Meta runs the world's largest social graph — Facebook, Instagram, Messenger and WhatsApp reach 3.60 billion family daily active people (DAP). It sells advertising against that attention (97.5% of trailing-twelve-month (TTM) revenue), priced by an increasingly automated auction: impressions +14%, price per ad +12% in Q2 2026. A second segment, Reality Labs (RL), sells Quest virtual-reality hardware and builds augmented-reality glasses; it loses roughly $19B a year. Since 2025 the center of gravity has shifted to artificial intelligence (AI): a superintelligence build-out taking 2026 capital expenditures (capex) to a guided $130–145B — origin, twelve quarters ago: $30–35B for 2024.

The split that matters: the Family of Apps (FoA) produced $106.0B of TTM operating income at a 52% segment margin in FY2025, while RL lost $19.1B TTM. Everything here is a question about how much of the first number the second act is allowed to spend.

What the price underwrites is a conversion: $92.4B of TTM capex must convert into durable, high-return capacity before the depreciation and the debt come due, while ~$19B/yr of RL losses and a stock-based compensation (SBC) bill of $25.1B (11.0% of revenue) keep running. Management's doctrine is to overbuild on purpose — Zuckerberg, Q2 2024 call (July 31, 2024): "I'd rather risk building capacity before it is needed, rather than too late, given the long lead times for spinning up new infra projects." Every annual capex guide was raised within its year and delivered inside or above its final range — never below it. Mid-teens owner-earnings growth is achievable if capex intensity peaks around 2026 and ad growth holds. That "if" — never underwritten in words — is the thesis.

• 🐻1 — Free cash flow has cracked. Q2 2026 free cash flow (FCF) was $784M on $60.8B of revenue — capex consumed 97.5% of operating cash flow in the quarter. TTM FCF fell −20.4% year over year while revenue rose +27.6%; TTM FCF is 16.6% of operating income, versus 94% in FY2024. Over the full five-year window, reported owner earnings have shrunk (−16.8%/yr, FY2021 → TTM) even as revenue compounded +15.8%/yr — the supercycle, so far, has consumed the compounding.
• 🐻2 — The timeline is management's own warning. Zuckerberg, Q1 2024 call (April 24, 2024): "I also expect to see a multi-year investment cycle before we've fully scaled Meta AI, business AIs, and more into the profitable services I expect." The generative-AI revenue line still does not exist in any filing, and Li (Q2 2025 call) put the products "much, much earlier on the return curve," not a meaningful revenue driver "this year or next year."
• 🐻3 — The balance-sheet regime has changed. Debt $28.83B (Dec 2024) → $83.66B (Jun 2026), $54.8B issued in eighteen months; Li told the Q4 2025 call Meta may "eventually maintain a positive net debt balance." Plus $46.03B maximum venture exposure — argued in full under Balance Sheet.
• 🐻4 — Margin compression, not leverage. Operating income fell −8.2% on +28.0% revenue (margin 30.9% vs 43.0%): research and development (R&D) $21.66B (+67.3%), $2.40B legal charges, $1.18B severance; FoA operating income −6.3% despite +28.0% segment revenue. Management's ex-charges framing (operating income +9% excluding the items) is accurate arithmetic — and an admission that "one-time" charges now recur.
• 🐻5 — Buybacks are zero and the count is rising. Repurchases: $0 in Q4 2025, Q1 2026 and Q2 2026. Shares outstanding rose from 2,529.6M (Jan 23, 2026) to 2,547.5M (Jul 24, 2026) — dilution is the demonstrated pace. The Feb 2024 framing — buybacks as the primary return channel — has been suspended three quarters without a word of explanation on any call.
• 🐻6 — The RL losses are structural until the peak statement is proven. $85.6B cumulative, 2021 through Q2 2026. The January 2026 "peak" language is one sentence against twelve quarters in which every forward cost statement — capex, expenses, RL losses — was later revised upward.
• 🐻7 — Legal overhang is live, not theoretical. The Federal Trade Commission (FTC) appealed its November 2025 trial loss to the D.C. Circuit (January 2026, backed by 29 state and territorial attorneys general — States' amicus brief, D.C. Cir. No. 26-5028, filed May 29, 2026), and Meta's Q2 2026 outlook warns youth-related trials "may ultimately result in a material loss" — repeated on four consecutive calls (Q3 2025–Q2 2026) and in four consecutive release outlooks.

• Answers 🐻2, in part — core-AI returns are measurable. Q2 2026: revenue $60.80B (+28.0%), impressions +14%, price per ad +12%, Advantage+ at a $75B run rate. Zuckerberg, Q2 2026 call (Jul 29, 2026): "these AI investments are paying off." Concedes 🐻2's other half: no generative-AI revenue line exists in any filing.
• Answers the reliability half of 🐻6's pattern claim. Revenue guides met or beaten 12 of 12 quarters — the full record is under Guidance.
• Answers 🐻1's permanence claim — the overbuild has paid before. The front-load logic built the Reels and core-AI capacity that now produces the ad numbers, and capacity, not demand, has been the binding constraint: "we do continue to be capacity constrained" (Li, Q4 2025 call). Concedes 🐻1's present: TTM FCF is down −20.4% today.
• Answers 🐻3 on level: gross debt 0.96× TTM operating income, no principal due before 2027, net cash still barely positive. Concedes 🐻3 on trajectory: $54.8B issued in eighteen months, more financed off balance sheet.
• Concedes 🐻5's present — $0 for three straight quarters — and answers its permanence. Dividend and the $50B authorization raise landed the same day (Feb 1, 2024) the capex guide first stepped up; Li framed the dividend on that call as "a nice complement to the existing share repurchase program," with buybacks the primary channel — designed not to compete with the build. The FY2024 program retired $30.13B.
• Answers 🐻6. The Jan 28, 2026 peak sentence exists, and H1 2026 delivery tracks it ($8,647M vs $8,739M). Concedes: the evidence is young — two quarters against twelve of upward revisions.
• Answers 🐻7's antitrust half. Boasberg found no monopoly; the market is certified contested and Meta still compounds at +28.0%. Concedes the youth-trial half: live, on four straight calls, unquantifiable.

Exhibit 3 — Q2 2026 vs Q2 2025 ($M, except per-share and DAP)

Source: Form 8-K earnings release, Q2 2026 (Jul 29, 2026).

Q1 2026, for context: revenue $56,311M (+27.1%), operating income $22,872M, net income $26,773M — flattered by an $8.03B benefit (Treasury Notice 2026-7, corporate alternative minimum tax (CAMT) relief), partly offsetting Q3 2025's $15.93B One Big Beautiful Bill charge. Net one-time tax drag in TTM net income: −$7.90B (normalized TTM net income ≈ $76.0B, ≈24.7×).

Exhibit 4 — FY2021 to TTM ($M; signed year-over-year change in every cell)

The rows that decide the thesis

Revenue & segments

Cash & the build

Per-share & reach

Source: Forms 10-K FY2021–FY2025; Form 10-Q Q2 2026 (TTM through Jun 30, 2026).

Read the table the way the reverse discounted cash flow (DCF) does: revenue never stopped compounding (+15.8%/yr across the window) and Family of Apps operating income compounded +14.8%/yr — and reported owner earnings fell −16.8%/yr, because capex went from 16.4% of revenue (FY2021) to 40.5% (TTM) while SBC compounded ~+25%/yr across the last four years. The entire investment question lives in the capex and SBC rows.

At June 30, 2026 (Form 10-Q Q2 2026): cash and marketable securities $90,260M; long-term debt $83,664M carrying; total assets $449,956M; equity $261,221M; gross debt 0.96× TTM operating income — net cash still, barely. The level is covered; the trajectory is the flag: $54.8B issued in eighteen months, and management told the Q4 2025 call it may "eventually maintain a positive net debt balance."

Exhibit 5 — The debt, enumerated

$84,000M principal in 26 tranches, 5 series, less $336M unamortized discount/costs = $83,664M carrying. All fixed-rate, no financial covenants, fair value $79.75B. No principal due before 2027; $10,250M due 2027–2030; $73,750M thereafter.

Source: Form 10-Q Q2 2026, Note 8; Form 424B2 pricing supplements.

Capital return: dividend $0.525/share/quarter (initiated Feb 2024 at $0.50; +5% Feb 2025); $5.37B paid TTM. Buyback authorization remaining: $25,030M at Dec 31, 2025, no expiration, unused since Q3 2025. Taxes paid on net share settlement: $8,704M in H1 2026.

The data-center venture — what the filing says vs what management said

The filing record (Form 10-Q Q2 2026, Note 5). Meta is a variable-interest holder in an unnamed unconsolidated venture building a data center — not the primary beneficiary, does not consolidate it. Maximum exposure to loss: $46.03B at June 30, 2026 (equity carrying value $2.92B + lease commitments + estimated future funding + a maximum residual value guarantee (RVG) threshold) — 2.46% of enterprise value (EV); 1.22× TTM free cash flow. Other variable interest entities (VIEs) add $6.41B (total $52.44B = 2.80% of EV). The exposure sits deliberately off the EV bridge: a maximum-loss measure, not a debt obligation.

The call record. Li pre-announced partner financing on the Q2 2025 call (Jul 30, 2025), and named the structure on the Q3 2025 call (Oct 29, 2025): the Blue Owl Capital joint venture, Louisiana (Hyperion), owned 80/20 — "the construction cost of the data center will not be recorded in CapEx"; Meta's 20% runs through other investing cash flows. A second venture was named on the Q2 2026 call: 1 GW (gigawatt) with BlackRock, El Paso.

Why it matters: reported capex — the input to every Reverse DCF lens — now excludes the largest site's construction cost; the $130–145B guide is not the whole cash build.

Concentrations, sized against EV

• One profit pool funds everything. FoA operating income of $106,027M TTM is 122.0% of consolidated operating income ($86,926M) — RL's losses are the difference. Advertising is 97.5% of TTM revenue (≈$222.5B). There is no second engine at scale.
• Reality Labs' cumulative claim on that pool: $85,604M of losses, 2021–Q2 2026 = 24.3% of FY2021–25 FoA operating income ($352,061M).
• The financed build: the $46.03B maximum VIE exposure above (2.46% of EV; 2.80% including other VIEs) plus $84,000M of senior-note principal — funding is increasingly contractual and off-/para-balance-sheet, which is why reported capex understates it.
• GPU supply — a disclosed dependence with no disclosed numbers. The FY2025 10-K names no supplier ("NVIDIA" appears zero times) beside $131.05B of aggregate non-cancelable commitments with no vendor split; per-supplier spend is an explicit gap. Call-record color, labelled as such: ~350,000 H100s / ~600,000 H100-equivalents by end-2024 (Q4 2023 call); a >100,000-H100 Llama 4 cluster (Q3 2024 call); Andromeda "built with NVIDIA" (Q4 2024 call); >1 GW Broadcom custom silicon + AMD chips (Q1 2026 call, Apr 29, 2026).

Q3 2026 guide — Chief Financial Officer (CFO) outlook, Q2 2026 release, verbatim

"We expect third quarter 2026 total revenue to be in the range of $61-64 billion. Our guidance assumes foreign currency is an approximately 1% headwind to year-over-year total revenue growth, based on current exchange rates."

The expense range's lower end was raised to absorb the $2.4B legal charges; capex was "narrowed from our prior outlook of $125–145 billion"; the tax guide rose from 13–16% to 15–17%.

Revenue guides — 12 of 12 met or beaten · Grade A−

Every quarterly revenue guide in the record was delivered inside its range or above it — above the top three times (Q4 2023, Q2 2024, Q3 2025), at or near the high end the rest (Q1 2026: $56.31B vs $53.5–56.5B guided; Q2 2026: $60.80B vs $58–61B). The most reliable guide in the coverage universe.

Exhibit 6 — Capex guides: the ratchet, scored from its origin

Source: CFO outlooks, Forms 8-K and call transcripts, Q3 2023–Q2 2026.

Three facts the exhibit makes unavoidable. (1) No annual capex guide was ever revised down — the only direction is up, within every year. (2) Delivery lands inside or above the final range, never below it — not sandbagging; the spending outruns each successive forecast. (3) The origin matters: the FY2026 midpoint ($137.5B) is 4.2× the FY2024 origin midpoint ($32.5B). Li's Feb 2024 tell was directionally honest; no guide ever contained the magnitude. Grade: C.

Expense guides · Grade B — and long-horizon frameworks

FY2025 expenses guided $114–119B → delivered $117.69B (inside). FY2026 guided $162–169B → $165–169B. Expenses land; capex doesn't.

Long-horizon: the "above 2025 operating income" commitment is the only multi-year P&L (profit-and-loss) commitment management has made; the RL peak statement (Jan 28, 2026) is the first long-horizon cost commitment — first test Oct 21.

Method: we work backwards from the price — solving for the growth rate the market price implies. That is the annual owner-earnings growth rate that makes $736.72 fair at a 10% required return: the solved rate holds for ten years, then eases gradually toward 2% over the following forty (an exponential fade), terminal value is struck at year 100, and there is no exit multiple. Owner earnings = free cash flow − stock-based compensation, solved on EV of $1,870.2B. Full workings: reverse_dcf_exp100.py + reverse-dcf-exp100-2026-10-17.json in the dossier (cliff predecessor: reverse_dcf_v3.py).

At the Oct 15 close, the market prices reported owner earnings compounding at about +22.7%/yr for a decade before the fade. Delivered FY2021–TTM, those earnings fell −16.8%/yr while revenue compounded +15.8%/yr — the price is not paying for the ad engine's record; it is paying for the build to end. If 2026 is the capex peak, the normalized-capex base ($53,838M) cuts the priced-in rate to about +11.2%/yr, inside the delivered revenue envelope. Across twelve quarters of calls, management has never promised that peak.

Exhibit 7 — The credited bar and the two memos that matter

Every other solve — the ex-Reality-Labs lens, the raw free-cash-flow memo, the normalization-window variants, per-share equivalents and assumed shrink legs — is computed identically and filed in the dossier results (reverse-dcf-exp100-2026-10-17.json); under the desk's bar budget they no longer print in this section.

Exhibit 8 — What you must believe: the priced-in growth rate on the governing base, by required return

Benchmark: the revenue compounding Meta actually delivered FY2021–TTM, +15.8%/yr. Every cell sits above it — no required return makes the reported base cheap against the delivered record; the normalized-capex memo in Exhibit 7 (+11.23%/yr at the house return) is the only printed bar below it.

• Competitive advantage period (CAP) — normalized lens: 10.8 years at +15%/yr growth; 22.0 years at +10%/yr; 7.3 years at +20%/yr. Reported lens: 21.4 years even at +20%/yr. (The CAP inversion is unchanged by the new fade shape.)
• Terminal value (TV) share of EV: 0.18% reported; 0.11% normalized — the terminal is struck at year 100, so effectively the entire answer sits in explicit forecast years (under the retired cliff the shares were 73.6% and 61.4%).
• Implied terminal EV/EBIT (enterprise value ÷ earnings before interest and taxes): 1.9× reported; 7.9× normalized — unchanged by the new shape (the strike-year multiple is 12.75× owner earnings in both); arithmetic, not a multiple anyone would pay.
• Actuals beside the bars (FY2021 → TTM, 4.5 years): revenue +15.8%/yr; FoA operating income +14.8%/yr; FCF −0.3%/yr; reported owner earnings −16.8%/yr. (The FY2022-trough frame is rejected, as in the dossier.)
• The record behind the credited bar: the demonstrated input is filed diluted weighted-average shares — 2,574M in FY2025 vs 2,614M in FY2024 (Form 10-K, filed Jan 29, 2026), a delivered +1.53%/yr shrink. The 2026 cover pace runs the other way (2,529.6M → 2,547.5M, Jan→Jul 2026; buybacks $0 for three straight quarters) and is filed as a measured-pace scenario in the dossier, not printed as the credited bar. Timing stays procyclical: FY2024 $30.13B; FY2025 $26.25B at ~$656/share (40M shares); 2021's $44.5B at ~$329/share preceded the 2022 collapse.

Plausibility verdict: the normalized memo bar (about +11.2%/yr for a decade) sits inside Meta's demonstrated revenue envelope if 2026 is peak capex intensity; the governing reported bar (about +22.7%/yr) is the market saying the supercycle must end. The verdict survives the normalization-window choice — those solves are filed in the dossier — and does not survive a structural intensity regime, under which the normalized lens collapses toward the reported one; that is why October 21's capex guide matters more than its revenue print. The call record cuts both ways: deliberate overbuild, no guide ever revised down — but capacity-constrained, not demand-constrained, for twelve straight quarters. The position of this page is a bet on a capex peak, not on the ad business. (The normalized lens is a desk normalization, not a guide.)