Meta Platforms

Phase A — Understand the business

Lens 1 · Company Overview

Meta runs the largest attention network ever assembled and monetizes it almost entirely through advertising. Two reportable segments ``:

• Family of Apps (FoA) — Facebook, Instagram, Messenger, WhatsApp, Threads. FY2025 revenue $198.76B, of which $196.18B is advertising (98.7%) and $2.58B is "Other" (WhatsApp paid messaging, Meta Verified subscriptions, Payments fees). FoA is 99% of total revenue and ~123% of total operating income (it carries the RL loss).
• Reality Labs (RL) — VR/AR consumer hardware (Quest), AI glasses (Ray-Ban / Oakley Meta), neural interfaces, Horizon. FY2025 revenue $2.21B (+3%), but an operating loss of −$19.19B ``.

Business model in plain terms: Meta gives away social products for free, harvests engagement (3.58B daily active people in Dec 2025, +7% YoY) , and sells targeted ad impressions to ~10M+ advertisers. Revenue = ad impressions × price per ad. In FY2025 both levers worked: **ad impressions +12%, average price per ad +9%** ; ARPP $57.03, +15% YoY . There is no take-or-pay or long-term contract structure on the revenue side — it is a real-time auction, spot-priced, with the largest single advertiser vertical being **online commerce** (the Temu/Shein/CN cross-border cohort) .

Key customers: a long-tail advertiser base; no customer concentration disclosed (customers.csv empty — no named concentration, consistent with a millions-of-SMB auction). Key suppliers: NVIDIA / AMD (AI accelerators), TSMC (via chip vendors + Meta's own MTIA silicon), the hyperscaler cloud providers it rents overflow capacity from, EssilorLuxottica (glasses), and the energy/data-center supply chain. Key competitors: Alphabet/Google (the other duopoly leg of digital ads), Amazon (retail-media ads), TikTok/ByteDance (engagement + short-video ads), Snap, Pinterest, Reddit, plus the AI-lab cohort (OpenAI, Anthropic, Google DeepMind) on the superintelligence axis.

The 2026 identity shift: Meta now frames itself as an AI company. Zuckerberg's stated mission is "personal superintelligence for everyone," and the company stood up Meta Superintelligence Labs (MSL) in 2025 ``. The thesis to underwrite is no longer "is the ad business durable" (it plainly is) — it is "does the AI capex supercycle create more value than it consumes."

Lens 2 · Supply Chain

Meta sits at the demand-aggregation end of two distinct chains:

Chain 1 — the ad/compute chain (the money engine): Silicon (NVIDIA H-series/Blackwell, AMD MI-series, TSMC-fabbed; Meta's own MTIA inference ASICs) → server OEMs / ODMs → Meta data centers (self-built + leased colocation) → ad-ranking & recommendation models (Andromeda ranking engine, Llama) → ad auction → advertisers (SMB long tail + the cross-border commerce cohort: Temu, Shein, AliExpress) → end consumers (3.58B DAP).

• Chokepoint / single-source dependency: leading-edge GPU supply (NVIDIA is the gating input for the whole industry) and the power + data-center construction bottleneck. The 10-K discloses $182.88B of leases not yet commenced and $237.67B of non-cancelable contractual commitments (mostly third-party cloud, servers, data centers) as of 2026-03-31 — that *is* the supply chain, contractually pre-committed. Meta also signed **+$24B more infrastructure contracts in April 2026** and carries **$14.72B of contingent cloud purchase obligations**.
• Mitigation: Meta is vertically integrating into custom silicon (MTIA) and signing multi-decade clean-energy PPAs (3–25 year terms, no fixed volume) `` to de-risk both the GPU and the power chokepoints.

Chain 2 — the Reality Labs hardware chain (the optionality engine): Component suppliers + EssilorLuxottica (frames, optics) → contract manufacturers → Ray-Ban / Oakley Meta glasses, Quest headsets, Neural Band → DTC + retail (Ray-Ban / Sunglass Hut / Meta Store).

• Named stakeholder: EssilorLuxottica is the load-bearing partner — in talks to double glasses production to ~20M units by end-2026, potentially 30M+ ``. This is a genuine single-source partner dependency for the breakout RL product.

This lens passes the names test: NVIDIA, AMD, TSMC, EssilorLuxottica, Temu, Shein are the actual nodes, not generic placeholders.

Lens 3 · Competitive Advantages (moats)

Meta's moat is multi-layered and among the most durable in tech, but the locus of the moat is shifting:

1. Network effects (the classic moat): Facebook/Instagram/WhatsApp each clear 1B+ users; the aggregate 3.58B DAP is a self-reinforcing graph that no challenger except TikTok has dented. Switching costs are social (your friends/photos/groups live here), not technical.
2. Data + ad-ranking scale (the compounding moat): the auction sees more conversion signal than any rival except Google, and Meta is pouring AI into ranking (the "Andromeda" engine, generative-AI ad tools). This is why price per ad rose 9–12% even as impressions grew — the system is getting better at matching ``. AI is widening this moat, not threatening it.
3. Bargaining power — asymmetric by counterparty:

• Over advertisers: very high. A fragmented base of millions of SMBs with no alternative of equal reach. Meta sets price via auction; advertisers are price-takers.
• Over suppliers (NVIDIA, TSMC, power): weak and weakening — Meta is a price-taker on GPUs (the Q1-2026 capex raise was explicitly blamed on "higher component pricing") ``. The MTIA custom-silicon program is the attempt to claw this back.

1. Capital / scale moat (the new moat being built): the willingness and balance-sheet capacity to spend $125–145B/yr on AI infrastructure is itself a barrier — only ~4 companies on earth can. If superintelligence is real, this is a moat. If it's a mirage, it's a value trap. That ambiguity is the whole investment debate.

Vulnerability: the moat is strongest in distribution and monetization, weakest in frontier model quality — Llama has lagged OpenAI/Google/Anthropic on the leaderboards, which is precisely why Meta paid nine-figure packages to rebuild the lab (Lens 9).

Lens 4 · Segments

Revenue by segment ``:

Operating income by segment ``:

Revenue by geography (Q1-2026) ``: US&Canada $21.27B (+29%), Europe $13.24B (+39%), Asia-Pacific $15.45B (+29%), Rest of World $6.36B (+40%). Europe is re-accelerating hardest despite being the most regulated market.

The trend that matters: total revenue growth accelerated — 22% (FY25) → 33% (Q1-26), driven by ad impressions re-accelerating (+12% FY25 → +19% Q1-26) and price per ad holding (+9% → +12%) . The cause: AI-driven ad-ranking improvements + a favorable FX swing + the cross-border commerce vertical. **FoA is the entire profit engine; RL is a −$19B/yr R&D line item financed by FoA** — management explicitly says RL's viability "is dependent on generating sufficient profits from other areas of our business" and guides 2026 RL losses "similar to 2025."

---

Phase B — Measure performance

Lens 5 · Earnings Result (latest print: Q1-2026, reported 2026-04-29)

Headline ``:

• Revenue $56.31B, +33% YoY (+29% constant currency) — a beat vs Street (~+25% expected) ``.
• Income from operations $22.87B, +30%, operating margin 41% (flat YoY — note: discipline held even through the capex ramp).
• Net income $26.77B, +61%; diluted EPS $10.44, +62%.
• EARNINGS-QUALITY FLAG (critical): the net-income/EPS beat is flattered by an $8.03B CAMT tax benefit (Treasury Notice 2026-7). Reported effective tax rate was −23%; ex-benefit it would have been 14% . Conversely, Q3-2025 net income was crushed to $2.71B (EPS $1.05) by a **one-time $15.93B OBBBA tax charge** (87% effective rate) . The tax line whipsawed ±$16B across two quarters; the clean read is the operating line, which compounded ~30%. Anyone anchoring on headline EPS growth is reading noise.
• Engagement: DAP 3.56B, +4% (decelerating from +7–8% — and management "missed on user growth," a stated reason for the selloff) ``.
• Margins/drivers: ad impressions +19%, price per ad +12% `` — both re-accelerated, the cleanest bullish signal in the print.
• Balance sheet: cash + marketable securities $81.18B; no buybacks in Q1-2026 ($25.03B remains authorized) — a tell that cash is being routed to capex over returns. Headcount 77,986, +1% (flat = efficiency intact).

The market reaction: despite the revenue beat, the stock fell ~7–8.5% post-print ``. Why: Meta raised 2026 capex guidance to $125–145B (from $115–135B), blaming higher component pricing, and missed on DAP. The print is a microcosm of the whole story — fundamentals beat, the stock fell on spend. That is the single most important fact in this dossier.

Lens 6 · Earnings Calls (sentiment trend)

Tracking the management narrative across the last several quarters ``:

• Through FY2024 → mid-2025: the dominant phrase was "efficiency" and "AI-driven ad performance" — the post-2022 discipline story. Tone: confident, ROI-grounded.
• H2-2025 → Q1-2026: the vocabulary pivoted hard to "personal superintelligence," "Meta Superintelligence Labs," "frontier models," "next computing platform." Zuckerberg is now selling a vision, not a near-term return.
• The tell: asked directly on the Q1-2026 call about signs of ROI on the $145B spend, Zuckerberg replied "that's a very technical question" and deflected . Bulls read confidence; bears read evasion. Management *stopped* leading with efficiency/discipline and *started* leading with ambition — a sentiment shift the market is pricing as risk (hence the multiple compression from ~26x to ~21x trailing) .
• What they keep saying: "we are investing now because we believe this will become the next computing platform." What they stopped saying: any concrete near-term RL or AI-capex payback timeline.

Lens 7 · Comps

Peer table — Meta vs the mega-cap ad/AI complex and the social-ad pure-plays. All multiples ``, dated; where not sourced, marked n/a.

5-yr avg ROE: n/a (do not fabricate; Meta's ROE is structurally very high — FY2025 net income $60.46B on a low-leverage equity base — but the precise 5-yr average is not sourced here).

The read: Meta is the cheapest name in the entire mega-cap AI/ad complex at ~18–20x forward — a discount to Alphabet (~30x) and Amazon (~29x) despite posting faster revenue growth (+33% vs peers) and a 41% operating margin. It trades below its own 4-quarter-average trailing P/E of ~26.4x (a ~20% de-rating) . Meta is "expensive vs the interactive-media industry average (~12.6x) but cheap vs the peer average (~27.7x)" . The comp argument is the core of any bull case: you are paying a below-market multiple for above-market growth — the discount is the market's price for capex/ROI uncertainty.

Lens 8 · Stock-Price Catalysts (>5% moves, 5-yr pattern)

``:

• 2022 (−65%+, ~$500B mkt cap erased): Apple ATT privacy changes gutted ad targeting + TikTok competition + metaverse losses; FCF collapsed to a $173M quarterly trough. The market punished spend without discipline.
• 2023 (+178–194%, best-in-S&P): the "Year of Efficiency" — 14% headcount cut, AI-Reels monetization, the Temu/Shein ad surge. The market rewarded discipline + AI-driven ad gains.
• 2024–2025: steady compounding to an all-time high near $660 (Dec 31, 2025 close) `` on accelerating ad revenue.
• Q1-2026 (−7–8.5%): the capex raise to $125–145B + DAP miss ``. De-rated to ~$600 by mid-June 2026.
• The pattern (the single most important behavioral fact for this name): Meta's stock does not primarily react to ad fundamentals (which have been strong throughout) — it reacts to the spend-vs-discipline narrative. 2022 punished spending; 2023 rewarded discipline; 2026 is punishing spending again. The market will re-rate Meta the moment the AI capex shows a discipline/ROI payoff — exactly as it did in 2023. This is a narrative-sensitive, not a fundamentals-sensitive, stock at the margin.

---

Phase C — Judge people & books

Lens 9 · Management

Mark Zuckerberg (Chairman & CEO, Founder).

1. Track record: built a $1.5T company from a dorm room; survived the 2022 near-death experience and engineered the 2023 turnaround (one of the great corporate comebacks). Has repeatedly made large, contrarian, correct bets (mobile pivot 2012, Instagram $1B/2012, WhatsApp $19B/2014, the 2023 efficiency reset). Also made the metaverse bet that has burned >$70B cumulative in RL operating losses with `` ~$19B/yr ongoing and no payback in sight — the case against his capital allocation.
2. Tenure & skin in the game: founder, ~21 years; controls the company via Class B super-voting stock (342M Class B shares carry control despite ~2.19B Class A outstanding) ``. This is a dual-class, founder-controlled structure — Zuckerberg cannot be overruled by public shareholders. Maximum alignment of incentive; minimum accountability of governance.
3. Capital-allocation history: aggressive reinvestment is the house style. FY2025: $72.22B capex, $26.26B buybacks, $5.32B dividends, $18.33B into non-marketable equity investments — the Scale AI $14.3B stake is the marquee item . ROE/ROIC is structurally very high on the FoA business and structurally destroyed on RL. The 2026 capex doubling is the single largest allocation decision in the company's history and the crux of the whole thesis.
4. Red flags: the dual-class control means board "independence" is nominal — note the FTC's proposed remedy literally sought "changes to the composition of our board" ``. The nine-figure AI pay packages (below) are a governance-light, founder-driven splurge. No related-party self-dealing of note; the issue is concentration of decision rights, not theft.
5. Founder vs professional manager: archetype-pure founder-operator with absolute control. For this stage (a multi-decade platform bet), that's arguably the right archetype — only a controlling founder can spend $145B/yr against a 10-year payoff without being fired. It also means there is no circuit-breaker if he's wrong.

Key AI leadership — the 2025 rebuild: Meta paid up to reconstitute its frontier-model team after Llama fell behind ``:

• Alexandr Wang (ex-Scale AI CEO, 28) → Chief AI Officer, co-leads Meta Superintelligence Labs; arrived via the $14.3B Scale AI investment.
• Nat Friedman (ex-GitHub CEO), Ruoming Pang (ex-Apple foundation-models head, package reportedly >$200M), plus researchers poached from OpenAI (~40% of MSL), DeepMind (~20%), Anthropic, Apple.
• Comp packages reportedly $10M–$100M+ annually, nine-figure signing bonuses for stars. RSU grant-date fair value in the books stepped from ~$500 to ~$599/share Q/Q , and SBC ran **$6.03B in Q1-2026** (R&D $5.33B of it). There was a reported **hiring halt** after the initial blitz . Assessment: this is a high-conviction, high-cost bet to buy frontier talent. It signals Meta knows model quality is its weakest link — and is willing to spend whatever it takes. Whether $200M packages produce a leading model is unproven.

Lens 10 · Forensic Red Flags

Acting as a forensic analyst. Meta's accounting is clean by the numbers — but there are real aggressiveness vectors and a staggering legal tail.

Accounting-quality observations ``:

• Server useful-life extension (the margin lever to watch): effective Jan 1, 2025, Meta extended estimated useful lives of "most servers and network assets to 5.5 years" ``. This lowers annual depreciation and flatters current margins — a legitimate but earnings-favorable estimate change, and one to monitor as the AI-server fleet (which arguably has a shorter economic life given the GPU upgrade cadence) balloons. This is the #1 forensic flag: depreciation policy on rapidly-obsolescing AI hardware. If the 5.5-yr assumption is too generous, future write-downs or a snap-back in D&A could compress margins.
• Cash flow vs earnings: FY2025 OCF $115.80B vs net income $60.46B — OCF exceeds net income by a healthy margin (D&A $18.62B, SBC $20.43B, deferred taxes $18.74B are the non-cash adds) . **No divergence red flag** — earnings are well cash-backed. But **FCF FELL to $43.59B (from $52.10B in FY2024)** because capex ($69.69B P&E + $2.52B finance leases) outran OCF growth . The FCF erosion is the bear case in one number.
• SBC flattering nothing improperly — Meta reports GAAP; the $20.43B SBC is fully expensed. But 40M anti-dilutive shares were excluded from Q1-2026 diluted EPS (vs 1M a year earlier) `` — option/RSU overhang from the AI hiring spree is growing and will pressure share count.
• Tax volatility (already flagged in Lens 5): the ±$16B OBBBA/CAMT swing across Q3-25/Q1-26 makes GAAP EPS comparisons meaningless quarter-to-quarter. Use operating income.
• Off-balance-sheet commitment build: $237.67B non-cancelable contractual commitments + $182.88B leases-not-yet-commenced ``. These are real future obligations (cloud, servers, data centers) sitting largely off the balance sheet — the true forward capital commitment is far larger than the reported debt of ~$59B.

Regulatory findings ``:

• SEC Litigation Releases: none. SEC AAERs: none. Meta has no SEC accounting-enforcement history (verified via EDGAR EFTS, period 2021-06-17 → 2026-06-17). Clean.
• Item 3 Legal Proceedings (10-K, the company's own disclosure) `` — this is where the risk lives, and it is enormous. Meta states the aggregate potential monetary damages/penalties across all proceedings "could amount to an aggregate of up to hundreds of billions of dollars and... could be material to the financial condition of the company":
• FTC antitrust (Instagram/WhatsApp divestiture): Meta WON at trial — judgment in its favor Nov 18, 2025; FTC filed notice of appeal Jan 20, 2026. The existential structural risk is substantially de-risked at the district level (a major positive that post-dates much bearish positioning).
• AI copyright (Kadrey v. Meta): Meta WON fair-use summary judgment June 25, 2025 as to named plaintiffs — a landmark validation of training-data use; a distribution claim remains, trial mid-2027.
• Youth / social-media-addiction MDL + state AG cases: the live tail risk. First personal-injury trial began Jan 27, 2026 (LA Superior Court); first state-AG trial (New Mexico) scheduled Feb 2, 2026; school-district bellwether June 15, 2026. Damages claimed "up to the high tens of billions," plus 100,000+ mass-arbitration demands.
• EU fines (largely accrued/appealed): €798M Marketplace antitrust, €200M DMA "subscription for no ads," €1.2B GDPR data-transfer, €542M Spain unfair-competition (AMI). DSA preliminary findings on minors + content moderation pending. New front: EU antitrust investigation opened Dec 4, 2025 over WhatsApp Business API banning third-party AI chatbots `` — a fresh probe not in the FY2025 10-K.
• Historical settlements (closed): $725M Cambridge-Analytica consumer (final May 2025), $5.0B FTC penalty (2020), Delaware derivative settled in principle July 2025.
• Non-SEC web check ``: confirms the EU/FTC/youth-litigation matters above; no new material accounting-fraud allegations surfaced.

Net forensic verdict: the books are clean (no SEC enforcement, cash-backed earnings, GAAP reporting). The risks are (1) depreciation policy on AI hardware, (2) a legal tail the company itself sizes at "hundreds of billions," and (3) FCF erosion from capex. None is an accounting-integrity flag; all are real economic risks to monitor.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection (FY2026 / FY2027 / FY2028 EPS)

Built bottom-up from FY2025 actuals () + Q1-2026 run-rate + Street anchors (). All outputs `` with arithmetic shown. FY2025 base: revenue $200.97B, op income $83.28B, diluted EPS $23.49 (GAAP, distorted by the OBBBA charge; normalized ~13% tax rate FY2025 EPS ≈ ~$26–27).

Revenue path ``:

• FY2026: Q1 ran +33%; assume deceleration to +22–25% full-year as comps stiffen and DAP growth slows → **$248–253B** (aligns with Street ~$253B ``).
• FY2027: +15–18% → **$290–298B.**
• FY2028: +13–15% → **$330–342B.**

Margin path ``: operating margin held at 41% in Q1-2026 despite the capex ramp, but the depreciation wave from $125–145B capex hits the P&L with a lag (the 5.5-yr server life spreads it, but the absolute base is exploding). Base case: operating margin drifts 41% → ~38% (FY26) → ~36% (FY27) as D&A + RL losses + AI opex bite, then stabilizes as AI monetization scales.

EPS path (base / bull / bear) ``, using a normalized ~14–15% tax rate (post-CAMT noise), ~2.55B diluted shares, modest buyback offset to RSU dilution:

Base FY2026 ~$30–31 brackets the Street's ~$30–32.81 ; FY2027 base ~$35 aligns with Erste's $34.4 .

Valuation cross-check: at ~$600 and base FY2026 EPS ~$30, the forward P/E is ~20x — consistent with the web ~18–20x. On base FY2027 ~$35, ~17x. A re-rate to even the peer-average ~27x on $35 FY2027 EPS implies ~$945; a bear de-rate to 16x on $27 implies ~$430. The asymmetry is wide and hinges entirely on the capex-ROI verdict.

(Per --watchlist unattended rules: forecast.ts create is SKIPPED — no Brier forecast logged in breadth mode.)

Lens 12 · Bull vs Bear

Bull case. You are buying the cheapest mega-cap in the AI complex (~18–20x fwd) growing revenue +33% with a 41% operating margin and a fortress ad-duopoly moat. AI is already paying off inside the core: ad impressions re-accelerated to +19% and price/ad to +12% because the ranking models got better . The optionality is free — Ray-Ban glasses are seeing "unprecedented demand" with production doubling to 20M+ units , business-messaging/WhatsApp monetization is +50–74%, and if any of agents/superintelligence/glasses hits, it's pure upside not in the multiple. The FTC divestiture risk just collapsed (won Nov 2025) and the AI-copyright fair-use ruling went Meta's way (June 2025). Capital returns continue ($25B buyback authorized, growing dividend). The 2023 playbook says: when the spend narrative flips to discipline/ROI, this stock re-rates violently upward.

Bear case (2–3 permanent-impairment risks):

1. The capex-ROI trap (the dominant risk): $125–145B/yr — nearly 2× FY2025 capex and more than 2024+2025 combined — against an undefined payback. If superintelligence monetization doesn't arrive, the **depreciation wave permanently compresses margins and FCF** (already fell to $43.59B from $52.10B), and the multiple stays suppressed for years. JPMorgan's Doug Anmuth downgraded to Neutral on exactly this ``. Zuckerberg's "that's a very technical question" deflection on ROI is the bear's exhibit A.
2. Server-life / write-down risk: the 5.5-yr depreciation assumption on AI hardware that may have a 2–3 yr useful economic life is an accident waiting to happen; an impairment or D&A snap-back would hit earnings hard.
3. The legal tail: the youth-addiction MDL (damages "high tens of billions," trials live in 2026) + the EU regulatory grind could force product changes that impair the European ad business ``.

Pre-mortem (it's Dec 2027, the thesis broke): Meta spent ~$280B on AI infrastructure over 2026–27; ad growth decelerated to low-teens as the AI-monetization "second leg" never materialized at scale; the depreciation wave dragged operating margin to ~33%; FCF went negative in a quarter; an adverse youth-litigation verdict + EU DSA remedy forced costly product changes; and the stock de-rated to ~14x as the market concluded Meta is a capital-incinerating utility, not a compounder. The "very technical question" became the epitaph.

Are multiples too high? No — they are arguably too low relative to growth and margin, if you believe the capex earns its return. The ~18–20x forward is a discount, not a premium; the market is explicitly paying Meta a lower multiple than slower-growing peers as insurance against the capex bet. This is the rare mega-cap where the bear case is "the multiple is right" and the bull case is "the multiple is wrong on the cheap side."

Contrarian view (what the market refuses to see): The market is treating the capex raise as a cost overrun (component-price-driven, ROI-unclear). But the same balance-sheet capacity to spend $145B/yr is itself the moat — it is precisely the barrier that keeps OpenAI/Anthropic dependent on outside capital while Meta self-funds from a $115B OCF base. If superintelligence is real, Meta is one of ~4 entities that can afford to build it and has 3.5B distribution endpoints to deploy it into. The market is pricing the spend as pure risk and the distribution+balance-sheet advantage as zero.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case.

• Where revenue is concentrated: ~99% of profit is FoA advertising, and a meaningful slug of growth has come from the cross-border China commerce cohort (Temu/Shein) ``. A trade-war escalation, a Temu/Shein pullback, or a US de-minimis crackdown could knock out the marginal growth buyer overnight. The "diversified advertiser base" hides a concentrated growth dependency.
• Why the moat is weaker than bulls think: Meta's moat is in distribution, not AI capability. On the frontier-model axis it is a follower — Llama lagged badly enough that Meta paid $200M packages to rebuild the lab. If the next computing platform is AI-native and Meta's models stay second-tier, its distribution moat could be disintermediated by whoever owns the best assistant (OpenAI, Google).
• Most dangerous underestimated competitor: Google. It has the model (Gemini), the distribution (Search/Android/YouTube/Chrome), the custom silicon (TPU — a real cost advantage Meta lacks), and the same ad-auction machine. In an AI-ad world, Google is better-positioned on compute economics than Meta, which is a GPU price-taker.
• Worst capital-allocation moves: >$70B cumulative incinerated in Reality Labs with ~$19B/yr ongoing and a tiny $2.2B revenue base ; the **$14.3B Scale AI deal + nine-figure pay packages** to buy talent that may not deliver a leading model . The dual-class structure means no one can stop him.
• Assumptions that must hold for ~$600: ad growth stays >15%, operating margin defends >36% through the depreciation wave, the youth-litigation tail stays sub-catastrophic, and AI capex eventually shows some return signal. Break any one and the thesis wobbles.
• What happens if growth disappoints 20–30%: if FY2026 revenue comes in at ~$220B (vs ~$250B) and margin compresses to 33% on the depreciation load, normalized EPS could be ~$22–24 instead of ~$30; at a fear-multiple of 15x that's ~$330–360 — a ~40–45% drawdown from $600. The 2022 precedent (−65%) proves this stock can halve on a spend-without-payoff narrative.
• The single scenario that permanently impairs the business: sustained AI capex (~$130B+/yr) with no monetization second-leg, such that the depreciation base structurally exceeds incremental AI-driven profit — converting Meta from a 40%-margin compounder into a low-growth, capital-intensive infrastructure operator. Plausibility: moderate. It's the bear's whole thesis and it is not crazy — but the Q1-2026 ad re-acceleration is real evidence against it.

Lens 14 · Management Questions (15, ordered by information value)

1. On the $125–145B 2026 capex: what specific, measurable signals (revenue per GPU-hour, AI-attributable ad lift, agent revenue) will tell us by year-end whether this is earning its cost of capital — and at what point would you cut it?
2. What economic useful life are you actually assuming for the AI-accelerator fleet, and how exposed is operating margin to a downward revision of the 5.5-year server-life assumption?
3. Walk us through the path to a second revenue leg beyond advertising (agents, business messaging, AI services) — dollars, timeline, and the milestone that proves it's working.
4. How much of 2025–26 ad growth came from the cross-border commerce cohort (Temu/Shein et al.), and how exposed is the growth rate to a trade-policy or de-minimis shock?
5. Llama has trailed frontier peers — what is the concrete evidence that Meta Superintelligence Labs closes that gap, and by when do you expect a clearly-leading model?
6. What is the expected return on the $14.3B Scale AI investment and the nine-figure talent packages — how do you measure ROI on a $200M researcher?
7. Reality Labs has lost >$70B cumulatively. What is the revenue/operating-loss path that would justify continued ~$19B/yr funding, and what's the kill criterion?
8. On the FTC appeal of the Nov 2025 ruling and the youth-litigation MDL — what is the realistic range of aggregate financial exposure you're reserving against?
9. The EU DMA/DSA regime and the new WhatsApp-chatbot probe — quantify the revenue at risk if the "less personalized ads" model is forced further or the API restrictions are unwound.
10. FCF fell to $43.6B as capex outran OCF. At what capex level does FCF go negative, and are you prepared to let it?
11. Custom silicon (MTIA): what share of inference/training moves off NVIDIA by 2027, and what's the cost-per-token advantage versus your current GPU price-taker position?
12. You paused glasses' international rollout on "unprecedented demand" — what's the realistic 2026–27 unit and gross-margin trajectory for the EssilorLuxottica partnership?
13. With the dual-class structure giving you unilateral control, what governance circuit-breakers exist if the AI bet is wrong — or is the board structurally unable to course-correct?
14. The 40M anti-dilutive shares now excluded from EPS reflect a growing RSU overhang — how do you balance the AI talent-comp arms race against share-count discipline and buybacks?
15. If superintelligence does not arrive on your timeline, what does Meta look like in 2030 — and is there a version of the strategy that returns more capital and spends less?

---