Meta Stock Analysis: A 33.5% Crash, Negative Cash Flow and a $140 Billion Bet on Personal Superintelligence

Meta is not a business in trouble. Its revenue grew 28.8% in the first nine months of 2026, 3.64 billion people use one of its apps every day, and its advertising grows faster than Google's or Amazon's. The stock is another matter. On 29 October 2026 Meta closed at $624.52, 20.9% below its all-time-high close of $790.00, after falling as far as $525.72 in March. What the market punishes is not the business but what Meta does with its cash: it is pouring it into data centres, chips and power, first at a pace that kept raising its spending guidance and then, in the third quarter of 2026, at a pace that left nothing over.

This article follows the argument of the video: the crash and where the valuation stands, the spending that caused it, why the market fears a repeat of the broadband bust, how Meta earns the money it spends, its health, its acquisitions, the metaverse bet, the shift from "Meta" to AI, whether the strategy makes sense, and finally why Zuckerberg tends to get the direction right and the product wrong. Market figures are anchored to the 29 October 2026 close, except Amazon's reaction to its report that evening, which is a premarket quote from 30 October; financial figures run to the third quarter of 2026 unless another date is given.

The Crash

The fall came in two legs. Meta's highest close was $790.00 on 12 August 2025. By 27 March 2026 it had fallen 33.5% to $525.72, while free cash flow was still positive: $14.1 billion in Q4 2025 and $12.4 billion in Q1 2026. What drove that first leg was the spending outlook, raised at almost every report: the stock fell 11.3% the day after the Q3 2025 report on 30 October 2025, 8.5% on 30 April 2026 and 8.0% on 30 July 2026. In between it recovered, to $777.59 on 24 September 2026. The second leg came with the cash: on 22 October, the day after the third-quarter report showed negative free cash flow, it fell 7.0%, from $697.52 to $648.85, and on 29 October it closed at $624.52, 19.7% below the September high. All three 2026 earnings reports were followed by a fall of 7% to 9%, although revenue beat guidance every time.^[1][3]

Big falls are nothing new for Meta. This is its eighth drop of 20% or more since the 2012 IPO. The worst came in 2021-2022, when the stock lost 76.7%, from $382.18 in September 2021 to $88.91 on 3 November 2022, a period that brought Apple's tracking change, falling ad prices and the first wave of metaverse and AI spending. It took 442 days from that low to reach a new high. Meta is also more volatile than its peers: since 2021 its annualised volatility was 43.9%, against 31.1% for Alphabet and 27.1% for Microsoft, and it spent only 44.9% of the last ten years within 10% of its high, against 71.3% for Microsoft.^[1]

What the valuation says now

At $624.52 Meta trades at 21.4 times its trailing earnings per share of $29.25, which is trailing net income over the latest diluted share count; on the sum of the last four reported EPS figures, $29.33, it is 21.3. The P/E was between 20 and 35 on about 77% of trading days since 2017, and it was higher than today on about 86% of them. Seven of the nine year-ends since 2017 fell in that band; the exceptions were crisis points: 19.6 at the end of 2018 and 11.2 at the end of 2022. The all-time low was 8.3 on 3 November 2022 (8.5 on reported EPS). Trailing earnings are distorted in both directions: a $15.93 billion one-time tax charge in Q3 2025, an $8.03 billion tax benefit in Q1 2026 and the $9.89 billion legal charge in Q3 2026. Analysts expect 2027 earnings per share of $33.96, which puts the stock on 18.4 times forward earnings, and revenue of $311.5 billion, up 21.4%. On cash the picture is far less cheap: $25.1 billion of free cash flow over the last four quarters is a yield of only 1.6%.^[2]

Cause: Capex Galore

The trigger of the latest leg is cash. In the third quarter of 2026 Meta's free cash flow was minus $2.1 billion, the first negative quarter in the 2021-2026 data. Meta now spends more on capex than its business brings in: capex, including principal payments on finance leases, was $34.9 billion, against operating cash flow of $32.8 billion. A year earlier free cash flow was $10.6 billion, so it fell by 120%. Over the first nine months of 2026 free cash flow fell 62.6%, from $29.5 billion to $11.1 billion.^[2][3]

The reason is simple arithmetic. Capex grew 80.2% in a year, from $19.4 billion in Q3 2025 to $34.9 billion in Q3 2026. Operating cash flow grew only 9.3%, from $30.0 billion to $32.8 billion. In 2021 capex took 33.4% of operating cash flow; in the first nine months of 2026 it took 88.6%.^[2]

And it is getting bigger. Meta guides 2026 capex to $135-145 billion, up 87% to 101% from $72.2 billion in 2025 and equal to 52% to 57% of guided 2026 revenue. In 2021 capex was $19.2 billion, 16.3% of revenue. With $85.8 billion spent by September, the fourth quarter alone needs $49.2-59.2 billion to reach the guidance, against $32.8 billion of operating cash flow in the third quarter. Unless cash flow jumps, the fourth quarter will be deeper in the red. Capex comes on top of running costs: total 2026 expenses are guided to $175-179 billion, and in the first nine months capex plus expenses came to about $212.9 billion against $181.7 billion of revenue.^[3]

What the money buys

Meta says its capex goes to "servers, data centers, and network infrastructure". It does not split the amount, but its balance sheet shows where the assets land: of the $69.1 billion rise in gross property and equipment in 2025, 43% was servers and network equipment, 34% construction in progress, which Meta describes as data centres, network infrastructure and servers, and 12% buildings. In the first nine months of 2026 the mix moved further towards construction, at 42% of a $92.8 billion increase. These are balance changes rather than a cash breakdown, because finished construction moves into servers and buildings.^[2][3]

The spending on the balance sheet is only part of it. Meta's non-cancelable commitments, mostly third-party cloud capacity, rose 13.8-fold, from $32.8 billion at the end of 2024 to $454.5 billion in September 2026. On top come $355.7 billion of leases that are signed but have not started yet, covering data centres, energy infrastructure, colocation and network infrastructure from late 2026 to 2036. On 14 October 2026 Guggenheim raised its 2027 Meta capex estimate to $225 billion, and Zuckerberg himself described the size of the AI check as "$100 billion, $200 billion", written before he had seen the models it would produce. The October podcast gives the unit cost: about $15 billion a year to operate a gigawatt of compute and $50-60 billion to build one. Meta also rents from neoclouds, with deals of $21 billion with CoreWeave and $12 billion with Nebius, and it holds 20% of data centre ventures in Louisiana and Texas that it leases back, with a maximum exposure to loss of $68.4 billion, including $41.5 billion of residual value guarantees. Those ventures are not consolidated on its balance sheet.^[2][9][19]

The other bill: lawyers

Legal costs have moved into the tens of billions at the same time. In August 2026 Meta reached an agreement with 52 attorneys general over youth harms worth up to about $18.59 billion, of which about $12.7 billion is payable over ten years and about $5.3 billion is due only if peers such as YouTube and TikTok join. It took a $9.89 billion charge for it in the third quarter, which cut the quarter's operating margin to 20.1% from 35.4% without it. On 21 October 2026 a New Mexico court ordered a separate $25 billion civil penalty in a privacy case, with 15% interest a year, about $3.75 billion, while it remains unpaid. Meta plans to appeal and has not disclosed how much it has accrued for it. Legal accruals rose from $6.87 billion at the end of 2025 to $11.58 billion on 30 September 2026. More is pending: personal injury trials, where the first verdict was $6 million with 70% assigned to Meta, a claim of up to $1.7 billion from Washington DC, and EU fines that already include €1.2 billion over data transfers, €798 million over Marketplace and €200 million under the Digital Markets Act.^[2][8]

The Market Fear: Remember the Broadband Build-Out and the Dot-Com Crash?

Meta's fall is part of a wider nervousness about the AI build-out. On 14 October 2026 CoreWeave, one of the "neoclouds" that rent out AI compute, fell 15.5% as yields on its bonds rose to about 15%; Nebius fell about 10% and Iren about 12%. Meta fell 3.3% the same day. On 25 October Elon Musk wrote that debt markets have "almost entirely shut off" for companies without substantial revenue outside AI data centres, and that most of the 200-plus neoclouds face "death and consolidation". Zuckerberg rejects the word: "I don't even know if I'd call it a bubble because that implies that it's overvalued".^[6][7]

Negative cash flow is what gets punished after earnings

Meta's peers show the same pattern in the same quarter. Amazon's free cash flow was about minus $33.5 billion in Q3 2026, with net capex of $75.9 billion against operating cash flow of $42.4 billion, and minus $42.4 billion over twelve months. Capex of $77.3 billion before asset sales was far above the $58.7 billion analysts expected, Amazon also narrowly missed revenue and earnings expectations and gave a cautious sales outlook, and Andy Jassy said the company expects to "spend meaningfully more in 2027". The stock traded 6.9% lower before the market opened on 30 October, at $227, 20.1% below its high of $284.02 from 3 August. Alphabet's free cash flow was minus $9.1 billion, where analysts expected minus $2.7 billion, with capex of $51.4 billion against operating cash flow of $42.3 billion. It also missed revenue and earnings expectations and said capex would rise significantly in 2027. On 29 October its stock fell 12.6%, its worst day ever, which erased about $500 billion of value, and closed at $283.82, 29.5% below its high of $402.62. Slowing ads added to the pressure: Google's ad revenue grew 10.6% in the third quarter, from $74.2 billion to $82.0 billion, against 25.9% at Meta.^[4][1]

Microsoft did not crash after its last earnings, despite heavy investment, because its cash flow stayed positive: $18.6 billion of free cash flow in the same quarter, with capex of $39.4 billion against operating cash flow of $58.0 billion. Its stock rose as much as 4% after hours, then closed 0.2% lower at $530.14 on 29 October, 2.4% below its closing high of $543.45 from 13 October. Meta, with minus $2.1 billion, had fallen 7.0% after its own report. So the three companies with negative free cash flow fell after earnings, and the one that stayed positive did not. It is not about cloud growth. In the July to September quarter AWS grew 39% to $46.0 billion and Azure 48% to $33.2 billion, while Google Cloud roughly doubled, up about 103% to $30.7 billion.^[4][5]

Why the dot-com comparison sticks

Investors remember the broadband build-out of the late 1990s: enormous hype, fibre laid far ahead of demand, then a market implosion. AI is different in one important way. The compute being built is used, and companies cannot get enough of it: Amazon and Microsoft both report capacity shortages. Meta itself has kept its Muse agent to the US and Canada, partly to serve it efficiently and partly to tailor it to each market.^[3]

The risk is the debt that pays for it. A miscalculation about future demand can hurt even established companies. Meta issued $30 billion of bonds in November 2025 and $25 billion in May 2026, the latter at coupons of 4.55% to 6.45%, and on 30 September 2026 its $83.7 billion of debt slightly exceeded its $82.4 billion of cash and securities. An independent board committee warned that heavy borrowing could hurt its credit rating. Meta has also created a new class of non-voting Class C shares, distributed one for one to existing holders and trading from January 2027, which the CFO calls effectively a two-for-one stock split; the prices in this article are before that split. The distribution itself raises no cash, but it lets Meta sell shares later without diluting Zuckerberg's voting control, which stood at 60.6% in October 2026. Meta tried a similar plan in 2016 and withdrew it in 2017 after a shareholder suit; Google did it in 2014. The capital return has already stopped: Meta spent $148.6 billion on buybacks from 2021 to 2025, cutting its share count by 7.1%, and has bought back nothing in 2026, while about $20 billion a year of stock pay pushed the count up 1.1%.^[2][3]

If AI demand pulls back or a downturn hits ad budgets, all that debt and all those commitments become far heavier. That is the market fear.

Funding the Madness: How Meta Makes Money (the Business Units)

Everything Meta builds is paid for by one business. In 2025 Meta's revenue was $201.0 billion, and 97.6% of it, $196.2 billion, came from advertising. Other Family of Apps revenue, mostly WhatsApp paid business messaging, Meta Verified and subscriptions, brought in $2.6 billion (1.3%), and Reality Labs, the Quest headsets and glasses, $2.2 billion (1.1%).^[2]

The two reported segments could not be more different. Family of Apps, which covers Facebook, Instagram, Messenger and WhatsApp, made $102.5 billion of operating profit in 2025, a 51.6% margin. Reality Labs lost $19.2 billion, about $8.70 for every dollar of revenue it brought in.^[2]

Ad revenue is the number of ads shown times the price per ad. In the third quarter of 2026 impressions grew 10% and the average price per ad 14%, a price increase Meta credits to better AI ranking and recommendations. Reach is enormous: 3.64 billion people used at least one Meta app every day in September 2026, 3% more than a year earlier. Growth in users has slowed; growth in what each user is worth has not. Threads has passed 500 million monthly users and, according to Zuckerberg, overtaken X. Nearly 10 million advertisers used Meta's generative AI creative tools in the third quarter, partnership ads with creators grew 97%, and Meta's own marketing and sales cost fell from 13.1% of revenue in 2022 to 6.0% in 2025. Revenue from Asia-Pacific and the rest of the world doubled from 2021 to 2025, although in Q3 2026 the US and Canada grew fastest, at 33.0%.^[3]

Meta is the world's second-largest ad business, with $196.2 billion in 2025 against Google's $294.7 billion and Amazon's $68.6 billion. It is also the fastest-growing of the three: in the first half of 2026 its ad revenue grew 30.1%, against 25.1% at Amazon and 14.9% at Google. It does not lead everywhere: teenagers still prefer TikTok (48% against 31% for Instagram in Piper Sandler's survey, via a secondary source), and YouTube pays creators far more, over $100 billion since 2021 by its own count, against nearly $3 billion from Facebook in 2025. That is the fight behind Instagram's new long-form video test and its TV app.^[4]

Business Metrics / Health

The core business is in excellent shape. Revenue rose 70.4% from $117.9 billion in 2021 to $201.0 billion in 2025, faster than Microsoft (67.6%, fiscal years to June), Pinterest (63.8%), Alphabet (56.4%) and Snap (44.1%) over the same years. In the first nine months of 2026 revenue grew another 28.8%, from $141.1 billion to $181.7 billion, with growth of 33.1%, 28.0% and 26.1% in the three quarters. Meta has never reported revenue below its own guided range in 19 quarters, and it guides fourth-quarter revenue to $73-76 billion, up 21.9% to 26.9%.^[2][3]

Margins are where the spending and the lawyers show up. The operating margin was 41.4% in 2025, below the 2024 peak of 42.2% for a full year and 48.3% in Q4 2024. It fell to 30.1% in the first nine months of 2026, hit by legal charges and severance: without them the margin would have been 36.8% in the second quarter and 35.4% in the third. Depreciation, the delayed cost of all that capex, rose 43.6% in the same nine months and will keep rising as new data centres come online.^[2]

Meta has been here before. In 2022 revenue fell 1.1%, the price per ad fell 16% and the operating margin dropped to 24.8%. Part of the price fall came from Reels, which earned "much less than feed" until it began adding to net revenue by the end of 2023. Zuckerberg answered with the "Year of Efficiency": in 2023 revenue grew 15.7% while costs grew only 0.6%. Headcount fell 22.8%, from 86,482 at the end of 2022 (the quarterly peak was 87,314 in September 2022) to 66,775 in September 2026. The major cuts were more than 11,000 jobs in November 2022, about 10,000 in 2023, about 3,600 performance-based cuts in early 2025, about 600 roles in the older AI groups in October 2025, about 10% of Reality Labs in January 2026 and about 8,000 in May 2026.^[2][15]

Acquisitions and Divestments: What's the Strategy?

Today's Meta was built by three early mega-deals: Instagram in 2012 for about $1 billion, and WhatsApp and Oculus in 2014 for about $19 billion and $2 billion, as widely reported. The FTC later tried to force Meta to sell Instagram and WhatsApp. Meta won in November 2025, and the FTC appealed in January 2026.^[8]

After that, big takeovers stopped. In 2024 Meta spent just $270 million of cash on acquisitions and intangible assets. In 2025 it switched to stakes and hiring deals instead. It paid about $14.3 billion for 49% of Scale AI, which brought in founder Alexandr Wang as Chief AI Officer. It offered to buy up to 49% of the holdings of NFDG, the fund of Nat Friedman and Daniel Gross, as both joined Meta. And it bought smaller AI companies: PlayAI and WaveForms for voice AI, and Limitless, maker of an AI wearable. Cash spent on acquisitions rose to $4.23 billion in 2025, 15.7 times the 2024 figure, plus $450 million paid in stock.^[2][12]

Not every approach worked. Meta's 2025 talks with Thinking Machines, Perplexity, Safe Superintelligence and Runway did not lead to deals, and informal 2025 talks with Kalshi, the prediction market, went nowhere, though the two signed a Threads integration deal in March 2026. In September 2026 Meta bought Stilla.ai, a Swedish start-up building AI agents for business messaging. Its $2 billion-plus purchase of Manus, an AI agent company announced in December 2025, was ordered unwound by China in April 2026. No reports show talks to buy a coding company, a video or TV company or a data centre software firm; Cursor, the most prominent coding tool, went to SpaceX's AI unit in a deal reported at $60 billion.^[12]

The pattern is clear: practically every deal since 2025 is about AI. The divestments and shutdowns are the misses of earlier strategies, and only one of them, the forced Manus unwind, involves AI. Kustomer, a customer-service software company bought for about $1 billion, was spun out in 2023 at a valuation of about $250 million, with Meta keeping a stake. Giphy was sold on a UK competition order. Portal, Workplace, Horizon Workrooms, the Spark AR filter platform, the Novi wallet and the Diem currency were shut down or sold.^[12]

Betting the House on Mark's Gut

In October 2021 Facebook renamed itself Meta, a bet on Zuckerberg's conviction that the metaverse would be the next computing platform. The rebrand and the metaverse focus drew ridicule, and five years later the numbers explain why. Reality Labs has lost $89.7 billion from 2021 to September 2026, against $12.0 billion of revenue. Its yearly loss grew 88.3%, from $10.2 billion in 2021 to $19.2 billion in 2025, while its revenue went down 2.9%, from $2.27 billion to $2.21 billion. Meta guides to $18-19 billion of Reality Labs losses in 2026 and about $15 billion in 2027. Family of Apps earns the money; Reality Labs burns it.^[2][3]

VR shrinks, glasses grow

Oculus VR, bought in 2014, is shrinking. Omdia counts a 38.7% fall in shipments of XR headwear, VR headsets and XR glasses together, in the first half of 2026, and Meta's share fell from 71.3% to 48.2%, down 23.1 points. Meta now says it will ship new VR devices only for "major breakthroughs". The bright spot is AI glasses without displays, made with EssilorLuxottica under the Ray-Ban and Oakley brands. IDC counted 167% growth in glasses without displays in the first quarter of 2026, against 86% for display eyewear, and expects about 13.6 million units this year. Counterpoint put Meta's share of display-less glasses at about 84% in the first quarter and 94% in the first half.^[13]

Even the glasses are a long way from paying for the division. In our illustrative model, with IDC's 27.3 million smart glasses in 2030, a 70% Meta share and a three-year life, about 57 million Meta glasses would be in use. Covering Reality Labs' costs at $100-150 of profit per user a year would take about 113-170 million active users. These are our assumptions, not Meta's. Robotics, which Bloomberg reported as a Reality Labs team in February 2025, appears in no Meta call or filing from 2025 or 2026.

Why keep paying?

All of this burns money with little obvious upside, subsidising devices that sell in the millions while Meta's apps reach billions. The reason is platform ownership. Apple's App Tracking Transparency change cost Meta an estimated $10 billion of revenue in 2022, according to its then CFO. Whoever owns the platform makes the rules, and Meta lost that fight on phones, tablets, PCs and every other form factor. Zuckerberg has framed AR and VR as "the next computing platform" since 2021, and in October 2026 he put it this way.^[9]

Meta: Drop the "Meta", Just "AI", It's Cleaner

Meta is slowly shedding its metaverse identity. Reality Labs is now pitched around AI glasses while Quest sales fall, and in his October 2026 interview Zuckerberg did not mention Quest or VR headsets once. Personal superintelligence has become Meta's new identity, replacing the metaverse.^[9]

The turn came after Llama 4, released in April 2025, disappointed. Zuckerberg later admitted "we were off the trajectory that we needed to be on". What followed was one of the most expensive hiring sprees in tech: Meta Superintelligence Labs was built around Wang and the Scale AI stake, and Meta poached researchers from OpenAI, Google, Apple and Thinking Machines with reported packages of $200 million to $1.5 billion, all spread over several years. Bank of America estimates the top 50 hires cost about $1 billion a year, about $20 million per person. Two of the three largest reported packages walked out within a year: Ruoming Pang left for OpenAI after seven months and Andrew Tulloch after eleven, reportedly for Anthropic.^[14]

The research core is small. In the October 2026 podcast with Zuckerberg and Wang, the Superintelligence Labs research team was described as "less than 200 people", "closer to 100", next to a larger infrastructure staff; the whole lab had just under 3,000 people after the October 2025 cuts. About 600 roles in the older AI groups were cut in October 2025, and Yann LeCun, Meta's chief AI scientist since 2013, left at the end of 2025 to found his own lab.^[9][14]

Muse: back at the frontier

The Muse model family is meant to bring Meta back to the frontier. Muse Spark, the first model from the new lab, launched in April 2026 and now powers Meta AI across the apps. Muse Wonder, Meta's first frontier model and the much bigger pre-train code-named Watermelon, followed in October 2026, and Meta says it is already working on larger models. Beside it came open-weights models called Muse Glimmer and Muse Spark Open. On version 4.3 of the Artificial Analysis Intelligence Index, in use at launch, Wonder Full scored 56, two points behind Claude Opus 5.5 at 58 and level with Claude Sonnet 5.5; version 5 replaced that scale on 27 October. The flagship is closed: the open-weights strategy of the Llama years has flipped.^[3][16]

The Muse agent is the consumer product. More than 10 million people in the US and Canada set it up in six weeks, and about 25% of them use it daily. It is not available elsewhere yet, partly for efficiency and partly because Meta wants to tailor it to each market. Muse reaches Meta's glasses from November, and a pocket device, Muse Charm, is about to ship. Meta positions privacy as its edge, with a confidential cloud computer that Zuckerberg says even Meta cannot see into.^[3]

The build-out

Meta is building data centres and buying energy and chips at a scale that matches its ambition. Prometheus in Ohio is above 1 gigawatt and is training the next models; Hyperion in Louisiana is planned for up to 5 gigawatts. In January 2026 Meta signed nuclear agreements for up to 6.6 gigawatts by 2035, on top of a 1.1-gigawatt Constellation agreement from 2025. Its chip deals are sized by the power the chips will draw: up to 6 gigawatts of AMD GPUs, more than 1 gigawatt of Broadcom-built MTIA chips, Meta's own design, plus a deal with NVIDIA. Most of the nuclear power arrives in the 2030s.^[17][3]

Meta is punching above its weight. It is the smallest of the four by market value and invests the most relative to it: its 2026 capex guidance equals 8.8% of its market value, against 8.4% at Amazon, 5.8% at Alphabet and about 4.4% at Microsoft. In dollars it spends the least of the four: Alphabet guides $195-205 billion, Amazon about $220 billion and Microsoft about $175 billion. Unlike all three, Meta has no cloud business to resell spare capacity. On owned compute it holds about 1.51 million H100 equivalents per $1 trillion of market value, second only to Google's 1.58 million since Alphabet's fall, and well ahead of Amazon (0.96 million) and Microsoft (0.91 million).^[4][18]

This also maps to the reactions after the last earnings: the more you invest, the more you crash. Meta, at 8.8%, fell 7.0%, Amazon, at 8.4%, traded 6.9% lower before the open, and Microsoft, at 4.4%, slipped 0.2%. Alphabet, at 5.8%, is the exception, falling 12.6% on a free cash flow miss and weaker results. Which leaves the question the market keeps asking: why all of this investment, and what is the product?

The Strategy: Does It Make Sense?

Meta's stated goal is personal superintelligence: an AI agent for each of its billions of users. Rather than becoming a neocloud or mainly an API provider, Meta wants to be the superintelligence provider for the people already on its apps. Enterprise sales exist but are still emerging: a model API and Muse Code, a coding agent, sold through a new Meta Enterprise Platform. Zuckerberg says each is "growing quite quickly", but Meta discloses no separate revenue. Non-ad, non-hardware revenue was $2.99 billion in the first nine months of 2026, 1.6% of the total. The plan also includes business agents for the at least 500 million small businesses on its apps, and in time agents that trade with each other. The October podcast frames all of it as a race between "a small handful of companies" able to build frontier superintelligence: "is it even really a choice", and "we're going to burn the boats".^[3]

The agent as the point of sale

Meta is building an economy around agents, where the agent is the point of sale. Shopping is already Muse's top use: more than 25% of weekly users shop through it, with an average purchase above $200. The free tier gives each user 100 million tokens a week and a personal cloud computer. Meta plans to earn mainly from businesses, plus subscriptions for advanced models and heavier use; the CFO says monetisation starts with subscriptions, and Meta One already has more than 15 million subscriptions and trials.^[3][10]

Meta expects revenue per Muse user to exceed the "several hundreds of dollars annually" that those same users already bring Meta through its apps today. No revenue target has been given.^[3]

Why not sell compute or coding?

The obvious alternatives are the businesses that are already cash cows: renting out compute, hosting data centres and coding agents. Anthropic's revenue run-rate was reported at $65 billion and OpenAI's at about $50 billion in 2026, though both rent most of their compute, with reported commitments of at least $518 billion and $750 billion, and Anthropic's draft filing shows a first-half net loss of $80.7 billion. Meta has the option: Zuckerberg said in July 2026 that Meta was "getting a lot of offers for compute at a significant premium over what we paid for it", as part of what Meta itself calls its "Meta Compute effort". Bloomberg reported that month that Meta plans to sell capacity, with Bloomberg Intelligence estimating more than $50 billion of revenue by 2028, and the New York Times reported talks with Anthropic worth up to $10 billion over two years, but no customer is confirmed. On coding, Zuckerberg says Meta is "very focused on coding" as a model capability, while the October podcast expects coders to be "a minority of all of our users".^[3][10][9]

What the price assumes

The good news for shareholders is that the price does not demand much. Our reverse valuation asks what revenue growth today's $1.62 trillion enterprise value implies. If Meta's free cash flow margin returns to 30%, close to its 2021, 2023 and 2024 levels, the price needs only 6.5% revenue growth a year for ten years, at a 9% discount rate. That is far below the 14.3% a year of 2021-2025 and the 21.4% analysts expect for 2027. With no growth at all, normalised cash flow would be worth $712-854 billion, so roughly half of today's value is paid for growth. The catch is the margin: at the roughly 10% free cash flow margin of the last four quarters, the price would need very high growth. Everything depends on capex falling back as a share of revenue.^[11]

So is Meta running towards a cliff? Not on these numbers, as long as the ad business keeps growing and the build-out eventually slows. The danger lies in the commitments that cannot be cancelled if it does not.

Directionally Right

Meta has a habit of pivoting to the next big opportunity, so what Zuckerberg says today will not fully show where Meta goes. Our view is that he is very good at knowing the direction and bad at knowing the destination. He is bad at guessing the product: the metaverse, personal superintelligence, even Reels, which began as a catch-up copy of TikTok before it became a success. He is great at investing intuitively in the right direction, which lets Meta pivot upwards when the product fails: GPUs and AI, recommendation algorithms, data collection.

Facebook is his one great founding idea. In 2006 he refused to sell it for $1 billion, and his reason was not a lack of ideas.

In our reading, Meta has since bought or copied more than it has invented. It bought Instagram, WhatsApp and Oculus. It built Reels while "catching up to what TikTok was doing". Zuckerberg himself says "most of the times where we make some decision that ends up seeming good is because we messed something up before". The pattern repeats with Reels, the metaverse, VR and AR, open models and now agents. The best example is the GPUs: Meta ordered them because Reels was short of capacity against TikTok, "but let's also double that", and they became the base of its AI effort.^[11]

Watch what Meta buys, not what it says

The most telling signal is what Meta buys, because capex shows what Meta can do, not what it says it will do. It is buying data centres, huge amounts of energy and chips: $454.5 billion of commitments, up to 6.6 gigawatts of nuclear power from the January deals and about 2.41 million H100 equivalents of compute already owned. Our thesis is that this makes Meta an intelligence provider one way or another, at a scale far above its weight, and if it works, with returns far above its weight. Whether that will be personal superintelligence paid for by commissions on purchases is another question. We doubt it.

Zuckerberg is often seen as very logical, even robotic. His own words suggest someone highly intuitive, driven less by a plan than by a compulsion to build.

Outlook

In the long run, history is on Meta's side. Seven of its eight falls of 20% or more since 2012 ended at a new high; only the current one has not yet. Since the end of 2020 the stock is up 128.6%, against 105.5% for the S&P 500, even after this year's fall. The ride has been violent, from a 64.2% fall in 2022 to a 194.1% gain in 2023. The analysts' mean target of $850.51 is 36.2% above the 29 October close. Our view is that the long-term upside is real, because Meta has always come back and is buying the capacity to matter in AI whatever the final product turns out to be. That is our investment thesis, not a certainty.^[1]

In the short run the downside is concrete. Fourth-quarter capex needs to reach $49.2-59.2 billion, so free cash flow will go deeper negative unless operating cash flow jumps. The next catalysts are the fourth-quarter report with the first 2027 capex outlook, the start of Class C trading in January 2027, the youth litigation trials that began on 28 October 2026 and the appeal of the $25 billion New Mexico penalty.^[3][8]