Meta Q3 earnings: advertising props up 97% of revenue, Muse still burning money

By AIX Finance, Author | Wang Hanxing, Editor | Wei Jia

Meta in Q3 may be the biggest turnaround among the major tech companies. Three months ago, it was still a "loser" that only drew AI "pie in the sky," with high spending and no output, and had fallen behind across the board in the large-model race. Now it has transformed into a pioneer of Personal Agents and returned to the center of the AI competition stage.

The variable lies in Muse, the personal agent launched in September, which surpassed 730,000 downloads in just 5 days and 10 million downloads within 40 days of launch.

In Meta's just-released Q3 earnings report, Muse's impact is not yet reflected in the data, but even without Muse's boost, this report is still impressive. Its overall revenue and advertising business performance both exceeded market expectations, with Q3 revenue of $64.6 billion, up 26% year over year; Q4 revenue guidance is $73 billion to $76 billion, both above market expectations.

But the impact of Muse's upfront investment has begun to show. During the period, Meta's free cash flow turned negative at $2.12 billion, mainly due to high capital expenditures.

In the earnings report, Meta management said it will continue to raise the full-year Capex guidance, with the lower bound raised from $130 billion to $135 billion and the upper bound maintained at $145 billion. In the first three quarters of 2026, Meta has raised capital expenditure expectations 3 times: overall guidance was raised by $10 billion in Q1, and the lower bound of guidance was raised by $5 billion in Q2 and Q3 respectively.

A brokerage analyst who has long followed US stocks told "AIX Finance" that raising only the lower bound of guidance is a common phenomenon among current tech companies and companies related to the AI industry chain. This reflects higher costs, mainly rising storage costs, while an unchanged upper bound can basically be understood as unchanged expectations.

However, the market has not turned bearish on Meta because of its negative free cash flow, and Muse's commercialization prospects remain favored.

After the earnings release, Meta's US shares rose more than 6% in after-hours trading at one point, and since Muse's release, the company's stock price has cumulatively risen more than 13%.

97% of revenue comes from advertising, AI returns are not yet visible

Of Meta's $64.6 billion in third-quarter revenue, advertising contributed $63.06 billion, up 26% year over year, accounting for 97.6% of total revenue. The growth driver was "rising volume and price." During the period, ad impressions increased 10% year over year, and the average price per ad rose 14% year over year.

The expansion in volume corresponds to the user base Family daily active people (DAP) reaching 3.64 billion in September, up 3% year over year; Meta attributed the price increase in the earnings report to AI-driven improvements in recommendation and monetization efficiency, and said that "AI investments have already delivered considerable incremental advertising revenue through stronger user engagement, better recommendations and more efficient monetization."

Beyond advertising, other business revenue was $1.1 billion, up 59% year over year, and Reality Labs revenue was $443 million, down 5.7% year over year.

On the expense side, Q3 total costs and expenses were $51.61 billion, up 68% year over year, but this included a one-time legal expense of $9.89 billion. In August, Meta reached a settlement with the attorneys general of 52 states over litigation concerning teen mental health and social media, with total consideration of up to about $18.59 billion. This expense was recorded in administrative expenses, directly pushing that item from $3.51 billion in the same period last year to $14.86 billion.

Excluding the settlement expense, total expenses were about $41.72 billion, an increase of about 36% year over year, still not low. This was mainly due to R&D expenses of $20.05 billion, up 32% year over year, and cost of revenue of $13.37 billion, up 45% year over year; the former corresponds to investment in model R&D, and the latter to depreciation and operations of data centers.

In addition, it is worth noting that as of September 30, Meta's total number of employees was 66,775, a decrease of about 12,000 from the same period last year. Cutting people while spending heavily on computing power, Meta's costs are being reallocated.

In terms of profit, operating profit during the period was $12.99 billion, down 37% year over year, and the operating margin shrank from 40% in the same period last year to 20%, but after excluding the impact of the settlement expense, operating profit was $22.88 billion, corresponding to a margin of about 35%, basically at the same level as the same period last year.

Q3 net profit was $9.86 billion, up 264% year over year, and diluted EPS was $3.83. This growth rate mainly came from a low base; in the same period last year, Meta recorded about $18.95 billion in income tax expense due to changes in US tax law, with an effective tax rate as high as 87%, pushing net profit down to $2.71 billion, while the effective tax rate this quarter fell back to 24%.

A better reflection of the true profitability level is the adjusted basis: after excluding the settlement expense, EPS was $7.40, with an effective tax rate of 17%.

Cash flow is the only "red light" in this earnings report. Q3 net cash provided by operating activities was $32.79 billion, up 9% year over year, still strong, but purchases of property and equipment were $33.94 billion, up 80% year over year, dragging free cash flow into negative territory.

Against the backdrop of negative free cash flow, Meta simultaneously announced a capital structure move. The board approved the establishment of non-voting Class C common stock, effective January 19, 2027, at which time bonus shares will be distributed to Class A and Class B shareholders on a 1:1 basis, with an effect equivalent to a stock split but without changing any shareholder's voting rights.

Chief Financial Officer Susan Li said the new shares can be used for financing, employee compensation and acquisitions, while preserving the voting rights of existing Class A and Class B shareholders and maintaining the founder-led structure. She also stressed that the company currently has no specific financing plan to announce, and this stock dividend distribution itself does not directly bring in financing cash.

This is preparation for future issuance of non-voting equity financing.

The only new product named by Zuckerberg in this earnings report is Muse. "This quarter we reached a major milestone on the path to delivering personal superintelligence to everyone, including launching Muse and Muse Wonder, our first frontier models," he said in the earnings statement. "Our investments are paying off, our core business is strong, and every front is accelerating."

Muse's commercialization currently relies mainly on subscription fees; subscribers enjoy more services than free users, and the future commercialization potential lies in transaction commissions. According to Zuckerberg, since launching on September 8, shopping has been Muse's most frequent use case, with over 25% of users having used the shopping feature, at an average of over $200 per transaction; in the first week after launch, because transaction volume exceeded expectations, Meta had to coordinate with payment partner Stripe to raise its credit limit.

In terms of product design, Meta equips each Muse user with an independent cloud virtual machine (Muse Secure VM); the user's files, login states, and app connections all run inside this isolated computer, and in the payment step Stripe generates a one-time card number, so real bank card information is not exposed to Muse.

But cooperation and confrontation appear at the same time. Shopify chose to integrate, allowing its merchants' products to be discovered by Muse and checked out; Instacart and Expedia have also reached partnerships; while Amazon, which holds about a 40% share of US e-commerce, blocks Muse from performing operations on its website, a move Morgan Stanley interprets as a negotiating posture rather than the endgame.

What Muse ignited is a global race to catch up. On September 30, OpenAI launched the personal agent Dots at its annual developer conference; on October 8, Google released the Gemini agent; on October 22, Claude launched the personal AI assistant Claude Duet. Major domestic companies are also following up intensively: Alibaba announced at its September 22 Yunqi Conference that Qwen is accelerating the building of a Personal Agent, and on October 13 announced the upcoming release of "Qwen Go!"; the new version beta of Ant Group's "Lingguang" is likewise benchmarked against Muse; Tencent and ByteDance are also advancing related product development.

Previously, the team of Brian Nowak, head of internet research at Morgan Stanley, modeled Muse's UE model; depending on token usage intensity, the direct service cost per user per month is about $3 to $130, and under the baseline scenario of using 25% of the weekly token quota, the monthly service cost per user is about $37.12, including both GPU inference and the CPU sandbox used for task execution.

Against Meta's global annual ARPU of about $70, even if a user uses only 5% of the free quota, the annual service cost is about $124, already exceeding what Meta earns from an ordinary user in a year.

The cloud virtual machine is another major expense beyond computing power costs, which means any major company that wants to follow up must be prepared to burn money over the long term. Nowak expects that Muse's true commercialized form will not land until late 2027 to 2028 at the earliest.

But the valuation premium Muse brings is equally enticing; after Muse launched and began going viral on social media, Meta's stock price rose nearly 13% within a week, posting its best monthly performance in more than 13 years. Many market institutions believe Muse may be the most influential AI product since OpenAI's ChatGPT burst onto the scene, and have raised Meta's target price one after another.

In the more than one month from its September 8 launch to the Q3 earnings release, Meta's stock price rose 13% cumulatively, while over the same period Alibaba's Hong Kong shares fell 11% and Tencent's Hong Kong shares fell 12.5%, which may explain why major companies are eager to replicate a "Muse," even though it is still a product form with immature commercialization that requires large amounts of sustained investment.

At least Meta has already returned to the AI competition table with it; three months ago, Meta was still the company that needed to prove it was not a "loser"; three months later, it is others' turn to answer that question.