Meta Platforms

IThe business

Meta sells advertising. In the twelve months to 30 September 2026 revenue was $241.6B, and 97.5% of it was ads on Facebook, Instagram, Messenger and WhatsApp C (calculated).²³⁴ Each day 3.64 billion people use at least one of those apps, up 3% in a year.⁵ The fiscal year ends on 31 December, so TTM on this page means the four quarters to 30 September 2026.

The ad business is very good and getting better. Q3 ad revenue rose 26% to $63.1B: 10% more impressions at a 14% higher price per ad. Average revenue per person was $17.72 in the quarter, up 23%, and the United States and Canada grew 33%.³ Those are large-company numbers that look like a small company’s.

The rest of the page is about what Meta is doing with that money. It spent $85.8B of capital in nine months, up 71%, and it guides to $135B to $145B for the year.⁵ Free cash flow was −$2.1B in Q3.⁵ On 26 October it launched an equity raise: about $37.5B now, and a plan for up to $60B more in 2027.¹⁰⁶¹¹

What it sells

• Family of Apps. Ads, plus $3.0B of other revenue in nine months, up 67%, mostly WhatsApp paid messaging and subscriptions. Segment operating income was $98.2B in the TTM, a 41% margin after legal charges C (calculated).³²⁴
• Reality Labs. Quest headsets and AI glasses. Revenue was $1.3B in nine months and the operating loss $12.8B. The loss since 2019 is $101B C (calculated).³²¹² Management expects about $15B next year.⁵
• Muse. A personal AI agent launched in September. The company says more than 10 million people in the United States and Canada onboarded in the early weeks, and that shopping is the top use, by over 25% of weekly users.¹⁰ On the call management said about 25% of those who set it up use it on a given day. It is free for most people and has no reported revenue line.¹³
• Enterprise. A model API, a coding tool and business agents. More than 2 million businesses use Business Agent each week.¹⁰ Management says offers to rent its compute arrive at a premium to its cost, and that it sees higher margins in selling intelligence than in selling compute.¹⁴

Where the ad growth comes from

Two things drive it: more time spent, and better matching of ads to people. Management reports double-digit growth in Instagram time per user, and a ranking model that lifted app-ad conversions 9.6% in Q3.¹³ Advantage+, its automated campaign suite, passed a $75B annual run rate in Q2.¹⁴ Those figures are company claims. What the filings confirm is the outcome: the 10-Q attributes the price increase mostly to better ad performance from its targeting and measurement tools.³

Meta added more ad dollars in Q3 than Google or Amazon: $13.0B against $7.8B at Google and $1.9B at Amazon C (calculated).³⁷¹⁵ The table under The competitive advantage has the detail.

The new shape of the costs

Costs and expenses rose 68% in Q3 to $51.6B. $9.89B of that was a legal charge. Without it, Q3 operating margin would have been 35%, against 40% a year earlier.⁵¹³ Headcount fell 15% to 66,775.⁵ So people are not the reason. Infrastructure is: depreciation was $19.5B in nine months, up 48%, and the 10-Q also names third-party cloud services and third-party AI token costs as drivers.³ Meta now pays other companies for computing and for tokens. Neither figure is disclosed.

The ad business earns the money. Everything hard on this page is about what the money is being turned into.

IIThe moat

I think Meta’s protection is three things working together: the habit of 3.64 billion daily users, the data those users produce, and an ad system that turns the data into results for millions of small advertisers. It is wide. Its direct social rivals are each under a tenth of its size. I think it is widening against other ad sellers and uncertain against the thing that matters most, an AI agent that sits between a person and the apps.

The evidence is in the price. Price per ad rose 14% while impressions rose 10%, and impressions grew mostly in places that pay less.³ Advertisers paid more per ad on a falling mix. They do that when ads work. The franchise also survived a real test: after Apple’s 2021 privacy change Meta lost signal, and revenue fell 1% in 2022. Revenue has since grown from $116.6B to $241.6B C (calculated).¹²³

Evidence it is widening

Share of growth. In Q3 Meta’s ad revenue grew 26%, Google’s 10.6% and Amazon’s 10.9% C (calculated).³⁷¹⁵ Amazon’s nine-month growth was 19.8%, so one quarter understates it C (calculated).¹⁵

The court. On 18 November 2025 a federal judge ruled for Meta in the FTC’s case seeking to unwind Instagram and WhatsApp. The FTC appealed in January 2026.³

Reach for new products. Instagram has 2 billion daily users and Threads passed 500 million monthly users.¹⁴ When Meta ships something it can put it in front of more people in a week than most rivals reach in a decade.

Evidence it is narrowing

The agent. If people ask an agent to shop for them, they scroll less. That is the threat, and it is why Meta is building Muse. I can’t judge how this ends and neither can anyone else yet. The 10-K says Meta expects new competition from AI, and management’s own view is that shopping will shift to agent-to-agent interactions, which is a different place from a feed of ads.²¹³

Teenagers. The 10-Q says rival products have cut some users’ engagement and that competition is sharpest among younger users. The settlement with 52 attorneys general requires product changes for users under 18 in the United States.³

Europe. Less personalized ads were launched under regulatory pressure, and the 10-Q warns further changes could have a significant effect on European revenue. Europe was 23% of nine-month revenue by customer address C (calculated).³

The franchise is wide against ad rivals. Against an agent that replaces the feed, nobody knows yet, including Meta.

IIIThe competitive advantage

Meta’s advantage is scale in one thing: closed-loop ad performance on apps people open every day. Google has more ad dollars, and intent from search. Amazon has purchase data. Meta has attention and the small advertiser. It says at least 500 million self-identified small businesses use its apps, from a survey, and nearly 10 million advertisers used an AI creative tool in Q3.¹⁰¹³

Where Meta wins

Meta wins on growth and on margin. Its Family of Apps margin before the legal charge was about 42% in Q3 C (calculated), and Google Services, which carries Search, is the only rival of similar scale.³⁷ It wins with advertisers who want a sale and not a brand impression, because it can show the sale.

Where it loses

It loses on owning the rails. Meta has no phone operating system and no browser. Apple’s 2021 change showed what that costs, and the 10-Q still says that change hurts budgets.³ It is behind on frontier models by its own timeline: Muse Wonder is described as its first frontier model and shipped this year.⁵ And it lacks a cloud business. Alphabet, Amazon and Microsoft rent spare capacity to outside customers, so a wrong guess on AI demand can land on someone else’s bill. Meta’s build has to pay for itself mostly through its own products.

Meta has the best ad growth of the large platforms and the weakest fallback use for a data center.

IVManagement

Mark Zuckerberg founded the company in 2004 and is chairman and chief executive. He controls it. At 1 October 2026 he held 60.6% of the votes and 13.4% of the economics.⁸ At $614.99 that stake, about 342.4M shares, is worth about $211B C (calculated).⁸¹ Few chief executives have that much of their own money in. Fewer can’t be removed by the people who own the other 86.6%.

How he is really paid

His reported pay for 2025 was $25.1M: $1 of salary and the rest security and travel-related costs. He gets no bonus and no stock awards.⁹ The pay table tells you almost nothing about him. Three other things do.

First, his stake. A 1% move in the stock moves his wealth by about $2.1B C (calculated). He is paid like an owner, so far as the price goes. He is also the owner who can wait longest, because no vote can hurry him.

Second, the Founder Alignment Agreement of 20 October. The Class C dividend gives every holder one non-voting share for each share held. Zuckerberg can sell 20% of his Class C freely, and the rest is released at 2.5% a quarter while he stays full-time. So he can raise cash without giving up votes for the first time. And each time the stock doubles from its 10 September price of $580.60, adjusted for the split, with at least a 10% yearly return, a further dividend goes to all holders.²⁰ That extra dividend goes to everyone, so other holders lose nothing by it. What the agreement pays him to do is stay, and keep control while taking some money out.

Third, the options. In March Meta granted options on 19M shares to executives and employees, about 20M by September, at an average strike of $2,788, about 4.5 times today’s price C (calculated), for a company value above $7T C (calculated).³²¹ The filings do not say who holds them. They vest on service and market conditions, and they pay nothing unless the stock rises enormously. The people who hold them are paid for a very large outcome, and these options pay nothing if the AI build merely works.

The board, and who can say no

Nobody can. Shareholders approved the Class C structure by written consent the day after the board did, using Zuckerberg’s own votes, and Class A holders got no meeting.⁸²⁰ At the May meeting a proposal to end dual-class shares drew 1.31B votes for and 3.65B against. Take out his 3.42B Class B votes and about 85% of the rest voted for it C (calculated).²³⁹ Say-on-pay comes once every three years, next by 2028.⁹ Hock Tan and Tracey Travis left the board in May.²⁴ Dina Powell McCormick left the board in December and came back in January as President and Vice Chairman, on a $60M stock grant, and now co-leads Meta Compute.²²⁹

Capital over a full cycle

This is a record with big wins and big losses, and one habit. Instagram and WhatsApp were bought cheap and became the business. Reality Labs has lost $101B. In 2022 the market forced a turn: Brad Gerstner’s open letter that October asked for at least a 20% cut in headcount expense, the stock had collapsed, and headcount went from 86,482 to 67,317 in a year.²⁵¹² It worked. The stock has since risen 6.9 times from its 2022 low close of $88.91 C (calculated), and the lesson management seems to have drawn is that it can always cut later.

The buyback record is what I would look at hardest. Meta bought $174.0B of stock from 2017 to 2025 C (calculated), and share counts did fall, from 2.96B diluted to 2.57B C (calculated).¹²² But it paid more as the price rose, and stopped as the need for cash rose: $173 a share in 2022, $218 in 2023, $458 in 2024 and $656 in 2025 C (calculated).¹²² No shares were bought in 2026.³ On 26 October it sold forward 40M shares at $625, below the 2025 average.⁶ In round figures it bought back $26B of stock in 2025 and agreed to sell $25B a year later at a lower price. That is what happens when a company’s cash needs turn in twelve months.

He owns about $211B of it and answers to nobody. That cuts both ways, and the same person made both Instagram and Reality Labs.

VFinancials

Earnings are strong and cash is not, because the money is going into the ground. Four things need watching: depreciation that has not arrived yet, obligations that are not on the balance sheet, tax and stock pay.

Revenue grew from $27.6B in 2016 to $201.0B in 2025, 24.7% a year C (calculated). Operating margin ran between 25% and 50% over the decade: 25% in the 2022 trough, 41% in 2025, 32.9% in the TTM.¹²²³

What an owner can take out

Reported net income in the TTM was $75.2B, or $29.31 a share C (calculated).²³⁴ That is not the number I care about. An owner can spend only what’s left after stock pay and after the capital needed to stand still.

• Free cash flow as the company defines it was $25.1B. Take out $26.9B of stock pay and it is −$1.8B C (calculated).²³⁴ On that measure an owner took out nothing.
• That is unfair, because most of the capital is for growth. So I rebuild it. Operating income before the $11.89B attorneys general charges, taxed at 17%, is $75.8B. Add back $24.9B of depreciation. Deduct what it would take to keep today’s assets going: about $31B for servers replaced every 5.5 years and buildings, or about $43B once the $89.5B of construction in progress is in service. That leaves $58B to $70B, or $22 to $26 a share E (estimate).³²

So the cash an owner could take out, if Meta stopped growing, is $22 to $26 a share against reported earnings of $29.31. That is a yield of 3.6% to 4.3% at the price E (estimate), below the 4.94% ten-year Treasury.²⁶ The stand-still figure is my estimate, which is why I give a range, but neither end gets close to the $24.9B of depreciation.

Obligations that are not on the balance sheet

This is the biggest thing on the page. At 30 September Meta had $355.7B of leases that have not started and $454.5B of non-cancelable contractual commitments. Together that is $810.2B, up from $234.8B nine months earlier, or 3.4 times TTM revenue C (calculated).³² Of the commitments, $45.9B falls due in the rest of 2026 and $121.4B in 2027. A further $14.7B of cloud capacity is contingent.³

The total is not all new spending on top of my capital path, since much of it is the path, and leases run up to 30 years. But most of it can’t be undone, and that changes what the capital guide means. Management stressed on the Q2 call that long-lived foundations leave server decisions open for 2028 and beyond.¹⁴ The 10-Q’s non-cancelable number rose $105B in the following quarter.

Some of the rise is price, not volume. In April management lifted the 2026 capital guide mostly on higher component costs, particularly memory, and said it is rolling out more than 1 GW of its own custom silicon with Broadcom to cut the bill.²⁸

Two more items sit off the balance sheet. Meta holds 20% of two data-center ventures, in Louisiana and Texas, that it does not consolidate. Its maximum exposure to loss is $68.4B, of which $41.5B is residual value guarantees.³ And $10.8B of money market funds now sits in escrow for suppliers, released between 2028 and 2030.³

Earnings quality

Depreciation to come. Construction in progress is $89.5B, none of it depreciating yet, and property rose $76.8B in nine months.³ In January 2025 Meta extended server lives to 5.5 years.² If lives prove shorter, profit is overstated today. Each half year off the server life adds about $2.4B a year on the current fleet C (calculated).

Tax. The nine-month tax rate was 2%, because of an $8.03B benefit from Treasury Notice 2026-7. The Q3 rate was 24%, inflated by non-deductible legal costs, and Q4 is guided at 15% to 17%.³⁵ Cash tax was only $2.9B in nine months.³ I use 17%. The IRS has also asserted $15.9B of additional tax for 2017 to 2019, plus interest and penalties and before offsets, which Meta is contesting in the Tax Court, and gross unrecognized tax benefits are $20.8B.³

Stock pay. The expense was $26.9B in the TTM, 11.2% of revenue. It understates the cost. RSUs that vested in nine months were worth $29.2B at vesting, against $20.9B of expense, and Meta paid $12.8B in cash to settle employees’ taxes on them, money that never touches free cash flow.³ Unrecognized expense is $75.6B, and new grants came at $608 a share.³ Cutting 15% of staff has not cut the stock bill, which rose 45% in nine months C (calculated).³⁴

Legal. Legal accruals were $19.9B at 30 September, from $7.3B at year end C (calculated).³ They are in a section of their own below, because they belong in the valuation.

The balance sheet

Meta had $82.45B of cash and securities and $84.0B of notes at 30 September, so it is about level on net debt. A year earlier it had $44.5B of cash and $28.8B of debt.³⁴ It has borrowed $55B in twelve months C (calculated), keeps Aa3 and AA- ratings, and has no financial covenants.³¹⁰ Interest expense was $977M in Q3, against $227M.³ Non-marketable investments of $32.6B include Scale AI, bought for $13.8B.³²

Then came equity. The preferred pays 6.50% and converts in August 2029 into 17M to 20M shares; it raised about $12.5B net. The forward sale covers 40M shares at $622.92 and pays about $24.9B when Meta settles, by 19 November 2027.⁶¹¹ NVIDIA, AMD and Broadcom indicated interest in up to $5.0B of it, not binding, and all three sell to Meta.²⁹ Underwriters have 30 days to take 6M more forward shares and more preferred, which would add about $5.6B; I count neither.⁶¹¹ The $60B at-the-market plan isn’t signed, and would start in Q1 2027 in non-voting Class C stock.¹⁰

From January, per-share figures on all sources will roughly halve: the Class C dividend works like a two-for-one split.⁵³ Every per-share figure on this page is before it.

In nine months the off-balance-sheet obligations rose by $575B, which is more than twice TTM revenue.

VIValuation

How this goes badly wrong

Before any numbers I ask what would most likely wreck this, and work back. I see three ways.

1. The capacity outruns the uses. Meta signs up for $810B and the ad gain and the new products don’t arrive to fill it. It has no cloud business to rent the excess to at scale, the depreciation arrives anyway, and it raises equity at a lower price to fund the rest. This is the most likely of the three, and I build my bear case on it.
2. The law changes the product. A $25B New Mexico penalty, trials through 2027 and plaintiffs’ demands that have at times reached more than a trillion dollars are costs, but cost isn’t the worst of it.³ The worst is court-ordered design changes that cut time spent by the young, the users who matter most in ten years. That has started: an August judgment in New Mexico ordered changes to how Meta runs its services in the state, and it is stayed while Meta moves to vacate it.³
3. The agent takes the feed. If someone else’s assistant becomes where people decide what to buy, an ad system built on scrolling loses its place. Muse is the defense, and it is why Meta can’t stop spending.

All three share a feature. The ad business is doing fine today, and none would show in this year’s accounts. So I read the valuation for what it pays for, and how it would look if one of these came true.

What I can’t judge

I can’t tell you what Muse costs to serve, because Meta won’t say; management says only that it is compute-intensive and expects a tenfold efficiency gain within a year.¹³ I can’t tell you how much of the ad growth comes from the new compute and how much would have come anyway. I can’t tell you the return on $810B, or the final legal bill. Those are the inputs that decide the value. So I show a wide range instead of a neat number, and the $439 below is the middle of a wide range. It is not a measurement.

What the price implies

At $614.99 Meta’s market value is $1,619B on 2.633B shares, which count the diluted Q3 average of 2.573B, the 40M forward shares and 20M from the preferred at its maximum C (calculated).⁵⁶¹ The close is confirmed on a second feed.¹³⁰ The price is 21.2 times Nasdaq’s 2026 consensus of $29.01 and 18.1 times 2027’s $33.91, with the basis unstated.³¹ That looks cheap for a business growing 26%, and it is why many people I respect like it. My base case returns 8.9% a year at this price, against the 10.5% I require E (estimate).

The usual reverse DCF breaks here. TTM cash after stock pay is negative, and no growth rate turns a negative start into a positive one. So I run it from three starts, and the spread is the point.

If you think Meta is really earning $76B after tax and the capital is optional, the price needs 8.8% a year for ten years, which looks easy. If you count what it takes to stand still, it needs 12.4%. If you start from cash actually left over in 2025, $23.2B, it needs 25%. The debate about Meta is which of those is the real number. The capital spending settles it, and the capital spending isn’t known.

At the price, the ad business alone is not the question. The question is whether roughly $1T of new capacity earns its cost.

The discount rate

I use 10.5%. The capital asset pricing model with the 4.94% ten-year Treasury and Damodaran’s 4.20% premium gives 10.2% on Meta’s ten-year and five-year betas of 1.26, and 11.8% on the three-year beta of 1.64 C (calculated).^2632331 Betas use month-end price returns against SPY, without dividends. The longer windows describe a company that funded itself. This one has just sold stock to fund a build with $810B of fixed obligations, so I lean above 10.2%. I take 10.5%, a beta of 1.33. That is also my rate for Broadcom. One point on the rate moves the base by about a fifth E (estimate).

Three cases

Each case is built by year, 2027 to 2036, valued at 1 January 2027: ad revenue, other Family revenue, new AI revenue, Reality Labs, then employee costs and other costs as a share of revenue, depreciation from the capital path, and 17% tax. Stock pay is a cost. Cash after that and after capital spending is owner free cash flow. Reality Labs’ loss sits inside those lines.

The base case takes the good news at face value for a while, then assumes a large business slows.

• Ads grow 17% in 2027, 14% in 2028 and fade to 5% by 2036, to $611B, 9.3% a year E (estimate). Nine months of 2026 grew 29%; management says growth stays strong through 2027 and gives no number.⁵³
• New AI revenue goes from nothing to $100B by 2036 E (estimate). One billion Muse users at $80 a year, about a quarter or less of what management says American users already bring in through the apps, plus $20B of enterprise. That is a judgment. I have no accounts to test it against.¹³
• Capital spending is $185B in 2027 and about $205B a year to 2032, then falls to 28.5% of revenue E (estimate). That 2027 figure is below the $216B a year the Q4 guide implies, so I am treating part of Q4 as timing, and I can’t prove that. Management said in July its plans aim to maximize 2026 and 2027 capacity. If 2027 holds the Q4 rate, my base is $430 E (estimate).⁵¹⁴ Commitments of $121.4B fall due in 2027 alone, and cloud deals in them are expenses, not capital.³ Meta has guided nothing for 2027.¹³
• Depreciation follows: $51B in 2027 and $121B in 2030, using 62% servers at 5.5 years, the rest over 22 E (estimate). So operating margin falls from 41% in 2025 and about 36% in 2026 before the attorneys general charges to 24% in 2029, then rebuilds to 36% by 2036 as revenue outgrows the build E (estimate).
• Costs include about one point of revenue for continuing legal expense, and I assume Reality Labs keeps narrowing after the $15B guided for 2027, inside the cost lines E (estimate).

My 2027 earnings are about $29.69 a share against the $33.91 consensus, about 12% lower, and the gap is mostly depreciation and cloud costs E (estimate).³¹

The bear case is way one: ads slow to 3% in 2029 in a weak year, new revenue stays small, capital falls only as fast as contracts allow, and margin bottoms at 16% in 2029 before recovering. It also pays the $25B New Mexico penalty in full, the $5.3B contingent settlement, the $15.9B IRS claim and $20B of other legal costs E (estimate). I don’t net these against accruals, because the amount accrued for New Mexico isn’t disclosed, so the bear leans conservative, though it leaves out the 15% interest while the appeal runs.³ The bull case is what management is reaching for: ads at 21% in 2027, a Muse and enterprise business of $250B in 2036, and revenue above $1T E (estimate).

Weighted 25%, 50% and 25%, the cases give $455 E (estimate). At 20%, 50% and 30% it is $477, and at 30%, 50% and 20% it is $433 E (estimate). The value after 2036 is 81% of my base C (calculated), so most of the answer rests on a business a decade from now that I can only sketch.

What the money does

Meta can’t fund this from cash flow. Cumulative owner free cash flow from Q4 2026 reaches about −$145B in my base by 2029 E (estimate). Add dividends of about $5.5B a year, and the $37.4B raised and $82B of cash are spoken for quickly.³⁶ The $60B plan is therefore likely to be used. I keep it out of the share count because new shares at a fair price don’t change value. Sold at $614.99, they’d lift my base to $445, because the price is above it. Sold at $480, about $441 E (estimate). Selling stock above its value helps the holders who stay. It is also a signal: management may be saying that at $625 equity is a fair price to pay for money.

What the last build earned, and what this one must

The record is good. From 2022 to 2025 Meta spent $171.6B of capital and took $54.0B of depreciation, so it added about $118B of net assets. Operating income before the attorneys general charges rose from $46.8B in 2021 to $91.3B in the TTM. After 17% tax that is $37B a year more, about 31% on the capital added C (calculated).¹²²³ The layoffs flatter that number, and cloud deals and leases don’t appear in it. It is still a good record.

The next build is harder. My base has $1,005B of capital from 2027 to 2031. At my server and building mix that is about $131B a year of depreciation. Say half the capital is tied up on average over its life. A 10.5% return on that half is $53B after tax, or $64B before it. So this capital needs about $194B a year of new profit before depreciation E (estimate). In my base that profit rises $154B from 2026 to 2031 and $227B by 2033, so the hurdle clears around 2033 E (estimate). That credits every dollar of ad growth to the build, and some of it would have come anyway. So the last build can look good and today’s price can still look full.

What closes the gap

No single simple change gets my base to the price. It takes one of these, on its own: ads 13.6% a year for ten years, to $895B, about 2.7 times Google’s advertising at its Q3 rate C (calculated); new AI revenue of $368B by 2036; capital spending 22% below my path; or a 2036 margin of 47%, a level Meta has passed only in 2017 E (estimate).⁷¹² Of those, I think the capital number is the most reachable, and it is entirely up to one person.

This isn’t a simple argument. Two changes together, ads two points faster every year and capital 10% lower, get to $603, about the price E (estimate). So the price needs ads to beat my base for a decade, and capital to come in lower, while revenue is growing faster, and the legal bill to stay near what’s accrued. Say 40% on ads, 35% on capital and 70% on legal. Multiplied, that is 10%, or about one in ten E (estimate). They aren’t independent, since great ad growth makes everything easier, so call it one in six or seven. These odds are my judgment and I can’t defend the second digit. What I can say is that three things must go right at once, which is why the price looks fuller than the 18 times earnings suggests.

The margin of safety

What does the same dollar earn elsewhere? The base return at $614.99 is 8.9% a year, against 4.94% on a ten-year Treasury that takes no business risk E (estimate).²⁶ That is a 4.0-point premium, for equity exposed to all three failures above. The closest comparable on this site is Alphabet: my page has it 43% above its base, over a longer forecast period, and it has search, a cloud business that can rent spare capacity and its own chips. Meta is 40% above its base, the least stretched of the four large AI spenders on this site: Alphabet is 43% above, Microsoft 44% and Amazon 47%. Those pages use different rates, 10.5% here and for Amazon, 10.0% for Alphabet and 9.0% for Microsoft. On the rate at which each base meets its price the order holds for the two I can read: 8.9% here, a little over 8% for Alphabet from its sensitivity table, about 7.5% for Microsoft. It also has the faster ad business. Uber and Tencent sit 24% below their base values on my pages, with very different risks. So on my numbers Meta sits closer to its value than its nearest peers, and further from it than the names below their base.

How good the business is and how good the price is are two questions. How good is the business? Very. I’d put the ad franchise with the best in the world. How good is the price? Not good on my numbers. It is 40% above my base, and the stock bought at $614.99 has to earn its way through a build I can’t underwrite. This looks like a great business at a price that already pays for much of the good outcome. A low earnings multiple doesn’t merge the two answers.

• Price$614.99Close, Q4 2026 [1]
• Base value$439My estimate E (estimate)
• Margin of safety below$340set from my bear case C (calculated)

I set the margin of safety so that my bear value of $255 is at least three quarters of the price paid, which gives $340, 23% under my base C (calculated). I use the bear rather than a flat discount because the risk here is one large, lumpy decision and not an ordinary cycle. At $340 my base returns about 11.9% a year and the bear about 9.1% E (estimate). That price is 35% below the lowest close of the past year, $525.72, so it would take a real scare.¹ The cushion is for what the model doesn’t hold: court-ordered design changes, a wrong call on the agent, and my own error. At $614.99 there is no margin of safety. The price sits 81% above that level. I don’t cut the value again for the risks already in the cases.

Where this could be in 5 and 10 years

For the base to be right at five years, ads have to grow about 9% a year to $448B in 2031, capital spending has to run near $205B a year, and operating margin has to be back to 26%. Cash is thin until then: −$26 a share to 2031. At ten years I need ads of $611B, AI products earning $100B and a margin of 36%. The year-10 value is almost all cash after 2036, so it leans on my 3% long-run growth and 10.5% rate.

The bear and bull differ on ads and capital. In the bear, ad growth drops to 3% in 2029 and Meta cuts capital to $115B a year, so margin reaches only 32%. In the bull, ads grow 21% next year, AI products earn $250B by 2036, and margin reaches 41% on capital of $285B a year by 2031.

The most likely way my base is wrong is capital. I put it near $205B a year. If the build needs 10% more, today’s base value falls from $439 to $360. If it needs 10% less, it rises to $518.

These are estimates of value, not forecasts of the share price. The market price can sit above or below value for years.

VIIThe case for the price

What most people miss

The legal charges hide how fast the core is growing. Q3 operating income fell 37% as reported. Without the $9.89B charge it rose 11%, in a quarter when depreciation rose 44%, and that is after $7.3B of stock pay C (calculated).⁵³

Meta got far smaller while growing. Headcount is down 15% in a year and revenue up 26%, so revenue per employee is up about half C (calculated), and headcount in engineering and other technical functions is down 16%.⁵³ Most coverage treats the layoffs as a cost story. To me they say the tools are working.

The depreciation shock is also partly a choice. Management said in July that its longer-term plan lays down data center and network foundations for 2028 and beyond, and leaves the server decisions for later.¹⁴ Long-lived shells are the cheaper part to be wrong about.

The bull case

Zuckerberg’s own argument is the strongest. Compute is short and the core has more uses for it than it can fill. Meta is building capacity to serve more than a billion people with Muse, and it expects Muse to pass the several hundred dollars a year American users already bring in through the apps. These are company claims, and only the supply statement comes with numbers.¹⁴¹³

Pershing Square disclosed a position of about $2B in February, 10% of capital, saying the share price ‘underappreciates’ the long-term AI upside; that is a press report of its deck.³⁴ Its 13F showed 2,660,861 shares at 31 March, and its June filing is a notice saying the holdings are reported by another filer, which I did not find.³⁵ Brad Gerstner reportedly called Muse Meta’s next 10x moment on CNBC on 24 September, and the same report says he sees the price as low against Meta’s earnings growth. That is a press paraphrase of a TV interview, so secondhand.³⁶ Altimeter’s 13F showed 1,354,990 shares at 30 June, 7.8% of the filing, down from 1,953,373 at 31 March.³⁷ Gavin Baker’s Atreides went from none at 31 March to 402,770 shares and calls on 1.2M at 30 June.³⁸ I found no public view from him on Muse, so the filing is all I can report. ARK held 238,973 shares at 30 June, 0.9% of its filing, and Tiger Global 2,823,180.³⁹⁴⁰ Those filings are months old and predate the raise.

Against reality

If ads run two points above my base every year and capital spending comes in 10% lower, the price is possible. That path is worth $603 E (estimate). It needs ads of $298B in 2027, up 19%, and capital of about $167B against my $185B.

Today, ads are ahead of that: 26% in Q3 and a Q4 guide of about 24% in total revenue C (calculated).⁵ Capital is not. The 2026 guide of $135B to $145B implies about $54B in Q4 alone C (calculated), or more than $215B a year.⁵³ Obligations have risen every quarter this year. And the company has just raised equity, which isn’t what a firm does if it plans to spend less. So one half of the case is on track and the other half is moving the wrong way.

What a 10x would take. Ten times $614.99 is about $6,150, or $16.2T of equity on today’s share count and more after any new issues. By 2036 that is about 26% a year C (calculated). At 25 times earnings it needs about $650B of net income. At a 40% operating margin and 17% tax that is about $1.95T of revenue, almost twice my bull case’s 2036 figure and about 2.4 times Amazon’s annualized sales C (calculated).¹⁵ Today’s largest companies are about $5T: Apple has gone about five times from $1T since 2018, and NVIDIA about five times since 2023 C (calculated).⁴¹⁴² I found none that has gone ten times from $1T. Meta’s start is $1.6T, and the $2,788 options are a bet on part of this path. It is close to impossible in ten years.

Where I come down

The ad numbers and the headcount moved me: the core is better than a 20% reported margin suggests. They did not move my $255, $439 and $688, because the commitments kept rising and the equity raise tells me the cash is short. I would raise my values if the January guide puts 2027 capital near $167B and ads stay above 19%. At ads two points higher alone, my base would be $515 E (estimate).

VIIIRisks, and what would change my view

1. Capacity. $810.2B of obligations, $41.5B of residual value guarantees and no large outside customer for spare compute.³
2. Litigation. In New Mexico a jury found more than 43 million violations of the state’s Unfair Practices Act, and on 21 October the court set a $25B penalty at 15% post-judgment interest, about $3.75B a year C (calculated). Meta will appeal and may have to post a bond; the amount accrued is undisclosed. Earlier New Mexico rulings set a $375M jury penalty and a $567M abatement award, and the court has stayed the judgment while Meta moves to vacate it. Florida, New Mexico and the US Virgin Islands stayed out of the attorneys general settlement. A consolidated bellwether trial was set for 28 October, then more in 2027, and counsel for over 200,000 claimants have sent mass-arbitration demands.³
3. Regulation. The EU’s Digital Markets Act and Digital Services Act, the WhatsApp AI interim measure, the FTC appeal and the FTC’s attempt to change its 2020 order.³
4. Agents. A rival assistant, or Muse failing to earn its compute.
5. Funding. Up to $60B more equity, at prices no one knows, in non-voting stock. New holders get no vote, and buybacks are now linked across both classes.¹⁰²⁰
6. Advertising cycle and concentration. Substantially all revenue is ads, and the 10-Q names a Middle East conflict and trade policy as pressures. A small number of resellers serve China-based advertisers, and that revenue is meaningful by the 10-K’s own word.³²
7. Platform dependence. Apple and Google control the phones.³
8. Key person. One person controls the votes. The agreement rewards him for staying eight years, and I found no named successor.²⁰
9. Tax. $15.9B of asserted tax for 2017 to 2019, and a rate lifted by non-deductible charges.³
10. AI copyright suits, with statutory damages per work and trials set from May 2027.³
11. Regulators abroad. Meta had to unwind Manus in August 2026 on a Chinese order.³

What would raise my base value: a 2027 capital guide near $167B, ads above 19%, or real revenue from Muse and enterprise. What would lower it: capital above $205B, more non-cancelable commitments, a New Mexico outcome that changes the product beyond one state, or equity sold under my base. The 10-Q says plainly that the company expects to seek more capital from time to time.³