Meta (META) — The 2027 Bill Is Already Signed

Meta's third quarter grew revenue 26% and operating income 11% before a $9.9 billion settlement charge. The 10-Q filed the next day schedules $121 billion of non-cancelable commitments for 2027, up $40 billion in three months; management gave no 2027 capital-spending number and announced a non-voting share class that can be used, among other stated purposes, to raise capital. On our July method the quarter would have lifted fair value from $370 to $386. We now value Meta at $312.11, mainly because we no longer weight a price-to-earnings value while its earnings are not turning into cash; each change is set out for the reader. At the $697.52 close on the day of the release the stock traded 123% above it.

TON618 Capital Research · As of 21 October 2026 · Download PDF · Download model

In July we valued Meta as a company that pays for its machines twice: in cash when it buys them, and in earnings as they depreciate. The third quarter enlarged both payments. The 10-Q schedules $121 billion of commitments for 2027, mostly cloud capacity, servers and data-center energy, not all of it capital spending; depreciation grew 44%, on a server life the company left at five and a half years. The quarter also brought a third instrument. Meta has bought back no stock this year, ended September with net debt near zero after a $25 billion bond, and on 21 October announced a non-voting Class C stock described as usable for "capital raising, employee compensation, acquisitions, other general corporate purposes." This note asks what the bill for this build is worth to a shareholder at $697.52.

Terms used throughout, defined once. Capex: capital expenditures including principal payments on finance leases, the basis on which Meta guides. Ex-charge: a figure with the disclosed one-time legal and severance charges removed, as Meta itself presents them. The dose: capex divided by net income for the same period, the ratio our depreciation series carries for every company. Free cash flow to equity (FCFE): net income plus depreciation, less capex, plus net new borrowing; share-based compensation stays an expense and is never added back. Re-mark: the house restatement of server depreciation onto a three-year life, against Meta's five to 5.5 years; a model estimate, not a restatement, and Meta's reported figures comply with GAAP. Growth stages: the forecast is split into periods, each lasting as long as the thing that drives it; every length is argued in the stage table. ROIC: after-tax operating income over invested capital (equity plus debt less cash), trailing twelve months, against the weighted average cost of capital (WACC). Terminal growth: the rate a business grows at forever after the forecast ends, set here from long-run GDP projections. Fair value: the FCFE value, probability-weighted across three scenarios at 25 / 50 / 25; the price-to-earnings value July weighted at 40% is shown as a cross-check (§4).

Revenue was $64.60 billion, up 26% on both a reported and constant-currency basis. Costs and expenses were $51.61 billion, up 68%, of which $9.89 billion was the charge for the August agreement with 52 attorneys general across US states, territories and the District of Columbia to resolve teen mental-health litigation. Reported operating income was $12.99 billion, a 20% margin. Excluding the charge it was $22.88 billion, a 35.4% margin, against 40.1% a year earlier and 36.8% in the June quarter on the same basis. Net income was $9.86 billion, or $3.83 a diluted share; excluding the charge, $7.40.

The beat, like for like. The brokerage feed shows $7.40 against a $5.74 consensus, a 29% beat. The two figures are not on the same basis. The settlement was announced in August, and the compiled figures show analysts split on whether to model the charge: Investing.com carried a GAAP-basis consensus of $3.57, CNBC reported $5.75 and said it was "not clear" the $3.83 was comparable, and MarketBeat showed $6.54. A pool between $3.57 and the ex-charge level mixes both treatments and cannot measure a surprise. The charge is $3.57 a share after tax, so analysts behind the $3.57 GAAP consensus who carried it would have been near $7.14 ex-charge, close to the June quarter's $7.18 consensus (our inference, not a compiled number): a beat of about 26 cents, roughly 4%. MarketBeat's pool can sit only at or below a pure ex-charge consensus, which caps the beat at 13%. Either way it is a fraction of 29%; on revenue, where the basis is clean, the beat was 1.9% against LSEG. Two more distortions: the charge is mostly non-deductible (a 24% tax rate reported, 17% without it), so it cost $9.19B after a tax shield of only $0.70B; and the reported 265% EPS growth compares with a year-ago $1.05 that carried a one-time tax charge (an 87% tax rate that quarter). Against the year-ago $7.25 excluding that charge, ex-charge EPS grew 2.1%.

The cash statement shows what the earnings cost. Operating cash flow was $32.79 billion and capex $34.91 billion, up 80% on the year, so free cash flow was negative $2.12 billion, the first negative quarter of the AI build. Share-based compensation was $7.36 billion, 11.4% of revenue; the house expenses it. No shares were repurchased, for a third quarter running, and the $25.03 billion authorization is untouched. Cash and marketable securities fell to $82.45 billion from $90.26 billion in June, with a further $12.83 billion restricted, $10.80 billion of it in escrow under infrastructure purchase agreements until 2028–30. Against $83.67 billion of debt, net debt is about $1.2B.

The reaction, one data point. The stock closed at $697.52 on 21 October, before the release, traded at $712.05 at 8:29 p.m. ET (+2.1%) after reaching $748.80, and closed the next session at $648.85, −7.0%, on 40.6 million shares against 29.2 million the day before: the third consecutive earnings session with a sharp fall, after −8.6% on 30 April and −8.0% on 30 July. The house values the company on the anchor close. One session says more about positioning than about the bill, which is where the quarter's real news sits.

The 2026 capex guide has moved one way all year: $115–135 billion in January, $125–145 billion in April, $130–145 billion in July and, on 21 October, $135–145 billion, "narrowed from our prior outlook." The ceiling has not moved since April; the floor has risen $20 billion. Nine months of spending, $85.82 billion, leave $49.2B to $59.2B for the fourth quarter alone, $54.2B at the midpoint, 55% above the third quarter's $34.91 billion. Even the floor implies a 41% step in one quarter.

For 2027, management gave words, not a number. The written outlook says the company is "still developing our budget and capacity plans for next year, but we expect to continue investing aggressively to support both our core business and emerging AI initiatives," and that the investments in Meta Superintelligence Labs "will take longer to generate returns than those in our core business." On the call the chief financial officer said the company is "evaluating a lot of investment proposals that have attractive return profiles over different time horizons," with nothing "more specific or quantitative to share until we get to the Q4 call."

The 10-Q supplies the numbers. Non-cancelable contractual commitments were $454.52 billion at 30 September, up $105 billion from $349.31 billion at 30 June; $121.41 billion fall due in 2027, against $81.65 billion three months earlier. They "mostly relate to third-party cloud capacity arrangements and investments in servers and network infrastructure, energy service agreements for data centers, and consumer hardware products in Reality Labs," so not all of it is capex: third-party cloud is an operating cost, and the CFO named it as one driver of the growth in infrastructure costs. Leases signed but not yet started reached $355.73 billion, from $278.99 billion; the June 10-Q had disclosed about $68 billion of further data-center leases signed in July for 2027–28, and the rise is consistent with them, though the September filing does not itemize it. Construction in progress, assets bought but not yet depreciating, stood at $89.50 billion, from $50.52 billion at the start of the year.

The dose. Capex per dollar of net income was 3.54 on reported earnings in the quarter and 1.83 excluding the charge, against 1.96 reported and 1.65 normalized in June, and 1.19 for 2025 on the same basis (1.15 on cash purchases alone, as July reported it). For the full year the guide midpoints give 1.91 reported and 1.83 on earnings the house normalizes (charges added back and the year taxed at 16.5%, which also removes the first quarter's one-time tax benefit), about where July expected ("near 1.8"). The base case carries 1.94 into 2027, and nothing in the commitments schedule suggests it falls.

The second payment. Depreciation and amortization was $7.15 billion, up 44%; servers-and-network depreciation, reported separately in the 10-Q, was $5.34 billion, up 48% on the year and 16% on the quarter. The 10-Q discloses no change in estimate and no material change to accounting policies, so the five-to-5.5-year server life in the 2025 10-K stands, as it did in June. A fourth quarter growing at the third's rate puts 2026 at $27.6B, on the ~$27.5 billion the July note carried. On the house three-year clock, re-run with this year's capex, the 2027 depreciation the longer schedule defers rises to $15.17B pre-tax from $13.55 billion in July: $4.89 a share of 2027 earnings after tax, from $4.38. Both payments are growing faster than the earnings that should cover them, which raises the question of who pays.

In 2025 operating cash covered the build with $43.6 billion of free cash flow to spare, and the company bought back $26.2 billion of stock. In 2026 it does not: Meta issued $24.91 billion of notes in May, has repurchased nothing, and has $11.05 billion of free cash flow for nine months. On the guide midpoints the fourth quarter consumes about $26B of free cash flow to equity before any borrowing (the house measure, with share-based compensation charged as a cost; about $18B on the company's own definition). The September 10-Q hardens a sentence from June: where the company "may also seek" to raise capital, it now says, "From time to time we expect to seek to raise additional capital through debt, equity, or other financing arrangements." The Class C stock gives the equity part of that sentence an instrument. Its stated purposes include "capital raising, employee compensation, acquisitions, other general corporate purposes"; it carries no vote, will be distributed one-for-one to existing holders (in effect a two-for-one split) once the amended charter takes effect on 19 January 2027, and was approved by shareholder written consent, which the founder's majority of the vote (a risk factor the 10-Q restates) makes possible. The CFO said: "We don't have any specific financing plans to announce at this time."

The model funds each year's shortfall as July's did: net borrowing on a stated path, with any remaining negative free cash flow charged against equity, as if raised at fair value. On that basis the base case needs $69B of external capital across 2027–29 before borrowing, the bear case $116B, and the bull case only $9B. Issuing Class C above the house value would, on the house's numbers, transfer value from new buyers to existing holders; issuing below it would do the reverse. The note does not forecast an issue; the first one is a tripwire.

The settlement is a claim on the same cash. About $12.7 billion is payable to the states in annual installments over ten years; a further $5.3 billion is contingent and not accrued, payable state by state only if peers such as YouTube and TikTok adopt the same teen protections and each such peer with more than $10 billion of annual profit is obliged to pay the same amount. Legal accruals rose to $19.88 billion from $7.32 billion at year-end. The model deducts the accrued charge after its tax shield ($9.19B) in every case and adds the contingent amount in the bear case. With the bill sized and its funding read, the question becomes what the shares are worth.

The engine is the July note's cash-flow-to-equity model. Share-based compensation is expensed and the diluted share count (2,573 million) is held flat, which is neutral here: buybacks, or Class C shares issued, at fair value move no value a share. The cost of equity is rebuilt on the 21 October ten-year Treasury yield of 5.13% (from 4.67%), with the July beta of 1.35 and a 4.5% equity risk premium: 11.21%. The forecast now runs from 2027 to 2033 with every stage length argued, and the return on capital is set against its cost quarter by quarter. Three things changed in how we value Meta, and the reader should be able to separate each from what the quarter did.

How long each part of the business grows faster than normal. Each length is an argument, not a convention. The model workbook's Stages sheet carries the year-by-year paths.

Why these lengths, and why these fades. Once its driver stops, above-average growth rarely lasts: across US companies from 1951 to 1998, sales growth stayed above the median five years running for 6.3% of firms and ten years running for 0.9% (Chan, Karceski and Lakonishok, 2003). Hence five years of fade wherever stage 1 ends, in a straight line, because nothing in the filings dates a faster or slower step down. Advertising's two years rest on what the company has itself dated; a third year is the single input most in the price's favor, and it is priced below. The capital build's one year is the shortest the filings support; a longer build is the assumption that most hurts value, and it is priced too.

Weighted 25 / 50 / 25, fair value is $312.11. Terminal value is 85% of the base case (81% to 89% across the three), which is why terminal growth and terminal capex are priced below. The earnings cross-check gives $506 on re-marked and $598 on reported 2027 earnings; the server clock is worth $92 a share in that lens, from $82 in July, because more server spending sits on the long schedule.

Two levels the reader should see argued. The cross-check's 16–22 range is July's, and it brackets what the market itself has paid on our base-case 2027 earnings: 17.9 times at the July close, 21.8 times on 21 October. Read on a single stage, 18.5 times next year's earnings implies about 5.8% growth in perpetuity at the 11.21% cost of equity, 22 times about 6.7% and 16 times about 5.0%; on the house's own two-stage path the base-case multiple is about 18 times, so the lens is set aside for conversion, not for growth. Terminal capex is the other level. At 22% of revenue, against depreciation of 13.5% and 4.0% growth, the base case's new capital earns about 13.1% in perpetuity, a little above the 11.21% cost of equity: modest excess returns persist. The bull case's 16% lets them run at about 35%; the bear case's 26%, at 3.0% growth, leaves them below the cost of capital, at 7%. A reader who thinks the build's returns persist should look at the bull column.

What moved the number, and what did not. Each row below re-runs the same engine with one factor changed from July's value to today's, in order, so the rows add to the move. The first ten rows apply the quarter's data on July's method; the next two change how the cash-flow model is calculated; the last is the separate decision on weighting.

Read it plainly. The 2026 base year, nine months of actuals and the fourth-quarter guide, added $15.45, and the 2027–28 revenue path reset on the guide and the written 2027 frame added $23.29: the case the company made on 21 October, credited. A 46-basis-point rise in the ten-year Treasury yield took $13.19 off; debt and cash at their September levels $7.67; the settlement $2.45; a heavier 2027 capital path, net of funding, $2.47; the larger depreciation deferral $3.93. On July's method that is $385.51. The cash-flow changes net to +$4.19, and the weighting decision takes the value to $312.11; the base case moved from $364 to $295.39. Our number fell because it no longer counts earnings that are not becoming cash, not because the quarter disappointed.

Return on capital: strong, and fading. Meta's return on invested capital is 33.5% on a trailing basis against a 10.9% cost of capital: by the house's mechanical rule, a strength regime turning to fading, still far above the hurdle but with the spread down 4.4 points in four quarters. With the one-time charges added back the decline is steeper, from 45% to 34%, as average invested capital grew 50% in a year and operating income did not. A return this far above its cost is why the price's case deserves a hearing; its direction is why the house does not extend the stage that produces it.

What the price requires. Hold every other base-case input and solve for the one that makes the cash-flow value equal $697.52. Revenue would have to compound 25.0% a year from 2027 to 2033, against 10.9% in the base case: $1,219B of revenue in 2033, against $528B (Exhibit 5). Or the cost of equity would have to be 7.6%, an equity risk premium near 1.8% on today's Treasury yield. Or perpetual growth would have to be 8.3%, far above long-run nominal growth of the economy. No level of terminal capex gets there, even zero. Scaling every scenario's growth together, the base case has to compound at 23.1%; at a cost of equity of 10.2% the price still needs 2.0 times the base case's growth rates, and at 12.2%, 2.6 times. At the next session's $648.85, the revenue requirement is 23.7%. The price rests not on one heroic input but on the 2026 growth rate continuing for most of a decade at a scale where it has never been sustained, or on the new AI lines becoming a second business the size of the first.

The inputs a reader will want to move. At July's Treasury yield the fair value would be $340, and a full point either way gives $377 or $263. Advertising's stage 1 at one year or three: $294 or $332. The capital build stage at none or two years: $329 or $295. Terminal capex four points lower or higher: $381 or $243. Terminal growth half a point either way: $290 or $338. July's 60 / 40 weighting of the price-to-earnings lens: $390. Without the settlement claim: $316. No single input in that list carries the value to the price: the most generous, terminal capex four points lower, still leaves the price 83% above it, and July's weighting, a method choice rather than an input, leaves it 79% above.

The July note armed six triggers on Meta; all six are graded against the release and the 10-Q.

Each tripwire carries its basis, its tolerance and what the model does with the answer, so the next update applies rules written now. The fourth-quarter release is expected around 27 January 2027 (unconfirmed on the brokerage calendar).

1. Fourth-quarter capex. Basis: the Q4 release's capex including finance leases, against the $49.2B–$59.2B the full-year guide implies. Inside the range: the base case's 2027 build stage holds. Below $49.2B: the year finishes under the floor set in October. The base case's 2027 capex then moves halfway toward the bull's 48% of revenue, and the house reads the floors as intentions, not commitments. Above $59.2B: the build stage is re-run at two years ($295).
2. The 2027 capex guide (Q4 call). Basis: the guided range against base-case 2027 revenue of $307B. At or below 40% of it, about $123B, with a stated glide path: the July trigger to move toward a neutral view fires and the build stage is removed ($329). Between that and $148B (the bull case's 48% of revenue), a guided midpoint moves base-case 2027 capex halfway toward the bull's ($316; one year of capex moves little). From there to $160B: the base case holds. Between $160B and the bear case's 55% of revenue (about $169B), a guided midpoint moves base-case 2027 capex halfway toward the bear's ($309). Above 55%: the build stage moves to two years ($295). A guide with no range — "invest aggressively" again — is read as above 55%.
3. Commitments due in 2028. Basis: the FY2026 10-K's commitments schedule (filed about late January to February 2027). If the commitments due in 2028 exceed the $121 billion now due in 2027, the build has a second signed year and the stage moves to two years ($295). Below $80 billion, the build has no second signed year and the stage is removed ($329). Either reading applies whatever the 2027 guide says.
4. The first Class C sale for cash. Basis: an 8-K, a prospectus supplement, or the financing section of a 10-Q after 19 January 2027. Record the proceeds and the price against the house value at the time. Proceeds above 2% of the market value in any twelve months mark the funding path as equity-led; the model's net-borrowing path is then cut by the proceeds. Fair value changes only by the difference between the sale price and the house value on the shares sold. Grants to employees in Class C are not a sale.
5. Ex-charge operating margin, fourth quarter and the 2027 expense guide. Basis: the Q4 release (the guide implies about 33%) and the 2027 expense outlook. A 2027 expense guide that implies operating income growing more than ten points slower than revenue moves the base case's 2027–28 EBITDA margin to the bear's path. Two consecutive quarters at or above 40% excluding charges move it to the bull's. The July concession test (45%+ with ad growth above 20%) stays armed.
6. The youth-safety line. Basis: the 10-Q and 10-K legal notes and any 8-K. The consolidated personal-injury bellwether opening 28 October 2026, the school-district case in February 2027 and the bellwethers of 16 March and 6 August 2027 are the dated events. Any accrual of the $5.3 billion contingent payment moves it from the bear case into every case. Disclosed legal charges above $2 billion in 2027 complete the July test: the legal line becomes a recurring cost, and the base margin is cut by the 2027 amount spread over five years.
7. The server clock. Basis: the FY2026 10-K useful-life note. Any change in estimate re-runs the re-mark. A move to four years or less would close most of the $92 spread between reported and re-marked earnings in the earnings cross-check; the cash-flow value depends on the book useful life only through the tax line.
8. The price. Fair value $312.11. A close at or below it resolves the gap. At $234, a 25% margin of safety, the house would look for an entry if the operating tests above are intact (this replaces the July note's $278).
9. Earnings turning back into cash. Basis: free cash flow (Meta's definition) over net income, quarterly, and the buyback line. Two consecutive quarters with free cash flow at or above 60% of net income, or a 2027 guide that takes capex back below twice depreciation, or buybacks resuming at more than $5 billion a quarter: the condition for an earnings multiple holds again, and the price-to-earnings value returns at July's 40% weight ($389.71 at today's inputs). Between 20% and 60%: unweighted, but two consecutive quarters above 40% restore half the July weight, 20% ($351). Below 20% for two more quarters: the cross-check stays unweighted and the note says so.

Fair value $312.11 a share (bear $114.48 / base $295.39 / bull $543.18 at 25 / 50 / 25), from $370 on 31 July. At the $697.52 close of 21 October the expected return to fair value is −55%; at the next session's $648.85, −52%. The quarter earned Meta a higher number, $385.51 on July's method, because its revenue engine is stronger than we assumed. It did not earn a smaller bill: $121 billion of the 2027 bill is already signed, not all of it capital spending; it is growing faster than operating income; and the company has built a share class that can help pay it. Until earnings turn back into cash, we value Meta on the cash it produces rather than on its earnings, and the price assumes the bill earns a return the business has not yet shown. The fourth-quarter call will put the first number on 2027, and the 10-K a second year of commitments. The Fund holds no position in META.