META 10-K Deep Dive — Meta Platforms, FY2025

Demo type: Analyze a financial report with the skills in this plugin. Source document: META_2026_10-K.pdf — Meta Platforms, Inc. Annual Report on Form 10-K for the fiscal year ended December 31, 2025, filed with the SEC on January 28, 2026. 138 pages. Skills used (in order): 10k-digest → financial-report-analyst → fundamental-analysis → chart-master → bear-case → stock-valuation / dcf-valuation → catalyst-calendar → position-ladder → result-validator Generated by Claude Code using the InvestSkill us-stock-analysis plugin. All figures are US dollars, in millions unless stated.

⚠️ Naming trap, read this first. The file is called META_2026_10-K.pdf because it was filed in 2026. It reports fiscal 2025. Every “this year” below means FY2025 (calendar 2025). Getting this wrong is the single most common error when analyzing a January-filed 10-K.

⚠️ No live market data was used. The only price in this analysis is the one Meta itself discloses in the filing — $660.09, the Class A closing price on December 31, 2025 (Item 5, p. 58). Valuation commentary is anchored to that date and is illustrative. Educational analysis only — not investment advice.

Table of Contents

1. How to reproduce this demo
2. What we could and could not read
3. 10k-digest — the filing in one page
4. Document orientation — where the value is buried
5. financial-report-analyst — the income statement
6. financial-report-analyst — the balance sheet
7. financial-report-analyst — the cash flow statement
8. The off-balance-sheet page nobody reads
9. fundamental-analysis — accounting quality
10. chart-master — the five charts that tell the story
11. What changed in the language of the filing
12. Prior-year promise tracker
13. bear-case — the pre-mortem
14. stock-valuation + dcf-valuation
15. catalyst-calendar — dated events from the filing
16. My comprehension — what this 10-K actually says
17. position-ladder — what I would actually do
18. result-validator — confidence audit
19. Thesis invalidation & Investment Signal

1. How to reproduce this demo

In Claude Code with the plugin installed:

# 1. Point the digest at the local PDF
/10k-digest  ~/finance_data/10-k/meta/META_2026_10-K.pdf

# 2. Full statement-by-statement read
/financial-report-analyst  ~/finance_data/10-k/meta/META_2026_10-K.pdf

# 3. Ratios, trends, earnings quality
/fundamental-analysis  META  --from-filing ~/finance_data/10-k/meta/META_2026_10-K.pdf

# 4. Adversarial pass — try to break the bull case
/bear-case  META

# 5. Value it three ways
/stock-valuation  META
/dcf-valuation    META

# 6. Turn the filing's dated disclosures into a watchlist
/catalyst-calendar  META

# 7. Size it
/position-ladder  META

# 8. Grade the whole analysis
/result-validator

The same prompts work outside Claude Code — copy prompts/10k-digest.md, prompts/financial-report-analyst.md, etc. into Cursor, Gemini CLI, Copilot Chat, or ChatGPT and paste the filing text after them.

Comparative tip used below: the strongest single technique in 10-K analysis is reading two filings side by side. Everything in §11 and §12 comes from diffing this filing against META_2025_10-K.pdf (FY2024). Do this every time.

2. What we could and could not read

Honest accounting of the source material, because it affects what you can trust below.

The chart problem, and why it doesn’t matter here. Pages 64–68 of the filing carry Meta’s headline visuals — revenue by user geography, DAP, ARPP, ad impressions growth, average price per ad. In this PDF those five exhibits render as broken image placeholders, so no data can be lifted from the pictures. Rendering the pages to PNG and reading them visually confirms it: the boxes are empty.

That turns out to be survivable, because Meta states every number from those charts in the surrounding prose or in Note 2. The ARPP series is even printed as a caption line under its chart:

“ARPP: $12.33 $11.20 $11.89 $12.29 $14.25 $12.36 $13.65 $14.46 $16.56” — Item 7, p. 66

Nine quarters, Q4-2023 through Q4-2025. Sum of the last four = $57.03, which reconciles exactly to the stated annual figure. So the charts are reconstructed from text in §10 rather than read from pixels.

Method note: when a filing’s exhibit won’t parse, don’t guess and don’t silently skip it. Say so, then find the same number in the text and reconcile it. An unreconciled figure is worse than a missing one.

3. 10k-digest — the filing in one page

Abstract

Meta grew revenue 22% to $200.97 billion and operating income 20% to $83.28 billion — its second consecutive 22% growth year at a scale where that should not be possible. Yet net income fell 3% to $60.46 billion and free cash flow fell 16% to $43.59 billion. Neither decline reflects a weakening business: net income was hit by a $15.93 billion non-cash tax charge from the One Big Beautiful Bill Act, and free cash flow was consumed by $72.22 billion of capital expenditure, up 84% year over year. The single most important disclosure in the document is one sentence on page 77: capex guidance of $115–135 billion for 2026. The most material risk is not competitive — it is that this filing commits Meta to spending more on infrastructure in one year than it generated in operating cash flow in 2025, against an AI payoff the filing itself declines to quantify. Tone: confident on the core business, deliberately unquantified on the returns from the spend.

Key metrics

Source: Consolidated Statements of Income (p. 89), Balance Sheets (p. 88), Cash Flows (pp. 92–93), Item 7 (pp. 60–79).

The three sentences that matter most

“We anticipate making capital expenditures of approximately $115 billion to $135 billion in 2026 to support our AI efforts and core business.” — Item 7, Liquidity, p. 77

What it really says: the midpoint, $125 billion, is 62% of 2025 revenue and 108% of 2025 operating cash flow. Put differently: the bottom of the range is more cash than Meta’s entire operating business generated in 2025. My read: this is not “an increase in capex.” It is a change of business model — from an asset-light advertising network into a capital-intensive compute utility that happens to own an advertising network. Every other number in this filing is downstream of that sentence.

“We also have $131.05 billion of contractual commitments as of December 31, 2025 … In addition to lease liabilities included in our consolidated balance sheets, we have leases that have not yet commenced, with total lease obligations of approximately $103.77 billion.” — Item 7, p. 79 / Note 7, p. 109

What it really says: $234.8 billion of future obligations that appear nowhere in the $148.8 billion of total liabilities on the balance sheet. The balance sheet understates Meta’s committed spending by more than the balance sheet’s entire liability side. My read: this is the most under-discussed line in the filing. It is not a scandal — the accounting is correct and the disclosure is clear — but any leverage ratio computed off the balance sheet alone is meaningless for Meta now.

“Absent any changes to our tax landscape, we expect our effective tax rate for the full year 2026 to be in the range of 13-16%.” — Item 7, p. 77

What it really says: the 29.6% rate that crushed 2025 EPS is a one-off. My read: this is the most reliable positive in the entire document, and the market’s 2025 EPS optics ($23.49, down year over year) badly understate the earnings base going into 2026. Normalizing 2025 for the charge gives $29.68 of adjusted EPS. See §5.

4. Document orientation — where the value is buried

A 138-page filing is not read front to back. Here is the value density map for this specific document.

PAGE   SECTION                              VALUE   WHY
─────────────────────────────────────────────────────────────────────────────────
 6– 11 Item 1  Business                     ●●●○○   Read the first 3 paragraphs only.
                                                    "personal superintelligence" (p.6) is new.
12– 47 Item 1A Risk Factors                 ●●●●○   36 pages. Don't read linearly —
                                                    DIFF it against last year (see §11).
48– 50 Item 1C Cybersecurity                ●○○○○   Clean. No material incidents in 2025.
   51  Item 2  Properties                   ●●○○○   "We own 30 data center locations."
51– 57 Item 3  Legal Proceedings            ●●●●○   One extraordinary sentence on p.51.
                                                    FTC antitrust WIN on p.54.
   58  Item 5  Market for Common Equity     ●●●●●   Price anchor $660.09. And the
                                                    quietest bombshell: no Q4 buyback.
60– 79 Item 7  MD&A                         ●●●●●   THE section. pp.77–79 are the payload.
80– 81 Item 7A Market Risk                  ●●●○○   Equity-price risk exploded (see §6).
83– 86 Auditor's Report + CAMs              ●●●●●   A NEW critical audit matter appeared.
88– 93 Financial Statements                 ●●●●●   Read the cash flow statement first.
94–127 Notes 1–15                           ●●●●●   Note 5 (p.106) is the best page in
                                                    the filing. Almost nobody reads it.
  128  Item 9A Controls                     ●●○○○   Effective. Unqualified ICFR opinion.

If you have ten minutes: page 58, pages 77–79, page 86, page 106. Those four locations contain roughly 80% of what changed about Meta this year.

5. financial-report-analyst — the income statement

The verbatim table

CONSOLIDATED STATEMENTS OF INCOME (in millions, except per share amounts) — p. 89

	2025	2024	2023
Revenue	$200,966	$164,501	$134,902
Cost of revenue	36,175	30,161	25,959
Research and development	57,372	43,873	38,483
Marketing and sales	11,991	11,347	12,301
General and administrative	12,152	9,740	11,408
Total costs and expenses	117,690	95,121	88,151
Income from operations	83,276	69,380	46,751
Interest and other income, net	2,656	1,283	677
Income before provision for income taxes	85,932	70,663	47,428
Provision for income taxes	25,474	8,303	8,330
Net income	$60,458	$62,360	$39,098
Diluted EPS	$23.49	$23.86	$14.87

Line-by-line, and what I think of each

Revenue +22.2%. Two years in a row at exactly 22%. At $201 billion of base, that is $36.5 billion of incremental revenue in a single year — more than Meta’s entire annual revenue as recently as 2016 ($27.6 billion). Compounding at 22% off a $165 billion base is the rarest thing in this filing. The composition matters:

“Advertising revenue in 2025 increased $35.54 billion, or 22%, compared to 2024 due to increases in ad impressions delivered and average price per ad. In 2025, ad impressions delivered increased by 12% … the average price per ad increased by 9%.” — Item 7, p. 73

My read: a 12/9 volume/price split is a high-quality growth mix. Price-only growth eventually meets advertiser resistance; volume-only growth means you are stuffing more ads into the same feed. Twelve and nine, with the price increase attributed to “an increase in advertising demand, which we believe is mostly due to ongoing improvements to our ad performance from our ad targeting and measurement tools” (p. 73) — that is the signature of a platform whose ranking models are genuinely getting better. This is the strongest evidence in the filing that Meta’s AI spending is already earning its keep in the core business, even though the filing never claims a dollar figure for it.

One caution the filing supplies itself: impressions grew “especially in Asia-Pacific”, and the price increase was “partially offset by a higher number of ad impressions delivered, especially in geographies and in products, such as Reels, that monetize at lower rates.” Mix is diluting price. The 9% is a net number fighting a headwind.

Cost of revenue +20%, held at 18% of revenue. Flat as a percentage — but read the reason:

“The increase was mainly due to higher operational expenses related to our data centers and technical infrastructure, which included decreases in the depreciation growth rate due to an extension in the useful lives of servers and network assets, effective January 1, 2025.” — Item 7, p. 74

My read: the flat 18% is partly manufactured. Without the useful-life change, cost of revenue would have been ~$2.9 billion higher and the ratio ~19.5%. Quantified in §9. Not deceptive — it is disclosed three separate times — but you must adjust for it.

R&D +31% to $57.4 billion, now 28.5% of revenue (up from 26.7%). This is the line where the AI story lives, and it grew 40% faster than revenue. Of it, $17.5 billion is share-based compensation (Note 12, p. 120) — 30% of the R&D line is stock, not cash.

“The higher employee compensation was primarily from an 8% growth in employee headcount from 2024 to 2025 in engineering and other technical functions and an increase in share-based compensation expense.” — Item 7, p. 74

My read: 8% engineering headcount growth cannot explain a 31% R&D increase. The gap is price per engineer. Corroborating evidence from Note 12: RSUs granted in 2025 carried a weighted-average grant-date fair value of $661.57 per share versus a $302.27 carrying value on the units outstanding at the start of the year. Meta is paying roughly double, per share, for new equity grants. That is the superintelligence talent war showing up in the accounts. It is real money and it recurs.

G&A +25% — the reason is unflattering but honest:

“The increase was mainly due to higher legal-related costs, which included lapping of a $1.55 billion decrease in accrued losses for certain legal proceedings that benefited the prior year.” — Item 7, p. 75

My read: 2024’s G&A was flattered by a legal accrual release; 2025’s is not. Legal accruals on the balance sheet rose from $5,523M to $6,867M (Note 9, p. 110), +24%. The legal cost base is structurally rising, not spiking.

The tax line is where the reported year breaks.

“On July 4, 2025, the One Big Beautiful Bill Act (OBBBA) was enacted … As a result, we recorded a $15.93 billion charge in the third quarter of 2025, of which $14.03 billion was a valuation allowance against our U.S. federal deferred tax assets.” — Item 7, p. 76 / Note 14, p. 121

The statutory reconciliation (Note 14, p. 122) makes the mechanics explicit:

	$M	%
U.S. federal statutory rate	18,046	21.0%
R&D tax credits	(3,912)	(4.6)
U.S. foreign tax credits	(1,405)	(1.6)
Valuation allowances (primarily OBBBA)	11,974	13.9
Changes in unrecognized tax benefits	3,127	3.6
Excess tax benefits from SBC	(4,307)	(5.0)
Other / state / foreign	1,951	2.3
Effective tax rate	25,474	29.6%

My read — and this is the most important analytical adjustment in the whole demo: the charge is non-cash and non-recurring, and its cash effect is the opposite sign. OBBBA gave Meta immediate expensing of domestic R&D and certain capex; cash taxes paid actually fell to $7.58 billion from $10.55 billion (p. 79) despite pre-tax income rising 22%. The accounting charge is the cost of recognizing that Meta will pay the 15% Corporate Alternative Minimum Tax instead of using deferred tax assets — so the DTAs got written down. Cash improved; GAAP earnings took the hit.

Anyone who looks at Meta’s headline 2025 EPS and concludes “earnings went backwards” has been fooled by a tax accounting entry. Underlying earnings grew about 24%. Conversely, anyone who anchors on $29.68 as a clean run-rate should remember the 2026 tax guidance of 13–16% is itself unusually low and depends on the stock price holding up (excess SBC tax benefits contributed −5.0 points).

6. financial-report-analyst — the balance sheet

CONSOLIDATED BALANCE SHEETS (in millions) — p. 88, condensed

	Dec 31, 2025	Dec 31, 2024	Δ
Cash and cash equivalents	$35,873	$43,889	−$8,016
Marketable securities	45,719	33,926	+11,793
Accounts receivable, net	19,769	16,994	+2,775
Total current assets	108,722	100,045	+8,677
Non-marketable equity investments	27,524	6,070	+21,454
Property and equipment, net	176,400	121,346	+55,054
Operating lease right-of-use assets	20,404	14,922	+5,482
Goodwill	24,534	20,654	+3,880
Total assets	$366,021	$276,054	+$89,967
Total current liabilities	41,836	33,596	+8,240
Operating lease liabilities, non-current	22,940	18,292	+4,648
Long-term debt	58,744	28,826	+29,918
Long-term income taxes	21,005	9,987	+11,018
Total liabilities	148,778	93,417	+55,361
Total stockholders’ equity	$217,243	$182,637	+$34,606

Three things happened to this balance sheet

1. It grew 33% in twelve months, and the growth is almost entirely physical. Property and equipment, net rose $55.1 billion. Inside Note 6 (p. 107), the composition is more dramatic than the net figure:

	2025	2024
Servers and network assets	$98,040	$68,397
Buildings	55,568	47,076
Construction in progress	50,521	26,802
Property and equipment, gross	233,726	164,663
Less: accumulated depreciation	(57,326)	(43,317)

My read: $50.5 billion of construction in progress — up 88% — is not yet being depreciated. Only 24.5% of gross PP&E has been depreciated at all. This is a loaded spring. As CIP converts to in-service assets through 2026–2027, depreciation steps up hard. 2025 depreciation was $18.0 billion on average net PP&E of ~$149 billion (a 12.1% effective rate). Apply the same rate to a 2026 average net PP&E of roughly $224 billion and you get ~$27 billion of depreciation — a $9 billion incremental operating-expense headwind in 2026 alone, before any 2027 step-up. And the useful-life extension that softened 2025 cannot be repeated.

2. Meta stopped being a fortress and became a normally-financed company.

Net cash halved in one year. Meta issued $30 billion of senior unsecured notes in November 2025 (Note 10, p. 111) — the largest raise in its history, in six series maturing 2030–2065, with no financial covenants. Total face debt is now $59 billion against $57.22 billion of fair value.

My read: at roughly 0.6× EBITDA ($59 B face against ~$102 B of operating income plus D&A) this is not a solvency question, and issuing very long-dated paper — maturities out to 2065 — at 4.20–5.75% to fund assets with a 5.5-year depreciable life is at worst an odd duration match and at best cheap permanent capital. What it does mean is that the “Meta can never be forced to do anything” argument is weaker than it was. If 2026 free cash flow is near zero (see §14), the buyback, the dividend, and the capex plan cannot all be funded from operations — one of them gives, or more debt gets issued. The filing already shows which one gave first: see the Q4 buyback in §7.

3. A $21.5 billion venture portfolio appeared out of nowhere. Non-marketable equity investments went from $6.07 billion to $27.52 billion. Note 5 (p. 106) names them:

“Our non-marketable equity investments accounted for under the measurement alternative mostly consist of our minority investments in Scale AI for $13.80 billion, which was closed during 2025, and our investment in Jio Platforms Limited of $5.82 billion.”

And Item 7A (p. 81) flags what that does to risk:

“The carrying value of the non-marketable equity investments accounted for under the measurement alternative was $20.08 billion and $6.02 billion as of December 31, 2025 and 2024.”

My read: Meta has quietly become one of the larger private-technology investors in the world. These holdings are carried at cost-plus-observable-adjustments, not marked to market — which means their carrying value will move only when there is a down round or an impairment trigger. Cumulative upward adjustments to date are just $429 million on $20.3 billion of cost. The asymmetry is unfavourable: you get almost no credit if Scale AI triples, and a visible earnings hit if it halves. Watch this line for the first material impairment; it would be the earliest hard evidence that the AI capex cycle is turning.

7. financial-report-analyst — the cash flow statement

This is the statement that tells the truth about 2025.

Selected lines, CONSOLIDATED STATEMENTS OF CASH FLOWS (in millions) — pp. 92–93

	2025	2024	2023
Net income	$60,458	$62,360	$39,098
Depreciation and amortization	18,616	15,498	11,178
Share-based compensation	20,427	16,690	14,027
Deferred income taxes	18,738	(4,738)	131
Net cash provided by operating activities	115,800	91,328	71,113
Purchases of property and equipment	(69,691)	(37,256)	(27,045)
Purchases of non-marketable equity investments	(18,330)	(11)	(1)
Payments for held-for-sale assets	(2,432)	—	—
Proceeds from Venture distribution	2,554	—	—
Acquisitions of businesses and intangibles	(4,231)	(270)	(629)
Net cash used in investing activities	(102,003)	(47,150)	(24,495)
Repurchases of Class A common stock	(26,248)	(30,125)	(19,774)
Taxes paid on net share settlement of RSUs	(18,400)	(13,770)	(7,012)
Payments for dividends	(5,324)	(5,072)	—
Proceeds from issuance of long-term debt, net	29,906	10,432	8,455
Net cash used in financing activities	(20,370)	(40,781)	(19,500)

Free cash flow, in Meta’s own words

“We define FCF as net cash provided by operating activities reduced by purchases of property and equipment and principal payments on finance leases.” — Item 7, p. 78

	2025	2024	2023
Net cash provided by operating activities	$115,800	$91,328	$71,113
Purchases of property and equipment	(69,691)	(37,256)	(27,045)
Principal payments on finance leases	(2,524)	(1,969)	(1,058)
Free cash flow	$43,585	$52,103	$43,010

Operating cash flow grew 27%. Free cash flow fell 16%. Free cash flow is now back at its 2023 level on 49% more revenue.

FREE CASH FLOW MARGIN (FCF ÷ revenue)
2023  31.9%  ████████████████████████████████
2024  31.7%  ████████████████████████████████
2025  21.7%  ██████████████████████            ← 1,000 bps of erosion in one year

My read on the quality of that $115.8 billion of operating cash flow: it is good cash, but it is flattered by two large non-cash add-backs that deserve naming. $20.4 billion of share-based compensation is a real economic cost paid in shares (offset by $18.4 billion of cash spent withholding taxes on vesting RSUs, which sits in financing — a classic place where SBC’s cash cost hides). And $18.7 billion of deferred income taxes is the mirror image of the OBBBA charge. Strip both to their economic substance and the 27% OCF growth is real but less spectacular than it looks. Cash conversion — OCF minus SBC, over net income — is still comfortably above 1.0×.

The quietest line in the entire filing

“There was no share repurchase activity during the three months ended December 31, 2025.” — Item 5, p. 58

Eight words. Meta has bought back stock every quarter for years — $26.3 billion in 2025, $30.1 billion in 2024, $19.8 billion in 2023 — and $25.03 billion remained authorized (p. 79). The authorization was not exhausted. Management simply chose not to use it in the quarter immediately after guiding to $115–135 billion of 2026 capex, and immediately after raising $30 billion of debt.

My read: this is the single most informative disclosure in the document, and it is one sentence with no commentary attached. It is management telling you, through action rather than language, that capital is now spoken for. When a company that has been a reliable buyer of its own stock stops buying while the authorization is still live, at a price it thought was attractive all year, the message is that the marginal dollar has found a better home — or that liquidity is being conserved. Either reading is a change of regime. If you track one thing in Meta’s next three 10-Qs, track whether buybacks resume.

8. The off-balance-sheet page nobody reads

Note 5 (p. 106) is the best page in this filing, and it is 400 words long.

"In October 2025, we entered into an arrangement to co-develop a data center campus in Louisiana (the Venture) … At Venture formation, we contributed $4.30 billion of held-for-sale assets, net of liabilities, and we received a one-time distribution of $2.55 billion. We hold a 20% membership interest … The parties have committed to fund their respective pro rata share of approximately $27 billion in total estimated development costs.

We also entered into lease agreements with the Venture for the use of properties on the data center campus, which will commence in 2029. The aggregate initial lease commitment is approximately $12.31 billion … In addition, we have provided residual value guarantees (RVG) with an aggregate threshold of approximately $28 billion that decreases over time … As of December 31, 2025, RVG payments are not probable and therefore, no liability has been recorded.

We do not have the power to direct the activities that most significantly impact the Venture’s economic performance. Therefore, we are not the primary beneficiary and do not consolidate the variable interest entity (VIE). Our maximum exposure to loss related to the Venture was $45.95 billion as of December 31, 2025, consisting of $1.83 billion carrying value of our equity investment, the lease commitments, our estimated future fundings, and the maximum RVG threshold."

Ernst & Young considered this important enough to make it a new critical audit matter — the third one in the report, and the only new one this year:

“Consolidation accounting for a variable interest entity … Auditing the Company’s determination of the primary beneficiary of the VIE was especially challenging due to the significant judgment required in determining the activities that most significantly affect the VIE’s economic performance.” — Report of Independent Registered Public Accounting Firm, p. 86

What this really is, in plain language

Meta wanted a giant data centre campus without putting a giant data centre on its balance sheet. So it put the assets into a joint venture, kept 20%, took a $2.55 billion cash distribution on the way in, and will lease the capacity back starting 2029. Because Meta does not control the JV’s key decisions, the JV is not consolidated. The 80% of the campus that Meta does not own — and the debt financing it — sit outside Meta’s financial statements.

My read. Three points, in order of importance:

1. The accounting is legitimate and the disclosure is genuinely good. Meta discloses the maximum exposure figure ($45.95 billion), names the residual value guarantee ($28 billion), and the auditor elevated the judgment to a critical audit matter. This is not an Enron structure. Structures like this become dangerous when they are hidden; this one is on page 106 with a number attached.

2. But $45.95 billion of maximum exposure to loss is 21% of Meta’s shareholders’ equity, and it is invisible in every screening tool on earth. No debt-to-equity screen, no EV/EBITDA calculation, no credit metric picks this up. Add the other unconsolidated VIEs ($5.58 billion), the leases not yet commenced ($103.77 billion), and the contractual commitments ($131.05 billion), and Meta carries roughly $286 billion of committed or contingent obligations against $148.8 billion of recognized liabilities.

3. The residual value guarantee is the part to actually worry about. A $28 billion RVG means that if Meta walks away from those leases and the campus is worth less than the guaranteed threshold, Meta pays the shortfall. That is a bet that AI data centre assets hold their value into the 2030s. If AI compute demand disappoints — the precise scenario in which Meta would want to walk away — the RVG is exactly the obligation that bites. It is a short position in the residual value of AI infrastructure, taken at the top of the AI infrastructure cycle.

Why this matters as a technique: the structure is a template, not an exception. Several hyperscalers are financing 2026–2028 AI capacity through JVs, SPVs, and neocloud leases. When you analyze any of them, go straight to the VIE note and read the “maximum exposure to loss” sentence. It is often the largest number in the filing that appears in no financial statement.

Full obligation bridge

Note: these buckets overlap somewhat — some Venture lease commitments are also inside the not-yet-commenced lease figure — so treat ≈$433 B as an upper bound, not an additive total. The point stands regardless of the precise number.

9. fundamental-analysis — accounting quality

The useful-life change, quantified

Meta discloses the impact three times, which is more transparency than most companies offer:

“In January 2025, we completed an assessment of the useful lives of property and equipment, which resulted in an increase in the estimated useful lives of most servers and network assets to 5.5 years, effective January 1, 2025. Based on the servers and network assets placed in service as of December 31, 2024, the financial impact of this change in estimate included a reduction in depreciation expense of $2.92 billion and an increase in net income of $2.59 billion, or $1.00 per diluted share, for the year ended December 31, 2025.” — Note 1, p. 94

My read: three separate judgments here.

• Is it defensible? Yes. Five-and-a-half years is in line with what large cloud operators have moved to in recent years, and the filing’s own useful-life table (p. 98) shows the range as “Five to 5.5 years” — an extension at the margin, not a leap.
• Is it disclosed well? Exceptionally. The dollar impact, the EPS impact, and the effective date are all given. Compare that to companies that bury a useful-life change in a policy paragraph with no quantification.
• Does it change my numbers? Yes, in two ways. First, $1.00 of 2025 EPS is non-repeatable — you cannot extend useful lives again next year. Second, and more importantly, extending server lives at exactly the moment you are buying $70 billion of servers per year maximizes the deferral of the cost of the AI buildout into future periods. The 2026–2028 depreciation wall is now steeper than it would otherwise have been. That is a timing choice, not an accounting abuse, but it means reported operating margins in 2026–2027 have a known, mechanical headwind that is already baked in.

Accounting quality score: 8.0 / 10. Clean audit, no restatements, minimal non-GAAP games, unusually good disclosure of the items that are judgmental. Marked down two points for the sheer scale of the off-balance-sheet structure and the well-timed useful-life extension — neither of which is wrong, both of which flatter the current year and burden later ones.

10. chart-master — the five charts that tell the story

Reconstructed from the filing’s text and note tables (the source PDF’s chart images do not carry extractable data — see §2).

1. Revenue and operating income — the scale is the story

REVENUE ($B)                                    each █ ≈ $10B
2023   $134.9  █████████████▌
2024   $164.5  ████████████████▍             +22.0%
2025   $201.0  ████████████████████          +22.2%

INCOME FROM OPERATIONS ($B)                     each █ ≈ $5B
2023    $46.8  █████████▍                    34.7% margin
2024    $69.4  █████████████▉                42.2% margin
2025    $83.3  ████████████████▋             41.4% margin

2. The scissors — operating cash flow vs. free cash flow

$B
120 ┤                                    ╭─── 115.8   Operating cash flow
    │                        ╭─── 91.3 ──╯
 90 ┤            ╭─ 71.1 ────╯
    │
 60 ┤                        ╭─── 52.1 ─╮
    │            ╭─ 43.0 ────╯           ╰── 43.6     Free cash flow
 30 ┤
    │
  0 ┼──────────────────────────────────────────
        2023          2024          2025

    THE GAP = capex.   2023: $28.1B    2024: $39.2B    2025: $72.2B

This is the whole 2025 story in one picture. The top line is a business getting better. The bottom line is what the shareholder actually receives. They separated this year for the first time.

3. Where the money goes — 2025 cost structure

% of revenue
Cost of revenue      18.0%  ██████████████████
R&D                  28.5%  ████████████████████████████▌   ← +180 bps YoY
Marketing & sales     6.0%  ██████
G&A                   6.0%  ██████
─────────────────────────────────────────────────
Operating margin     41.4%  █████████████████████████████████████████▍

4. ARPP — the monetization engine is accelerating, not fading

AVERAGE REVENUE PER PERSON, quarterly ($)
Q4'23  $12.33  ████████████▎
Q1'24  $11.20  ███████████▏
Q2'24  $11.89  ███████████▉
Q3'24  $12.29  ████████████▎
Q4'24  $14.25  ██████████████▎
Q1'25  $12.36  ████████████▍         +10.4% vs Q1'24
Q2'25  $13.65  █████████████▋        +14.8% vs Q2'24
Q3'25  $14.46  ██████████████▌       +17.7% vs Q3'24
Q4'25  $16.56  ████████████████▌     +16.2% vs Q4'24   ← record
                                          Annual 2025: $57.03 (+14.9%)

My read: the YoY comparison accelerated through the year (+10.4% → +14.8% → +17.7% → +16.2%). With DAP growing 6.9%, roughly two-thirds of Meta’s revenue growth is now coming from monetizing existing users better rather than adding new ones. That is the more durable of the two sources at 3.58 billion users, because user growth mathematically must decelerate and monetization does not have to.

5. Segment reality check

FAMILY OF APPS                          REALITY LABS
Revenue    $198,759 M                   Revenue      $2,207 M
Op income  $102,469 M                   Op loss    $(19,193) M
Margin         51.6%                    Margin       (870)%

FoA is 98.9% of revenue and 123% of operating income.
RL destroys $8.70 of operating income for every $1.00 of revenue.
Three-year cumulative RL losses (2023-25): $53.0 billion.

“In 2025, our RL segment reduced our overall operating profit by approximately $19.19 billion, and we expect our 2026 RL operating losses to remain similar to 2025.” — Item 7, p. 63

My read: compare this year’s wording to last year’s — the FY2024 10-K said losses would “continue to increase in 2025.” This year says “remain similar.” That is the first time in the Reality Labs era that Meta has guided the loss to stop growing. It is a small sentence with real money attached: at 8% annual growth, RL losses would have been ~$20.7 billion in 2026; “similar to 2025” implies ~$19.2 billion. About $1.5 billion of implied restraint. Not a pivot, but the first sign of a ceiling.

11. What changed in the language of the filing

Diffing FY2025’s 10-K against FY2024’s surfaces a strategic pivot that no financial statement shows.

The clearest single sentence, from the very top of Item 1:

“We are innovating in artificial intelligence (AI) technologies to build transformative experiences and capabilities across our Family of Apps and new platforms, and to advance our vision to deliver personal superintelligence for everyone.” — Item 1, p. 6

And the definition, offered without hedging:

“We are also working to develop the next generation of AI models and advance our vision to build superintelligence, which we define as AI that surpasses human intelligence. Although it is inherently difficult to predict when superintelligence may be achieved, we are investing now because we believe this has the potential to begin a new era of individual empowerment.” — Item 1, p. 6

My read: in a legal document where every adjective is negotiated by counsel, putting “AI that surpasses human intelligence” in Item 1 of a 10-K is a deliberate act. Two years ago the same section was about the metaverse. The corresponding new risk factor is worth reading in full because it is unusually candid:

“There are significant risks involved in developing and deploying AI and there can be no assurance that the usage of AI will enhance our products or services or be beneficial to our business, including our efficiency or profitability.” — Item 1A, p. 19

That sentence, sitting alongside $115–135 billion of committed 2026 capex, is the entire investment debate in two lines of the same document.

Risk factor structure

The FY2025 filing runs 36 pages of risk factors (pp. 12–47), a modest expansion on the prior year. The substantive additions are AI-specific: model liability, chatbot scrutiny (“inquiries and investigations by the FTC, members of Congress and state attorneys general”, p. 19), open-source model misuse, AI talent competition, and dependence on third-party compute:

“our ability to continue to develop and effectively deploy AI technologies is dependent on access to specific third-party equipment, technology, and other technical and physical infrastructure, such as processing hardware, network capacity, models, computing power, and related energy requirements, as to which we cannot control the availability or pricing.” — Item 1A, p. 19

My read: Meta has added a supplier risk factor. For a company whose historical risks were all about users, advertisers, and regulators, acknowledging dependence on GPU availability and electricity pricing is a structural change in the risk profile. It is also, quietly, a disclosure that Meta’s cost of AI is not under Meta’s control.

12. Prior-year promise tracker

The accountability audit: what the FY2024 10-K said would happen in 2025, versus what this filing reports.

Management credibility read: 4 of 6 clean, 1 partial, 1 clear miss — and the miss is the one that matters. Meta overshot its own capex guidance by more than $7 billion in a year when capex was already the single most-watched number in the story. That is directly relevant to how you should treat the 2026 guidance: a company that beat its capex ceiling by 11% last year has just given you a $20 billion-wide range for next year. Assume the top half. Modelling $125 billion is the polite assumption; $135 billion is the prudent one.

On the tax miss, I score it Partial rather than Missed: the OBBBA was enacted in July 2025 and no reasonable January-2025 forecast would have anticipated it. Absent the law change, Meta would have landed inside its 12–15% guide.

13. bear-case — the pre-mortem

The discipline here: assume it is January 2029 and the stock is down 50% from $660. Write the history using only evidence available in this filing.

1. The depreciation wall arrives before the revenue does. $50.5 billion of construction in progress converts to in-service assets across 2026–2027. Depreciation goes from $18 billion to $27 billion to $38 billion. The useful-life extension that cushioned 2025 cannot be repeated. Advertising revenue grows a perfectly respectable 12–15% but operating margin falls from 41% to the low 30s because the cost base grew faster. The market re-rates a mid-teens grower with compressing margins at 15× rather than 28×. Evidence: Note 6 (p. 107), Note 1 (p. 94).

2. Free cash flow goes negative and the buyback does not come back. At the top of the capex guidance ($135 billion) and a modest deceleration in operating cash flow, 2026 FCF is negative (see §14). Meta funds the gap with more debt. A company with no net cash, negative free cash flow, and a rising interest bill is not valued the way a fortress balance sheet is. Evidence: capex guidance p. 77; $30 B raise Note 10 p. 111; zero Q4 buyback p. 58.

3. The AI spend does not produce a new revenue line. Read the filing carefully and notice what is absent: there is no disclosed revenue from Meta AI, from the Llama models, or from any AI product. Every dollar of AI benefit shows up as better ad ranking inside an existing business. That is genuine value, but it is incremental margin on ads, not a second engine. If ad ranking improvements saturate — and there is a ceiling on how relevant an ad can be — the $115–135 billion buys diminishing returns. Evidence: Note 2 (p. 102) discloses exactly two revenue sources, advertising and “other”; the AI risk factor (p. 19) explicitly declines to assure profitability.

4. The residual value guarantee is called. The bear scenarios for AI infrastructure and the scenario in which Meta wants out of the Louisiana leases are the same scenario. A $28 billion RVG, currently carried at zero liability because payment is “not probable,” becomes probable precisely when Meta can least afford it. Evidence: Note 5 (p. 106–107).

5. Regulatory tail risk is genuinely uncapped, and the filing says so. This is the most extraordinary sentence in the document:

“The maximum aggregate monetary damages or penalties sought across our various legal 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.” — Item 3, p. 51

Against a $6.87 billion legal accrual. Live matters include EU DSA preliminary findings of infringement (October 2025), a €798 million EC Marketplace fine under appeal, a €542 million Spanish unfair-competition judgment under appeal, the New Mexico AG trial scheduled for September 8, 2026, and the FTC’s appeal of its antitrust loss filed January 20, 2026. Evidence: Item 3 pp. 51–57, Note 11 pp. 112–118.

6. The talent cost is permanent, not a phase. $54.81 billion of unrecognized SBC to be expensed over three years — a mechanical ~$18 billion/year floor before new grants. New RSUs are being issued at $661.57 average grant-date value against $302.27 for the existing pool. If the stock falls, Meta must issue more shares to retain the same people, and dilution accelerates exactly when the stock is weak. Evidence: Note 12, p. 120.

Bear-case fair value: see §14 — roughly $274, or −58%.

What the bear case must ignore to work: 22% revenue growth, accelerating ARPP, 51.6% Family-of-Apps operating margin, a 13–16% tax rate arriving in 2026, an FTC antitrust win, and the fact that Meta has been called overextended on capex in 2018, 2022, and 2023 and compounded through each one.

14. stock-valuation + dcf-valuation

Anchor

Multiples at the filing-date price

My read: the P/E looks reasonable — 22× on normalized earnings for a 22%-grower is not demanding. The P/FCF of 38× is the number that should stop you, and the gap between the two is the entire debate. You are being asked to pay an earnings multiple while receiving a free-cash-flow yield of 2.6%, on the promise that capex normalizes later.

Three-scenario DCF

Assumptions stated in full, because a DCF is only as honest as its inputs. Discount rate 9%, five-year explicit forecast, exit multiple on terminal-year FCF, net cash added.

The base case reproduces the year-end price to within 0.3%. I did not tune it to do that — the inputs are the ones the filing supports. The conclusion is uncomfortable and worth stating plainly:

At $660, Meta was priced for the base case to work. No discount for the possibility that capex stays elevated. No premium for the possibility that AI creates a new revenue line. The stock offered no margin of safety and no free option — the entire distribution of outcomes was in the price at fair value.

That is a very different statement from “Meta is expensive” or “Meta is cheap.” It is: you are not being paid to take this risk, but you are also not overpaying for it. The asymmetry, at that price, was roughly symmetric — and for a thesis whose central variable ($125 billion of annual capex against an unquantified return) is genuinely unknowable, symmetric is not good enough.

The FY2026 free cash flow question, isolated

Everything hinges on one arithmetic problem:

2026 operating cash flow  =  ???
2026 capex (guided)       =  $115B – $135B
                             ─────────────────
2026 free cash flow       =  the answer

2025 operating cash flow was $115.8B.
The BOTTOM of the capex guidance equals ALL of 2025's operating cash flow.

Meta needs operating cash flow to grow roughly 8% to break even on free cash flow at the midpoint of its own capex guidance, and roughly 17% to break even at the top of it. It grew 27% in 2025 — but 2025’s OCF was helped by an $18.7 billion deferred-tax swing that does not repeat in the same form. My honest expectation: 2026 free cash flow lands somewhere between −$10 billion and +$20 billion. For the first time since 2011, Meta may not generate meaningful free cash flow. That fact is knowable today, from this filing, and it is not what “Meta trades at 22× earnings” leads you to expect.

FY2026 EPS, the other side of the coin

The same filing supports a much happier earnings picture, because of tax:

My read: 2026 will very likely be a year in which EPS rises 15–45% while free cash flow falls to near zero. Those two facts will be reported in the same press release, and the market will have to decide which one it believes. That divergence — not the growth rate, not the competition, not the regulators — is the actual 2026 investment question for Meta.

15. catalyst-calendar — dated events from the filing

Every one of these dates is disclosed in the 10-K itself. No external sourcing.

The one to circle: Q1 2026 earnings. Not for the revenue number — for two disclosures. (1) Did the buyback resume? If yes, management is confident cash covers the plan. If no, capital is fully committed and equity holders are last in line. (2) Did capex guidance move? Meta entered 2025 guiding $60–65 billion and finished at $72.2 billion. An upward revision to the $115–135 billion range would tell you the top end was never a ceiling.

16. My comprehension — what this 10-K actually says

Stripped of everything else, here is my honest reading of the document.

Meta ran two different companies in 2025, and this filing is the seam between them.

The first company is Family of Apps: $198.8 billion of revenue, 51.6% operating margins, 22% growth, 3.58 billion daily users, and monetization per user that is accelerating rather than maturing. It is one of the finest businesses ever assembled. It gets better every year because its ranking models get better every year, and 2025 is the clearest evidence yet that AI investment feeds directly back into ad performance — 12% more impressions at 9% higher prices is not something a mature ad network does. This company, on its own, is worth a great deal and is not seriously threatened by anything disclosed in the filing. Even the regulatory picture improved this year: Meta won the FTC’s break-up case on November 18, 2025.

The second company is an infrastructure build the size of a national utility program. $72 billion of capex in 2025, $115–135 billion guided for 2026, $50.5 billion of construction in progress, $103.8 billion of leases signed but not started, $131.1 billion of contractual commitments, $30 billion of fresh debt, a $46 billion off-balance-sheet joint venture, and a $28 billion residual value guarantee on data centre assets in 2029. This company has no disclosed revenue. Not “small revenue” — no separately disclosed revenue at all. Its entire justification, in the filing’s own words, is a vision to “deliver personal superintelligence for everyone.”

The first company is paying for the second. That is the whole story, and the filing states it almost explicitly:

“Our ability to support these investments is dependent on generating sufficient profits from other areas of our business.” — Item 1A, p. 26

The four things I would want a colleague to remember

1. The reported 2025 numbers are misleading in both directions, and they cancel out badly. Net income fell 3% — that is a tax-accounting artifact; underlying EPS grew 24%. Operating cash flow grew 27% — that is flattered by the mirror image of the same tax item and by a useful-life change worth $1.00 of EPS. The truest single number in the filing is free cash flow of $43.6 billion, down 16%, at a 21.7% margin versus 31.7% a year ago. That number has no adjustments and no story attached. It is what the business actually produced for owners.

2. The balance sheet no longer describes the company. $148.8 billion of recognized liabilities against roughly $286 billion of additional committed and contingent obligations. Every conventional leverage metric on Meta is now wrong by a wide margin. If you run screens, Meta will look like a fortress; it is not one any more. It is a well-capitalized company with an enormous forward spending commitment.

3. Management is telling you where its conviction is — through actions, not adjectives. They raised $30 billion of debt. They stopped buying back stock in Q4 while $25 billion of authorization sat unused. They extended asset lives at the exact moment depreciation was about to explode. They put “superintelligence” in Item 1 and cut “metaverse” mentions in half. They guided Reality Labs losses flat for the first time ever. Read together, these are the moves of a management team redirecting every available dollar toward AI infrastructure and being unusually honest that they cannot promise it will work.

4. The bull and bear cases share every fact and differ only on one unknowable. Both agree revenue is compounding at 20%+. Both agree capex is $115–135 billion. Both agree the tax rate falls to 13–16%. The only disagreement is whether $125 billion a year of GPUs and data centres produces a return above cost of capital. Nothing in this 138-page document answers that, and the filing’s own AI risk factor concedes as much: “there can be no assurance that the usage of AI will enhance our products or services or be beneficial to our business.”

What I actually think

I think the Family of Apps business is stronger at the end of 2025 than at the start, and I think the market’s focus on the −3% net income headline was wrong. I also think the market’s implied valuation at $660 was — remarkably — right: my base-case DCF lands at $658 using assumptions drawn from the filing.

A stock priced exactly at its base case, whose base case depends on an unquantifiable return from the largest capital program in the company’s history, is not a compelling risk-reward. It is a fair price for a fine business making an enormous bet. That combination deserves a HOLD — own it if you own it, and demand a discount before adding, because the discount is where your compensation for the capex risk has to come from.

The thing that would change my mind fastest is not a revenue number. It is the first quarter in which Meta discloses a revenue line that did not exist before — an AI product, a Meta AI subscription, a compute or model business. That would convert the second company from a cost centre with a story into a business with a P&L, and the entire valuation framework would change. Until then, the capex is a bet financed by the ad business, and it should be valued as one.

17. position-ladder — what I would actually do

Illustrative construction only, anchored to the filing-date price of $660.09. Not advice, and not calibrated to any real portfolio.

Position sizing note: because the central risk here is a capital allocation risk rather than a competitive one, it does not diversify away against other mega-cap tech — Microsoft, Alphabet, and Amazon are running the same playbook with the same suppliers into the same demand assumption. Treat Meta’s AI-capex exposure as correlated with, not diversifying against, the rest of a large-cap tech book. Sizing should reflect the aggregate exposure, not Meta’s alone.

18. result-validator — confidence audit

Overall confidence: 82 / 100 — HIGH on facts, MEDIUM on conclusions.

Where this analysis is most likely to be wrong, in order:

1. The 2026 depreciation estimate (~$27 B). Derived by applying 2025’s 12.1% effective rate to an estimated average net PP&E. Meta discloses neither the in-service timing of construction in progress nor an asset-mix breakdown, so the error bar is wide — plausibly ±$5 B, which is ±2 percentage points of operating margin.
2. The DCF’s terminal exit multiples (14× / 18× / 22×). These are judgment, not derived from the filing. They drive 60–70% of the equity value in every scenario. Change the base-case exit multiple from 18× to 15× and fair value drops to roughly $560.
3. Whether 2026 capex lands at the midpoint. History says assume the top (§12). Every $10 B of capex above plan is roughly $4/share of DCF value at a 9% discount rate.
4. The valuation anchor is 7+ months stale as of this demo’s publication. The $660.09 close is the filing’s own figure; the market has traded since. Verify a live quote before drawing any conclusion about price.

What would raise confidence to 90+: Q1 and Q2 2026 10-Qs (to see actual capex pacing, buyback resumption, and the depreciation step-up), plus the FY2026 proxy statement (for executive incentive metrics — if compensation was re-tied to capital efficiency rather than growth, that would be a strong signal about internal discipline).

19. Thesis invalidation & Investment Signal

Re-run this analysis when any of the following occurs:

• [ ] Meta files its Q1 2026 10-Q — check buyback resumption and capex pacing
• [ ] 2026 capex guidance is revised (in either direction) outside the $115–135 B range
• [ ] Any quarter reports a new disclosed revenue line from AI products (thesis-changing, upside)
• [ ] Free cash flow is negative for two consecutive quarters (thesis-changing, downside)
• [ ] A material impairment is recorded against the $27.5 B non-marketable equity portfolio
• [ ] The Venture residual value guarantee moves from “not probable” to an accrued liability
• [ ] FTC antitrust appeal is decided
• [ ] New Mexico AG trial (Sept 8, 2026) concludes
• [ ] The FY2026 10-K is filed (≈ late January 2027)

Bull case in one line: the best advertising business ever built is compounding at 22% with accelerating monetization and a tax rate about to halve, and it is self-funding an option on the most important technology of the decade.

Bear case in one line: an asset-light 32% free-cash-flow-margin business converted itself, in eighteen months, into a 22%-and-falling free-cash-flow-margin business with $286 billion of off-balance-sheet obligations, in exchange for a return the filing itself declines to promise.

Net read: both are true. The price at the filing date already contained both. That is what a HOLD looks like.

╔══════════════════════════════════════════════╗
║              INVESTMENT SIGNAL               ║
╠══════════════════════════════════════════════╣
║ Signal:      NEUTRAL                         ║
║ Confidence:  MEDIUM                          ║
║ Horizon:     LONG-TERM                       ║
║ Score:       5.8 / 10                        ║
╠══════════════════════════════════════════════╣
║ Action:      HOLD                            ║
║ Conviction:  MODERATE                        ║
╚══════════════════════════════════════════════╝

Score Guide: 8.0–10.0 Strongly Bullish | 6.0–7.9 Moderately Bullish | 4.0–5.9 Neutral | 2.0–3.9 Moderately Bearish | 0.0–1.9 Strongly Bearish Confidence: HIGH (strong data, clear signals) | MEDIUM (mixed signals) | LOW (limited data, conflicting signals) Horizon: SHORT-TERM (1 week–3 months) | MEDIUM-TERM (3 months–1 year) | LONG-TERM (1+ years)

References

1. Meta Platforms, Inc., Annual Report on Form 10-K, fiscal year ended December 31, 2025. Filed with the SEC on January 28, 2026. 138 pages. Local copy: ~/finance_data/10-k/meta/META_2026_10-K.pdf. SEC EDGAR: https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001326801&type=10-K
2. Meta Platforms, Inc., Annual Report on Form 10-K, fiscal year ended December 31, 2024 — used for the year-over-year language diff (§11) and the promise tracker (§12). Local copy: ~/finance_data/10-k/meta/META_2025_10-K.pdf.
3. Item 1 — Business (pp. 6–11) · Item 1A — Risk Factors (pp. 12–47) · Item 1C — Cybersecurity (pp. 48–50)
4. Item 2 — Properties (p. 51) · Item 3 — Legal Proceedings (pp. 51–57) · Item 5 — Market for Common Equity (p. 58)
5. Item 7 — MD&A (pp. 60–79) · Item 7A — Market Risk (pp. 80–81)
6. Item 8 — Financial Statements and Supplementary Data (pp. 82–127), including Report of Independent Registered Public Accounting Firm (Ernst & Young LLP, pp. 83–87) and Notes 1–15
7. Item 9A — Controls and Procedures (p. 128)

All quotations are verbatim from the filing. Page numbers refer to the printed page numbers within the 10-K, not PDF page indices.