loss on an unconsolidated data centre venture, including a residual value guarantee threshold of approximately $28 billion (C26). That sits outside the debt figure this claim compares cash against, so the direction it describes is understated rather than overstated.
- 2026-10-21editedQ3 2026 balance sheet: net cash down to about $1.5B from about $6.6B at June.
A company that books $2.4 billion of legal charges in one quarter and warns of more trials in the same document is describing a recurring cost, not an exception - which undercuts the adjusted operating income figure in C05. The New Mexico verdict raises the stakes by a step: liability is now found in a case where the state says it will seek up to $62.85 billion, against $83.3 billion of operating income in 2025. Against that, $62.85 billion is the state's stated ceiling rather than an award, and the same Attorney General's first jury verdict was $375 million; the amount is unknown and the claim does not guess it. The reason it stays in accepted risks rather than invalidation is that the operating-profit criterion already separates charge-bearing quarters from the business.
- 2026-09-27editedThe New Mexico Cambridge Analytica case, where the 10-Q says the state intends to seek up to $62.85 billion, ended in a liability verdict on September 25, 2026; added with the earlier $375 million verdict and the $942 million court-order headline from the state's own release list.
- 2026-10-02editedAt the October 1 penalty hearing New Mexico asked for $35 billion to $40 billion, a step down from the $62.85 billion the 10-Q said it intended to seek and the $219 billion statutory maximum; Meta proposed $345 million to $3.45 billion. Both are requests, no penalty is set, and the ruling is expected by late October.
- 2026-10-21editedPenalty set: $25B ordered by the New Mexico court (appeal promised), plus a $9.89B quarterly charge for the 52-state settlement.
- 2026-10-22editedQ3 2026 10-Q: interest rate on the $25B order, accrual, possible appeal bond.
What management did with the money
It is paying for the build partly with borrowed money. Debt rose about $24.9B in six months and net cash fell from $22.9B to $6.6B. Share buybacks, $26.26B in 2025, have stopped. It is also on the hook for up to $45.95B on a data-centre venture it does not put on its balance sheet.
Insiders in 2026Q2: 56 purchases and 125 sales, from their Form 4 filings.
The evidence under this - 2 claims
- Net cash of $6.6 billion on a company earning $83 billion of annual operating income is not a solvency question, and the chief financial officer frames the debt deliberately: 'we've also been evolving our capital structure to include a greater mix of debt to bring down our cost of capital'. The finding is the direction rather than the level. In six months net cash fell 71%, the buyback stopped, and the capital that used to retire shares now buys servers - which is a real transfer from the current shareholder to a future one, whatever the return on it turns out to be.
- Net cash of $6.6 billion is also not the whole commitment. The 2025 10-K reports $45.95 billion of maximum exposure to loss on an unconsolidated data centre venture, including a residual value guarantee threshold of approximately $28 billion (C26). That sits outside the debt figure this claim compares cash against, so the direction it describes is understated rather than overstated.
- 2026-10-21editedQ3 2026 balance sheet: net cash down to about $1.5B from about $6.6B at June.
The structure is disclosed rather than hidden, the guarantee is contingent on Meta walking away from a campus it built to use, and $45.95 billion of maximum exposure is a little over six months of 2026 guided capital spending at a company earning roughly $83 billion of annual operating income. The finding is where it sits rather than how large it is: none of it appears in the quarterly release, the call or the Q2 2026 10-Q, which name neither the Venture nor its counterparty, so a reader following the quarter rather than the annual report sees a balance sheet that is $45.95 billion less committed than it is.
The price, and where we disagree
In short. Fair. Meta is still cheaper than its rivals on profit, but after the September rise the price pays for some growth, not just for today's business.
The report is in. Sales of $64.6B beat the $61B to $64B guide, but free cash flow was negative and a $25B New Mexico order landed the same day, so the question is whether the spending pays back.
The next report, in early 2027, is the next test. Management guides $73B to $76B of sales and 2026 spending of $135B to $145B. A revenue line for the new AI products would answer the question the price is asking.
The evidence under this - 4 claims
A capital budget roughly doubling year over year with no 2027 figure offered is the single largest uncertainty in this file, and the refusal is not evasion - the company says planning is dynamic and that it is 'gearing our current infrastructure plans toward maximizing capacity in 2026 and 2027'. Two things cut the other way. The company still guides operating income above last year's while spending this, which bounds the damage to the income statement. And it states that 'the industry has historically underbuilt for the wave of AI adoption, making existing capacity, including our own, extremely valuable' - a claim about scarcity that, if true, makes the spending an option rather than a sunk cost.
- 2026-10-21editedQ3 2026 release: capex range narrowed to $135-145B, expenses raised for the charge, Q4 revenue guided $73-76B, no 2027 capex number.
- 2026-10-22editedQ3 2026 10-Q: a 2027 figure exists even though no 2027 capital spending number was guided.
The absence is the finding and it cuts both ways. Against the company: a $130-145 billion capital programme with no quantified revenue attached is the reason the market is refusing to capitalise it, and this thesis cannot argue that the market is wrong on evidence the company has not produced. For the company: the largest single use of the compute is improving the advertising business, and that return is already visible in the disclosed 8.3% increase in ad clicks and 15.7% uplift in conversions on Facebook - so 'no revenue attached' understates it, because some of the return arrives as advertising revenue rather than as a new line.
- 2026-10-21editedQ3 2026 release still attaches no revenue to the spending.
Free cash flow of $784 million on operating cash flow of $31.86 billion means capital spending consumed 97.5% of the cash the business produced in the quarter. The offsetting argument is that this is an accounting distinction rather than an economic one: the spending buys servers and data centres with multi-year lives, and a company that expensed the same amount would show identical cash flow and far lower profit. Which reading is right is precisely what the thesis is about, and neither the filings nor the call resolve it - the company has attached no revenue to the spending.
- 2026-10-21editedQ3 2026 results: free cash flow turned negative at $2.12B on $34.91B of capital spending.
A company that books $2.4 billion of legal charges in one quarter and warns of more trials in the same document is describing a recurring cost, not an exception - which undercuts the adjusted operating income figure in C05. The New Mexico verdict raises the stakes by a step: liability is now found in a case where the state says it will seek up to $62.85 billion, against $83.3 billion of operating income in 2025. Against that, $62.85 billion is the state's stated ceiling rather than an award, and the same Attorney General's first jury verdict was $375 million; the amount is unknown and the claim does not guess it. The reason it stays in accepted risks rather than invalidation is that the operating-profit criterion already separates charge-bearing quarters from the business.
- 2026-09-27editedThe New Mexico Cambridge Analytica case, where the 10-Q says the state intends to seek up to $62.85 billion, ended in a liability verdict on September 25, 2026; added with the earlier $375 million verdict and the $942 million court-order headline from the state's own release list.
- 2026-10-02editedAt the October 1 penalty hearing New Mexico asked for $35 billion to $40 billion, a step down from the $62.85 billion the 10-Q said it intended to seek and the $219 billion statutory maximum; Meta proposed $345 million to $3.45 billion. Both are requests, no penalty is set, and the ruling is expected by late October.
- 2026-10-21editedPenalty set: $25B ordered by the New Mexico court (appeal promised), plus a $9.89B quarterly charge for the 52-state settlement.
- 2026-10-22editedQ3 2026 10-Q: interest rate on the $25B order, accrual, possible appeal bond.
How we read the multiple
The whole company, debt included, costs 23.1 times its yearly operating profit, against 31.6 for comparable rivals: still the cheapest. On the cash left after its building spending it costs 46.7 times, dearer than every rival but Alphabet. When we wrote this, those figures were 17.5 and 34.8.
The gap is the data-centre spending. If it earns a return, the profit is real. If not, the cash figure is right, and profit is overstated until the yearly charge for wearing out servers catches up.
Against its own history. Now dearer than its own normal on every measure, while the business grows faster than it used to, so the old discount for the spending has been given back.
The evidence under this - 2 claims
The two multiples describe the same company and cannot both be right about it, which is the thesis. The case against reading the profit multiples as cheap is that operating profit is stated after depreciation on an asset base that is roughly doubling, so today's depreciation charge understates tomorrow's - meaning the 23.1 times figure is struck on a profit number the capital programme will itself reduce. That is the strongest argument against this thesis and nothing in the filings refutes it; what bounds it is the company's own guidance that 2026 operating income will still exceed 2025's. The second argument is new: no quarter has been reported since August, so the whole move from 17.5 to 23.1 times is the market paying more for the same profit, and the discount the thesis was written to exploit is now smaller than it was.
- 2026-09-27editedRe-struck at the 2026-09-25 close of $751.66 from the refreshed research pack, 34.5% above the $559.02 it was written at; Meta is no longer the dearest of the cohort on free cash flow.
Amazon moved the opposite way on the same asset class in the same month, shortening a subset of servers and networking equipment from six years to five effective January 1, 2025 and attributing it to 'the increased pace of technology development, particularly in the area of artificial intelligence and machine learning'. Both are the companies' own estimates and both are defensible, but they cannot both describe the same hardware: one firm judged AI infrastructure to be lasting longer at exactly the moment the other judged it to be obsoleting faster, and the lengthening is the one that lowers near-term depreciation.
Priced against its rivals
The competitors its own annual report names, each confirmed against the rival's filing. Measured on company value for each $1 of operating profit, cheapest first:
| Ticker | Company | company value for each $1 of operating profit |
|---|---|---|
| META | Meta Platforms | 23.1× |
| GOOGL | Alphabet | 28.4× |
| RDDT | 34.9× | |
| AMZN | Amazon.com not in the median | 30.2× |
| PINS | Pinterest held out | not priced |
The middle of the other 2 is 31.6. Amazon.com, Inc. competes with it in one product line inside a much bigger company, so its price says little about this business and it is not counted in the middle. Meta Platforms, Inc. is cheapest of 3.
Rivals' figures come from a data provider, so read the spread of the group rather than any single pair.
Still the cheapest of its rivals on profit and near the dearest on cash. Both positions are Meta's own, and the gap between them is the argument.
The bet: what it pays if it works
The bet is on Reality Labs, the glasses and headsets, together with the AI products Meta has begun to sell beyond advertising: subscriptions, business agents and its models. Neither makes money yet, so advertising is counted separately, at what the closest listed rivals are valued at.
Counted that way, advertising alone is worth more than the whole company today, so the price pays nothing for either. If both take off, the company is worth about half as much again, nearly all advertising; if they never earn, nothing is lost.
| Reality Labs and Meta AI beyond advertising revenue in two to three years | Pairs of AI glasses needed | At the stated multiple |
|---|---|---|
| $4.50B (the big win) | about 11.5M at 80% kept | 1.5× |
| $6.00B | about 16.5M at 80% kept | 1.5× |
To pay 2× (a Fair bet) it would need about $1.01T of Reality Labs and Meta AI beyond advertising revenue (about 3.3B pairs of AI glasses at 80% kept); to pay 5× (a Cheap bet), about $7.05T of Reality Labs and Meta AI beyond advertising revenue (about 23.2B pairs of AI glasses at 80% kept), at the same multiple.
Each row is Reality Labs and Meta AI beyond advertising alone, plus the rest of the company (Family of Apps) counted at about $3.01T.
The steps it has to take
- Done More than seven million AI glasses sold in a year
- Done Meta's own glasses line launched with EssilorLuxottica
- Done Model API, business agents and Meta One launched
- Not started Reality Labs sales above $3B a year due 2027-12
- Not started Reality Labs' yearly loss smaller than 2025's $19B due 2027-12
- Not started Meta discloses revenue from the API, agents or Meta One due 2027-12
The detail behind the label
The evidence under this - 7 claims
EssilorLuxottica gives no revenue for the line and says the glasses weighed on its own margin; units are not Meta's revenue, and Reality Labs' 2025 sales rose only 3% as Quest headsets fell (C37).
Annualised, the Reality Labs loss is roughly $18.5 billion against a company operating income of $83 billion - a fifth of the profit, spent on a segment whose revenue is two-thirds of one percent of the total. The defence is that it is disclosed, deliberate and bounded: the company says plainly it will keep losing money and that it depends on profits elsewhere, so nobody buying this security is being surprised. It is recorded in the accepted risks rather than the invalidation criteria for that reason.
The two multiples describe the same company and cannot both be right about it, which is the thesis. The case against reading the profit multiples as cheap is that operating profit is stated after depreciation on an asset base that is roughly doubling, so today's depreciation charge understates tomorrow's - meaning the 23.1 times figure is struck on a profit number the capital programme will itself reduce. That is the strongest argument against this thesis and nothing in the filings refutes it; what bounds it is the company's own guidance that 2026 operating income will still exceed 2025's. The second argument is new: no quarter has been reported since August, so the whole move from 17.5 to 23.1 times is the market paying more for the same profit, and the discount the thesis was written to exploit is now smaller than it was.
- 2026-09-27editedRe-struck at the 2026-09-25 close of $751.66 from the refreshed research pack, 34.5% above the $559.02 it was written at; Meta is no longer the dearest of the cohort on free cash flow.
- Meta discloses no revenue for the assistant, the agents, the API or Meta One, so the figure has nothing filed under it (C10).
- Meta's own 2024 forecast, unsealed in court, put generative AI product revenue at $2B to $3B in 2025, and none has been disclosed since: its own forecasts have run ahead of what it reports.
- Gemini and ChatGPT already hold the assistant users Meta would charge (C14).
None of the three carries a disclosed revenue figure; C10 records that Meta attaches no revenue to the capital spending that funds the models and compute.
Where we disagree with the market
The capital spending is mostly money burned, so the reported profit deserves less than rivals get. The market has paid up since August, but still thinks so.
The advertising engine is accelerating. The market is still discounting a build the accounts treat as an asset, though far less than a month ago.
The core business is doing what a strengthening business does: more advertisements shown and a higher price for each, at the same time. The market still pays less for profit it doubts will survive the spending, though the gap is smaller than in August.
The case against us is real. No revenue from the spending has been disclosed, management will not guide 2027 spending, and debt is rising. Earnings have missed forecasts in two of the last four quarters.
What analysts say, and the evidence
The evidence under this - 6 claims
The chief financial officer attributes part of the 12% price increase to 'improvements in macro conditions relative to Q2 of last year and currency tailwinds' rather than to performance alone, so not all of it is pricing power. The offsetting reading is the combination: raising price 12% while raising volume 14% is not something a business without pricing power can do in the same quarter, whatever share of the price move is cyclical, and she notes the price gain was partially offset by strong impression growth from lower-monetising surfaces - meaning mix worked against the reported figure rather than for it.
- 2026-10-21editedQ3 2026 results: advertising +26%, impressions +10%, price per ad +14%.
The two multiples describe the same company and cannot both be right about it, which is the thesis. The case against reading the profit multiples as cheap is that operating profit is stated after depreciation on an asset base that is roughly doubling, so today's depreciation charge understates tomorrow's - meaning the 23.1 times figure is struck on a profit number the capital programme will itself reduce. That is the strongest argument against this thesis and nothing in the filings refutes it; what bounds it is the company's own guidance that 2026 operating income will still exceed 2025's. The second argument is new: no quarter has been reported since August, so the whole move from 17.5 to 23.1 times is the market paying more for the same profit, and the discount the thesis was written to exploit is now smaller than it was.
- 2026-09-27editedRe-struck at the 2026-09-25 close of $751.66 from the refreshed research pack, 34.5% above the $559.02 it was written at; Meta is no longer the dearest of the cohort on free cash flow.
The absence is the finding and it cuts both ways. Against the company: a $130-145 billion capital programme with no quantified revenue attached is the reason the market is refusing to capitalise it, and this thesis cannot argue that the market is wrong on evidence the company has not produced. For the company: the largest single use of the compute is improving the advertising business, and that return is already visible in the disclosed 8.3% increase in ad clicks and 15.7% uplift in conversions on Facebook - so 'no revenue attached' understates it, because some of the return arrives as advertising revenue rather than as a new line.
- 2026-10-21editedQ3 2026 release still attaches no revenue to the spending.
A capital budget roughly doubling year over year with no 2027 figure offered is the single largest uncertainty in this file, and the refusal is not evasion - the company says planning is dynamic and that it is 'gearing our current infrastructure plans toward maximizing capacity in 2026 and 2027'. Two things cut the other way. The company still guides operating income above last year's while spending this, which bounds the damage to the income statement. And it states that 'the industry has historically underbuilt for the wave of AI adoption, making existing capacity, including our own, extremely valuable' - a claim about scarcity that, if true, makes the spending an option rather than a sunk cost.
- 2026-10-21editedQ3 2026 release: capex range narrowed to $135-145B, expenses raised for the charge, Q4 revenue guided $73-76B, no 2027 capex number.
- 2026-10-22editedQ3 2026 10-Q: a 2027 figure exists even though no 2027 capital spending number was guided.
- Net cash of $6.6 billion on a company earning $83 billion of annual operating income is not a solvency question, and the chief financial officer frames the debt deliberately: 'we've also been evolving our capital structure to include a greater mix of debt to bring down our cost of capital'. The finding is the direction rather than the level. In six months net cash fell 71%, the buyback stopped, and the capital that used to retire shares now buys servers - which is a real transfer from the current shareholder to a future one, whatever the return on it turns out to be.
- Net cash of $6.6 billion is also not the whole commitment. The 2025 10-K reports $45.95 billion of maximum exposure to loss on an unconsolidated data centre venture, including a residual value guarantee threshold of approximately $28 billion (C26). That sits outside the debt figure this claim compares cash against, so the direction it describes is understated rather than overstated.
- 2026-10-21editedQ3 2026 balance sheet: net cash down to about $1.5B from about $6.6B at June.
Sixty-four positive ratings against two negatives, unchanged through a 24% fall to August and the recovery since, is a distribution that does not move with the price in either direction - a reason to treat the consensus as uninformative rather than to take comfort from it. The swings in surprise are also extreme for a company of this size, and the October 2025 miss is explained by a one-off tax charge rather than operations. The clean part is the adjustment line: no gap at all between the adjusted figure consensus is struck on and the filed diluted figure, which is unusual and means these multiples run on audited earnings.
- 2026-09-27editedRe-read from the refreshed research pack: the rating split and the earnings record are unchanged; the counter-evidence no longer describes the shares as down 24% over a year.
What could go wrong, and how we would know
The risks
- Spending that never pays back. $130B to $145B is guided for 2026 and no revenue from it has been disclosed.
- Debt, then new shares, replacing cash. Net cash fell to about $1.5B from $22.9B, and on 26 October Meta launched about $35B of share sales with about $60B more planned.
- A hidden commitment. Up to $45.95B of exposure to a data-centre venture sits off the balance sheet.
- Commitments already signed. $454.5B of purchase commitments are non-cancelable, $121.4B of it due in 2027, so next year's spending is largely set.
- Court cases and penalties. A New Mexico court ordered Meta to pay $25B and Meta will appeal; a $9.89B charge settled 52 states' claims; more youth trials are due.
- Slower user growth. Daily users grew 3% against 7% six months earlier.
The evidence under this - 8 claims
User growth has halved: 7% in December 2025 against 3% in June 2026. At 3.6 billion daily users out of roughly 5.5 billion people online, the measure is approaching saturation by arithmetic, so deceleration is expected rather than alarming - but it removes one of the two levers on revenue and leaves the whole growth case resting on monetisation per user. It also means the erosion test in the moat block matters more than it would for a company still adding users.
- 2026-10-21editedQ3 2026 release: users grew 3% to 3.64 billion.
A capital budget roughly doubling year over year with no 2027 figure offered is the single largest uncertainty in this file, and the refusal is not evasion - the company says planning is dynamic and that it is 'gearing our current infrastructure plans toward maximizing capacity in 2026 and 2027'. Two things cut the other way. The company still guides operating income above last year's while spending this, which bounds the damage to the income statement. And it states that 'the industry has historically underbuilt for the wave of AI adoption, making existing capacity, including our own, extremely valuable' - a claim about scarcity that, if true, makes the spending an option rather than a sunk cost.
- 2026-10-21editedQ3 2026 release: capex range narrowed to $135-145B, expenses raised for the charge, Q4 revenue guided $73-76B, no 2027 capex number.
- 2026-10-22editedQ3 2026 10-Q: a 2027 figure exists even though no 2027 capital spending number was guided.
The absence is the finding and it cuts both ways. Against the company: a $130-145 billion capital programme with no quantified revenue attached is the reason the market is refusing to capitalise it, and this thesis cannot argue that the market is wrong on evidence the company has not produced. For the company: the largest single use of the compute is improving the advertising business, and that return is already visible in the disclosed 8.3% increase in ad clicks and 15.7% uplift in conversions on Facebook - so 'no revenue attached' understates it, because some of the return arrives as advertising revenue rather than as a new line.
- 2026-10-21editedQ3 2026 release still attaches no revenue to the spending.
- Net cash of $6.6 billion on a company earning $83 billion of annual operating income is not a solvency question, and the chief financial officer frames the debt deliberately: 'we've also been evolving our capital structure to include a greater mix of debt to bring down our cost of capital'. The finding is the direction rather than the level. In six months net cash fell 71%, the buyback stopped, and the capital that used to retire shares now buys servers - which is a real transfer from the current shareholder to a future one, whatever the return on it turns out to be.
- Net cash of $6.6 billion is also not the whole commitment. The 2025 10-K reports $45.95 billion of maximum exposure to loss on an unconsolidated data centre venture, including a residual value guarantee threshold of approximately $28 billion (C26). That sits outside the debt figure this claim compares cash against, so the direction it describes is understated rather than overstated.
- 2026-10-21editedQ3 2026 balance sheet: net cash down to about $1.5B from about $6.6B at June.
A company that books $2.4 billion of legal charges in one quarter and warns of more trials in the same document is describing a recurring cost, not an exception - which undercuts the adjusted operating income figure in C05. The New Mexico verdict raises the stakes by a step: liability is now found in a case where the state says it will seek up to $62.85 billion, against $83.3 billion of operating income in 2025. Against that, $62.85 billion is the state's stated ceiling rather than an award, and the same Attorney General's first jury verdict was $375 million; the amount is unknown and the claim does not guess it. The reason it stays in accepted risks rather than invalidation is that the operating-profit criterion already separates charge-bearing quarters from the business.
- 2026-09-27editedThe New Mexico Cambridge Analytica case, where the 10-Q says the state intends to seek up to $62.85 billion, ended in a liability verdict on September 25, 2026; added with the earlier $375 million verdict and the $942 million court-order headline from the state's own release list.
- 2026-10-02editedAt the October 1 penalty hearing New Mexico asked for $35 billion to $40 billion, a step down from the $62.85 billion the 10-Q said it intended to seek and the $219 billion statutory maximum; Meta proposed $345 million to $3.45 billion. Both are requests, no penalty is set, and the ruling is expected by late October.
- 2026-10-21editedPenalty set: $25B ordered by the New Mexico court (appeal promised), plus a $9.89B quarterly charge for the 52-state settlement.
- 2026-10-22editedQ3 2026 10-Q: interest rate on the $25B order, accrual, possible appeal bond.
The structure is disclosed rather than hidden, the guarantee is contingent on Meta walking away from a campus it built to use, and $45.95 billion of maximum exposure is a little over six months of 2026 guided capital spending at a company earning roughly $83 billion of annual operating income. The finding is where it sits rather than how large it is: none of it appears in the quarterly release, the call or the Q2 2026 10-Q, which name neither the Venture nor its counterparty, so a reader following the quarter rather than the annual report sees a balance sheet that is $45.95 billion less committed than it is.
A commitment is a purchase promise and not yet a payment: some cloud capacity can be resold ('may be reduced if the cloud service provider is able to sell such capacity to other customers'), the lease figure runs over ten years, and the chief executive says offers for Meta's compute come at a premium to what it paid (C10). Against that, $121.41 billion due in 2027 is not far below 2025's whole operating cash flow of $115.80 billion (C08), so 2027 spending is largely signed whatever the company says on the Q4 call.
The company is not short of cash today: about $82 billion held, $97 billion of operating cash flow in nine months and top credit ratings, and it stresses that equity 'diversifies' its funding. The stock sale can be read as pre-funding a build it could still carry with debt, not as an emergency; what changes is that Meta chose to sell shares at a price 14% below the 25 September close rather than borrow more, and that the ATM part is only a plan.
- 2026-10-27editedPricing term sheet and release: common offering upsized to $25.0B at $625, preferred terms fixed; total about $37.5B.
What we still do not know
- What the capital spending earns. Meta says buyers have offered more for its computing power than it paid, and has put a number on none of it.
- How much it will spend in 2027. The finance chief declined to say, though $121.4B of purchase commitments already fall due that year.
- How fast its new data-centre spending wears out. Servers are written off over 5.5 years, lengthened in 2025.
- How much stock it will sell. About $35B is launched and about $60B is planned with no agreement signed.
The evidence under this - 5 claims
A capital budget roughly doubling year over year with no 2027 figure offered is the single largest uncertainty in this file, and the refusal is not evasion - the company says planning is dynamic and that it is 'gearing our current infrastructure plans toward maximizing capacity in 2026 and 2027'. Two things cut the other way. The company still guides operating income above last year's while spending this, which bounds the damage to the income statement. And it states that 'the industry has historically underbuilt for the wave of AI adoption, making existing capacity, including our own, extremely valuable' - a claim about scarcity that, if true, makes the spending an option rather than a sunk cost.
- 2026-10-21editedQ3 2026 release: capex range narrowed to $135-145B, expenses raised for the charge, Q4 revenue guided $73-76B, no 2027 capex number.
- 2026-10-22editedQ3 2026 10-Q: a 2027 figure exists even though no 2027 capital spending number was guided.
The absence is the finding and it cuts both ways. Against the company: a $130-145 billion capital programme with no quantified revenue attached is the reason the market is refusing to capitalise it, and this thesis cannot argue that the market is wrong on evidence the company has not produced. For the company: the largest single use of the compute is improving the advertising business, and that return is already visible in the disclosed 8.3% increase in ad clicks and 15.7% uplift in conversions on Facebook - so 'no revenue attached' understates it, because some of the return arrives as advertising revenue rather than as a new line.
- 2026-10-21editedQ3 2026 release still attaches no revenue to the spending.
Amazon moved the opposite way on the same asset class in the same month, shortening a subset of servers and networking equipment from six years to five effective January 1, 2025 and attributing it to 'the increased pace of technology development, particularly in the area of artificial intelligence and machine learning'. Both are the companies' own estimates and both are defensible, but they cannot both describe the same hardware: one firm judged AI infrastructure to be lasting longer at exactly the moment the other judged it to be obsoleting faster, and the lengthening is the one that lowers near-term depreciation.
A commitment is a purchase promise and not yet a payment: some cloud capacity can be resold ('may be reduced if the cloud service provider is able to sell such capacity to other customers'), the lease figure runs over ten years, and the chief executive says offers for Meta's compute come at a premium to what it paid (C10). Against that, $121.41 billion due in 2027 is not far below 2025's whole operating cash flow of $115.80 billion (C08), so 2027 spending is largely signed whatever the company says on the Q4 call.
The company is not short of cash today: about $82 billion held, $97 billion of operating cash flow in nine months and top credit ratings, and it stresses that equity 'diversifies' its funding. The stock sale can be read as pre-funding a build it could still carry with debt, not as an emergency; what changes is that Meta chose to sell shares at a price 14% below the 25 September close rather than borrow more, and that the ATM part is only a plan.
- 2026-10-27editedPricing term sheet and release: common offering upsized to $25.0B at $625, preferred terms fixed; total about $37.5B.
What would change our view
- Attractive→NeutralThe operating multiple closes toward its peers
It needs:
- ○The whole company is priced at about 25 times operating profit, most of the way to its rivals, so the build is no longer discounted.about ≈$813.48 today, 8% above the last close
- Attractive→NeutralThe build stops being investment
It needs any of:
- ○Spending planned for 2027 is higher than 2026's, with no revenue from it disclosed.watched as it happens
- ○Operating profit falls from a year earlier in a quarter with no legal or severance charges.checked at the next results, due in Jan 2027
- ○Advertising revenue growth drops below 10% from a year earlier, so the engine paying for the build is slowing.checked at the next results, due in Jan 2027
We would also be wrong if
- Daily users stop growing or fall.
- Net cash turns into net debt and the dividend is cut to pay for the build.
What is coming
- October 2026 New Mexico court rules on the Cambridge Analytica penalty: Whether the award lands nearer Meta's $3.45B ceiling or the state's $35B to $40B request; tens of billions would test the accepted legal risk and the net cash position.
Every dated event ahead
- October 2026 New Mexico court rules on the Cambridge Analytica penalty: Whether the award lands nearer Meta's $3.45B ceiling or the state's $35B to $40B request; tens of billions would test the accepted legal risk and the net cash position.
- 28 October 2026 Q3 2026 results: Revenue against the $61-64B guide, price per ad still rising, 2026 capital spending held within $130-145B, and any first revenue figure attached to the computing build.
- January 2027 Q4 2026 results and first 2027 capital spending outlook: Whether 2027 capital spending rises again and buybacks resume; a large step-up with no disclosed revenue would deepen the narrowing cost leverage and the move to debt.
Since we wrote this
- 27 OctNewsMeta's share sale prices at $37.5B, a little above launchThe common offering was upsized to $25B at $625 a share. It confirms the funding plan already scored; view and conviction hold.
- 26 OctNewsMeta launches $35B of share offerings, plans $60B moreOutside money is now part of the plan, so financial position and per-share capture each fall one level. View and conviction held; the price is lower.
- 22 OctThesis reviewConviction lowered to LowThe owner adopted the recommendation: the advertising case holds but cash flow, legal costs and funding make the cushion thinner.
- 22 OctEarningsThe 10-Q shows $121B of 2027 commitments already signedNext year's spending is largely set: $454.5B of non-cancelable commitments, $121.4B due in 2027, plus the $25B New Mexico order's 15% interest. View and conviction held.
- 21 OctEarningsSales beat, but cash flow turned negative and legal costs jumpedRevenue grew 26% and ad prices rose 14%, yet free cash flow was negative $2.12B, a $9.89B settlement charge hit profit and a $25B New Mexico order followed. We recommend lower conviction.
- 11 OctThesis reviewReality Labs declared an early business, with a bet on itThe owner asked every thesis to name the lines its story rides; Reality Labs is a filed segment losing $4.6B a quarter.
- 11 OctThesis reviewThe bet now covers the AI products as well as the glassesA thesis carries one bet, so the owner asked that its early lines be folded into one business and priced together.
- 7 OctNewsOfcom investigates Meta's safety checks for Instagram Instants; Meta joins AI-bot standards group. View heldOfcom opens Instagram Instants investigation; Meta joins an AI-agent commerce standards group.
- 6 OctNewsNorway bans camera glasses in some public places; Meta contests UK data demands. View heldNorway restricts camera glasses in public; Meta, TikTok and X contest UK regulator's data demands.
- 5 OctThesis reviewThe long case now sits under the verdictAfter the five-name pilot, the owner asked for the long case on every thesis.
- 4 OctThesis reviewOwner's checklist: revenue streams, cost of growth, management, catalystsThe owner's checklist found the method silent on who runs the company, what growth costs, where the money comes from and what is dated ahead; this pass fills those four blocks from the proxy and filings.
- 3 OctNewsApple will flag AI data requests after Muse complaints; Dutch chain drops Meta glasses. View heldPrivacy backlash around Meta's Muse Mac app and its AI glasses.
- 2 OctNewsNew Mexico asks for $35B to $40B; view heldThe state asked a judge for $35B to $40B of penalties after its jury win; Meta offered up to $3.45B. A ruling is due by late October. It is a request, not an award, so the view held.
- 29 SepNewsMeta hired a software chief and rented AI capacity; no figures came. View heldMeta hired MongoDB's chief executive to lead a business AI platform, and took AI computing capacity in Southeast Asia from Firmus.
- 27 SepNewsJury finds Meta liable; shares up 34.5%, price now fairA New Mexico jury found Meta misled users in a case where the state seeks up to $62.85B; no penalty is set. The rise since August leaves Meta cheapest of its rivals on profit, but no longer cheap.
- 25 SepNewsMeta expanded its AI wearables lineup; no revenue is disclosed yet. View heldMeta's Connect conference launched Muse (a personal AI agent), a camera-free AI glasses line and the Muse Charm pendant, with Amazon blocking Muse from its own shopping surfaces and Shopify, PayPal and Instacart agreeing to carry it.
- 21 SepNewsAdoption arrived for the new product; a revenue line for it still has not. View heldA consumer AI agent at the top of the App Store, and a German liability ruling.
- 15 SepThesis reviewA $45.95B data-centre commitment found. View heldThe annual report showed up to $45.95B of exposure to a venture kept off the balance sheet. It sharpens a worry we already had; the advertising case and the price comparison did not move.
- 25 AugThesis writtenAttractive, medium convictionA month after the June-quarter results, when operating profit fell 8% on 28% more sales and cash after spending dropped to $784M.
Also checked 10 times with nothing new to act on.
Sources 35
Sources - 35
Every claim above cites one of these, or states the reasoning that derived it. The quote is frozen at the date it was read: a link can rot, and an agent fetch returns a reading of a page rather than the page. Backs counts the whole claim ledger, so it can name a claim this page does not surface.
- S01primarycompany filing2026-01-29backsC01C02C03C07C25C26C27C30C33
- S02primarycompany release2026-01-28backsC04C06C07C08C11C22C24C37C38
- S03primarycompany release2026-07-29backsC01C04C05C06C08C09C10C11C12C18C24C37C41
- S04primarytranscript2026-07-29the expense bridge, the free-cash-flow collapse, the quantified ranking returns, the refusal to guide 2027 capital spending, the unquantified compute offers and the deliberate shift toward debt
Q2 total expenses were $42 billion, up 55% compared to last year and included $2.4 billion in charges related to legal proceedings and $1.2 billion in severance expenses [...] Capital expenditures, including principal payments on finance leases, were $31.1 billion [...] Free cash flow was $784 million. We ended the quarter with $90.3 billion in cash and marketable securities and $83.7 billion in debt. [...] Combined with our GEM model for ads ranking and sequence learning, these advancements generated an 8.3% increase in ad clicks and a 15.7% uplift in conversions on Facebook. [...] we aren't providing a specific outlook for 2027 CapEx at this time. Infrastructure planning remains highly dynamic. [...] We have quite a number of offers at a meaningful premium over what we paid for the compute. [...] we've also been evolving our capital structure to include a greater mix of debt to bring down our cost of capital
backsC04C05C06C08C09C10C16C17C18 - S05primaryvendor data2026-09-27Portfolio-management research pack (market tier: price, returns, volatility, analyst split, insider filings, theme co-movement, segment tagging)content/META/research-pack.jsonbacksC01C12C19C20C23C24C25C28C44
- S06primaryregulator2025-11-18FTC v. Meta Platforms, Inc., No. 20-3590 (JEB) - Memorandum Opinion, United States District Court for the District of Columbia United States District Court for the District of Columbiathe only adversarially-tested independent finding on whether Meta's attention is defensible - and it says the wall has broken down, which is why the moat is recorded as scale showing up in pricing rather than as a network effect
While it once might have made sense to partition apps into separate markets of social networking and social media, that wall has since broken down. [...] at a minimum includes TikTok and YouTube, fierce competitors for users' time and attention in this space. [...] The Court ultimately concludes that the agency has not carried its burden: Meta holds no monopoly in the relevant market. Judgment must therefore be entered in its favor.
backsC13 - S07primaryindustry data2026-04IAB/PwC Internet Advertising Revenue Report, full-year 2025 results Internet Advertising Bureau and PricewaterhouseCoopersthe independent, measured base for the prize and the rate for the driver, by format - and the geography mismatch that is why the prize does not compose. The headline total is quoted above; the format figures used in C15 ($117.7bn social at +32.6%, $114.2bn search at +11%, $63.4bn commerce media at +18.0%) sit in the report's format tables, and each was verified present in the document text rather than taken from a summary of it - the tables extract non-contiguously, which is why they are not in the quoted sentence
continued strong growth in U.S. digital advertising revenues, which rose 13.9% year over year (YoY) in 2025 to a record $294.6 billion
backsC15 - S08primarycompany filing2026-02-12the naming Meta's own filing withholds, from the other side of the table - and the identification of a conversational assistant as a competitor for the same attention
competitors such as Amazon, Meta (including Facebook, Instagram, Threads and MetaAI), Google (including Gemini, Lens and YouTube), OpenAI (including ChatGPT), Snap, Reddit, TikTok and X, many of which are larger and have significantly greater financial and human resources
backsC14 - S09secondarycompany filing2026-08-25Portfolio-management filed-competitor roster (data/reference/competitors.json, META entry) and the peer filings it citesdata/reference/competitors.jsonbacksC14C20C21C22
- S10secondaryvendor data2026-09-27Portfolio-management research pack (vendor_fundamentals tier: peer_valuation cohort and the ratio series)content/META/research-pack.jsonbacksC20C21
- S11primarycompany filing2026-05-04the instrument behind the $25 billion debt increase, with its maturity ladder and coupons - which the results release reports only as a balance-sheet total
META PLATFORMS, INC. $3,000,000,000 4.550% SENIOR NOTES DUE 2031 $2,000,000,000 4.875% SENIOR NOTES DUE 2033 $6,000,000,000 5.250% SENIOR NOTES DUE 2036 $4,000,000,000 6.200% SENIOR NOTES DUE 2046 $6,000,000,000 6.300% SENIOR NOTES DUE 2056 $4,000,000,000 6.450% SENIOR NOTES DUE 2066 UNDERWRITING AGREEMENT April 30, 2026
backsC11 - S12primaryregulator2026-09-15FINRA Consolidated Short Interest, META, settlement date 2026-09-15 Financial Industry Regulatory Authoritythe positioning figure the thesis recorded as unreachable - Rule 4560 regulatory data, not a vendor's resale of it
settlementDate 2026-09-15, currentShortPositionQuantity 30990107, previousShortPositionQuantity 28807174, changePercent 7.58, averageDailyVolumeQuantity 19680893, daysToCoverQuantity 1.57
backsC28 - S13primarycompany filing2026-02-06the other hyperscaler's judgement on the same asset class in the same month, from the opposite direction - which is what makes Meta's extension a choice rather than a convention
Effective January 1, 2025 we changed our estimate of the useful lives of a subset of our servers and networking equipment from six years to five years . The shorter useful lives are due to the increased pace of technology development, particularly in the area of artificial intelligence and machine learning.
read, but no claim rests on it - S14primarycompany filing2026-07-30the two New Mexico Attorney General cases: the $375 million jury penalty of March 24, 2026 and the $953 million abatement sought in the youth-safety case, and the up to $62.85 billion the state intends to seek in the Cambridge Analytica casebacksC18
- S15primaryregulator2026-09-25New Mexico Department of Justice press-release list (State of New Mexico v. Facebook, Inc.) - list entries only; the release pages refused the fetch New Mexico Department of Justice (Office of the Attorney General)the counterparty's own record of the September 25, 2026 liability verdict, and the headline of an August 7, 2026 court order the thesis had not carried
Jury Finds Facebook Violated New Mexico Consumer Protection Law, Faces Billions in Potential Civil Penalties September 25, 2026 Verdict finds company made deceptive statements on data privacy, misinformation, and hate speech following the Cambridge Analytica scandal [...] Court Orders Meta to Pay $942 Million and Overhaul Protections for Children on Facebook and Instagram in Landmark New Mexico Ruling August 7, 2026
backsC18 - S16primarycompany filing2026-09-2334 open-market sales of 91,252 shares, every one marked as made under a Rule 10b5-1 plan, and no open-market purchasebacksC19
- S17secondarynews2026-10-01backsC18
- S18secondarynews2026-10-01backsC18
- S19primarycompany filing2026-04-16beneficial ownership and voting power at April 1, 2026, shares outstanding by class, controlled-company status, the 2025 Bonus Plan priorities and payout, the RSU-only equity design and the CEO's $1 salarybacksC31C32C34
- S20primarycompany filing2026-01-16start date, prior roles and pay terms of the new President: $1,000,000 salary, 200% bonus target, $60,000,000 of RSUs vesting quarterly over four yearsbacksC32C34
- S21primarycompany filing2026-07-30revenue by source (Advertising, Other revenue, Reality Labs), no repurchases in the first half of 2026 with $25.03 billion still authorised, $8.70 billion of stock-plan tax payments, $24.91 billion of new debtbacksC29C30C33
- S22primarycompany release2026-02-11EssilorLuxottica Q4/FY 2025 results release (carried on BioSpace; the company's own page is a script this repository cannot run) EssilorLuxotticathe manufacturer's own count of Meta AI glasses sold in 2025, from the partner that makes and sells them
AI-glasses selling more than 7 million units in the FY, with all the regions and brands contributing
backsC35C38 - S23primarycompany release2025-09-17the list price of the main glasses line
Learn more about Ray-Ban Meta glasses (Gen 2) and get yours today, starting at $379 USD.
backsC36C38 - S24primarycompany release2025-09-17the list price of the display glasses
Starting at $799 USD, which includes both the glasses and Meta Neural Band, Meta Ray-Ban Display will let you experience, learn about, and interact with the world
backsC36 - S25primarycompany filing2026-02-05a listed rival's own account of the augmented-reality glasses it sells to developers
Spectacles, or Specs, are our wearable AR glasses, which overlay computing over the world and extend the immersive AR Lenses experience beyond Snapchat. Spectacles are powered by Snap OS
backsC40 - S26primarytranscript2026-07-29management's account of the glasses line, business agents and the model API, none of it with a revenue figure
Our glasses remain one of the fastest-growing consumer electronics of all time [...] Early sales have been strong, exceeding our expectations [...] there are already more than 1 million businesses using them to talk to their customers or complete sales every week [...] we also expect to grow a large business serving large customers as well. We've built our API.
backsC39C41 - S27secondarynews2025-04-30Meta's own 2024 forecast for generative AI product revenue, against which no revenue has since been disclosed
Meta made a prediction last year its generative AI products would rake in $2 billion to $3 billion in revenue in 2025, and between $460 billion and $1.4 trillion by 2035, according to court documents unsealed Wednesday .
backsC41 - S28primarycompany release2026-10-21Q3 revenue, costs, segment figures, cash flow, balance sheet, guidance and the Class C announcement
Excluding this charge, the third quarter 2026: Income from operations of $12.99 billion would have increased by $9.89 billion to $22.88 billion.
backsC04C05C06C08C09C10C11C18C42 - S29primarycompany filing2026-10-21the non-voting Class C stock, the one-for-one stock dividend, the Founder Alignment Agreement terms
Mr. Zuckerberg generally cannot obtain liquidity in a significant majority of the non-voting Class C common stock that he will receive in the Stock Dividend without a corresponding reduction in his voting power.
backsC31C42 - S30secondarynews2026-10-21the $25 billion New Mexico order, 43 million violations at $5,000, and Meta's statement that it will appeal
a judge ordered the social media giant to initiate its biggest payout yet: $25 billion to the state of New Mexico for the tech giant’s role in deceiving resident users over its data privacy practices
backsC18 - S31primarycompany filing2026-10-22non-cancelable commitments by year, leases not yet commenced, restricted escrow funds, the New Mexico order's interest and bond language, and the Class C dividend accounting
As of September 30, 2026, we had $ 454.52 billion of non-cancelable contractual commitments, comprising both short-term and long-term arrangements.
backsC09C18C42C43 - S32primarycompany filing2026-10-26250,000,000 depositary shares at a $50 liquidation preference each, a mandatory conversion into common stock on 15 August 2029, a 37,500,000-share option, the $22.5 billion of Class A stock sold through forward sale agreements with no proceeds at the start, and cornerstone interest from Nvidia, AMD and Broadcom (second 424B5, same date)
We intend to use the net proceeds from this offering for general corporate purposes.
backsC44 - S33primarycompany filing2026-10-26the financing plan of about $35 billion now and about $60 billion later, the stated funding mix, $97 billion of nine-month operating cash flow, about $82 billion of cash and the Aa3/AA- ratings, with no ATM agreement yet signed
The Company plans to launch an approximately $60 billion at-the-market (“ATM”) program upon the commencement of trading of the Company’s previously announced Class C common stock
backsC42C44 - S34primarycompany filing2026-10-27Meta Platforms Rule 433 pricing term sheet, concurrent offerings, 26 October 2026 (filed 27 October)the common offering upsized to $25.0 billion at $625.00 a share against a $630.56 last sale, settling 29 October, with the forward sale agreements held by Morgan Stanley and Goldman Sachs
Meta Platforms, Inc. has increased the size of the Common Stock Offering to $25.0 billion, consisting of 40,000,000 shares of Class A Common Stock
backsC44 - S35primarycompany release2026-10-26the 6.50% Series A mandatory convertible preferred at $50 a depositary share, converting into 0.0681 to 0.0800 Class A shares each on or about 15 August 2029, about $12.5 billion of net proceeds
The Common Stock Offering was upsized to $25.0 billion from the previously announced offering size of $22.5 billion.
backsC44