What it does · how the money flows · why it matters
Every day, 3.56 billion people — nearly half of humanity — open Facebook, Instagram, WhatsApp or Messenger. While they scroll, Meta's machines study what holds their attention, and auction that attention, billions of times a day, to ten million advertisers who bid for the predicted probability that a specific person will buy a specific thing. That sentence is the whole business: Meta does not sell ads so much as it sells predicted customers, and it does so at margins — 82% gross, 41% operating — that most industrial companies would mistake for a misprint.
In the year ended December 2025 this machine produced $201 billion of revenue (up 22%) and $60.5 billion of profit, and it is accelerating: the most recent quarter grew 33%, the fastest since 2021 (approx. — company release). Roughly 99% of the revenue comes from the "Family of Apps"; the other 1% — Reality Labs, the glasses-and-headsets laboratory — loses about $19 billion a year, a fact we will weigh honestly rather than wave away. The stock, at $681, costs about 24 times trailing earnings and 21 times next year's — for context, we recently paid 28× forward for Amazon and passed on Tesla at 204×.
"I got this wrong, and I take responsibility for that." — Mark Zuckerberg, layoff letter, November 2022
Keep that sentence nearby, because it separates this founder from most. In November 2022 this stock traded at $88 — the market had decided the metaverse spending was a bottomless pit and priced Instagram, WhatsApp and three billion daily users as if they were free. Zuckerberg apologised in writing, cut 21,000 jobs, declared a "Year of Efficiency," and the business promptly tripled its operating income in three years. The lesson for every reader of this shelf: when a great cash machine trades at a panic price because of an optional spending program, read the cash-flow statement, not the headlines. We will need that lesson again before this report is over.
| Founded | 2004 · a Harvard dorm room · IPO May 2012 at $38 |
| Sector / Industry | Communication Services · Internet Content & Information |
| CEO | Mark Zuckerberg — founder, ~61% of votes with ~13% of the equity (approx.) |
| Makes money from | Advertising (~98%) · business messaging & devices (small, growing) |
| Revenue (FY2025 · TTM) | $201.0 B · ~$215.0 B |
| Market capitalisation | ~$1.73 T · 3.56B daily users (approx.) |
Every crisis, the same arc: panic → pivot → monetise
Meta's history is one repeating pattern, and knowing it is worth more than any forecast: the market declares the business broken, the founder pivots, and the machine comes back bigger. It has now run four times.
| Year | Milestone |
|---|---|
| 2004–06 | 'TheFacebook' launches from a Harvard dorm. Two years later a 22-year-old Zuckerberg turns down Yahoo's ~$1B — the first sign he optimises for decades, not exits. |
| 2012 | IPO at $38 flops — the stock halves to ~$17.55 on fears Facebook can't monetise mobile. Pivot #1: rebuild mobile-first; mobile goes from ~0% of ad revenue to the vast majority within three years. Same year: Instagram bought for $1B — 13 employees, no revenue, widely mocked. |
| 2014 | WhatsApp bought for $19B (~$22B at close) — ~450M users, 55 employees. The monetisation switch waits twelve years (it is being flipped now). |
| 2018–19 | Cambridge Analytica: data on up to 87M users misused; congressional testimony; a record $5B FTC fine plus a privacy consent order. The reputational nadir. |
| 2021–22 | Apple's ATT privacy change vaporises ~$10B of revenue (CFO's own estimate); TikTok panic; metaverse rename and spending spook everyone. The stock falls 77% — $382 to $88.09 (Nov 2022) — pricing the apps as worthless. |
| 2022–23 | Pivot #3: 'I got this wrong' — 21,000 layoffs, the Year of Efficiency. The stock rises ~194% in 2023. Feb 2024: the company that swore it never would declares its first dividend; the market adds ~$196B in a day. |
| 2024–26 | The AI era: Llama makes Meta the open-source standard-bearer; Llama 4 stumbles (Apr 2025); pivot #4 — $14.3B for 49% of Scale AI, a 'Superintelligence Labs' under 28-year-old Alexandr Wang, $100M+ hiring packages, and (Apr 2026) 'Muse Spark', Meta's first closed frontier model. Capex guided to $125–145B, part-funded by a $30B bond. Q1 2026 revenue +33% — fastest since 2021. |
Two lessons for an owner. First, the two acquisitions may be the best in business history: Instagram, bought for $1B, is estimated to produce north of $80B a year in revenue (approx. — third-party estimates; Meta doesn't disclose); WhatsApp's $19B looked absurd for a decade and its business-messaging revenue is only now compounding past a $2B run-rate. Judging deals at announcement is a mug's game. Second, the founder self-corrects faster than any executive we track — mobile, Stories, Reels, efficiency, and now open-to-closed AI. That record is the bull case for tolerating his 61% voting control. The bear case is arithmetic: nobody can stop pivot number five if it is wrong, and the current one is the most expensive bet in the company's history.
How it works — and how confidently you can know its future
The machine, in five steps:
Buffett never owned this company — by the standard reconstruction, because he doubted anyone could project social-media behaviour ten years out. Fair enough in 2015. But note what changed: the question is no longer "will people still scroll?" (half of humanity does, daily, through every scandal) but "does AI capex pay?" — the exact question we already answered for Microsoft, Alphabet and Amazon. Meta is the same test with a better report card: its AI spending shows up in today's revenue line, not just tomorrow's promise.
One engine, 98.9% of the fuel · and a $19B laboratory
Meta reports two segments, and the honesty of the split is brutal:
Three engines inside the engine deserve a sentence each. Advantage+ and GEM: Meta's generative ad models — trained at frontier scale — automate campaigns and lifted conversions ~3–5% in a year (approx. — company engineering blog); this is where the AI capex already pays. WhatsApp: after twelve years, the $19B acquisition is being switched on — business messaging past a $2B annual run-rate and ads introduced into the Updates tab (~1.5B daily users) in 2025. Threads: 400M monthly users bootstrapped free off Instagram's graph, now carrying ads. And one structural beauty an accountant will appreciate: with no inventory and advertisers paying promptly while Meta pays its own bills slowly, the cash-conversion cycle runs at about −96 days — like Amazon, its counterparties finance the machine.
Four network effects — and a crisis that made them deeper
Meta's moat is filled from four springs, and — unusually on our shelf — the recent story is of widening:
Network effects, times four. Facebook (3B+ monthly), Instagram (2B+), WhatsApp (3B+), Messenger — plus Threads, spun up to 400M users at near-zero cost off Instagram's social graph, a live demonstration of the flywheel. No competitor has ever replicated a portfolio of interoperable graphs; users who tire of one app drift to another owned by the same landlord.
The data-and-AI flywheel — deepened by its own crisis. Apple's 2021 privacy change (ATT) tore away Meta's off-app tracking and cost ~$10B in a year. Watch what happened next: Meta rebuilt targeting on its own first-party signals with frontier-scale AI — and the smaller ad platforms that lacked the data and the capex never recovered. The wound became the moat. Today's +12% impressions / +9% pricing is the receipt.
The advertiser ROI machine. Ten million mostly-small advertisers, self-serve, measured to the dollar. Ad budgets go where returns are proven; outside China, only Google's auction rivals this one's depth.
Scale economics on AI. $100B+ a year of compute, amortised over 3.5 billion daily users — a per-user cost no challenger can approach. The same "capex as moat" logic we credited to Amazon and Alphabet.
And TikTok? The 2020–24 existential threat has been de-fanged: forced by US law into a joint venture under American ownership (January 2026), while Reels reached parity and monetises Western attention better. The irony is exquisite — TikTok's existence is also what saved Meta in court, when a federal judge ruled Meta holds no monopoly because TikTok and YouTube compete so fiercely (Part XI). Where it could crack: not at the core, but at the platform layer above it — Meta still rents its distribution from Apple's and Google's app stores, and if AI assistants become the new front door to the internet, Meta must own one or pay tolls again. That is what the glasses (7M+ units in 2025) and the superintelligence lab are really for. I score the moat a 9: wide, and — rare on this shelf — widening at the core.
One number pays for the other. For now.
Meta's capital spending was ~$39B in 2024, $72.2B in 2025 (approx.), and is guided to $125–145 billion for 2026 — nearly double last year, more than 2024 and 2025 combined, part of a pledged ~$600B of US AI infrastructure by 2028. Gigawatt-scale campuses (Prometheus in Ohio; Hyperion in Louisiana, expanded to 5GW just this week) are under construction. And the funding mix changed character: buybacks fell to zero in Q4 2025, a $30B bond was sold (the year's largest), a $27B off-balance-sheet vehicle finances one datacenter, and our own cash-flow data shows net debt issuance of +$27.4B in FY2025 — the first time Meta has built with borrowed money.
Split the bet in two, because the halves deserve different grades. The defensible half funds the ad models — GEM, Advantage+, the ranking systems — and its return is not a projection; it is visible in the current revenue line (+22% FY2025, +33% last quarter). This is the best-evidenced AI-capex payback among the four giants we have analysed. The moonshot half is Superintelligence Labs: $14.3B for 49% of Scale AI and its 28-year-old founder installed as Chief AI Officer, signing bonuses reported up to $100M+ ("missionaries will beat mercenaries," sniffed Sam Altman, whose researchers were being poached), the Llama 4 stumble — a departing chief scientist admitted benchmark results were "fudged a little bit" — and the April 2026 pivot from open-source Llama to the closed "Muse Spark" model. This half has no revenue line at all. It is faith in one man's judgment, financed, for the first time, partly with debt.
| The bear case | The bull case |
|---|---|
| Metaverse II at 5× scale — Reality Labs burned $80B+ with little to show; the superintelligence lab has no revenue line either | The ads payback is visible TODAY — +33% revenue last quarter; impressions and pricing both rising on AI targeting |
| The funding turned to debt — buybacks zero in Q4 2025, a $30B bond, an off-balance-sheet SPV; FCF fell to $46B | The balance sheet can afford it — net debt ~0.6× EBITDA, interest covered 51×, $82B of cash and investments |
| Org churn at the top of AI — chief scientist out, GenAI team dissolved, 'fudged' benchmarks on the record | The founder throttles failures fast — he cut the metaverse spend when it mattered; the pivot record is five-for-five |
| Depreciation from ~$600B of pledged plant compresses EPS mechanically, whatever demand does | Glasses are working — 7M+ units in 2025 (approx.) — the first credible post-smartphone distribution Meta has ever owned |
Our framing, consistent with the rest of this shelf: at ~21× forward earnings, you are paying for the machine — the moonshot is thrown in for nothing. If it works, it is a free call option on the next platform. If it fails, the founder's own record says he will cut it — he has done exactly that once already, in writing, with an apology. The scenario that hurts is the middle one: a decade of $100B+ spending that neither pays nor gets cut, eroding margins one depreciation schedule at a time. Watch one line each quarter: capex guidance against ad-revenue growth. The day the second stops covering the first, this thesis changes.
The duopoly's quieter half — with the bigger open flank
| Arena | Who | Where Meta stands |
|---|---|---|
| Social attention | TikTok (US JV, Jan 2026) · YouTube Shorts · Snap | De-fanged threat; Reels at parity, monetising better |
| Digital advertising | Google ~$295B · Amazon ~$69B ads | #2 of the big three; together ~62% of global digital ads (approx.) |
| Business messaging | Apple iMessage · Telegram · RCS | WhatsApp's 3B users dwarf all — monetisation just beginning |
| AI assistants / next platform | OpenAI · Gemini · Apple · Anthropic | The real fight — Meta AI has reach, not yet preference |
| Distribution (the old scar) | Apple & Google app stores | Still a tenant; the glasses are the escape plan |
Read the table against our Alphabet report and the shape is striking: Google's moat is contested at its front door (AI answers versus ten blue links), while Meta's front door is safer than it has been in six years — the TikTok war ended in a negotiated draw, and nobody disrupts habit-based social graphs by being slightly better. Meta's open flank is one level up: it has never owned the device or the operating system its apps live on. Apple proved the cost of tenancy in 2021 (~$10B, one privacy toggle). Everything strange about Meta today — the glasses, the assistant, the superintelligence rhetoric — is one strategy: never rent the front door again.
A founder-king with a five-for-five pivot record
One person controls this company, so the management section is really a character study with a balance sheet attached.
Capital allocation — read the sequence, it is eloquent. For its first twenty years Meta returned nothing and diluted nobody's attention with dividends. Then, in the efficiency era: buybacks at scale ($44.5B in 2021, $28B in 2022 — much of it near the lows, our cash-flow data shows) that shrank the share count 13.5% over the decade — the only AI hyperscaler on our board whose share count went down while building. February 2024: the first dividend, now $2.10/year (a 0.3% yield, 7.6% of earnings — a token, but tokens signal). Then 2025's counter-signal: buybacks paused to zero in Q4, $27.4B of net new debt, everything redirected to the AI build. Management is telling you, in the universal language of cash, that it believes the compute earns more than the stock. They were right about that once before, at $88. The honest caveats: stock compensation runs at a rich 10.4% of revenue, the Llama-benchmark episode ("fudged a little bit," per the departing chief scientist) is a genuine integrity scuff, and the political repositioning of 2025 — fact-checking ended, policy chiefs swapped — shows a company that trims its sails to every administration. Seven out of ten: brilliant, self-correcting, unaccountable.
A record-margin machine — with three honest asterisks
| Metric | Value | Read |
|---|---|---|
| Revenue (FY2016 → FY2025) | $27.6B → $201.0B | ▲ ~25%/yr for a decade; +22% in FY2025 itself |
| Operating income (FY2022 → FY2025) | $28.9B → $83.3B | ▲ ~2.9× off the 2022 trough — the efficiency dividend |
| Gross margin · operating margin | 82% · 41.4% | ▲ software-royalty economics |
| Return on equity · invested capital | 33.2% · 20.0% | ▲ elite — despite the capex drag |
| R&D / revenue | 29.3% | ◆ $57B a year — half machine, half moonshot |
| Capex / revenue (TTM) | 35.2% | ▼ the heaviest ratio on our board (MSFT ~30%, GOOGL ~26%, AMZN ~20%) |
| Free cash flow (FY2024 → FY2025) | $54.1B → $46.1B | ◆ falling as the build accelerates |
| Balance sheet | Net debt ~0.6× EBITDA · Altman-Z 8.6 | ◆ fortress-adjacent — but $27.4B of net debt ADDED in FY2025 |
| Stock-based compensation / revenue | 10.4% | ▼ $20.4B/yr — a real cost, honestly counted |
| Shares outstanding (decade) | 2.92B → 2.52B | ▲ −13.5% — buybacks beat the SBC |
| Cash conversion cycle | −96 days | ▲ counterparties finance the machine |
Three asterisks an honest analyst must attach. One: reported EPS was flat from 2024 ($23.86) to 2025 ($23.49) — but that is a tax illusion: the 2024 rate was an unusually low 12% and 2025 normalised to 30%, while operating income grew 20%. Judge the operating line. Two: Reality Labs' ~$19.2B annual loss is inside these margins — the Family of Apps alone earns even more absurdly than the consolidated numbers show. Three: Piotroski scores this a 4 — not because the business weakened, but because the financing changed shape (new debt, thinner FCF). That is the moonshot's fingerprint on an otherwise pristine set of accounts.
The cheapest growth-per-dollar among the giants — if the capex behaves
| Yardstick | Today | Forward | Read |
|---|---|---|---|
| P/E — reported earnings | ~24x | ~21x (FY26) · ~17x (FY28) · ~12x (FY30) | modest for 20%+ growth |
| P / Free cash flow | ~36x | — | capex-distorted — the familiar pattern |
| EV / EBITDA | 15.8x | — | reasonable |
| Price / Sales | 8.0x | — | for 33% net margins — fair |
| Dividend | $2.10 · ~0.3% | payout 7.6% | a token with room to grow 10× before it strains |
The pattern you now know by heart: the automated DCF says $273 (−60%) because it treats a $125–145B construction year as a permanent condition — the same trap that said Alphabet was worth $132 and Amazon $75. Ignore it for the same reason. The anchor is the earnings path: $32.91 next year (21×), $40.22 in 2028 (17×), $56.43 in 2030 (12×) — consensus of 34–44 analysts. Set that beside its peers on this shelf: Amazon cost us ~13× its 2030 estimate with FCF negative; Alphabet ~27× current earnings with its front door contested; Tesla wanted 43× a 2030 dream. Meta at ~12× the 2030 consensus, with margins at records, the core moat widening, and both ad levers rising, is — per unit of growth — the cheapest quality asset among the giants we have analysed. What the discount buys you is the risk shelf: the courtroom in August, the debt-financed moonshot, the one-man governance. Margin of safety at $681: modest but real against the machine alone; the buy zone deepens toward ~$620 (19× forward) — and remember what this shelf teaches: this stock visits panic prices roughly twice a decade, and the last one was a 90% off sale.
An August courtroom, an appealed acquittal, and a debt-funded bet — verified July 2026
Verified the week of publication. The one to watch is August: California, Colorado, Kentucky and New Jersey go to trial in Oakland (Judge Yvonne Gonzalez Rogers) alleging Instagram and Facebook were engineered to addict minors — and Meta's own filing says the states seek up to $1.4 trillion in penalties, "a sanction of that size has no analog in the history of consumer protection enforcement." Context cuts both ways: a New Mexico jury already awarded $375M on similar claims (March 2026), TikTok and Snap settled rather than fight, and Meta is the last major platform to take a jury verdict risk — but trillion-dollar demands are negotiating positions, not verdicts, and headline risk into August is guaranteed either way. On antitrust, the news is better than the headlines: Judge Boasberg ruled in November 2025 that Meta is not a monopolist (TikTok and YouTube being "fierce" substitutes) — no Instagram or WhatsApp divestiture — though the FTC filed its appeal in January 2026 and the D.C. Circuit will likely hear arguments this autumn. In Europe, the €200M pay-or-consent fine is paid and the January 2026 "less-personalised ads" option has been received as broadly compliant — the residual cost is quieter: thinner ad targeting on a continent that supplies ~23% of revenue. The remaining pills are the ones this report already priced: the debt-assisted moonshot, and a governance structure in which your only rights are economic.
On the fourth of November, 2022, the market offered one of the great businesses of the age for eighty-eight dollars a share — roughly eight times its earnings, with Instagram, WhatsApp and three billion daily customers priced, in effect, at zero. I did not buy it, and neither did most professionals, because the headlines said the founder had lost his mind in a virtual world wearing ski goggles. The people who simply read the cash-flow statement — who noticed that a money machine with an optional spending problem is still a money machine — made five times their capital in three years. I begin with that confession because this letter's job is to decide whether today's version of the same argument still holds at six hundred and eighty-one dollars.
The machine first, because the machine is the investment. Nearly half of humanity opens one of this company's four applications every day, out of pure habit, through every scandal and every administration. Their attention is auctioned to ten million advertisers with a precision that produced, last year, two hundred and one billion dollars of revenue at forty-one percent operating margins and thirty-three percent returns on equity — and it is accelerating, not maturing: the latest quarter grew thirty-three percent, the fastest in five years, because the company's AI spending shows up directly in its ad prices. Unlike its neighbour Alphabet, whose front door is under genuine siege, Meta's core has not been safer in six years — the TikTok war ended in a regulated draw, and the crisis that was supposed to kill it (Apple's privacy change) forced it to build targeting machinery its smaller rivals could never afford. The moat did not merely survive; it got deeper at the bottom.
Now the rubs, and they are three. First, the founder is spending a hundred and twenty-five to a hundred and forty-five billion dollars this year — part of it, for the first time, borrowed — on a "superintelligence" laboratory with no revenue line, staffed by hundred-million-dollar hires, after an episode in which his own departing chief scientist said benchmark results were "fudged a little bit." The buyback stopped cold last quarter to pay for it. I have watched this man burn eighty billion dollars on a virtual world, and I have also watched him apologise in writing and cut it — the only founder of his generation with a documented, five-for-five record of killing his own mistakes. Second, you have no say: sixty-one percent of the votes attach to thirteen percent of the economics, and every shareholder proposal this year died on arrival. Third — and nearest — a courtroom in Oakland this August, where four states will tell a jury this company engineered its products to addict children, and demand up to one-point-four trillion dollars. That number is an opening position, not a verdict; a New Mexico jury settled for 375 million on the same theory, and every other platform paid to stay out of the room. But headlines will be ugly, and juries are juries.
Against those rubs, the price. Twenty-four times trailing earnings; twenty-one times next year's; twelve times the 2030 consensus — for the fastest-growing, highest-margin franchise among the four giants on this shelf. We paid thirteen times 2030 for Amazon with its cash flow underwater; we called Alphabet fair at twenty-seven times with its moat contested; we passed on Tesla at forty-three times a dream. By the only yardstick that lets you compare them — price per unit of durable earning power — this is the cheapest quality asset we have examined this year. The discount exists because of August, because of the debt, because of the ski goggles. In other words: you are being paid, modestly but genuinely, to hold the risks this report just spent eleven parts measuring.
So the decision: buy the machine. At today's price you are paying a fair-to-attractive multiple for the advertising engine alone — the courtroom discount is your margin of safety, and the superintelligence moonshot comes with the ticket for approximately nothing. Take a position here, sized for a headline-heavy autumn; add with real conviction toward six hundred and twenty dollars, and if August produces one of this stock's periodic panics — it has halved twice in fourteen years and made new highs within thirty months both times — remember the eighty-eight-dollar lesson and act like you remember it. Watch one number quarterly, capex guidance against ad growth, and one date, the trial. The company that swore it would never pay a dividend now pays one; the founder who lost the plot found it again within a year. I have stopped betting against people who correct their mistakes faster than I correct mine.