Meta Platforms, Inc. (META)

1Business snapshot

Meta operates two reportable segments. Family of Apps (Facebook, Instagram, WhatsApp, Messenger, Threads) generated $198.76bn of FY2025 revenue and $102.47bn of operating income — a 52% margin FY2025 10-K. Reality Labs generated $2.21bn of revenue and lost $19.19bn FY2025 10-K.

The revenue model is narrow and transactional. Advertising was $196.18bn of FY2025's $200.97bn, or 97.6% FY2025 10-K. Ads sell through a real-time auction, not under contract, so there is no meaningful backlog. Revenue is impressions × price per ad — in Q2 2026, impressions rose 14% and price per ad 12% Q2 2026 10-Q.

The unit is one Family daily active person (DAP): 3.60bn on average for June 2026, up 3% Q2 2026 10-Q. Growth is therefore monetisation, not user acquisition. FoA revenue per DAP rose to roughly $16.8 from roughly $13.5 a year earlier, but FoA operating income per DAP fell to roughly $6.50 from roughly $7.13 (both computed) — more revenue and less profit per user.

Two non-advertising lines are emerging but small: FoA "other revenue" reached its first $1bn quarter (+73%, on WhatsApp paid messaging and subscriptions) Q2 2026 10-Q, and Meta has begun selling model access through a public API and a new subscription, Meta One Q2 2026 call.

2Core debate

Not whether the advertising business works — it does. The debate is what the AI infrastructure build does to the economics of owning it.

Bulls read capex as re-investment: the ad gains are AI-driven and measurable, compute is scarce, and Meta is buying an asset it can monetise several ways. Bears read a regime change in capital intensity, in which the free cash flow that defined Meta as an investment has become depreciation, debt and off-balance-sheet lease obligations, with no disclosed return metric. Both sides accept the same facts; they disagree on whether the margin and FCF compression is a J-curve or a new baseline.

3Base thesis

Meta owns the only Western consumer attention pool of its scale and is converting that attention into revenue faster than the attention itself is growing. Ad revenue grew 27% in Q2'26 while users grew 3% Q2 2026 10-Q — the gap is ranking quality, ad load and new surfaces (Threads globally, WhatsApp Status), all internally controlled rather than market-dependent. Management reports concrete gains: an 8.3% increase in ad clicks and 15.7% uplift in conversions on Facebook from new ranking models, and Advantage+ at over $75bn annual revenue run-rate Q2 2026 call. If those gains persist and the build is demand-led, operating income compounds through the investment period and the current margin trough is a bottom, not a trend.

Assumptions this rests on:

1. Price per ad holds double-digit growth. It has run +6% to +12% over the last five quarters and was +12% in Q2'26 Q2 2026 10-Q. This is the single most fragile input.
2. Ad load and new-surface expansion still have runway. Meta cites "ad load optimizations" as an impression driver Q2 2026 call; the current level and headroom are not disclosed.
3. Capex converts to revenue rather than to depreciation alone. $80.35bn sits in construction in progress and is not yet depreciating Q2 2026 10-Q; that lag flatters today's margin and reverses.
4. The 2026–27 capacity is used, not stranded. Management's plan is explicitly to "maximise 2026 and 2027 capacity" with flexibility deferred to 2028+ Q2 2026 call.
5. Youth-harm litigation resolves at a cost that is large but absorbable relative to ~$87bn of LTM operating income.

4Main bear case

The bear case does not require the ad business to break — only that the cost of running it has permanently risen. Q2'26 operating margin was 31% against 43% a year earlier; FoA margin fell to 39% from 53% Q2 2026 10-Q. Quarterly free cash flow was $784m Q2 2026 call, against $52.1bn for the whole of 2024 (computed). Buybacks, $30.1bn in 2024 and $26.2bn in 2025, were zero in H1 2026 despite $25.03bn remaining authorised FY2025 10-K, Q2 2026 10-Q. Long-term debt rose from $59.0bn to $84.0bn in six months, the May 2026 issue priced as high as 6.45%. Beyond the balance sheet sit $278.99bn of leases not yet commenced, roughly $68bn more signed in July 2026, and $349.31bn of non-cancelable commitments Q2 2026 10-Q. A shareholder who owned Meta for its cash returns no longer owns that company.

How it breaks: those obligations are fixed and long-dated (lease terms to 30 years) while the revenue justifying them is a variable auction. If ad growth decelerates to mid-teens while depreciation from $80bn of construction in progress lands and lease payments commence, operating income can decline for years without anything dramatic happening to the business. The Louisiana venture alone carries maximum exposure to loss of $46.03bn against a $2.92bn carrying value Q2 2026 10-Q.

5Business quality

Genuinely high, but two grades lower than it was two years ago on capital intensity rather than on competitive position.

What is durable: scale, a two-sided network across four apps, an ad system whose targeting and measurement loop improves with volume, and a diffuse advertiser base (9m small businesses use at least one AI creative tool Q2 2026 call). No customer concentration is disclosed.

What is not: the moat is contested. In FTC v. Meta the court granted judgment in Meta's favour on 18 November 2025; the FTC appealed on 20 January 2026 Q2 2026 10-Q. Reality Labs has lost $92.2bn cumulatively since 2020 (computed from FY2022, FY2024 and FY2025 10-Ks and the Q2 2026 10-Q) against $2.21bn of FY25 revenue — persistent evidence that capital allocation outside the core has been poor.

The economics have changed: Meta converted ~32% of revenue to free cash flow in FY2024 and 1.3% in Q2'26 (computed).

6Key value drivers

1. Price per ad — the purest read on whether AI ranking gains are real and captured by Meta rather than passed to advertisers.
2. Ad impressions — engagement, ad load and new surfaces, not users.
3. FoA operating margin — where cost growth lands first.
4. Capex-to-revenue — 51% in Q2'26 alone (computed). Whether it returns toward historic levels is the thesis.
5. Non-advertising revenue — FoA other, API, Meta One, Business Agents; ~1.7% of FoA revenue today.
6. Legal charge run-rate — $2.4bn in Q2'26 alone Q2 2026 call.

7Financial profile

Revenue compounded from $134.90bn (FY23) to $164.50bn (FY24) to $200.97bn (FY25), with operating margin 35% → 42% → 41% FY2025 10-K. H1 2026 revenue was $117.11bn, +30%, but operating income grew only 10% to $41.65bn Q2 2026 10-Q.

Cash generation has inverted. Operating cash flow is rising ($115.80bn FY25; $64.09bn H1'26), but capex including finance leases went $28.10bn (FY23) → $39.23bn (FY24) → $72.22bn (FY25) → $50.92bn in H1'26 alone. Free cash flow: $43.01bn, $52.10bn, $43.59bn, then $13.17bn in H1'26 (computed).

The balance sheet is strong in absolute terms — $90.3bn cash and marketable securities against $83.7bn of debt Q2 2026 call — but net cash has effectively gone. Property and equipment net is $225.72bn, of which $80.35bn is construction in progress not yet depreciating Q2 2026 10-Q. Depreciation is already accelerating: $6.00bn in Q2'26 against $4.28bn a year earlier.

One accounting item deserves attention: Meta extended the estimated useful lives of most servers and network assets to 5.5 years effective 1 January 2025 FY2025 10-K. The benefit is not separately quantified, but it flatters reported FY25 and FY26 margins relative to the prior policy.

FY2026 guidance: total expenses $165–169bn; capex $130–145bn; tax rate 15–17% for the remaining quarters; operating income above 2025 Q2 2026 call.

8Valuation

On reported earnings Meta does not look expensive; on cash it does. Trailing twelve-month operating income is roughly $86.9bn against TTM free cash flow of roughly $37.9bn (both computed from FY2025 10-K and Q2 2026 10-Q). The gap between those two numbers is the investment question: buying Meta today is buying a claim on operating income the company is choosing not to convert into cash.

The multiple is undemanding against earnings and demanding against cash. A view on Meta at this price is largely a view on when — and whether — capex intensity normalises.

9Management & capital allocation

Mark Zuckerberg controls a majority of voting power through the dual-class structure and can determine the outcome of all matters submitted to stockholders FY2025 10-K. There is no external check on capital allocation — the company's strength (fast, large, unconsensus bets) and its most expensive weakness (Reality Labs).

The current stance is unambiguous: infrastructure first, shareholder returns second. Buybacks went to zero in H1 2026 while the dividend continued at ~$1.35bn a quarter Q2 2026 10-Q. Debt was raised twice in eight months ($30bn November 2025, $25bn May 2026), and capital is increasingly sourced through partnerships — Blue Owl, and a new BlackRock venture for a 1GW El Paso campus Q2 2026 call. Susan Li describes operating cash flow as "the primary pillar" of funding, with 2027 options still under evaluation Q2 2026 follow-up call. Cost discipline runs alongside the spend: roughly 8,000 employees were affected by a May 2026 reduction, with $1.2bn of severance in Q2 Q2 2026 call.

Claims to verify over time:

• That compute demand exceeds supply and Meta is receiving offers "at multiples of what we paid for it" Q2 2026 follow-up call — asserted, never quantified, and Li explicitly declined to give a timeline for return on invested capital.
• That the venture backstops decline in value over time and provide flexibility every four years Q2 2026 follow-up call — testable against the disclosed RVG thresholds.
• That operating income in FY2026 exceeds FY2025's $83.28bn Q2 2026 call.

10Risks ranked by damage

11Metrics to monitor

12Thesis tripwires

13Open questions

• What is the actual return on the AI infrastructure — no metric is disclosed, and management declined to provide a timeline (see §9).
• What is the maintenance-versus-growth split of capex? Not disclosed. Without it, normalised free cash flow cannot be estimated.
• What is current ad load and how much headroom remains? Not disclosed.
• What quantified benefit did the 5.5-year server useful-life extension deliver to FY25 and FY26 margins? Not disclosed.
• What is the run-rate opex effect of the ~8,000-person reduction net of AI hiring? Not quantified.
• What is the outcome of the New Mexico trial commencing 8 September 2026, and of the MDL state-AG trial that began 12 August 2026? Both post-date the latest filings in sources/.
• Flagged, unverified: prior project notes reference a broad state-AG settlement reported in late August 2026 with an expected ~$10bn Q3 legal charge. This is not in sources/ and post-dates the Q2 10-Q. It must be confirmed against a primary filing (8-K or Q3 10-Q) before it is treated as fact or written into the thesis.

14Update log

15Source map