X-Ray Analyses›Communication Services›Meta Platforms, Inc.
X-RayNº 11422 November 2026
◆ Follow-upSecond X-Ray of Meta Platforms, Inc.. Our previous, dated 15 Jul 2026 at $681.31, concluded “Buy the Machine”. This report checks that call against what has happened since. Read it →
X-Ray Analysis◆The Buffett Lens
M

Meta Platforms, Inc.

NASDAQ: META·Internet Content & Information·United States·Explore META live ↗

The machine held, the free ride ended

◆ The Buffett LensIn July we told readers to buy Meta at $681.31 because the advertising machine paid for everything else: the courtroom risk was a discount, and the AI moonshot rode free. Four months later the machine is still the best of its kind, with revenue up 26% and the price of an advertisement up 14%. But both reasons for conviction have failed. The courts sent bills, and owners are now paying for the moonshot in new shares, preferred stock and bonds: about $66 billion raised since July, with free cash flow below zero. Our July report also got six facts wrong. Hold what you own. Add below $480.
◆ Educational analysis & opinion — not investment advice. Figures as of 22 November 2026. See full disclaimer below.
9
Moat
5
Management & Capital
6
Financial Strength
7
Growth
6
Valuation
6.6
◆ The Scorecard · one-second read
"The machine paid for the moonshot until July. Since then the owners have."
−17.7% since July · revenue +26%, price per ad +14% · capex $135–145bn, Q3 free cash flow −$2.1bn · ~$66bn raised since July · $455bn of commitments · up to $18.6bn to the states, $25bn in New Mexico (appeal) · 16.4× FY27
◆ Type · Compounder in a capital-spending phaseDividend · $0.525 a quarter — 0.4%, frozen in 2026★ Follow-up · six July errors corrected; level $540 → $480
Price & total return · live
The Tape, since we wrote
The gold pin marks the price we used. Daily closes, live.
—
Drag across the chart to measure
Live price history is momentarily unavailable · price at this X-Ray: $560.84.
Go deeper — the Trading Desk, 15 years of financials, DCF & peers for METAOpen META →
◆
Part
I

What we said in July — and what happened

The machine held; the two reasons for conviction did not; six facts were wrong

On 15 July we published Meta at $681.31 under the title "Buy the Machine". The argument had two halves. The advertising engine, we said, was the best ever built and was being sold at a fair-to-attractive price. And the two things scaring investors were really gifts: the courtroom risk was the margin of safety, and the enormous AI spending was a moonshot that "rides free" on the engine's cash. We said to buy, and to add with conviction toward $620. The shares closed below $620 eight days later and touched $539 on 30 July. They rose to $777.59 in September and closed at $560.84 on 20 November. On 17 November we wrote a postscript lowering the verdict to "Hold the Machine — Add Near $540". This report is the full account.

What we said on 15 JulWhat happened by 20 NovGrade
'Buy the machine' at $681.31; add toward $620−17.7%. Below $620 by 23 July; a reader who added there has lost about 10%.✗ Wrong so far
The advertising engine is the best ever builtRevenue +28% in Q2 and +26% in Q3; price per ad +14% in Q3 with impressions +10%.✓ Right
'The courtroom discount is the margin of safety'Up to $18.59bn agreed with 52 attorneys general (a $9.89bn charge in Q3); a $25bn penalty in New Mexico, under appeal.✗ Wrong
'The moonshot rides free'Free cash flow −$2.1bn in Q3; about $66bn raised since July in shares, preferred and bonds; no buybacks in 2026.✗ Wrong
Capex 2026 $125–145bn; 'watch capex against ad growth'Guide raised to $130–145bn, then $135–145bn; revenue growth slowing from 33% to 26%.◆ The signal flashed
'Net debt ~0.6× EBITDA'Wrong when written. In March Meta had about $22bn of net cash ($81.2bn of cash and securities against $58.8bn of debt).✗ Wrong
The $30bn bond was 'the first time Meta has built with borrowed money'Wrong. Meta issued notes in 2022, 2023 and 2024. We also left out the $25bn of notes issued in May 2026, two months before our report.✗ Wrong
Flat 2025 EPS was 'a tax illusion', the rate 'normalised to 30%'The 2025 rate was abnormally high, inflated by a one-off $15.93bn tax charge; without it, about 13%. The conclusion (judge operating income) stands.✗ Wrong reasoning
Free cash flow '$54.1bn → $46.1bn'Our feed's definition. Meta's own, after finance-lease payments: $52.1bn → $43.6bn.✗ Wrong figure
'About 21× next year's earnings'The estimate we used ($32.91) was for 2026, the current year. 'Next year' was 2027.✗ Mislabelled
New Mexico 'settled for $375 million' (the Letter)It was a jury verdict, not a settlement. And YouTube, not only Meta, faced the March jury in Los Angeles.✗ Wrong

The errors, plainly. Six facts in our July report were wrong when we wrote them. Three came from trusting our data feed instead of the filings: the net-debt ratio, the free-cash-flow definition and the share count. Three came from careless reading: Meta had borrowed before, the 30% tax rate was the exception rather than the norm, and a jury verdict is not a settlement. Two of them made Meta look weaker than it was in July, which is the opposite of the usual bias. None of them would have changed the call. What changed the call is what happened after July, and the next three parts are about that. We have also corrected two numbers in our 17 November postscript: the forward share sale is about $28.7bn, not $25bn, because the underwriters took up their option.

Our July price → today
$681.31 → $560.84
−17.7%; the high in between was $777.59
Raised since July
~$66bn
shares, preferred and bonds
Our level
$620 → $540 → $480
July · postscript · today
◆
Part
II

★ Two round trips in four months

$681 to $539, up to $778, down to $561

Two round trips in four monthsMETA daily close, $, 1 Jul – 20 Nov 2026.500600700800July add level $620postscript $540our new level, $48012345678JulAugSepOctNov115 Jul · our X-Ray, $681.31230 Jul · Q2 results, −8.0%, $539.03326 Aug · settlement with the states421 Sep · Muse traction, +11.3%524 Sep · high, $777.59622 Oct · Q3 results, −7.0%729 Oct · forward share sale + preferred820 Nov · market sell-off, −5.9%, $560.84★ Our $620 was reached eight days after the X-Ray. Buying there has lost ~10%.Sources: our price history; Meta 8-Ks; press reports for the 21 Sep and 20 Nov moves.
DateWhat happenedMETA that day
30 JulQ2 results: earnings per share of $6.18 against about $7.19 expected, after $2.40bn of legal charges and $1.18bn of severance for about 8,000 roles.−8.0%
26 AugSettlement with the states over young users, filed and approved the same day (CNBC).+1.1%
9–24 SepLaunch of Muse, Meta's AI agent; the shares rise from about $614 to a high of $777.59, including +11.3% on 21 September (reported causes: early Muse traction and a broker upgrade).+27%
22 OctQ3 results: revenue ahead, but free cash flow negative, a $9.89bn legal charge, higher 2026 costs and the announcement of a non-voting share class.−7.0%
26–29 OctForward sale of 40m shares at $625 (later 46m) and $12.5bn of mandatory convertible preferred stock.$630 → $612
19–20 NovNo Meta news. Anthropic's listing week; Nvidia −14% on 20 November after a press report; the Nasdaq −5.3% and its worst week since March 2020 (CNBC).−3.3%, −5.9%

Two kinds of move. The falls on 30 July and 22 October were about Meta: results that showed the cost of the courts and of the build. The rally in September was also about Meta, and about the hope that Muse would turn the AI spending into a product. The last two days, −9.1% together, were about the market: nothing Meta said or did changed between 18 and 20 November. An owner should weigh the first kind and largely ignore the second.

The lesson for us is about our own level. We said to add "with conviction toward $620", and the price went through it within eight days. A level is only useful if it reflects a margin of safety, and $620 did not: it was about 19 times the earnings of the year we were in, for a company about to spend more than it earned. We set it too close to the price, and we will not do that again in this report.

◆
Part
III

★★ The bill arrives before the return

Capital spending has overtaken the cash the business produces

The bill arrives before the returnQuarterly, $ billions, 2026. Capex includes finance leases; Q4 = guidance mid-point.$19.84bn$12.39bnQ1revenue +33%$31.08bn$0.78bnQ2revenue +28%$34.91bn−$2.1bnQ3revenue +26%$49–59bnQ4 guiderevenue +22–27%Capex incl. finance leasesFree cash flow (company definition)★ 2026 capex guided at $135–145bn; 2027 'aggressive', with no number yet.Sources: Meta releases and 10-Qs, Q1–Q3 2026; Q1 free cash flow derived from the six-month figure.

In the third quarter Meta spent $34.9bn on data centres, servers and the equipment it leases, and generated −$2.1bn of free cash flow on its own definition. The full-year guide of $135–145bn implies $49–59bn in the fourth quarter alone, more than the company spent in the whole of 2024. For 2027 there is no number yet; the company said in October that it is "still developing" its budget and expects "to continue investing aggressively". The figure will come with the fourth-quarter results in late January, and it is the most important number of the year for an owner.

Over the last twelve months, capital spending including finance leases was about $108bn, some 45% of revenue, against company free cash flow of about $25bn. In July we wrote that the advertising engine paid for everything else. In the first quarter it did, with $12bn to spare. By the third it no longer did. The machine has not weakened; the spending has outgrown it.

Q3 2026Valuevs a year earlier
Revenue$64.60bn+26%
Ad impressions / price per ad+10% / +14%Q2: +14% / +12%
Family daily active people3.64bn+3%
Operating margin, reported / without the legal charge20% / 35%—
Reality Labs operating loss−$4.13bn2026: −$18–19bn; 2027: ~−$15bn (company)
Capex incl. finance leases / free cash flow$34.9bn / −$2.1bn—
Stock-based compensation$7.36bnNine months +45%
Q4 guide$73–76bn revenue+22–27%

Where the money goes. Most of the capital spending is for Meta Superintelligence Labs and for the AI systems that rank feeds and advertisements. The second use already pays: much of the rise in the price per advertisement reflects better targeting and better ad formats, which are AI products. The first use, the frontier models and the Muse agent, has no revenue line yet; Muse had more than ten million accounts by October, according to AP. The company itself warned in its third-quarter release that these investments "will take longer to generate returns than those in our core business". That is an honest sentence, and an owner should take it at its word.

◆
Part
IV

★★ How the build is being paid for

About $66bn raised in four months, and far more signed off the balance sheet

How the build is being paid for$ billions. Raised since May 2026 (top) and signed but off the balance sheet (bottom).Notes, May 2026$25bnForward sale of 46m shares, Oct$28.7bnMandatory convertible preferred, Oct$12.5bnDollar notes, Nov$15bnEuro notes, Nov (€9bn)$10.4bnTotal raised in six months: about $92bn, of which ~$66bn since the July X-Ray.Signed, not yet on the balance sheetDifferent scale. The first two may overlap.Leases not yet started (2026–36)$355.7bnNon-cancellable commitments$454.5bnResidual value guarantees (max)$41.5bn★ Net debt went from ~$22bn of net cash in March to about zero in September.Sources: Meta 10-Q (Q3 2026, notes 5, 9, 10), 8-Ks of 29 Oct, 5 and 12 Nov 2026; euro converted at ~1.16.

When a company spends more than it earns, someone else pays the difference. Since our July report Meta has raised about $66bn. In late October it agreed to sell 46 million new shares forward at $625 to the banks, about $28.7bn, which it can settle at any time until November 2027. It issued $12.5bn of mandatory convertible preferred stock with a 6.5% dividend, which converts into common shares in 2029 at between $625 and $734. In November it borrowed $15bn in dollars and, for the first time, €9bn in euros, at coupons of 4.3–7.5%, well above what it paid on its earlier bonds. Add the $25bn of notes issued in May, and the company has raised about $92bn in six months.

The balance sheet has changed accordingly. In March Meta had about $22bn more cash and securities than debt. By the end of September the two were roughly equal: $82.5bn of cash and securities against $83.7bn of notes, before the November bonds. Off the balance sheet, Meta has signed $355.7bn of data-centre and energy leases that have not yet started, has $454.5bn of non-cancellable commitments (the two may overlap), and guarantees up to $41.5bn of the residual value of data centres built by joint ventures in Louisiana and Texas. A further $10.8bn sits in escrow for equipment purchases.

Instrument (since July)SizeWhat an owner should know
Forward sale of Class A shares46m shares, ~$28.7bnCash arrives only on settlement, by 19 November 2027; about 1.8% more shares.
6.50% mandatory convertible preferred$12.5bn~$0.8bn a year of preferred dividends from February 2027; if they are missed, common dividends and buybacks are blocked.
Dollar notes (5 Nov)$15.0bn5.85% to 7.50%, maturing 2031–2056.
Euro notes (12 Nov)€9.0bnMeta's first euro issue, 4.30% to 6.40%.
Texas data-centre venture (July)$14bn projectMeta owns 20%, leases the site from 2028 and guarantees part of its value.

Is this dangerous? Not yet. Meta's operating profit before the legal charge was about $23bn in the third quarter alone, interest costs are small against it. But three things have changed since July. Owners are being diluted again, after years of buybacks; there have been none in 2026, and the share count rose 1.4% in a year. The company has committed to obligations that run for a decade or more whatever happens to advertising. And it has said, in its information statement to shareholders, that it "expects to use shares of Class C common stock to conduct equity offerings from time to time". The new share class (Part VIII) is, among other things, a financing currency.

◆
Part
V

The courts sent their bills

What was a discount in July is a cost in November

In July we described the youth-safety cases as a risk the market was overpricing. Here is what they cost. In August Meta agreed with 52 state and territorial attorneys general to pay up to $18.59bn: $12.7bn in ten annual instalments, about $0.5bn for privacy claims and $5.3bn that is contingent on rivals adopting similar terms and has not been accrued. Meta took a $9.89bn charge in the third quarter and agreed to changes in its products for users under 18, with an independent auditor. Florida, New Mexico and the US Virgin Islands are not part of it.

New Mexico went to trial. On 25 September a jury found more than 43 million violations of the state's unfair-practices law, and on 21 October the court imposed a civil penalty of $25bn, plus 15% a year of post-judgment interest. Meta says it will appeal and may have to post a bond. Its legal accruals rose from $7.3bn at the end of 2025 to $19.9bn in September; our reading of the third-quarter numbers is that the New Mexico accrual is far below $25bn, so an adverse appeal would mean a further large charge. All claims about harm to minors remain allegations except where a jury or court has ruled, and those rulings are under appeal or stayed.

MatterStatus, 22 November 2026Read
52 attorneys general (youth)Up to $18.59bn, approved 26 August; $5.3bn contingent.◆ Paid over ten years
New Mexico (unfair practices)$25bn penalty + 15% interest (21 Oct); appeal and bond to come.✗ The largest open item
New Mexico (youth)$375m jury penalty (March) and ~$567m abatement order (August); judgment stayed.◆ Stayed
Los Angeles personal-injury trials (JCCP)March verdict $6m (70% Meta, 30% YouTube), on appeal; a second trial under way, no verdict found; more trials in 2027.◆ Open
FTC v. Meta (Instagram, WhatsApp)Meta won in November 2025; the FTC's appeal (D.C. Cir. No. 26-5028) has no argument date we could find.✓ Low risk, slow
EU (DMA, DSA, WhatsApp AI access)€200m DMA fine under appeal; preliminary DSA findings on minors (fines up to 6% of turnover); an interim order in June to open WhatsApp to rival AI chatbots.◆ Open
◆
Part
VI

The moat, rechecked

Still the deepest in advertising; the question is what it is being asked to carry

The claim (July)Evidence nowVerdict
Network effects across four apps3.64bn daily people, +3%; WhatsApp business messaging and other revenue +67% in nine months.✓ Intact
Pricing power from targetingPrice per ad +14% in Q3 with impressions +10%: advertisers pay more for each ad.✓ Stronger
Scale in AI ranking and ad toolsThe part of the AI spending that already pays; Alphabet's revenue +25% shows the duopoly holding share.✓ Intact
Regulation as a manageable costSettlements, penalties and product changes for minors; EU orders on data and WhatsApp access.◆ Costlier

We keep the moat at 9. The strongest primary evidence in four months is that advertisers paid 14% more per advertisement in the third quarter while Meta showed only 10% more of them. A business that can raise its price faster than its volume, at this scale, has a moat. What has changed is not the moat but what the moat is being asked to finance: a laboratory, a decade of data-centre leases and a share class built for issuing equity. A wide moat around a castle that keeps borrowing to build towers is still a wide moat. It is no longer a margin of safety by itself.

How knowable is the next ten years?
3/5 — the advertising is knowable; the return on $455bn of commitments is not. We can describe what Meta sells and why advertisers stay. We cannot know what the superintelligence spending will earn, or when, and the company says it will take longer than the core.
What you must believe to buy at $560.84
  • That earnings per share grow about 11% a year from the 2027 consensus of $34.14 to 2030, despite the depreciation of the data centres now being built and about 2.6% of new shares.
  • That the courts' bills stay within what the cash flow can carry: the states' settlement over ten years, and a New Mexico appeal that reduces or spreads the $25bn.
  • That the 2027 capex guide in January shows a peak in sight, and free cash flow turns positive again in 2027.
◆
Part
VII

The competition

Google spending even harder, TikTok under new owners, assistants at the door

RivalWhere it meets MetaThe latestRead
AlphabetAdvertising and AI models.Q3 revenue +25%; capex $51.4bn and free cash flow −$9.1bn in the quarter; 2027 capex to 'increase significantly' (as reported).◆ Same bill, larger
TikTok (US joint venture)Time spent by younger users.US joint venture since January (Oracle, Silver Lake, MGX); Meta's 10-Q still names TikTok as reducing some users' engagement.◆ Steady pressure
AI assistants (OpenAI, Apple, others)Where people ask, shop and talk.Rival agents launched or planned; the EU's interim order forces WhatsApp to admit rival chatbots.◆ The new flank
TalentAI researchers.Options on 20m shares granted in 2026 at an average strike of $2,788; reported departures to rival labs.◆ Costly

Across our board. Alphabet's third quarter is the useful comparison. It too grew advertising strongly, and it too reported negative free cash flow in the quarter, after even larger capital spending. The duopoly is not losing to anyone; it is spending its cash on the same race. For an owner that is reassuring and sobering at once: Meta is not being outcompeted, but it cannot slow down alone without risking its position. The moat holds. The race is what costs.

TikTok, the competitor we worried most about in July, has become less of a political story and more of an ordinary one: a US joint venture retraining its recommendation system on American data. It remains the main reason Meta's daily users grow only 3% a year. The newer flank is assistants. If people begin to ask an AI agent what to buy instead of scrolling a feed, the advertising has to follow them there. Muse is Meta's answer; it is too early to judge it.

◆
Part
VIII

Management, control & capital allocation

A non-voting share class, options at $2,788, and insiders on the sell side

M
Mark Zuckerberg · Founder, Chair and Chief Executive
About 60.6% of the votes on about 13.4% of the shares. Signed a Founder Alignment Agreement in October; sold 27,474 shares (~$21.4m) on 24 September through charitable entities under a trading plan adopted in January.
S
Susan Li · Chief Financial Officer
Sold about $13.6m of shares since July, including $5.7m on 17 November at $620–626. No purchases.
C
Chris Cox · Chief Product Officer
Sold 80,000 shares, about $55m, between 9 and 21 September at $650–725, under a plan adopted in May.

The new share class. On 20 October the founder's entities approved by written consent a new class of non-voting Class C shares. Every holder will receive one Class C share for each Class A or B share on or about 19 January 2027, which works like a two-for-one split. Meta gave two purposes in its information statement: to let the founder sell shares without losing control (California's proposed wealth tax was cited; it was defeated in November, as reported), and to issue equity in future without diluting the vote. Under the accompanying agreement, the founder receives further one-for-one Class C dividends each time the share price doubles from a reference level, if returns reach 10% a year, and commits to eight more years at the company. The board's special committee expects Class C shares to "trade at some level of discount".

Capital allocation. We lower the score from 7 to 5. The record from 2023 to 2025, when Meta cut costs and bought back shares, earned the 7. In 2026 there have been no buybacks; $25bn of authorisation sits unused. Stock-based pay is running at about 11% of revenue, with $75.6bn of unrecognised pay still to come, and the stock plan now lets the company reprice underwater options without shareholder approval; this year's options for 20 million shares carry an average strike price of $2,788, about five times today's price. A new president received a $60m grant on joining. A company that issues new shares, preferred stock and bonds in the same quarter in which it removes shareholders' say over repricing has stopped treating its share capital as scarce.

Insiders. Since July there has been no open-market purchase by any director or officer, and about $134m of sales, most of it under pre-arranged plans: the founder, the product chief, the chief operating officer, the finance chief, the technology chief and a director. Sales under plans tell an owner little. The absence of a single purchase at $540 in July or at $560 now tells a little more.

Who owns it
★ The founder
60.6% of votes
About 13.4% of the economic interest (DEF 14C, October). Our feed still shows a 2024 filing at 13.6%.
Insiders
~$134m sold, none bought
Since 15 July, across 56 Form 4 filings; the rest are tax withholdings and transfers.
Analysts
Mean target ~$835
Our feed: median $850, range $595–1,000; 56 of 69 at buy.
The company
$0 of buybacks in 2026
$25.0bn of authorisation unused; basic share count +1.4% in a year.
◆
Part
IX

The numbers

Earnings intact, cash flow absorbed, and six faults in our feed

Two pictures of the same company. On earnings, Meta looks reasonably priced. Reported earnings per share over the last four quarters were $29.32, which puts the shares at 19.1 times; on the 2027 consensus of $34.14 they trade at 16.4 times, below their five-year average of about 23 times. Operating profit before the legal charges and severance was about $92.9bn over twelve months. On cash, Meta looks expensive. Company free cash flow over the same period was about $25bn, which puts the shares at about 57 times, the highest in the five years we checked. The gap between the two pictures is the capital spending, which is depreciated slowly in the earnings but paid at once in cash.

Which picture is right depends on whether the data centres earn a return. If they do, the earnings are the better guide and the shares are fair to cheap. If they do not, the cash is the better guide and the earnings will fall as the depreciation arrives. The market is splitting the difference: on the 2027 consensus, the earnings yield of about 6.1% implies only modest growth, while the cash yield of 1.75% would need a great deal. That is a reasonable place for the price to sit while the question is open.

★ Six faults in our feed — read these before trusting a screen
What the screen showsWhat is wrong, and the number to use
Net debt $100.9bn; total debt $112.1bnCounts operating leases as debt and ignores $71.2bn of securities. Debt is <b>$83.7bn</b>, cash and securities $82.5bn: net debt about <b>$1.2bn</b> (30 Sep).
Free cash flow FY2025 $46.1bn; Q3 −$1.16bnAdds back finance-lease payments. Company figures: <b>$43.6bn</b> and <b>−$2.12bn</b>. Our July report inherited this fault.
Non-operating income, Q3: −$13mSign flipped: interest and other income was <b>+$13m</b>. The same fault appears in our feed for Salesforce and Broadcom.
Product and region revenue, FY2024–25Rows duplicated (FY2024 product total $235.2bn). Actual revenue: <b>$164.5bn</b> (2024) and <b>$201.0bn</b> (2025).
Enterprise value snapshotDated 31 December 2025. At $560.84 the market value is about <b>$1.43tn</b> and the enterprise value, with operating leases, about $1.47tn.
Dividend summary: no paymentsEmpty, while the raw list shows four payments of $0.525. Trailing dividend: <b>$2.10</b>.
◆
Part
X

The dividend

Small, covered, frozen — tested

Meta pays $0.525 a quarter, $2.10 a year, a yield of 0.4%. It began paying in February 2024, raised the rate by 5% in March 2025 and has not raised it in 2026. The next declaration is due in early December. Two new features matter. From February 2027 the preferred shares receive about $0.8bn a year ahead of common shareholders, and if those dividends are ever missed, common dividends must stop. And the forward share sale lets the banks accelerate settlement if Meta pays a dividend above a specified amount, a soft limit on a large raise. After the Class C distribution in January, the dividend per share should halve mechanically, with the total unchanged; the board has not yet said so.

★ Proving the dividend — six tests
TestEvidenceResult
1. Cover on free cash flow$5.40bn paid over twelve months against ~$25bn of free cash flow: about 21%.Pass
2. The trend of that coverWorsening: free cash flow was negative in Q3 as capex rose.◆ Weakening
3. From operations or from paper?Over the year, from operations; in Q3, the dividend was paid while the company was raising capital.◆ Mixed
4. Balance-sheet roomNet debt about zero before the November bonds; $455bn of commitments; preferred dividends rank first from 2027.◆ Adequate
5. What would force a cutMissed preferred dividends, or a sustained collapse in advertising while capex is committed.Unlikely
6. The growth rate and its direction+5% in 2025, nothing in 2026.◆ Frozen

The dividend is safe. It is also beside the point for an owner of Meta, and the company is treating it that way: a token payment, held flat while the money goes into data centres. We would not buy Meta for its income, and nobody should.

◆
Part
XI

★ Valuation

A fair price for the machine; a margin of safety needs $480

Reported EPS, 12 months
$29.32
19.1×, after $9.89bn of legal charges and a tax benefit.
2027 consensus EPS ◆
$34.14
16.4×; 43 analysts. Our new level is 14×.
Company free cash flow, 12 months
~$25bn
~57×, the highest in five years, because of capex.
Operating profit before legal charges
~$92.9bn
Enterprise value about 15.8×.

Our method. We value Meta on its earnings per share in 2030, reduced by about 2.6% for the shares to be issued under the forward sale and the preferred conversion, at a multiple that suits the kind of business it will then be; we discount 4.1 years at 10% and add the dividends received on the way (about $7 a share). Earnings already deduct stock-based pay, so we do not subtract it twice. All figures are before the Class C distribution: after 19 January 2027, divide every per-share number in this report by about two.

CaseEPS 2030 (before dilution)MultipleValue today
The build earns nothing~$35 (flat from 2027)15×~$355
Steady returns ◆~$45 (+10% a year)18×~$545
The consensus$55.8220×~$745

The cases. "The build earns nothing" assumes that the depreciation of the data centres absorbs all the growth in advertising, so earnings stay at the 2027 level, and the market pays only 15 times for a business that has become capital-heavy. "Steady returns" grows earnings 10% a year from the 2027 consensus, slower than analysts expect, with the AI spending earning a modest return. "The consensus" takes the analysts' 2030 estimate of $55.82, from 38 of them, at 20 times.

The middle case moves to about $455 at 15 times and $635 at 21 times; to about $480 at $40 of earnings and $610 at $51; to $590 at an 8% discount rate and $505 at 12%. What is priced in? At $560.84, and 18 times, the market expects 2030 earnings of about $46.70, about 11% a year of growth from the 2027 consensus. That is a fair assumption for this machine, and it is no more than fair.

Margin of safety. The price is about 3% above our central value, so there is none. In our postscript we named $540: the July low, and about 15.8 times 2027 earnings on our feed's consensus. Having now read the commitments in full, we want more. Our level is $480, about 14 times 2027 earnings and 12% below our central case: the price at which the middle case survives a multiple of 16, or earnings growth of only 6%. It is a lower level than any we have named for Meta, because the company now carries obligations it did not carry in July, and an owner should be paid for them.

The build earns nothing ($355), our level ($480), the price ($561), the consensus ($745); our central value is ~$545 — Hold; add below $480
$355
$480
$561
$745
$330$770
★ Hold the machine — add below $480. The price is about 3% above our central value of ~$545. Analysts' mean target: ~$835. Halve all figures after the Class C distribution of 19 January 2027. → Interactive valuation on the company page
◆
Part
XII

Risks, lawsuits & controversies

Verified 22 November 2026 against the Q3 2026 10-Q, 8-Ks and court reports

Capex $135–145bn; 2027 unknownNew Mexico $25bn (appeal)$455bn of commitments~$66bn raised since JulyYouth trials in 2027 (allegations)EU DSA minors (fines up to 6%)Option repricing without a voteRevenue growth slowing (33% → 26%)AI copyright trial, May 2027

Legal. Beyond Part V: the District of Columbia's attorney general seeks up to $1.7bn in a privacy case; a jury found against Meta in the Flo Health app case in 2025, with damages still to be set; authors' copyright claims over AI training go to summary judgment in February 2027 and to trial in May 2027; the EU has fined Meta €798m over Marketplace (under appeal); a Spanish court awarded €542m to media publishers (under appeal). Meta's 10-Q describes all these matters. Claims not yet decided are allegations.

Corporate. The Class C distribution, the Founder Alignment Agreement and the option-plan changes were approved by written consent of the controlling shareholder, without a meeting; the information statement is dated 2 November 2026.

★ How we will know — six dated tests
TestTodayWhat would change our mind
Q4 results and the 2027 guide, late JanuaryCapex $135–145bn (2026)A 2027 capex guide with a visible peak, and free cash flow positive again, would support the middle case; another large step up without revenue to match would move us towards the first case.
Price per ad+14% (Q3)Holding above 10% says the AI spending already pays in the core; a fall towards zero would say the opposite.
New Mexico appeal and bond$25bn + 15%A bond posted in cash, or an appeal lost, would cost far more than the accrual; a reduced award would remove the largest open item.
Class C, 19 JanuaryDistributionA first Class C equity offering soon after would confirm the share class as a financing tool; we would lower the management score again.
The dividend, early December$0.525Another year without a raise is not a problem; a cut would be.
The price$560.84Below $480 (about $240 after the January distribution), with the 2027 guide no worse than feared: add.
◆ PART XIII · To our shareholders
The Letter ⓘ

In July I told you to buy Meta at six hundred and eighty-one dollars because the advertising machine paid for everything else. I said the courtroom fears were the discount and the AI spending rode free. Four months later the shares are at five hundred and sixty-one, and both halves of that argument have failed, though not the way the market feared.

The machine did its part. Revenue grew twenty-six per cent in the third quarter, and advertisers paid fourteen per cent more for each advertisement. Few businesses on earth can raise prices like that at this size. But the courts sent their bills: up to eighteen and a half billion dollars to fifty-two states, and a twenty-five-billion-dollar penalty in New Mexico that Meta is appealing. And the moonshot stopped riding free. In the third quarter the company spent thirty-five billion dollars on data centres and produced no free cash at all. To pay the difference it has sold shares forward, issued preferred stock and borrowed in two currencies, about sixty-six billion dollars since I wrote.

I also owe you six corrections. I said Meta had net debt when it had twenty-two billion dollars of net cash. I said its 2025 bond was its first borrowing; it was not. I called an abnormally high tax rate normal, used our data feed's definition of free cash flow instead of the company's, called this year's earnings next year's, and called a jury verdict a settlement. None of these changed the call. All of them should have been caught.

What has not changed is the quality of the core, and that is why I would not sell. At sixteen times next year's expected earnings you are paying a fair price for the best advertising business ever built. You are not being paid for the risk that the data centres earn less than they cost, or that the company keeps issuing shares to build them. In July I set my level too close to the price; within eight days it was gone. This time I want a margin. I will add below four hundred and eighty dollars, about fourteen times next year's earnings, and I will look first at the capital-spending guide for 2027 in January.

After 19 January every share becomes two, and all these numbers halve. The arithmetic will look different; the question will not. A machine that pays for its own future is worth a premium. A machine whose owners are asked to pay for it is worth a fair price, and no more.

— The Buffett Lens. A margin of safety you can walk past in eight days was never a margin.

▲The Bull Case
★★ The machine is intact — revenue +26% in Q3, price per ad +14% on impressions +10%, 3.64bn daily people; Q4 guided at $73–76bn (+22–27%).
★ Fair on earnings — 16.4× 2027 consensus, below the five-year average of ~23×; operating profit before legal charges ~$93bn over twelve months.
★ The courts are becoming numbers — the states' settlement is fixed at up to $18.6bn over ten years; the FTC lost at trial; the New Mexico award is under appeal.
▼The Bear Case
★★ Owners now pay for the build — Q3 free cash flow −$2.1bn; ~$66bn raised since July in forward shares, preferred and bonds; net cash of $22bn in March is gone; $455bn of commitments and $41.5bn of guarantees.
★ Share capital no longer scarce — no buybacks in 2026; non-voting Class C 'to conduct equity offerings'; options repriceable without a vote; stock pay ~11% of revenue.
★ The bills keep coming — New Mexico $25bn plus 15% interest; youth trials in 2027; EU DSA findings on minors; revenue growth slowing from 33% to 26%.
Hold the Machine —
Add Below $480
The best advertising business ever built, now spending more than it earns and asking owners to fund the difference. ★ At $560.84, 16.4× 2027 earnings, the price is about 3% above our central value of ~$545, within a range of ~$355 to ~$745. Hold; add below $480, about 14× 2027 earnings (about $240 after the Class C distribution of 19 January). The 2027 capex guide in late January is the test.
⚡About 14 times 2027 earnings, where the cost of the build is in the price. Add the $480 price trigger to your Watchlist (it halves to ~$240 after 19 January).
◆ The Buffett Lens · Dividend Line Research · As of 22 Nov 2026 · Price $560.84 (20 Nov close) · Follow-up to our 15 Jul 2026 X-Ray
Found this useful? Share the X-Ray.
◆ More on Meta Platforms, Inc.

Articles that take this apart

◆ Keep reading

More X-Rays from the same lens

Disclaimer: This is an editorial analysis for information and education, not investment advice, and not a recommendation to buy or sell any security. ⚠️ Price and market data are from our live data pull of 21–22 November 2026; the price used is the 20 November close. Results, capital spending, debt, leases, commitments, guarantees, the forward sale, the preferred stock, the Class C shares, insider transactions, the dividend and legal matters are from Meta's releases, Forms 10-Q and 10-K, 8-Ks, its information statement and Forms 4; Alphabet's figures from its Q3 release. The settlement approval, the causes of market moves, Muse usage, analyst actions and some legal updates are from press reports, as marked. Valuation cases are our own arithmetic. This report corrects six statements in our July X-Ray and two figures in our November postscript. Past performance does not guarantee future results. Do your own research or consult a licensed adviser.
Dividend Line · X-Ray Analyses — written in the house methodology