Meta Platforms (META.US) delivered a September-quarter report after the close on Wednesday that ramped up spending on both the revenue and cost sides of the ledger.
For the quarter ended September 30, revenue reached $64.60 billion, up 26% year over year and ahead of the $63.37 billion consensus compiled by LSEG, while fourth-quarter revenue guidance of $73 billion to $76 billion also topped expectations at the midpoint.
In the same filing, however, the company raised the lower bound of its 2026 capital expenditure guidance to $135 billion from $130 billion, keeping the upper end at $145 billion, and announced the creation of a new class of non-voting shares, opening a fresh financing channel for Zuckerberg's AI arms race in a quarter when free cash flow already turned negative.
Revenue side: advertising remains the money printer, Muse AI grows strongly
Advertising is still the main source of cash.
Third-quarter ad impressions rose 10% year over year and average ad prices climbed 14%, while family daily active people (DAP) reached 3.64 billion, up 3% year over year; the midpoint of fourth-quarter guidance at $74.5 billion implies roughly 24% year-over-year growth.
Reuters and Bloomberg both noted that the vast majority of Meta's revenue still comes from advertising, with AI's main role so far being to optimize ad targeting and recommendations and lift user engagement.
The real incremental story comes from Muse.
Zuckerberg said on the call that more than 10 million users in the United States and Canada have opened Muse accounts within a month and a half of launch and that momentum remains strong; he expects Muse to eventually serve "billions of people," users who "potentially can generate more revenue than the average Meta user."
For now Muse remains free for most people, with Meta earning money through transaction fees paid by enterprises, subscriptions from heavy users, paid users who want access to the most powerful models, and paid tools for businesses; Zuckerberg said that "if compute efficiency continues to improve rapidly over the next few years — which we think is very likely — Muse will remain highly profitable for many years."
Yet the payoff timeline for AI has been pushed far into the future.
CFO Susan Li said plainly in the report that "these investments will take longer to generate returns than investments in our core business."
Wells Fargo in an October 6 report projected Meta's 2027 operating expenses at roughly $212 billion, with meaningful AI revenue only arriving in 2028.
In addition, some enterprises have restricted access to Muse over security concerns.
A new enterprise-side pipeline is also being laid: on September 28 Meta established Meta Enterprise Platform and hired then-MongoDB CEO Chirantan "CJ" Desai as chief enterprise platform officer, reporting directly to Zuckerberg; MongoDB shares fell more than 17% that day, and the platform has yet to disclose pricing, availability regions or a customer list.
Cost side: lower bound of capex guidance raised
In the third quarter Meta's capital expenditures, including principal payments on finance leases, reached $34.91 billion; cumulative spending over the first nine months was about $86 billion.
On that basis, hitting the full-year range of $135 billion to $145 billion would require roughly $49 billion to $59 billion of spending in the fourth quarter.
The market had actually expected more: analyst models surveyed by FactSet had projected third-quarter capex of about $41 billion and fourth-quarter spending above $49 billion, and FactSet expected Meta's 2027 spending to reach $196 billion.
In other words, Meta raised the floor rather than the ceiling — it did not open the spigot wider, it merely lifted the "most conservative case."
Other expense lines are tightening.
The report showed headcount down 15% year over year; Reality Labs is expected to lose up to $19 billion this year and about $15 billion in 2027, with the savings "primarily used for AI."
Cash flow: from printing money to overdrafting, a one-time legal charge rewrites the income statement
The truly striking item is cash flow.
Meta's third-quarter free cash flow was negative $2.12 billion, versus positive $10.63 billion a year earlier; the figure was better than the roughly negative $9.4 billion the market had expected, and marked the first negative reading since the company went public.
At the same time, according to the report, third-quarter operating cash flow was $32.79 billion, and cash, cash equivalents and marketable securities totaled $82.45 billion as of September 30 — the company still has ample operating cash generation, with the shortfall on the capital expenditure side.
The income statement was distorted by a one-time factor: diluted earnings per share for the third quarter came in at $3.83 after including a $9.89 billion charge related to an August settlement with 52 state attorneys general; excluding that charge, EPS would have been $7.40.
The financing moves therefore follow naturally.
Meta will create a new non-voting Class C common stock, planned to take effect in January 2027, distributed as a stock dividend to existing shareholders on a 1:1 basis, effectively a two-for-one split; the new shares will be used for capital raising, employee compensation and potential acquisitions without diluting CEO Mark Zuckerberg's control, and the company said the move will ensure Zuckerberg continues to lead it "into the next era."
Class A shares will continue to trade on Nasdaq, and the Class C shares will also apply to list on Nasdaq.
External debt is likewise fuel for the machine.
Goldman Sachs data showed global AI-related debt issuance reached $489 billion by mid-2026, with the five largest hyperscalers accounting for nearly 40%; Meta issued about $30 billion of bonds earlier this year for data centers, and in July its Texas data center project sold $12.55 billion of debt at a 7.534% yield, a premium of about 287.5 basis points over the comparable 10-year U.S. Treasury.
Off-balance-sheet structures are expanding too: the Hyperion data center in Louisiana raised debt through an entity jointly owned by Meta and Blue Owl, reportedly about $27 billion, with Meta holding only a minority stake but committing to a long-term lease.
Against the backdrop of the neocloud selloff: Meta is both a buyer and increasingly a referee
This is where the quarter's report is most instructive about current "compute concerns."
Over the past two weeks the AI infrastructure sector has suffered a selloff: as of October 16, CoreWeave shares had dropped more than 15% in a single week and the yield on its 2030 bonds had jumped 500 basis points to nearly 16%, with the market fixated on its $50 billion of debt; CoreWeave sits in the middle of "circular financing arrangements" between chipmaker Nvidia and customers such as OpenAI and Meta.
Gartner expects that by 2030 more than 80% of neocloud providers will cease to operate independently, and estimates AI infrastructure spending could reach $7 trillion by 2030 while AI-related revenue will not exceed $3.6 trillion; AI-related bond issuance fell from $113 billion in June to $23 billion in September.
Meta's books offer three contrasting signals.
First, demand has not collapsed.
Muse needs enormous compute, which is one reason it has not yet launched globally — a company sitting on $82.45 billion of cash saying it does not have enough compute is itself the most direct footnote to demand; Musk also said this week on a SpaceX call that compute demand far exceeds supply.
Second, Meta is shifting from "selling compute" to "using compute itself."
While Meta launched its enterprise platform and is testing external compute output, Wells Fargo in October removed the projected 500 megawatts of 2027 compute resale — about $5 billion of high-margin revenue — from its model, precisely because of Muse's success: compute must first go to its own products.
For neoclouds this is both order support and a double squeeze from supply substitution.
Third, the divergence is not in AI demand but in the cost of capital.
Meta has $32.79 billion of third-quarter operating cash flow and $82.45 billion of cash as a cushion, making it a self-funded buyer; intermediaries like CoreWeave must roll over debt, and JPMorgan estimates their blended cost of capital at 9% to 11%.
When financing costs are repriced, the first to come under pressure are leveraged intermediaries rather than self-builders covered by cash flow — but Meta itself is also borrowing more and doing more off-balance-sheet arrangements, putting it on the same rate curve as the neoclouds, just further ahead.
Legal tail risk: a $25 billion fine starts a long race
On the same day as the earnings report, New Mexico judge Francis Mathew ordered Meta to pay the state $25 billion.
Mathew found that the amount warranted under state law exceeded $219 billion before reducing it to $25 billion; a jury in September found the company and its management made 26 misleading statements on issues including data privacy and hate speech, implying more than 43 million violations of the state's consumer protection law; state attorney Randi McGinn had sought $35 billion to $40 billion at an October 1 hearing, while Meta argued for a cap of $3.45 billion.
Meta chief legal officer C.J. Mahoney said the company is confident it will prevail on appeal and called the ruling something that "makes it nearly impossible to do business in New Mexico"; state Attorney General Raúl Torrez called it "one of the largest financial penalties against a corporation in American history."
Meta previously paid $5 billion to the U.S. Federal Trade Commission in 2019 over the Cambridge Analytica affair, and an August multistate settlement included $459.2 million from six states tied to Cambridge Analytica; earlier this year New Mexico also won $942 million across two related trials, with another trial set to open next week.
D.A. Davidson head of technology research Gil Luria believes the ruling is costly but unlikely to affect the company's valuation — "Meta has a market cap approaching $2 trillion, and it will likely keep appealing and drag the case out."
Emarketer senior analyst Minda Smiley warned that if AI agents become more mainstream, privacy and security questions will only intensify, and "given Meta's reputation in this area, it could ultimately hurt it."
Market reaction and institutional views
Data showed Meta shares had risen about 6% year to date through Wednesday's U.S. close.
The social media giant closed Wednesday at $697, with a market capitalization of $1.79 trillion and a price-to-earnings ratio of 26.52.
After the results, Meta rose 2.08% in after-hours trading to $712 as of publication.
Goldman Sachs reiterated a "buy" rating on Meta and maintained an $860 price target.
Goldman said it expects investors to react positively to Meta's September-quarter report.
The report showed quarterly revenue above the upper end of the company's prior guidance range, with ad revenue up 26% year over year.
Goldman noted that Meta management expressed optimism on the earnings call about scaling AI and its positive impact on the core business, while flagging continued higher investment ahead without giving specific quantified figures.
Meanwhile, Raymond James reiterated a "strong buy" rating on Meta Platforms with an $860 price target, stressing that the company's upcoming report is drawing intense market attention.
Citizens maintained an "outperform" rating with an $885 price target, citing smooth AI progress and strong advertising performance.
On the other hand, BMO Capital reiterated a "market perform" rating with a $580 price target, acknowledging improved AI product momentum but remaining cautious on the scale and timing of AI monetization.
The sell-side lineup was already crowded: BofA's Justin Post has a buy rating and $810 target, Deutsche Bank's Benjamin Black a buy and $820, Bernstein's Mark Shmulik a buy and $800, and Wells Fargo in October raised its target to $1,000 from $796; the average price target of 50 analysts compiled by Zacks is $824.56.
One line from senior analyst Jesse Cohen roughly captures the mood shift in this report: "The market has moved from punishing Meta for spending to rewarding it for having a compelling product story."
Comments