META › analyze
META · Analyze
A portfolio-specific passage was removed from the public build.
⚠️ READ FIRST — THE DECLINE HAS TWO SEPARATE CAUSES, AND MOST COMMENTARY CONFLATES THEM
Date Close What happened 07-29 $585.61 Pre-print 07-30 $539.03 −8.0% — the print. No 2027 capex guide, FCF $784M, third straight $0 buyback 08-03 $590.24 The earnings drop was FULLY ROUND-TRIPPED in three sessions 08-11 $599.12 Peak — above the pre-print level 08-12 $578.85 −3.4% MDL trial jury selection 08-17 $568.97 −3.5% New Mexico $567M public-nuisance ruling digested 08-18 $543.67 −4.4% MDL opening statements + Nasdaq −1.2/−1.6% + long-end yields The market priced the capex shock and then unwound it. What is being priced right now — −9.3% in five sessions — is litigation, not capital spending. Any thesis that reads today's $543.67 as "the market finally waking up to capex" is reading the wrong tape.
Table of Contents
- The one thing that decides this stock
- Fundamentals
- Moat
- Sentiment & Intelligence
- Valuation
- Tension Resolution
- Verdict
- Data traps caught in this run
- Key metrics summary
- Sources
1. The one thing that decides this stock
Meta has spent four and a half years growing revenue 93.5% and free cash flow 5.1%.
| 2021 | TTM (to 6/30/26) | Change | |
|---|---|---|---|
| Revenue | $117.9B | $228.2B | +93.5% |
| Free cash flow | $39.0B | $41.0B | +5.1% |
| FCF margin | 33.1% | 18.0% | −15.1 pts |
| Capex / revenue | 15.9% | 39.1% | +23.2 pts |
| FCF / share | $13.83 | $15.97 | +15.5% (+3.3%/yr) |
Everything else in this report is commentary on that table. The ad engine is working better than at any point in Meta's history — it just passed Google to become the largest advertising business on earth. The cash it produces is going into the ground faster than it comes out.
The question is not "is the ad business OK?" It is: is management earning a return on the marginal $100 billion — and has it given you any way to check?
The answer to the second half is documented and unambiguous. Asked directly by Morgan Stanley's Brian Nowak for a 2027 capex number, CFO Susan Li said:
"We aren't providing a specific outlook for 2027 CapEx at this time. Infrastructure planning remains highly dynamic… Generally, we believe near term capacity is more valuable than long term capacity and it remains a very dynamic planning process."
Direction without magnitude. The Street filled the vacuum itself: HSBC models $176B for 2027; Deutsche Bank says buy-side expectations are "low-to-mid $200bn" with its own estimate ~$210–215B. That $176B–$215B spread for a single year is the valuation problem.
2. Fundamentals
Dispatched: Fundamentals Analyst
2.1 Q2 2026 — the quarter, and the honest bridge
| Q2'25 | Q2'26 | YoY | |
|---|---|---|---|
| Revenue | $47.516B | $60.801B | +28.0% |
| Gross profit | $39.025B | $49.471B | +26.8% |
| Gross margin | 82.1% | 81.4% | −0.7 pt |
| R&D | $12.942B | $21.656B | +67.3% |
| Total opex | $18.584B | $30.696B | +65.2% |
| Operating income | $20.441B | $18.775B | −8.2% |
| Operating margin | 43.0% | 30.9% | −12.1 pts |
| Net income | $18.337B | $15.848B | −13.6% |
| Diluted EPS | $7.14 | $6.18 | −13.4% |
| D&A | $4.342B | $6.356B | +46.4% |
| Capex (incl. finance-lease principal) | $16.538B | $31.08B | +88% |
| Free cash flow | $9.023B | $0.784B | −91.3% |
| Buybacks | $10.167B | $0 | 3rd straight quarter |
Now the bridge the headline hides — and it cuts in Meta's favour. Q2 carried $2.40B of legal-proceeding charges and $1.18B of severance for the May 2026 headcount reduction — $3.58B of one-time items:
| Reported | Ex one-time items | |
|---|---|---|
| Operating income | $18.775B | $22.355B |
| Operating margin | 30.9% | 36.8% |
| Diluted EPS | $6.18 (−13.4%) | ~$7.36 (+3.0%) |
Susan Li confirmed the direction on the call: ex-both items, operating income would have risen 9% YoY rather than falling 8%. So the P&L is materially healthier than the headline, and the operating-margin decline is −6.2 points, not −12.1.
But no add-back fixes a capex line. Free cash flow of $784M on $60.8B of revenue is not adjustable. That is the number that broke the stock, and it is the number that stays broken.
2.2 Trailing twelve months
| Metric | TTM (Q3'25–Q2'26) |
|---|---|
| Revenue | $228.247B |
| Operating income | $86.927B (38.1% margin) |
| Net income | $68.098B (29.8% margin) |
| Operating cash flow | $130.301B |
| Capex | $89.325B — 39.1% of revenue |
| Free cash flow | $40.976B (18.0% margin) |
| SBC | $25.136B — 11.0% of revenue, 61.3% of FCF |
| FCF less SBC | $15.84B |
| Diluted EPS | $26.55 (normalised $28.82) |
✅ Cross-validated. My hand-computed TTM matches stockanalysis.com exactly ($130,301M OCF / −$89,325M capex / $40,976M FCF).
2.3 🔴 Capital allocation — the engine has been switched off
| Use of cash, TTM | $B | % of uses |
|---|---|---|
| Capex | 89.33 | 84.5% |
| Acquisitions | 4.64 | 4.4% |
| Dividends | 5.37 | 5.1% |
| Buybacks | 3.33 | 3.1% |
| Debt repaid | 3.10 | 2.9% |
- Capex alone consumed 69% of operating cash flow.
- Shareholder returns were $8.69B = 6.7% of OCF, against 44.4% in FY2024.
- The funding gap was closed with +$51.7B of net new debt in twelve months.
Shares outstanding — the reversal that matters most to an owner:
| Shares | QoQ | |
|---|---|---|
| Q2'25 | 2,516.0M | |
| Q3'25 | 2,521.2M | +0.21% |
| Q4'25 | 2,529.6M | +0.34% |
| Q1'26 | 2,538.4M | +0.35% |
| Q2'26 | 2,548.4M | +0.39% |
+1.29% YoY. The FY2022–25 trend was −1.6%/yr. Per the framework — growth funded by dilution is not growth for me — this is a direct reversal of the per-share tailwind, and it has now run four consecutive quarters.
2.4 Balance sheet
| 6/30/25 | 6/30/26 | |
|---|---|---|
| Total assets | $294.7B | $450.0B (+52.7%) |
| Gross PP&E | $213.0B | $316.9B (+48.8%) |
| Construction in progress | $35.9B | $80.3B (+124%) — not yet depreciating |
| Capital-lease obligations | $20.7B | $28.7B (+38.2%) |
| Total debt | $49.6B | $112.3B (+126.6%) |
| Net debt | $16.8B | $68.2B (+305%) |
| Cash + ST investments | $47.1B | $90.3B |
| Equity | $195.1B | $261.2B |
Debt/Assets 25.0%. Net debt / TTM EBITDA 0.62×. Leverage is still low in absolute terms — this balance sheet can carry the build. That is a genuine strength and the main reason this is a HOLD and not worse. But the direction is violent, and May 2026 paper priced at 4.55%–6.45%, with El Paso JV debt reportedly demanded at >7% — roughly 40bp wider than the October 2025 Blue Owl/Hyperion SPV.
2.5 🔴 The depreciation wall — corrected arithmetic
My first pass on this was wrong and the correction matters. Steady-state D&A converges toward the capex run-rate, not capex ÷ useful life.
Disclosed facts (Q2 2026 10-Q): servers and network assets gross went $98.04B → $119.68B in six months. Q2 depreciation on servers/network alone was $4.62B (+48% YoY). Useful life is 5.5 years, set January 2025 (from 4–5 years), and the 10-Q confirms no policy change since. A second extension would be a red flag; it has not happened.
Modelled as cohorts (65% short-life silicon / 35% long-life structures), capex plateauing at $100B:
| 2026 | 2027 | 2028 | 2029 | 2030 | |
|---|---|---|---|---|---|
| D&A ($B) | 39 | 51 | 61 | 71 | 81 |
| Op margin (rev +15%/yr, EBITDA margin held flat) | 30.6% | 28.5% | 27.7% | 27.3% | 27.6% |
| Op income ($B) | 70 | 75 | 84 | 95 | 110 |
This is the most important correction in the report, and it is less dire than the raw gap suggests. Depreciation of $55–70B/yr by 2028 is contracted arithmetic — but because revenue compounds, operating margin lands ~27–28% rather than collapsing, and operating income dollars still grow. Meta does not become unprofitable. It becomes a ~28%-margin, ~20%-ROIC business instead of a ~41%-margin, ~30%-ROIC one.
That is a re-rating, not an impairment. The distinction is the whole verdict.
2.6 FY2026 free cash flow, on the company's own guide
Guidance: FY26 capex $130–145B (narrowed up from $125–145B). H1 actual $50.92B → H2 implied $79–94B, a further +27% to +51% step up from Q2's $31.08B.
| H2 OCF | FY capex $130B | FY capex $145B |
|---|---|---|
| $68B | FY FCF +$2.1B | −$12.9B |
| $72B | +$6.1B | −$8.9B |
| $76B | +$10.1B | −$4.9B |
FY2026 free cash flow lands between roughly −$8B and +$16B. The mid case is breakeven. FY2024 FCF was $54.07B.
2.7 ROIC — the collapse the July report forecast, arriving a year early
NOPAT (normalised 15.5% tax) ÷ invested capital:
| Q2'25 | Q3'25 | Q4'25 | Q1'26 | Q2'26 | |
|---|---|---|---|---|---|
| ROIC | 30.9% | 31.1% | 30.3% | 25.6% | 18.4% |
TTM-average basis: 25.8%. The July report projected "~21% in 2026 and ~19% in 2027." Spot ROIC hit 18.4% in Q2 2026. Both numerator and denominator moved against it.
2.8 Fundamentals scorecard
| Test | Reading | Verdict |
|---|---|---|
| FCF & CAGR | $41.0B TTM; +1.1%/yr over 4.5 yrs | 🔴 |
| FCF/share | $15.97; +3.3%/yr | 🔴 |
| Capital allocation | 84.5% to capex; returns 6.7% of OCF | 🔴 |
| Shares outstanding | +1.29% YoY, 4 straight quarters of dilution | 🔴 |
| Debt/Assets | 25.0%; net debt/EBITDA 0.62× | 🟢 |
| Revenue growth | +28% YoY, accelerating sequentially | 🟢 |
| Gross margin | 81.4%, essentially unchanged | 🟢 |
| Operating margin | 30.9% reported / 36.8% adjusted | 🟡 |
| ROIC | 30.9% → 18.4% in four quarters | 🔴 |
| Earnings quality | Net margin < op margin — no mark inflation | 🟢 |
3. Moat
Dispatched: Moat Analyst — qualitative Phase 1, quantitative Phase 2
3.1 🟢 The July report's central bear datapoint was falsified by the filing
This must lead, because the prior analysis built its downgrade partly on it.
| July 2026 claim | Q2 2026 10-Q reality |
|---|---|
| "Facebook impressions −5% while CPM +13% — price-led growth on shrinking volume, the terminal signature of a maturing ad surface" (Tinuiti advertiser panel) | Family ad impressions +14% YoY; average price per ad +12% YoY. Volume grew faster than price. |
Adding inventory normally lowers price. Both rising simultaneously means advertisers bid up for a supply that also expanded — a demonstrably better matching engine, not ad-load stuffing. This is the hardest pro-moat number available, and it points the opposite way from the third-party panel that drove July's conclusion.
Corroborating: Meta passed Google in worldwide ad revenue for the first time — $243.5B / 26.8% share vs Google's $239.5B / 26.4%, while accelerating to +24.1% against Google's +11.9% (EMARKETER). Instagram global time spent +double digits YoY. Instagram DAU crossed 2 billion. Family DAP 3.60B, +3%.
3.2 Moat decomposition — what is owned vs. what is rented
| Pillar | Owned/rented | Strength | Trend |
|---|---|---|---|
| WhatsApp network effect | Owned | Very strong | → |
| Conversion-feedback loop (4M advertisers' outcome data) | Owned — the actual moat | Strong | ↓ slightly |
| Facebook social graph | Owned | Strong | ↓ slow (13–17 = 3.2% of users, was 5.7% in 2019) |
| Instagram creator supply | Owned | Strong | → |
| Scale economics / auction depth | Owned | Strong | ↓ (op margin 43%→31%) |
| Ranking-model quality | Rented | Moderate | ↓↓ |
| SMB tooling "lock-in" | Rented | Weak-moderate | ↓ |
| WhatsApp API monetisation | Being expropriated | Moderate | ↓↓↓ |
| Frontier model capability | Rented — literally | Weak | ↓ |
Two findings deserve emphasis:
The SMB "switching cost" is scale economics mislabelled. Advantage+ runs at a >$75B annualised run-rate and 9M SMBs use Meta's AI creative tools (from 4M a year ago). But performance advertisers reallocate on ROAS daily — nothing is locked in; the budget is re-won every quarter. Worse, Advantage+ actively destroys the one real switching cost that existed — the accumulated in-house skill of running Meta ads. Automate the campaign and you remove the reason a marketer's expertise is Meta-specific.
Meta is a net buyer of frontier intelligence from its own competitors. The 10-Q names "third-party AI token costs" as a driver of the 55% expense increase. Google restricted Meta's Gemini access in March 2026 because it could not supply the capacity Meta wanted to buy. A company spending $130–145B a year on compute was rationed by a rival on model access. That is not what a compute moat looks like.
3.3 Is the compute a moat or an expense?
Split the asset. Buildings ($61.0B, 20–30 year lives), power, land and grid interconnect are a durable barrier. Silicon ($119.7B gross and compounding, 5.5-year life) is a depreciating expense. Meta is paying ~85–90% of the bill for the perishable half.
And a barrier that admits every one of your competitors is not a barrier. The "only 4–5 firms can afford this" claim describes Google, Microsoft, Amazon, OpenAI/Anthropic, and xAI — the complete list of Meta's actual rivals. It excludes only entities that were never going to compete.
The commitment is also far larger than the capex line shows:
| FY2026 capex guide (what screens see) | $130–145B |
| Uncommenced leases signed, at 6/30/26 | $279.0B |
| Additional data-centre leases signed July 2026 | ~$68B |
| Total non-cancelable contractual commitments | $349.3B |
| El Paso JV residual-value guarantee | ~$13B |
| Cash escrowed under infrastructure agreements | $10.8B |
| Unrecognized SBC (~3yr) | $79.8B |
The El Paso structure — contribute $2.3B of assets for a 20% interest, take back $1B, lease the capacity back, guarantee up to $13B of residual value — is capex migrating off the capex line into structures no screener reads.
A portfolio-specific passage was removed from the public build.
3.4 Competitive landscape — every named attention rival is shrinking or sub-scale
| 2026 status | |
|---|---|
| TikTok US | JV closed Jan 2026; no user exodus (~95% of baseline, ~80 min/day). US ad rev $14.5B — Meta US ≈ 7×. Ban risk removed in both directions |
| YouTube | Q2'26 ads $11.1B, +13% — Meta out-growing it by 14 points |
| Snap | NA DAU −7%; cut 16% of staff Apr 2026 |
| Revenue +61% and the stock fell 21–23% — top-of-funnel is rented from Google | |
| X | Audited $1.8B 2025 ad revenue, −59% from 2022. Threads has passed it on DAU |
| Bluesky | 10.4M MAU, −27% YoY |
| Snap+Pinterest+Reddit+X+Walmart combined | 2.4% of worldwide digital ad spend |
Reddit is the instructive read-across. It grew revenue 61% and lost a fifth of its market cap because management conceded AI search does not replace what it displaces. Meta owns its distribution end to end. That contrast is the strongest available argument for Meta's moat — and it comes from a competitor's earnings call.
The AI-assistant attention thesis is not supported by 2026 data: all genAI platforms combined draw 16 minutes/day for US adults, against >90 minutes for social media and ~25 for Instagram video alone. eMarketer attributes record media time to genAI plus "resilient social media engagement" — genAI reads as additive. ChatGPT ads reached >$100M annualized — 0.04% of Meta's quarterly ad revenue. The Gemini app carries no ads at all.
"TikTok" appears zero times in the Q2 transcript. No analyst asked about competitive erosion. The sell-side's concern is entirely capex and ROI.
The real AI risk is narrower and later: commercial intent interception. AI-referred retail traffic is +393% YoY and converts 42–54% better than non-AI — off a ~1.08% base of visits. High quality, trivial volume. A 2029+ risk, not a 2027 revenue event.
3.5 Reality Labs
Cumulative operating losses ≈$92.2B since 2020 (Q2'26 alone: −$4.619B). RL revenue $431M in Q2. ~7:1 burn ratio over six years.
⚠️ The widely circulated "$83.6B cumulative" figure is correct only through 2025-12-31 and omits the $8.65B lost in H1 2026.
Quest is retreating (headset sales −16% YoY). Horizon Worlds is effectively abandoned on VR. Smart glasses are the one thing genuinely working — 7M+ units in 2025 (3× the 2023–24 combined), 69.2% category share, Ray-Ban Display's international rollout paused for demand.
But 7M+ units at $299–799 cannot be reconciled with RL's ~$1.7B annualized revenue — EssilorLuxottica books the bulk of the hardware. Meta is funding the category's creation and letting its partner take the hardware margin. The glasses cannot pay for Reality Labs at any plausible 2030 volume.
Management has never disclosed a hurdle rate, stop-loss, or kill criterion in six years and $92 billion. That is not an option purchase; it is an open-ended subscription. The governance failure, not the loss, is what should be underwritten.
3.6 Adversarial stress test — "$50B, no legacy. How do I attack Meta?"
Do not touch the graph. X lost 59% of its ad revenue trying; Bluesky lost half its users; Snap's NA DAU is going backwards. The graveyard is full and the cause of death is always the same.
Attack the layers above and below it — intent and measurement. 1. Buy the measurement layer. Meta grades its own homework. Sell provable incremental ROAS. This is exactly how Amazon took +67% DSP growth in Q2 2026. 2. Be upstream. Never build a feed; build the assistant that answers "what jacket should I buy." The query never reaches Meta's auction. 3. Litigate for distribution — and this already worked. A Dec 2025 complaint to DG COMP produced, by June 2026, an interim measure granting rival AI assistants free access to the WhatsApp Business API. Legal fees bought distribution across a 3.3-billion-user network. Similar investigations are now open in Africa, Brazil, and Turkey.
Why it mostly fails: $50B is six months of Meta's capex. The conversion-feedback loop is genuinely unpurchasable. Time-spent is not moving. Why part of it works anyway: moves 2 and 3 do not require beating Meta, and one of them has already paid off at essentially zero capital cost.
3.7 Evergreen rating: 6.5 / 10 (from 7.0 in July, 8.0 in March)
Why not lower: 3.60B daily people. Impressions +14% and price +12% in the same quarter. 82% gross margin. Largest ad business on earth while accelerating against Google. Every named competitor shrinking or a rounding error. The bear case has never been "the users leave," and 2026 gave no evidence that they are.
Why not 7 any more: FoA operating margin 53% → 39% in twelve months. Buybacks zero for three quarters with the share count rising. $279B of uncommenced leases. Depreciation of $55–70B/yr by 2028 already contracted. The EC interim measure landed on the fastest-growing revenue line Meta owns. RL at $92B with no kill criterion. Google rationed Meta's Gemini access.
The formulation I would defend: Meta's moat is intact and its capital allocation is now the dominant variable. For fifteen years the question was "can anyone take the users?" — still no. The question now is "is management earning a return on the marginal $100B?" and management has removed the investor's ability to check. An evergreen business is one you can stop watching. This one requires watching a decision rather than a franchise.
4. Sentiment & Intelligence
Dispatched: Sentiment & Intelligence Analyst
4.1 Guidance — verbatim
"We expect third quarter 2026 total revenue to be in the range of $61-64 billion… We now expect full year 2026 total expenses to be in the range of $165-169 billion… We continue to expect to deliver operating income this year that is above 2025 operating income… We anticipate 2026 capital expenditures, including principal payments on finance leases, to be in the range of $130-145 billion, narrowed from our prior outlook of $125-145 billion… we expect our tax rate for the remaining quarters of 2026 to be between 15-17%… we continue to see scrutiny on youth-related issues… and have a number of youth-related trials scheduled for this year in the U.S., which may ultimately result in a material loss."
🚩 Correct the record: multiple syndicated outlets print the 2026 capex guide as "$135–145B" or "$115–135B." Both are wrong. The primary source says $130–145B, narrowed from $125–145B — Meta raised the low end by $5B.
4.2 🔴 The legal docket — this is what is moving the stock today
Meta's own 10-Q states that maximum aggregate damages sought across its proceedings "could amount to an aggregate of up to hundreds of billions of dollars and, as a result, could be material to the financial condition of the company." That sentence is in the filing.
| Matter | Status |
|---|---|
| MDL 3047, N.D. Cal. (Judge Yvonne Gonzalez Rogers) | Trial began 08-12; opening statements TODAY 08-18. Claims by 4 states (CA, CO, KY, NJ) + a federal COPPA disgorgement claim by all 29 state AGs. 4–8 weeks; verdict expected Oct–Nov. Zuckerberg expected to testify. The jury is advisory only — Gonzalez Rogers issues the final ruling. |
| New Mexico #1 | Mar 2026 jury: $375M penalty. 08-06: judge declared Meta's platforms a public nuisance — first ever for social media — ordering $567M into an abatement fund PLUS sweeping product mandates (no overnight notifications, mandatory under-18 time limits). ~$942M total. |
| New Mexico #2 | Trial 2026-09-08 — AG intends to seek up to $62.85B |
| Tennessee AG | Trial began 07-20, ongoing |
| User bellwethers | 1st: Mar 2026, ~$4.2M to Meta (on appeal). Next two: 2026-10-28 |
| Mass arbitration | >200,000 individual demands over Instagram |
| FTC antitrust | Meta WON at district court 2025-11-18. FTC appealed 01-20-2026. Meta's response brief due 2026-08-20. No oral argument scheduled. Divestiture risk remote inside 12 months |
| FTC consent-order modification | Would ban commercial use of minors' data and limit product launches. Status update due 2026-09-08 |
| EU DSA minors | Preliminary findings 04-29 and 07-10-2026: under-13s present, addictive design. Max fine 6% of turnover → >$12B |
| EC WhatsApp API | Interim measure imposed June 2026 — free API access for rival AI providers. Meta appealing. Parallel cases in Africa, Brazil, Turkey |
| IRS | $15.89B asserted for 2017–19 transfer pricing; unrecognized tax benefits $18.74B |
On the "$1.4 trillion" headline: it is not a plaintiff demand. It is Meta's own characterization of theoretical maximum statutory-penalty exposure, which Meta calls "vastly disproportionate." Plaintiffs' counsel reportedly told the judge $200B is "more likely." Do not quote $1.4T as a claim.
The asymmetry that matters: injunctive relief, not dollars. A fine hits one quarter. A court-ordered change to how Meta ranks content for minors hits engagement and ad load permanently — pattern-ranking-function-as-policy-variable firing at maximum amplitude. New Mexico has already set that template.
4.3 Analyst positioning — targets slashed, ratings untouched
| Analysts | 57 · Recommendation mean 1.35 "Strong Buy" |
| Mean / median target | $754.14 / $750.00 |
| High / Low | $1,000 / $580 |
| Sell ratings | Zero, across every tracker |
🚩 At $543.67, META trades ~6% BELOW the lowest published sell-side target on the Street. Historically this signals stale models, not a floor.
Every post-print action was a price-target cut with the rating reiterated: HSBC $905→$830, Evercore $930→$820, JPMorgan $875→$800, Rosenblatt $1,015→$883, Mizuho $835→$750, Baird $830→$750, TD Cowen $800→$750, UBS $766→$715, Cantor $770→$680, DA Davidson $850→$700, Susquehanna $900→$650 (−28%, the largest). Erste cut FY2027 EPS $35.13→$33.92.
Read the direction, not the level. Universal one-directional cuts with zero rating changes is the classic "we still believe but we can't defend the number" posture. MarketBeat's observation is the tell: the percentage cut in targets exceeded the actual share-price decline — analysts cut faster than the stock fell, then the stock caught down to them and kept going.
Mizuho cut specifically because Meta announced a compute business with no contracts, no pricing, no named customers, no revenue. That is a fixable complaint, and it defines the catalyst.
4.4 Insider activity — and a pitfall firing live
🚩
pitfall-yahoo-insider-purchases-counts-rsu-grants— CONFIRMED, with a NEW mechanismYahoo currently reports for META: "Net Shares Purchased +1,093,984 · +41.1%." This is false as open-market buying. Every Form 4 since 2026-07-29 was pulled from EDGAR. There is not a single transaction code P. The "purchases" are (a) quarterly RSU vestings, and (b) — the new mechanism — Zuckerberg's 591,690-share Class B→Class A conversion (code C), which was the mechanical first leg of a charitable gift, not a purchase.
- Open-market buys since June 2026: ZERO. No officer or director stepped in with the stock 31% off its high.
- Sales are metronomic 10b5-1 selling, overwhelmingly one person (COO Javier Olivan, weekly clockwork) plus routine director diversification. Total 6-month sales ≈ 0.009% of the company. Zero acceleration into the decline — the same fixed share counts at $680, $600 and $543. No informational content.
- CFO Susan Li and CTO Bosworth last sold 2026-05-18 — no C-suite selling after the print.
- 🔴 Correcting a common premise: Zuckerberg has filed NO open-market sale (code S) since at least 2025-11-01. Every transaction in ~10 months is a charitable gift (G) or a conversion (C) enabling one. He has not sold into this decline.
4.5 Short interest and institutions
Short interest FELL 8.8% to 31.46M shares = 1.43% of float, 1.55 days to cover. Shorts covered into the drop. No squeeze setup, no short-driven pressure — this is long-side de-risking.
⚠️ Q2 13Fs (filed ~08-14) report positions as of 06-30 — a month BEFORE the print. They say nothing about the reaction. For what it is worth, the 6/30 picture was a rotation: Appaloosa +55%, Viking 1.06M→1.87M, Pershing Square 2.66M→3.2M added; Bridgewater −46%, Tiger −8.5%, Third Point exited entirely.
4.6 AI strategy
"Llama" appears zero times in the Q2 transcript. The family is now Muse from Meta Superintelligence Labs: Muse Spark 1.0 (Apr 2026, first closed-weight frontier model), Spark 1.1, Muse Image, Muse Glimmer 30B (open-weight Apache 2.0, 08-10). Third-party benchmarking puts Muse Spark 4th behind Gemini 3.1 Pro, GPT-5.4 and Claude Opus 4.6.
Zuckerberg conceded the position: the models are impressive "for the scale of the model, the kind of stage of development of the lab." And Susan Li pre-emptively hedged the entire frontier ambition — the most revealing line in the transcript for a company spending $130B+/yr:
"…our distribution advantages will give us the opportunity to serve AI products that are valuable for everyone… This should be true regardless of whether our models are on the frontier."
Meta AI usage: Meta has stopped disclosing absolute numbers. Only "+60% growth in daily people interacting." Any "1.2B MAU" figure in circulation is a third-party aggregate, not company-disclosed. That silence is itself a datapoint.
Meta Compute: announced, uncontracted. Zuckerberg says Meta is "getting a lot of offers for compute at a significant premium over what we paid for it" and floated "an efficient auction over our compute." No pricing, no named customer, no revenue, no timeline. Only "We'll have more to share on all of this soon." Most plausible venue: Connect, 2026-09-23.
5. Valuation
Dispatched: Valuation Analyst
5.1 🚩 First, the multiple everyone is quoting is wrong
| Yahoo forward P/E displayed | 15.58× → implies forward EPS $34.90 |
| FY2026 consensus EPS | $26.70 |
| FY2027 consensus EPS | ~$30.50 (Erste cut to $33.92; Zacks $30.98) |
$34.90 matches no published consensus year. H1'26 reported EPS is $16.62 (normalised ~$13.34); reaching $34.90 for FY26 needs $9.14/qtr in H2 against Q2's actual $6.18 — a +48% per-quarter jump into rising depreciation. Not plausible.
The honest forward multiples:
| Basis | EPS | P/E at $543.67 |
|---|---|---|
| TTM GAAP | $26.55 | 20.5× |
| TTM normalised | $28.82 | 18.9× |
| FY2026 consensus | $26.70 | 20.4× |
| FY2027 consensus | $30.50 | 17.8× |
The watchlist's "17.2x fwd" and the screen's "15.6x" both overstate the cheapness. This is pitfall-vendor-forward-eps-is-stale-on-the-day-of-a-guidance-cut persisting three weeks past the cut.
5.2 Where the market prices it
| Metric | Value |
|---|---|
| Price / market cap / EV | $543.67 / $1.395T / $1.463T |
| P/E trailing GAAP / normalised | 20.5× / 18.9× |
| P/FCF (TTM) | 34.0× |
| P/FCF less SBC | 88.1× |
| P/FCF (FY2026E, ~breakeven) | n/m |
| EV / TTM operating income | 16.8× |
| P/B | 5.3× · P/S 6.1× · FCF yield 2.94% |
| Dividend yield / payout | 0.37% / 8% |
| 52-week range | $520.26 – $790.80 (−31.2% from high) |
| 10yr avg forward P/E | ~26–27× |
The gap between 18.9× earnings and 34.0× free cash flow is the entire thesis. On earnings it looks cheap. On cash it does not.
5.3 Model 1 — Graham IV (weight 10%)
√(22.5 × EPS × BVPS) with EPS $28.82, BVPS $102.52 → $258. Graham says 53% overvalued.
The usual objection holds — the formula penalises businesses whose real assets never touch the balance sheet. But note the twist: BVPS went $95.99 → $102.52 in four months as Meta capitalises data centres. Graham becomes progressively MORE relevant as Meta turns asset-heavy. Treat as a floor reference, not a target. Weight 10%.
5.4 Model 2 — Bogle expected return (weight 15%)
| Scenario | Yield | Earnings growth | P/E change | Return |
|---|---|---|---|---|
| Bear — capex rises again, margin compresses | 0.37% | 4% | 20.4→16 (−4.7%/yr) | −0.4%/yr |
| Base — capex plateaus 2027, margin troughs ~28% | 0.37% | 10% | 20.4→19 (−1.4%/yr) | +9.0%/yr |
| Bull — capex flat, FCF snaps back, buyback resumes | 0.37% | 15% | 20.4→24 (+3.3%/yr) | +18.7%/yr |
Base ~9%/yr is adequate but not compelling against a ~4.3% 10-year. The bear case is dead money, not a loss. Weight 15%.
5.5 Models 3 & 4 — DYT and DDM: N/A
Yield 0.37%, payout 8%, ~2.5 years of dividend history. No yield band exists; a DDM on a company retaining 92% of earnings tells you nothing. Excluded rather than weighted for completeness. One observation: with buybacks at zero and net debt up 305%, a dividend raise is now a less likely use of capital than it was in March.
5.6 Model 3 — Earnings power on FY2027 (weight 55%, primary)
FCF-multiple valuation is unusable this cycle — FY2026E FCF is somewhere between −$8B and +$16B, and any P/FCF built on it is noise. So value on earnings, and express the capex question through the multiple.
| Scenario | FY27 EPS | Multiple | FV | Thesis |
|---|---|---|---|---|
| Bear | $26 | 14× | $364 | 2027 capex guided >$200B; D&A wall bites; op margin →25%; ROIC →15%; MDL yields injunctive relief |
| Base | $30.50 | 19× | $580 | Capex plateaus $130–145B; op margin troughs 28–30%; FCF recovers toward $60B by 2028; buyback resumes 2028 |
| Bull | $34 | 23× | $782 | Capex guided flat or down; Meta Compute lands real contracts; FCF snapback >$80B; buyback resumes 2027 |
5.7 Model 4 — Reverse DCF (weight 20%)
At a 9% discount and 3% terminal growth, EV of $1.463T requires steady-state FCF of ~$88B. TTM FCF is $41.0B.
The market requires Meta to roughly 2.1× its free cash flow and hold it.
⚠️ Note what happened since July. At $597.47 the same calculation required ~$106B against TTM FCF of $48.3B — a 2.2× improvement. At $543.67 it requires $88B against $41.0B — 2.1×. The stock fell 9% and the required improvement barely eased, because FCF fell almost as fast as the price. You are not being paid much for the decline.
Implied fair value on achievable FCF paths: ~$470.
5.8 Weighted fair value
| Model | Weight | FV | Contribution |
|---|---|---|---|
| Graham | 10% | $258 | $25.80 |
| Bogle | 15% | (return-based: base +9%/yr) | supportive, no point estimate |
| Earnings power FY27 | 55% | $580 | $319.00 |
| Reverse DCF | 20% | $470 | $94.00 |
Weighted central: ~$516 · Fair-value range $450 – $650
At $543.67, META trades ~5% ABOVE central fair value — fairly valued to marginally rich, with a wide band and high uncertainty.
That conclusion is worse than July's "roughly at fair value" despite a 9% lower price, and the reason is precise: the price fell 9%, but FY2026 FCF guidance fell from "somewhere positive" to "approximately zero," ROIC fell 720bps, and the 2027 capex number was refused. The denominator deteriorated faster than the numerator.
6. Tension Resolution
Manager arbitration
Tension 1 — "The ad moat is eroding" (July report) vs. the 10-Q
- July Moat Analyst: Tinuiti panel showed Facebook impressions −5%, CPM +13% — price-led growth on shrinking volume.
- Q2 10-Q: family impressions +14%, price per ad +12%.
- Resolution — the filing wins, decisively, and July was wrong. The panel measured a US-weighted, Facebook-surface advertiser mix; the company reports family-wide. Mixing the two produced an inverted conclusion. The demand side of the moat is stronger than the prior analysis assumed, and this argues for an upgrade to business quality. It is the main reason conviction falls only half a point despite everything in §2.
Tension 2 — "Operating margin collapsed 12 points" vs. "ex-items it rose"
- Fundamentals: op margin 43.0% → 30.9%, EPS −13.4%.
- Sentiment/company: $3.58B of one-timers; ex-both, op income +9% YoY, adjusted margin 36.8%, adjusted EPS +3%.
- Resolution — both are right, and the adjusted number is the fairer read of the P&L. Skepticism is the default in both directions; refusing a legitimate add-back is as sloppy as accepting a fake one. But the adjustment is irrelevant to the actual problem. Free cash flow of $784M contains no add-backs. The P&L is fine. The cash flow statement is the story. Anyone arguing this quarter on operating margin — bull or bear — is arguing about the wrong statement.
Tension 3 — "31% off the high and below the lowest sell-side target" vs. "fair value ~$516"
- Sentiment: 57 analysts, zero sells, mean target $754, lowest $580 — the stock is 6% below the lowest.
- Valuation: weighted FV ~$516; the stock is ~5% above it.
- Resolution — the sell-side is stale and the gap is evidence, not opportunity. MarketBeat's own note that target cuts exceeded the price decline shows analysts revising down faster than the tape. A mean target of $754 embeds forward EPS near Yahoo's phantom $34.90.
principle-down-a-lot-is-not-cheapapplies directly: the drawdown is measured against a $790.80 high set when FCF was $54B/yr and buybacks were running. That high was priced on a cash-generation profile that no longer exists. The discount is largely earned.
Tension 4 — Legal risk: tail event or thesis risk?
- Moat: ranks youth-safety litigation #2 by expected damage, citing injunctive relief.
- Sentiment: the $1.4T figure is Meta's own maximum-exposure framing, not a claim; plaintiffs' working anchor is ~$200B.
- Resolution — size it honestly and separate the two risks. The monetary exposure is survivable: Meta has $90.3B of cash, $130B of annual OCF, and even a $200B judgment paid over years would not break the balance sheet, though it would end the buyback debate permanently. The injunctive risk is the real one and it is not quantifiable. New Mexico already ordered product changes on 08-06 — the template exists. A federal ruling mandating how Meta ranks content for minors is a permanent constraint on engagement and ad load. This is correctly a conviction cap, not a valuation input — you cannot model it, so you demand a discount for it. That demand is why the entry zone sits well under fair value.
7. Verdict
HOLD — do not add here · Conviction 6.0 / 10 ⬇️ (from 6.5)
Fair value $450 – $650 · central ~$516 · Entry $455 – $495 · Strong buy < $440 · Trim 22× fwd
The case in five lines
- The business is better than the prior analysis assumed. Impressions +14% and price +12% simultaneously; Meta passed Google to become the largest ad business on earth; every named competitor is shrinking. July's central bear datapoint was falsified by the filing.
- The economics are worse than the prior analysis assumed. FY2026 FCF is guided-implied to roughly zero; ROIC fell 30.9% → 18.4% in four quarters; buybacks are off for three quarters and the share count is rising; net debt tripled to $68.2B.
- The single largest variable was explicitly refused. No 2027 capex number, with direction signalled higher. The Street's own range is $176B–$215B. You cannot underwrite that spread.
- At $543.67 the stock is ~5% above central fair value — and the reverse DCF still demands a 2.1× increase in free cash flow, barely easier than at $597 three weeks ago.
- A live federal trial with unquantifiable injunctive risk is in opening statements today.
Why 6.0 and not lower
The franchise is not impaired. Gross margin is unchanged at 81.4%, leverage is genuinely low (net debt/EBITDA 0.62×), revenue is accelerating sequentially, and the depreciation wall — modelled properly — compresses margin to ~28% while operating income dollars still grow. This is a business converting from a ~41%-margin toll collector into a ~28%-margin infrastructure-heavy compounder. That is a re-rating, not an impairment, and much of it is now in the price.
Why 6.0 and not higher
Three things a long-term owner is entitled to have been removed at once: the buyback, the falling share count, and the disclosure needed to check management's return on the marginal $100B. Add a live trial whose worst outcome is a permanent constraint on the ranking function. A business you must watch a decision on, quarter by quarter, is not a 7.
What to do
A portfolio-specific passage was removed from the public build.
The falsifiable tests
🟢 Upgrade to 7.0+: a disclosed 2027 capex number with an ROI framework · buybacks resume at scale · a named Meta Compute customer with a dollar figure (most likely at Connect, 09-23) · FoA operating margin stabilises ≥45% for two quarters · MDL resolves without injunctive relief.
🔴 Downgrade to 5.0 or below: 2027 capex guided >$200B · MDL yields injunctive relief constraining minor-facing ranking · ad impressions turn negative · FoA operating margin <35% · a second extension of server useful life (the accounting response to a depreciation wall) · EC WhatsApp interim measure survives appeal and replicates in India or the US · a fifth straight quarter of dilution with no buyback.
8. Data traps caught in this run
| # | Trap | Detail |
|---|---|---|
| 1 | Vendor forward EPS is phantom | Yahoo's 15.58× implies $34.90, matching no consensus year (FY26 $26.70, FY27 $30.50). Real forward is 20.4× / 17.8×. pitfall-vendor-forward-eps-is-stale-on-the-day-of-a-guidance-cut, 3 weeks past the cut |
| 2 | "Buybacks" means two different things | Yahoo TTM $3.33B vs stockanalysis $21.44B. The $18.1B gap is RSU tax withholding, not share retirement — which is why the count still ROSE 1.29%. NEW note written |
| 3 | Yahoo insider "purchases" +41.1% | Zero code-P transactions exist. RSU vestings plus a new mechanism — a Class B→A conversion (code C) enabling a charitable gift. pitfall-yahoo-insider-purchases-counts-rsu-grants extended |
| 4 | Third-party ad panel inverted the conclusion | Tinuiti Facebook-surface "impressions −5%" drove July's downgrade; family-wide 10-Q says +14%. NEW note written |
| 5 | Capex line understates commitment by ~3× | $130–145B guided vs $279B uncommenced leases + $68B July + $349B contractual + $13B RVG. pitfall-lease-financed-capex-hides-from-the-fcf-line extended to megacap scale |
| 6 | Yahoo capex excludes finance-lease principal | Yahoo Q2 FCF $1.746B vs the 10-Q's $784M — a 2.2× overstatement of the quarter's cash generation |
| 7 | "Reality Labs has lost $83.6B" | Stale — correct through 12/31/25 only. Current: ~$92.2B |
| 8 | "$1.4 trillion exposure" | Not a plaintiff demand. Meta's own maximum-statutory-penalty framing. Plaintiffs' anchor reportedly ~$200B |
| 9 | 2026 capex reported as "$135–145B" | Wrong. Primary source: $130–145B, narrowed from $125–145B |
| 10 | Q2 13Fs are pre-print | Filed 08-14, positions as of 06-30 — a month before the print. They say nothing about the reaction |
| 11 | Tax noise in both directions | Q3'25 carried an $18.95B one-time charge; Q1'26 a −$5.02B benefit. They roughly cancel: normalised TTM EPS $28.82 vs GAAP $26.55 |
Pitfall that did NOT fire — worth recording. pattern-net-margin-above-operating-margin-is-a-tripwire and pitfall-unrealized-equity-marks-break-headline-pe do not apply to META. Net margin (29.8%) sits below operating margin (38.1%), and interest income of $859M against ~$90B of cash and securities is a plausible 3.8%. Unlike GOOGL, AMZN and NOW, Meta's earnings are not inflated by private-AI-stake marks. Only the cash-destruction half of pattern-ai-build-inflates-earnings-while-destroying-fcf fires here — a meaningful quality distinction in Meta's favour.
9. Key metrics summary
| Value | Read | |
|---|---|---|
| Price / market cap | $543.67 / $1.395T | −31.2% from high |
| P/E ttm GAAP / normalised | 20.5× / 18.9× | |
| True forward P/E | 20.4× FY26 · 17.8× FY27 | not 15.6× |
| P/FCF ttm | 34.0× | 🔴 the real multiple |
| P/FCF less SBC | 88.1× | 🔴 |
| Revenue TTM / growth | $228.2B / +28% YoY | 🟢 accelerating |
| Operating margin (rep / adj) | 30.9% / 36.8% | 🟡 |
| Gross margin | 81.4% | 🟢 unchanged |
| FCF TTM / margin | $41.0B / 18.0% | 🔴 from 33.1% in 2021 |
| FCF CAGR 2021→TTM | +1.1%/yr | 🔴 revenue +15.8%/yr |
| FY2026E FCF | −$8B to +$16B | 🔴 ~breakeven |
| Capex / revenue | 39.1% TTM → ~55% FY26E | 🔴 |
| ROIC | 18.4% spot / 25.8% TTM | 🔴 from 30.9% |
| Debt/Assets · net debt/EBITDA | 25.0% · 0.62× | 🟢 |
| Shares outstanding | +1.29% YoY | 🔴 4 quarters of dilution |
| Buybacks | $0 for 3 quarters | 🔴 |
| Dividend / payout | 0.37% / 8% | |
| Short interest | 1.43% of float, −8.8% | 🟢 shorts covering |
| Analyst mean target | $754.14 (low $580) | ⚠️ stale |
| Graham IV | $258 | |
| Weighted fair value | ~$516 (range $450–650) |
10. Sources
Primary (company/SEC): Q2 2026 press release · 8-K Ex-99.1 · Q2 2026 10-Q · Q2 2026 earnings call transcript · Form 4 filings via EDGAR (CIK 0001326801, 0001548760) · El Paso / BlackRock JV
Data: Yahoo Finance MCP (get_stock_info, quarterly income/balance/cashflow, get_holder_info, get_historical_stock_prices) · .mcp/fin.py META · stockanalysis.com (FCF cross-validation) · roic.ai (free plan capped at 2 years — unusable for the 5–8yr CAGR; see pitfall-roicai-free-plan-caps-history-at-two-years)
Legal: CNBC — NM $567M · PBS · CNN — MDL opening statements · TechPolicy.Press · FTC appeal · EC DSA IP/26/920 · EC DMA commitments
Analyst/market: Barchart — PT cuts · CNBC — "ceiling is lower" · Yahoo — the one line · Yahoo — 2027 capex expectations · MarketBeat
Competitive: EMARKETER — Meta passes Google · CNBC — TikTok JV, no exodus · TechCrunch — Threads vs X, Bluesky · CNBC — glasses 3x · CNBC — LeCun departure
Knowledge base: pattern-ai-build-inflates-earnings-while-destroying-fcf · pitfall-lease-financed-capex-hides-from-the-fcf-line · pitfall-unrealized-equity-marks-break-headline-pe · pattern-net-margin-above-operating-margin-is-a-tripwire · pitfall-vendor-forward-eps-is-stale-on-the-day-of-a-guidance-cut · pitfall-yahoo-insider-purchases-counts-rsu-grants · pitfall-multiple-trim-inherits-the-broken-vendor-field · principle-down-a-lot-is-not-cheap · pattern-ranking-function-as-policy-variable · pitfall-roicai-free-plan-caps-history-at-two-years
Prior: Output/Stocks/Technology/META/analyze-2026-07-29.md · Output/Stocks/Technology/META/analyze-2026-03-21.md