Business

Know the Business

Meta is an attention-to-advertising monopoly: one business unit (Family of Apps) reaches 3.58B daily users and contributes $102.5B of operating profit at 52% margins, funding a second unit (Reality Labs) that loses $19B a year. The lever that actually matters right now is whether 2026's $115–135B capex translates into durable AI-driven ad-pricing gains before incremental depreciation lands on the P&L. The market is probably pricing the AI ad uplift correctly, underestimating how much Reality Labs is pivoting from VR burn to wearable optionality, and underestimating the 2026 margin compression as server depreciation ramps.

1. How This Business Actually Works

Meta is not a "social network" for the purposes of valuation. It is a two-sided auction: the supply side is 3.58B daily users generating ~12% more ad impressions each year; the demand side is ~10 million advertisers bidding for those impressions, with AI-driven targeting pushing the price per ad up ~9% annually. Revenue grows at roughly impression growth × price growth — FY2025 delivered exactly that (+12% × +9% ≈ +22%).

The two levers move independently. Impressions grew even in 2022 when price/ad collapsed 16% — that year's recession was a pure pricing story, driven by Apple's ATT framework obliterating ad measurement and advertisers pulling budgets. Post-2023, AI ranking (Advantage+, Lattice) restored price/ad growth; Reels monetization catch-up is the tail wind still pending.

The real economic engine is monopolistic: FoA owns ~23% of global digital ad share, competitor ad networks cannot replicate 3.58B cross-product identity graphs, and the marginal cost of serving an extra ad is approximately zero once the data centers exist. R&D spend is the bottleneck — $57.4B in FY2025, +31% YoY — because AI model training requires ever-more expensive compute. Bargaining power is split: Apple (iOS ATT), EU regulators (DMA consent mandates), and Alphabet (Android signals) each squeeze a slice of the targeting stack. The "moat" is not technology; it is two decades of social-graph data that only WhatsApp, Instagram, Facebook, Messenger, and Threads see. Threads alone (~400M MAUs) would be a top-5 social network.

2. The Playing Field

Meta is the fastest-growing of the mega-cap cohort and trades at a meaningful multiple discount despite the second-highest operating margin.

Only Microsoft earns a higher operating margin, and only at half Meta's growth rate. Alphabet is the only true comp — both are ad-duopoly, both serve global reach — but Alphabet's margin structure is ~10 points lower (YouTube content costs, Cloud at sub-scale margins, Other Bets losses). Amazon's ad business would fit in this peer set, but its consolidated financials drown in commerce mix. SNAP and PINS are tiny reference points — both struggle to monetize their audiences anywhere near Meta's ~$57 ARPP. The peer set reveals one thing clearly: the premium businesses in digital ads are the ones with a closed identity graph. Pinterest and Snap do not have one; that is the full explanation for their margin profiles.

The "advantage" Meta shows over Alphabet is cost discipline and auction density — it earns better incremental margins on each ad dollar because its feed format is pure direct-response (vs YouTube's mix of brand and DR). The "weakness" vs Alphabet is regulatory exposure: Meta has no search fallback revenue and 98% of its top line depends on a targeting stack that EU DMA rulings could meaningfully reshape.

3. Is This Business Cyclical?

Digital advertising is cyclical on price, not volume. Impression supply grows almost mechanically with engagement; ad price is the cycle variable.

FY2022 is the template for the cycle: price/ad fell 16% while impressions rose 18%, producing a -1% revenue year. The drivers were unique-but-instructive — Apple ATT cut signal quality, macro inflation squeezed DR advertisers, and TikTok short-form video split attention and priced its inventory to win share. Revenue recovered fast once AI-driven measurement tools (Advantage+, conversions API, Lattice) filled the ATT gap. The operating line is more volatile than revenue: operating income fell from $46.8B (FY2021) to $28.9B (FY2022), a 38% drop, because fixed R&D and infrastructure costs don't flex with ad pricing.

Working capital is not a cycle driver here — Meta is deferred-revenue heavy and has near-zero DSO. Capex is the cycle driver now: 2026 capex of up to $135B against FY2025 operating cash flow of $115.8B means free cash flow conversion compresses sharply regardless of top-line trajectory. This has already started — FCF fell from $52.1B in FY2024 to $43.6B in FY2025 despite 22% revenue growth.

4. The Metrics That Actually Matter

Forget EPS headlines and total revenue. Five numbers explain value creation here.

Price per ad (YoY %) — the cleanest single read on the business. When this number is positive, Meta's targeting stack is working and AI investment is earning its keep. When it turns negative (2022), no amount of impression growth saves the operating line. Better than DAU/engagement ratios because it measures monetization quality, not just reach.

Family DAP — the supply-side ceiling. At 3.58B already, growth must come from sessions per user, ads per session, and format shift (Reels) — not new humans. Watch the YoY rate: if it drops below +4%, incremental ad dollars have to come entirely from pricing, which raises execution risk.

ARPP — the monetization quality score. $57/year worldwide, $16.6/quarter in Q4 2025. Compare to Snap (~$11), Pinterest (~$8) to see the moat. Rising ARPP + rising DAP = the only combination that matters.

Reality Labs operating loss — the discretionary burn. FY2025: -$19.2B. Management guided "similar" in 2026. This is the line to watch if AI capex disappoints — RL is where spending gets cut first. Framing is shifting from VR (Quest decline) to wearables (Ray-Ban Meta, Display + Neural Band): ~70% of 2026 RL opex is wearables-directed.

FCF conversion — the capex discipline read. FY2024: $52B FCF on $69B op income = 75% conversion. FY2025: $44B / $83B = 52%. If 2026 conversion drops below 35%, the AI-spend / returns debate goes from theoretical to binding.

Superficial ratios to ignore: P/E (noisy because of OBBBA tax-rate volatility — FY2025 ETR was 30% headline, 13% ex one-time charge), and gross margin (meaningful only as R&D-ratio, since "cost of revenue" excludes the real cost base).

5. What I'd Tell a Young Analyst

Three things to watch, in priority order.

First, watch price-per-ad every quarter, not headline revenue. Meta's moat is pricing power on an ad auction, not user count. If the sequence breaks — impressions up, price down — the AI story is failing faster than the capex is slowing.

Second, assume Reality Labs is $20B/year of permanent opex, not an investment. Do not capitalize it in your model. The market keeps pretending wearables will inflect; seven years of evidence say otherwise. If RL ever shrinks materially, that's upside — don't underwrite it.

Third, the 2026 capex cliff is the biggest risk and the biggest fight. $115–135B in 2026 vs $72B in 2025 means roughly $10–15B of incremental depreciation flows through in 2026–2027, and more after. If Advantage+/Lattice AI pricing gains don't outrun that, margins compress even with 20%+ revenue growth. This is the one number that would change a thesis.

What the market may be underestimating: WhatsApp monetization (paid messaging + business API), which grew FoA "Other" revenue +50% in FY2025 off a small base and has a genuinely different margin profile than advertising. Thread ad rollout (barely monetized) and smart-glasses optionality are lottery tickets — worth tracking but not underwriting.

What the market may be overestimating: the EU DMA/DSA ring-fence. Meta itself warned the Less Personalized Ads mandate could have "significant negative impact on European revenue"; Europe is ~24% of revenue. Also overestimated: how much direct AI revenue exists — there is none. Meta AI has ~600M MAUs and zero dollars of direct revenue attached to it.

Thesis-changer tests: (a) price-per-ad goes negative for two consecutive quarters, (b) DAP growth falls below 4% YoY, (c) Reality Labs loss widens past $22B/year, (d) EU LPA enforcement takes effect unmodified. Short of those, this remains a business where every year spent worrying about TikTok or the metaverse has been wrong, and the simplest model — impressions × price × DAP × ARPP — has been the right one.