How to Read Meta's Numbers: The META Learning Companion

The META research family, published November 6, 2026 (page counts are those of the approved PDF editions):

• Deep Dive: Can Meta's capital program pay off? (54 pages). Public preview; the full report is open to verified Free Early Access members.
• Research Note: The business is proven; the capital is not (16 pages).
• Technical Appendix: Schedules and model conventions (29 pages).
• Learning Companion: How to read Meta's numbers (15 pages).

---

Worked examples from the 09-30-2026 quarter and the October and November 2026 financings, reconciled to the Deep Dive and its rating.

Where to start

This companion explains the ideas the Meta research actually uses, in the order you meet them. It teaches the reasoning rather than repeating the conclusion, and every example is taken from the same figures the Deep Dive uses, so you can carry a number from one document to the other without it changing meaning.

1. Start with what the company sells

Meta sells attention to advertisers. Of $64.60 billion of revenue in the September 2026 quarter, $63.06 billion was advertising on the Family of Apps, $1.10 billion was other Family of Apps revenue such as paid messaging and subscriptions, and $443 million was Reality Labs. Advertising is 97.6% of the total.

That concentration is why an analyst spends most of the work on one revenue line. It is also why the newer products deserve care rather than enthusiasm: several of them earn money, including artificial-intelligence glasses and paid messaging, but none has a disclosed revenue line of its own, so their scale and unit economics are undisclosed rather than absent. Those are different statements, and keeping them apart is most of the discipline in reading a company that is launching products faster than it reports them.

2. Break the revenue into links you can see

Advertising revenue is roughly four things multiplied together: how many people use the apps, how much time each person spends, what share of that time is filled with advertising, and what an advertiser pays for each advertisement.

The September quarter is unusually readable because the company disclosed both halves of the arithmetic. People grew 3%, impressions grew 10%, and the average price per ad grew 14%. Multiply 1.10 by 1.14 and you get 25.4%, against reported advertising growth of 25.9%. Because both inputs are rounded to whole percentage points, the implied product could sit anywhere between about 24.3% and 26.5%, so the identity holds within the precision the company publishes.

What you learn from the gap between 3% and 10% is where the growth came from. The audience barely moved, so the extra advertisements came from time spent per person, from ad load, and from the mix of surfaces. Ad load is the share of a session filled with advertising, and it is a company decision with a ceiling: past a point, more advertising reduces the engagement that creates the inventory. So a reader should treat part of this year's gain as a lever that has been pulled rather than a new capability. Meta does not disclose its ad load, which means you cannot tell how much of the lever is left.

3. Read a segment as an allocation decision, not a fact

Meta reports two segments, Family of Apps and Reality Labs, and evaluates them on revenue and operating income only. No assets or liabilities are reported by segment.

The important detail is in the accounting policy rather than the numbers. Data-center and technical infrastructure costs are allocated to segments based on estimated usage, and most of that lands in Family of Apps. Shared costs such as facilities and some research are allocated largely on headcount. The practical consequence is that infrastructure costs are allocated predominantly to Family of Apps, while Reality Labs also incurs infrastructure and third-party AI token costs. Reality Labs is not the AI program, and its $4.13 billion quarterly loss is not the program's cost.

This is also a good place to learn a general habit. Segment profit is not consolidated profit, and the two segments sum to the consolidated figure only because the company allocates nearly everything. If you preferred to charge model-training research to a central line, you would change both segment margins and change the consolidated figure not at all. When an allocation can move a number, say whose allocation it is. In the Deep Dive, the 42.1% Family of Apps margin excluding the third-quarter $9.89 billion charge is explicitly the analyst's allocation, because the company stated the charge at the consolidated level only. The $22,878 million consolidated figure excluding that charge does not depend on that choice, because the company gave it directly. Scope matters as well: the nine-month agreement-related charges of $11.89 billion are larger than the third quarter's $9.89 billion, by $2.00 billion recognized before the third quarter, and the second quarter carried $2.40 billion of legal-proceeding charges. The filings do not say how the $2.40 billion relates to that $2.00 billion, so the two are not mapped to each other, and any adjusted figure has to say which charge it removes.

4. A cost that has not started yet

When a company builds a data center, the money leaves before the cost appears in profit. Spending becomes an asset on the balance sheet called construction in progress, and it only begins to depreciate when it is placed in service.

At 09-30-2026 Meta's gross property and equipment was $326.49 billion, of which $89.50 billion was construction in progress and $3.76 billion was land. Land is not depreciated as a policy matter and can sit under buildings in full use; construction in progress is not depreciated because it is not yet in service. So 28.6% of the gross asset base is not being charged against profit, and 27.4% of it is not yet placed in service for accounting purposes. That is why today's margins do not yet carry the eventual fixed cost of the program.

Two cautions keep this from becoming a one-sided argument. First, composition decides the size of the future charge. If every dollar of that construction were a building depreciated over 27.5 years it would add about $3.3 billion a year; if every dollar were server and network equipment over 5.25 years it would add about $17.0 billion. Those are two hypothetical compositions, not a range, because the company does not disclose the mix or the in-service dates. Second, management names third-party cloud spending and AI token costs among the expense-growth drivers. Owned capacity could replace some rented capacity; the amount, workloads and timing are not established. Neither the rented amount nor the substitution rate is disclosed, so the net margin effect of commissioning cannot be estimated honestly from these filings. Holding both timing effects in mind at once is the skill.

5. Capital turnover, in plain language

The hardest idea in the Meta work is also the simplest to state: how much revenue does a dollar of productive capital support?

The research measures capacity as productive gross asset cost, excluding land and construction in progress and including finance-leased assets. At 09-30-2026 that was $233.23 billion. Annualizing the September quarter's revenue gives $258.40 billion, so revenue was 1.11 times productive capital, or capital was 0.90 times revenue.

The model then asks what that ratio settles at as the business matures while replacement and growth investment continue, testing 0.80, 0.65 and 0.50. Those are planning ratios used to size required capacity, and at 0.65 the planned capital is 1.54 times revenue. Three things about this are worth internalizing.

It is a financial proxy, not an engineering measurement. It says nothing about how many chips are installed or how busy they are, and no Meta disclosure provides a unit-capacity or utilization figure to calibrate it against.

The targets are not the outcomes. Told to plan for 0.65, the model realizes about 0.57 on the reference path in 2044, because capacity is bought roughly two years ahead of the revenue it serves and investment is not allowed to fall faster than 20% a year. When you compare an assumption with a current ratio, compare like with like: the realized column, not the target label.

Moving only the target from 0.50 to 0.80 changes the reference value per pre-distribution package from $301.72 to $563.90, an 87% difference with every other input held fixed. When one unmeasured assumption can nearly double an answer, the honest presentation is a range with the assumption named, not a single number.

6. Free cash flow depends on the definition you choose

Meta's own definition of free cash flow is operating cash flow less purchases of property and equipment less principal payments on finance leases. On that basis, free cash flow fell from more than 30% of revenue two years ago to 6% in the first nine months of 2026, and was negative $2.12 billion in the September quarter.

Here is a lesson in reading cash flow statements carefully. Equity compensation is added back in the operating section as a non-cash expense, which it is. But Meta also paid $12.77 billion in nine months to settle employees' tax obligations when awards vested, and that cash appears in the financing section. Deduct it and the nine-month figure becomes negative $1.72 billion. That is an analyst adjustment, not a correction: the company's definition is internally consistent, and the adjustment simply asks the cash to be counted where it left.

The valuation itself discounts neither of these figures. It values the operating business on unlevered free cash flow to the firm, before financing flows and discounted at a firm rate, and it charges share-based compensation as a real operating cost rather than adding it back. Legal payments are already deducted inside that operating value, as the section "Count every claim once, and date it" in this companion explains, and are not deducted again; the remaining claims against the business are then deducted once, in the equity bridge.

The forward burden matters too. $75.59 billion of restricted stock unit expense is granted but not yet recognized, expected over about three years subject to continued service, which averages roughly $25 billion a year before any new grant. That is an expense schedule conditional on service, not a fixed debt, and the annual profile will not be flat.

7. Count every claim once, and date it

A valuation is a value of the business minus the claims against it. Most errors in a report like this come from counting a claim twice or forgetting one.

The Meta work handles the legal exposure in one place. Dated legal payments run through a cash ledger, their present value of $17,144.17 million at the September reference is deducted inside the operating value, and the $19,881 million legal accrual on the balance sheet is then not deducted again, because that would charge the same exposure twice. One cross-check in the Deep Dive adds legal present value back to the denominator, and it does that only so a cash yield whose numerator excludes legal payments is compared with a denominator carrying the same exclusion. Matching definitions on both sides of a ratio is what makes the ratio mean anything.

The tax claim sits in the bridge instead, and it teaches the difference between present value and the cash on the day. A $5,000 million claim at the September reference, compounded forward at the 8.75% discount rate to an assumed 07-01-2028 payment date, posts $5,791.65 million of cash. The two numbers are the same obligation seen from two dates. If you see a bigger number in a cash calendar than in a claim table, check whether you are looking at compounding before you call it an inconsistency.

A judgment is not automatically a claim of its face amount. In October 2026 a New Mexico court imposed a $25 billion civil penalty on Meta with 15% post-judgment interest; Meta has accrued an undisclosed amount and intends to appeal. Three numbers have to be kept apart: the judgment, the accrual already on the balance sheet, and the cash already in the model's legal ledger. Subtracting $25 billion from the value would ignore the accrual; assuming the existing allowance covers it would ignore the size of the order. The research therefore shows conditions, not forecasts: at the reference 0.65 condition, $462.05 per package if the matter resolves within amounts already represented by the modeled legal allowance, and $451.27 to $454.68 if the judgment is paid two years after the order with interest. Posted collateral principal is not a second judgment loss either: if the judgment is reversed the principal comes back, and if it is upheld the principal is applied to the payment, so the research models it in the liquidity test and not as a loss of value. That is a convention about principal only. Refundable does not mean free: posting cash, a surety bond or a letter of credit can carry interest, fees or financing costs, can itself require collateral or use up credit capacity, and the research does not model any of that.

8. Accrued, contingent and asserted are three different things

The youth settlement is a short course in liability language. Of up to approximately $18.59 billion, roughly $12.7 billion is committed and payable in ten annual installments, $75 million covers the settling states' costs, and approximately $5.3 billion becomes payable state by state only if industry peers adopt comparable teen protections and each peer with annual profits above $10 billion is obligated to pay the same amount. That last tranche was not accrued, because the recognition threshold was not met at the reporting date.

Non-accrual does not mean the expected cost is zero. It means the accounting test was not satisfied. The research excludes the tranche as a judgment, for an economic reason rather than an accounting one: Meta pays it only in a world where its largest competitors are also paying and are also constrained on the same product surfaces. You are free to disagree, and the honest way to disagree is to add the claim with your own assumption about when the conditions would be met and discount it, rather than dividing a nominal figure by the share count.

The tax side adds a third word. A $15.89 billion notice is an amount an authority has asserted and the company is contesting. It is not an agreed liability, and it is also not irrelevant, because some of the exposure on those positions already sits inside recognized accruals and the filings do not map one onto the other. Four figures appear in that note, $20.82 billion of gross unrecognized benefits, $3.47 billion of interest and penalties, a $14.58 billion provision effect and the $15.89 billion assessment, and they are measured on different definitions and may overlap. Resist the urge to rank or add them.

9. New instruments, in the order they affect you

Three late-October 2026 capital-structure announcements and financings each teach a mechanism.

A stock dividend of a new non-voting class. One existing A becomes an A+C package; one existing B becomes a B+C package. The stock dividend itself does not mechanically reduce a holder's proportionate ownership, without guaranteeing market value or ignoring financing adjustments, which is why the research states every value per pre-distribution investor package: that keeps the comparison with the dated $605.08 price like for like. The charter was filed 10-21-2026 for 01-19-2027 effectiveness; the stock dividend is contingent on effectiveness, and its record and payment dates were not set at the cutoff. After issuance subject to the applicable conditions, Class C can be used to issue equity without altering votes. Future issuance would dilute economic ownership while leaving control untouched. A founder agreement filed with the announcement staples 80% of Mr Zuckerberg's dividend shares to his voting shares and releases them as 32 qualifying quarters of continued service accrue, not necessarily consecutive and so not a guaranteed eight years, with further release routes tied to additional dividends, so the separation of his economic ownership from his votes is gradual rather than immediate.

A mandatory convertible preferred. $14.375 billion at 6.50%, $12.5 billion at the October closing and $1.875 billion more when the underwriters exercised their option, converting around 08-15-2029 into between 1.3620 and 1.6000 shares per $1,000 of liquidation preference, before Class C and other applicable conversion adjustments. The payoff shape is worth learning: below the initial price the holder gets the maximum number of shares and takes the downside; between the initial price and a threshold about 17.5% higher the holder receives a fixed $1,000 of value; above that threshold the holder participates at the minimum rate. So a modest gain within that conversion band does not raise the principal conversion value; coupons and their timing are separate. The research carries the instrument at its issue value as a proxy, because no dated post-issue market mark was available and no conversion-aware model was built, and it says so. Nasdaq's approval of the shares for listing upon issuance is a different thing from a liquid market with a price you can use.

Forward sale agreements. Meta pre-agreed terms on 40 million Class A shares at an initial forward price of $622.91875, added 6 million more on the same terms when the underwriters exercised their option on 11-04-2026, and has received no cash for any of them. Settlement is at its discretion, by delivering shares, on or before 11-19-2027. The lesson is that an outside settlement date is a deadline rather than a scheduled receipt, and that a contract whose spread and price reductions are redacted in the filed exhibit can only be valued on assumed inputs.

One detail sits underneath all three. Both new contracts contain adjustments that activate if the non-voting class trades below Class A. They were drafted for that possibility, which is a reason not to assume the two classes will trade at parity merely because it is convenient to do so.

Two early-November developments teach two more lessons.

Notes, in two currencies. In early November Meta borrowed $15.0 billion in dollars, completed on 11-05-2026, and priced EUR 9.0 billion in euros for settlement on 11-12-2026. The lesson is the oldest one in a balance sheet: borrowed cash comes with its claim. The dollar notes raised $14,938 million and created a claim worth about $14,980 million at their offering price, so the value per share moves only by the costs, about two cents, while cash available for the funding test rises by billions. A euro claim also moves with the exchange rate, which is why the research states the rate it uses, $1.1224 per euro on 11-05-2026, and that no hedge is disclosed.

A planned at-the-market program. Meta has said it plans to sell up to about $60 billion of Class C stock over time once that class trades, and has also said that no agreement exists and that timing, size and terms are undetermined. A plan is not cash and not shares, so the research adds neither. What it teaches is how to read dilution: selling stock above what the business is worth helps existing holders, selling below it hurts them, and the plan cannot tell you which, because it depends on a value nobody can observe.

10. Conditional scenarios are not forecasts

The Meta valuation is a grid of unweighted conditions. No case is called most likely, no probability is attached, and no midpoint functions as a base case. That is not evasion; it is the only honest presentation when the deciding input is unmeasured.

Two related tools appear in the work and are often misread. A reverse calculation asks what combination of assumptions would match a price. Here it shifts growth across a declared domain and reports values from $400.01 to $609.39 at the 11-05-2026 valuation date. Against that day's close of $605.08 the top of the range reaches the price, at a growth shift of about +3.90 percentage points, but only if the New Mexico matter resolves within amounts already represented by the modeled legal allowance; with the judgment paid as entered, it does not. The right reading is that a crossing belongs to one dated price and one set of conditions. It is not a statement about what the price requires, and widening the domain until it reaches the price would prove nothing.

A second lens is better than a second pass through the same model. The maintenance cash yield asks what cash the business would produce if it stopped growing and merely replaced its capacity, and compares that with what the quote implies you are paying for the operating business, with legal cash excluded from both sides. At 4.32% in the separate reference maintenance cross-check, the quote-implied operating value is roughly 23.1 times operating maintenance cash; this is not an equity price/FCF multiple. On every path tested, the FY2027 cash yield is negative, which tells you what the price is not paying for.

The 8.75% rate is itself a condition. It is an analyst choice, not a rate calibrated to dated market data: the research does not construct it from a dated risk-free rate, equity risk premium and beta. Holding every other assumption fixed, the reference value at 0.65 turnover is $623.65 at 7.5% and $323.53 at 10.5%, against $462.05 at 8.75%, at the 11-05-2026 valuation date. Those are sensitivities to one input, not probabilities and not a fair-value range.

11. Value and financeability are separate questions

A company can be worth a great deal and still need money on a particular Tuesday. Meta's reference operating path spends about $114 billion more than it earns between late 2026 and the end of 2029, before legal, tax, venture or debt payments.

The funding test in the research is deliberately crippled: it allows only the suspension of forecast common dividends, adds no borrowing beyond the notes Meta has already sold or priced, defers nothing, and releases no restricted cash early. With the November notes and the larger forward in the ledger, on even intra-year timing the reference 0.65 case's minimum cash is $52,178 million and no additional funding is required against the modeled $10 billion buffer under that timing pattern. Redistribute the same annual flows adversely inside the year and the same path shows a $6,774 million shortfall against a $10 billion buffer in mid-2029; without the euro notes, which had not settled at the cutoff, it would be $15,871 million. Notice what moves it: borrowed and pre-sold cash, each dollar matched by a claim, not cash the business earned. The conclusion to draw is not that the company is in trouble, nor that it is now safe; it is that the result depends on an intra-year residual-cash and event-timing pattern not calibrated in this work; the reported quarterly operating cash flow figures do not identify that pattern. When a conclusion flips on an unmeasured input, say it is unresolved. Liquidity can also bind before value moves at all: posting $25 billion of cash collateral for any New Mexico appeal would by itself create a $23,959 million requirement on the uneven pattern, even if the judgment were later reversed; the test counts gross principal and leaves out interest carry, fees and facility effects.

The counterintuitive result is worth remembering. The largest funding requirement in the whole set belongs to a high-demand case, which needs $29.3 billion on even timing, even after the November financings, while being worth more per package than the reference case. Growth served with weak capital productivity consumes cash faster than it creates value.

12. A rating answers a different question again

A rating is not a valuation, and the difference is worth learning because the two are constantly confused.

A valuation answers what a business is worth under stated assumptions. The Meta work does that conditionally, as a grid, because the assumption that decides the answer is undisclosed. A rating answers a narrower and more practical question: at this price, on this date, over a stated horizon, is the opportunity adequate for the risks you can identify? That is why a rating has to be bound to a price and a session. The same business at a different price can deserve a different rating, with the operating story, the scenarios and the quality of the evidence all completely unchanged.

Three habits follow. First, separate the rating from the thesis. "The advertising engine is proven and the capital program is not" is a thesis; it does not move when the quote moves. "The price offers an inadequate margin for the dated risks" is a rating; it does. Second, do not reverse-engineer a fair value from a rating. A rating that names a future price objective is a price target wearing a different label, and a score produced by averaging a grid is arithmetic pretending to be judgment. Third, ask what would change it, and insist the answer be an observation rather than a mood. In the Meta research the list is short and specific: a disclosed capacity or utilization measure, any disclosed 2027 capital budget with a revenue outlook beside it, a change in disclosed useful lives, attributed revenue from the newer products, what the October and November financings actually deliver in cash against shares, and any appeal of the New Mexico judgment.

The discipline here is the same one this companion has been teaching throughout. Say which part is measured, which part is a reading and which part is your opinion, and make the opinion do real work rather than hiding behind the range.

13. Five habits worth taking with you

Separate what is measured, what is inferred and what is judged, and do it in ordinary sentences rather than by labeling every clause. Impressions up 10% is measured. That the like-for-like price rose more than the blended 14%, because management says mix was a drag, is a reading. That the share of growth attributable to the investment program is positive and smaller than the program's cost is a judgment.

Check what an absence actually means. These filings disclose no backlog, no remaining performance obligation and no minimum advertiser commitment, which tells you there is no disclosed contractual floor. It does not tell you that no advertiser is under any contract, and it does not tell you what notice an advertiser must give.

Never add differently defined disclosures. $355.73 billion of uncommenced leases, $454.52 billion of contractual commitments and a $68.40 billion maximum venture exposure overlap, are measured on different bases and are not a single liability. Read them one at a time.

Never treat a pledge as a rule. After the November 2026 elections, Texas's governor-elect pledged a moratorium on data-center approvals and Democrats in Congress called for new legislation; Texas already had temporary measures in place since August. An election result, a campaign pledge or proposal, an existing executive or agency measure and an enacted law are different things. A pledge or an election result does not by itself change what a company may do; an existing enforceable measure, such as the Texas halt on new grid connections, can constrain it now, and so can an enacted law once it takes effect. California's youth-feature law, AB 1709, adds one more state: an enacted law scheduled to take effect on 01-01-2027 that Meta and others are challenging in court, where a challenge is not an injunction. Meta's 20% El Paso venture shows why the details matter: it sits on a different electricity grid from the one the existing Texas halt covers, which narrows one risk without removing the others.

Ask which single assumption is doing the work, and then ask who measured it. In this case the answer is the mature ratio of revenue to productive capital, and the answer to the second question is nobody.

What this companion is, and its limits

It teaches the concepts the Meta Deep Dive and Research Note use, with examples reconciled to those documents. Company information runs through the Form 8-K filed after the 11-05-2026 close, the balance sheet is dated 09-30-2026, and the price reference is a dated vendor regular-session close of $605.08 on 11-05-2026, which is not an exchange-certified close and is not current as you read this. The sources are listed, with their documents and links, at the end of this companion. Nothing here is a recommendation. The research it accompanies carries a rating at a stated price and horizon, and it still assigns no probability, no confidence grade and no central fair value.

Blockhaven view

Blockhaven Rating: 2 / 5 — CAUTIOUS. Assessed at the November 5, 2026 vendor regular close of $605.08 per pre-distribution Class A investor package. Year-end 2029 is a reassessment window, not a calculated security-return horizon.

Sources

• S1 Meta Platforms third-quarter 2026 earnings release and CFO outlook commentary, dated 10-21-2026. https://s21.q4cdn.com/399680738/files/content_files/Meta-09-30-2026-Exhibit-99-1_FINAL.pdf (opens in a new tab)
• S2 Meta Platforms third-quarter 2026 earnings presentation, including the quarterly segment series. https://s21.q4cdn.com/399680738/files/content_files/Earnings-Presentation-Q3-2026.pdf (opens in a new tab)
• S3 Quarterly report on Form 10-Q for the period ended 09-30-2026, including Notes 4, 5, 6, 8, 9, 10, 11, 12 and 13. https://www.sec.gov/Archives/edgar/data/1326801/000162828026066773/meta-20260930.htm (opens in a new tab)
• S4 Annual report on Form 10-K for the year ended 12-31-2025, used for accounting policies, useful lives, segment allocation conventions and FY2023 to FY2025 cash flow. https://www.sec.gov/Archives/edgar/data/1326801/000162828026003942/meta-20251231.htm (opens in a new tab)
• S5 Third-quarter 2026 earnings call transcript, 10-21-2026, used for management statements, product metrics and experiment results. https://s21.q4cdn.com/399680738/files/content_files/META-Q3-2026-Earnings-Call-Transcript.pdf (opens in a new tab)
• S6 Current report on Form 8-K covering the October 26 to 29 financings, with the certificate of designations filed as Exhibit 3.1. https://www.sec.gov/Archives/edgar/data/1326801/000119312526438017/d183592d8k.htm (opens in a new tab)
• S7 Prospectus supplement for the mandatory convertible preferred offering, dated 10-26-2026. https://www.sec.gov/Archives/edgar/data/1326801/000119312526436405/d133631d424b5.htm (opens in a new tab)
• S8 Forward sale agreement confirmation filed as Exhibit 10.1 to S6, with the spread and scheduled price reductions redacted in the filed exhibit. https://www.sec.gov/Archives/edgar/data/1326801/000119312526438017/d183592dex101.htm (opens in a new tab)
• S9 Vendor price history used only for the dated regular-session close of 11-05-2026 ($605.08; open $632.30, high $634.99, low $583.80); Yahoo Finance chart data show the same close and low. https://stockanalysis.com/stocks/meta/history/ (opens in a new tab)
• S10 Current report on Form 8-K dated 10-21-2026, with the Founder Alignment Agreement of 10-20-2026 filed in full as Exhibit 10.1. https://www.sec.gov/Archives/edgar/data/1326801/000162828026066700/meta10202026-ex101.htm (opens in a new tab)
• S11 Preliminary information statement on Schedule 14C, 10-21-2026. https://www.sec.gov/Archives/edgar/data/1326801/000162828026066710/meta-20261021.htm (opens in a new tab)
• S12 Definitive information statement on Schedule 14C, 11-02-2026, including the Delaware litigation relating to the recapitalization. https://www.sec.gov/Archives/edgar/data/1326801/000162828026068935/meta-20261102.htm (opens in a new tab)
• S13 Form 8-A registering the Series A depositary shares on Nasdaq, 10-29-2026, accession 0001193125-26-438111. https://www.sec.gov/Archives/edgar/data/1326801/000119312526438111/d184794d8a12b.htm (opens in a new tab)
• S14 Nasdaq certification of approval for listing and registration upon official notice of issuance, 10-29-2026, accession 0001354457-26-001039. https://www.sec.gov/Archives/edgar/data/1326801/000135445726001039/METAM_8A_Cert.pdf (opens in a new tab)
• S15 CNBC report of 10-30-2026 on Meta's settlements with bellwether plaintiffs S.J. and K.D.B., terms undisclosed, used only for those reported facts. https://www.cnbc.com/2026/10/30/meta-settles-with-teen-during-trial-testing-claims-social-media-caused-depression-eating-disorders.html (opens in a new tab)
• S16 Quarterly report on Form 10-Q for the period ended 06-30-2026, accession 0001628280-26-050705, used for the second-quarter general and administrative discussion (MD&A, printed p44). https://www.sec.gov/Archives/edgar/data/1326801/000162828026050705/meta-20260630.htm (opens in a new tab)
• S17 Findings of fact, conclusions of law and judgment filed 10-21-2026 in State of New Mexico ex rel. Torrez v. Meta Platforms, Inc., No. D-101-CV-2021-00132, First Judicial District Court, Santa Fe County, as published by the New Mexico Department of Justice; its final page awards a $25,000,000,000 civil penalty and states 15% interest per annum until paid in full. https://nmdoj.gov/wp-content/uploads/Facebook-Judgement.pdf (opens in a new tab)
• S18 Bloomberg report of 11-02-2026, as published by Claims Journal, on Snap finalizing a settlement with bellwether plaintiffs S.J. and K.D.B. reached partway through jury selection, according to the plaintiffs' attorneys, terms undisclosed, used only for those reported facts. https://www.claimsjournal.com/news/national/2026/11/02/341087.htm (opens in a new tab)
• S19 Pricing term sheet for the $15.0 billion of senior notes, issuer free writing prospectus of 11-03-2026, accession 0001193125-26-445646. https://www.sec.gov/Archives/edgar/data/1326801/000119312526445646/d186765dfwp.htm (opens in a new tab)
• S20 Pricing term sheet for the EUR 9.0 billion of senior notes, issuer free writing prospectus of 11-04-2026, accession 0001193125-26-447522, settlement scheduled 11-12-2026. https://www.sec.gov/Archives/edgar/data/1326801/000119312526447522/d154804dfwp.htm (opens in a new tab)
• S21 Current report on Form 8-K dated 11-03-2026 and filed 11-05-2026, accession 0001193125-26-450074, accepted 4:58:46 p.m. Eastern time, reporting completion of the dollar notes and full exercise of the 6,000,000-share option with additional forward sale agreements; Exhibit 1.1 is the underwriting agreement and Exhibits 10.1 and 10.2 the additional forward confirmations. https://www.sec.gov/Archives/edgar/data/1326801/000119312526450074/d166273d8k.htm (opens in a new tab)
• S22 Prospectus supplement for the dollar notes dated 11-03-2026, accession 0001193125-26-448015, used for net proceeds, as-adjusted capitalization and the preferred over-allotment reported under Recent Developments. https://www.sec.gov/Archives/edgar/data/1326801/000119312526448015/d135326d424b5.htm (opens in a new tab)
• S23 Preliminary prospectus supplement for the euro notes dated 11-04-2026, accession 0001193125-26-446110, in which net proceeds are left blank and no hedge is described. https://www.sec.gov/Archives/edgar/data/1326801/000119312526446110/d44678d424b5.htm (opens in a new tab)
• S24 Issuer free writing prospectus of 10-26-2026, accession 0001193125-26-431036, announcing the October offerings and the financing plan, including the planned at-the-market program of approximately $60 billion of Class C stock. https://www.sec.gov/Archives/edgar/data/1326801/000119312526431036/d116268dfwp.htm (opens in a new tab)
• S25 Prospectus supplement for the Class A offering in connection with the forward sale agreements, filed 10-28-2026, accession 0001193125-26-436410, used for the $625.00 public offering price, the $622.91875 initial forward price and the 11-19-2027 settlement limit. https://www.sec.gov/Archives/edgar/data/1326801/000119312526436410/d94886d424b5.htm (opens in a new tab)
• S26 European Central Bank euro foreign exchange reference rates, used only for the 11-05-2026 rate of $1.1224 per euro. https://www.ecb.europa.eu/stats/eurofxref/eurofxref-hist-90d.xml (opens in a new tab)
• S27 Meta and BlackRock release of 07-28-2026 announcing the El Paso, Texas data-center venture. https://investor.atmeta.com/investor-news/press-release-details/2026/Meta-Announces-New-Strategic-Venture-with-BlackRock-to-Develop-Data-Center-in-El-Paso/default.aspx (opens in a new tab)
• S28 Reuters report of 11-04-2026 by Laila Kearney and Nathan Crooks on Texas data centers after the governor's race, used only for the reported measures and pledges. https://www.reuters.com/world/us/texas-data-centers-face-uncertain-future-after-hinojosa-wins-governors-race-2026-11-04/ (opens in a new tab)
• S29 Reuters report of 11-05-2026 by David Lawder, Steve Holland and Noel Randewich on the election results and the technology-stock decline, as republished by The Star. https://www.thestar.com.my/news/world/2026/11/05/democrats-vow-to-rein-in-trump-as-he-shows-off-white-house-ballroom (opens in a new tab)
• S30 Associated Press report of 11-03-2026 on the Texas governor's race. https://apnews.com/article/texas-governor-election-hinojosa-abbott-c9a0ffa1ec3cd599f9bc304ae26268a2 (opens in a new tab)
• S31 Associated Press report of 11-05-2026 on the governor-elect's first news conference and the divided Texas Legislature. https://apnews.com/article/democrat-gina-hinojosa-texas-governor-bipartisan-administration-5640fa066346ce909f21b719ab073ee2 (opens in a new tab)
• S32 NPR report of 11-03-2026 on control of Congress, based on Associated Press race calls. https://www.npr.org/2026/11/03/g-s1-149526/2026-midterms-congress-balance-of-power (opens in a new tab)
• S33 ERCOT presentation of 07-19-2022 on the Pecos West intertie, used only for the fact that El Paso is served from the Western Interconnection rather than the ERCOT grid. https://www.ercot.com/files/docs/2022/07/15/Pecos%20West%20Intertie%20ERCOT%20RPG%2020220719.pdf (opens in a new tab)
• S34 Reuters report of 11-05-2026 by Jonathan Stempel on the separate lawsuits filed by Google, Meta and TikTok on 11-04-2026 to block California Assembly Bill 1709, used for the filing, the law's terms as reported and its 01-01-2027 effective date. https://www.reuters.com/world/us/google-meta-tiktok-sue-block-california-youth-social-media-law-2026-11-05/ (opens in a new tab)
• S35 Associated Press report of 11-05-2026 by Barbara Ortutay on the same lawsuits, used for the federal courts in which they were filed and the scheduled effective date. https://apnews.com/article/589c0eabeabced2e07b8f9138ee2df79 (opens in a new tab)
• S36 Office of the Governor of California release of 09-10-2026 on the signing of child online safety legislation, including AB 1709. https://www.gov.ca.gov/2026/09/10/governor-newsom-signs-the-strongest-child-safety-chatbot-and-social-media-laws-in-the-nation/ (opens in a new tab)
• S37 Opinion of the California Court of Appeal, Second Appellate District, Division Two, filed 11-05-2026, certified for publication, in Social Media Cases, San Diego Unified School District v. Meta Platforms, Inc., B340661, and three related appeals (JCCP No. 5255), affirming the judgment for the defendants, as hosted by Justia. https://cases.justia.com/california/court-of-appeal/2026-b340661.pdf?ts=1793918139 (opens in a new tab)
• S38 Wall Street Journal report of 11-04-2026 by Meghan Bobrowsky on comments by Meta President Dina Powell McCormick at the WSJ Tech conference, used only for the reported remark that building data centers will be harder after the elections. https://www.wsj.com/tech/ai/meta-president-says-midterm-results-will-make-building-data-centers-harder-f1d27c93 (opens in a new tab)
• S39 Conductor announcement that it is joining Meta, with its service timeline beginning 11-04-2026; no transaction terms are stated. https://www.conductor.build/announcement (opens in a new tab)

Assumptions and disclosure

The Companion explains the research in the Deep Dive. Conditional values remain conditions, not forecasts; opinions may change as evidence changes; and no case is a promise of performance.

Important disclosures

Blockhaven research is provided for general informational and educational purposes. It is not individualized investment advice and does not take into account any reader's financial circumstances, investment objectives, risk tolerance, tax situation or portfolio.

Nothing in this report is an offer or a solicitation to buy or sell a security or other financial instrument. Reading or accessing Blockhaven research does not by itself create an investment-advisory, fiduciary, brokerage or client relationship.

Investing involves risk, including the possible loss of principal. Scenario values, terminal values, valuation ranges, sensitivities and forward-looking assumptions are analytical tools, not guarantees or promises of future performance, and actual results may differ materially.

Blockhaven uses information from company filings, company publications, market-data providers and other sources it believes to be reliable, but that information may be incomplete, delayed, revised or incorrect.

Opinions, ratings and interpretations reflect Blockhaven's research judgment as of the report date and may change as new evidence becomes available.

Positions and conflicts

As of November 5, 2026, Blockhaven LLC, its owner and the owner's spouse, and accounts or entities they control report no positions in Meta securities or derivatives. They report no compensation, advisory, sponsorship or advertising relationship with Meta and no other material financial or business conflicts involving Meta.

© 2026 Blockhaven LLC. All rights reserved.