To achieve computing power, Zuckerberg plans to burn through free cash flow

Meta CEO Mark Zuckerberg. Image generated by AI

Written by｜Su Yang

Editor｜Xu Qingyang

In the AI era, Meta is facing a difficult question that must be answered.

On October 21, local time in the United States, Meta released its Q3 2026 financial report. The company's quarterly revenue grew by 26% year-on-year, reaching $64.601 billion; however, due to large-scale infrastructure investments, free cash flow fell to negative $2.12 billion, compared to positive $10.625 billion in the same period last year.

Meta Q3 key financial data

Company founder and CEO Mark Zuckerberg stated that investments are starting to pay off, and CFO Susan Li further pointed out that the company will continue to increase investment in 2027. Meta also announced a new non-voting Class C share scheme in the financial report, attempting to reserve capital and governance space for cross-year investments.

For Meta, aggressive investment has become an inevitable trend. The current key issue is not only whether the advertising business can continue to grow, but also whether the new investments can generate sufficient revenue and efficiency improvements to cover the ever-expanding infrastructure costs.

01 Sky-high Lawsuits and Fines Weigh on Profits

Meta Q3 departmental business revenue

In Q3, Meta achieved total revenue of $64.601 billion, a 26% increase from $51.242 billion in the same period last year. Excluding the impact of a $113 million foreign exchange headwind, the revenue growth rate at constant exchange rates was also 26%.

This growth was primarily driven by the core advertising business, with advertising revenue for the quarter reaching $63.061 billion, up 26% from $50.082 billion in the same period last year.

Other revenue from the app family was $1.097 billion, bringing the total revenue of the app family to $64.158 billion. The Reality Labs department, focused on the metaverse and hardware, generated revenue of $443 million.

On the traffic side, the app family had 3.64 billion daily active users, a year-on-year increase of 3%. Ad impressions increased by 10% year-on-year, and the average price per ad rose by 14%, with the platform's ad monetization capability continuing to strengthen.

Management linked this performance to improvements in AI recommendation systems, believing that AI is helping to enhance user engagement and advertising efficiency.

In contrast to revenue growth, profits saw a significant decline, primarily due to high legal expenses.

In Q3, Meta recorded $9.89 billion in settlement costs related to lawsuits concerning adolescent mental health. As a result, consolidated operating profit fell to $12.988 billion, a 37% year-on-year decrease; reported net profit was $9.86 billion, with diluted earnings per share of $3.83.

On the same day the financial report was released, October 21, a judge in the U.S. District Court for New Mexico ruled against Meta in a case alleging misleading consumer data privacy, ordering Meta to pay up to $25 billion in damages to the state.

02 AI Begins to Test Cash Flow

AI investments cause Meta's free cash flow to turn negative

Although the advertising business continues to provide funding, infrastructure investments are changing Meta's cash flow situation.

According to official data, in Q3, Meta's operating cash flow was $32.786 billion, up from $29.999 billion in the same period last year.

On the expenditure side, cash expenditures for property and equipment purchases rose to $33.942 billion, compared to $18.829 billion in the same period last year. Adding $9.64 billion in principal payments for finance leases, total cash expenditures reached $34.906 billion.

According to Meta's official data, free cash flow for the period fell to negative $2.12 billion. In contrast, free cash flow for Q3 2025 was positive $10.625 billion, and the cumulative free cash flow for the first nine months narrowed from $29.508 billion in the same period last year to $11.049 billion.

The shift to negative free cash flow does not indicate a liquidity crisis for the company. Meta's operating cash flow remained high at $32.786 billion for the quarter, and the negative free cash flow primarily reflects that the growth rate of capital expenditures exceeded that of operating cash flow.

As of the end of Q3, the company's net property and equipment increased from $176.4 billion at the end of 2025 to $253.222 billion, and long-term debt rose from $58.744 billion to $80.922 billion.

The key moving forward is whether Meta's operating cash flow can continue to grow and whether new assets can gradually convert into revenue and profit.

03 Aggressive Investment Continues in 2027

Faced with cash consumption, Meta's management has not signaled any contraction.

For Q4 performance, the company expects revenue to be between $73 billion and $76 billion, with a midpoint guidance of $74.5 billion, representing a quarter-on-quarter growth of about 15.3%.

With resource reallocation, losses in the Reality Labs department are expected to be controlled, with full-year losses for 2026 projected to decrease to between $18 billion and $19 billion, and further narrowing to around $15 billion in 2027.

In terms of capital expenditures, Meta has raised the lower limit of its full-year 2026 capital expenditure guidance by $5 billion, updating it to between $135 billion and $145 billion. CFO Susan Li explicitly stated that although the 2027 budget is still being formulated, the company expects to maintain an aggressive investment pace.

Regarding the sources of investment returns, combining management statements and product planning, three levels can be identified.

In the short term, AI's enhancement of recommendation algorithms and advertising efficiency is bringing incremental revenue to the core advertising business. In the medium term, the personal AI assistant Muse and the cutting-edge model Muse Wonder launched in Q3 are gradually testing subscription models and developer interfaces, exploring commercialization directions for enterprise services and transaction scenarios. In the long term, Meta hopes to enhance internal operational efficiency and open up new commercialization opportunities through superintelligent laboratories and computing infrastructure construction.

Management stated that although the return cycle for investments in cutting-edge AI technology is relatively long, the company is optimistic about maintaining overall revenue growth as technology accumulates and products iterate.

04 Reserving Capital Space for Long-term Strategy

Data centers being built by Meta

To align with infrastructure construction and long-term strategic layout, Meta announced adjustments to its governance and capital structure in the financial report, with the board approving the establishment of a new non-voting Class C common stock.

According to the disclosed plan, the company intends to distribute stock dividends to existing Class A and Class B shareholders in the form of one share of Class C stock for every share held.

The company stated that the distribution of Class C shares will not change the existing voting rights structure of shareholders. The company's special committee has signed relevant agreements with Zuckerberg to maintain the existing governance arrangements.

In 2014, Google's parent company Alphabet completed a stock split by proportionally distributing non-voting Class C shares, allowing founders Larry Page and Sergey Brin to successfully lock in the company's control while continuously issuing stock for employee equity incentives and acquisitions.

The non-voting Class C shares will primarily be used for Meta's future employee equity incentives, potential strategic acquisitions, or other capital operations. The stock dividend itself will not directly bring new cash flow to the company. This arrangement reflects Meta's desire to reserve space for future equity incentives and capital operations while maintaining the existing control structure, mimicking mature capital tools used by other tech giants.

For the AI strategy that requires sustained investment over many years, a stable governance structure helps management advance long-term planning, but the actual effect still depends on the subsequent efficiency of capital use and investment returns.

Meta's Q3 financial report presents two curves moving forward simultaneously: the advertising business continues to generate revenue, while AI infrastructure continually expands capital demands. The growth in advertising revenue provides a funding basis, but it does not guarantee that AI investments will yield corresponding returns.

Meta has made it clear that it will continue to increase investments in 2027. For Zuckerberg, adhering to a long-term strategy is just the first step; as investments continue to expand, the company needs to prove that the new revenue and efficiency gains brought by AI are sufficient to support this ongoing infrastructure expansion.

Below is a refined version of the transcript from Meta's Q3 financial report analyst conference call:

Zuckerberg first commented:

In Q3, our app family had 3.64 billion daily users, with engagement on Facebook and Instagram reaching all-time highs. Threads' monthly active users have surpassed X, and WhatsApp message volume has also hit new highs.

We are rapidly advancing towards our goal of "providing personal superintelligence for everyone," and the launch of the Muse personal AI agent has been very successful. In just a month and a half since its launch in the U.S. and Canada, user sign-ups have exceeded 10 million, with about 25% of users using it daily. The most common use case is shopping (over 25%, with an average spending of over $200).

Due to the volume of transactions in the first week exceeding expectations, we had to temporarily increase credit limits in partnership with Stripe. Next month, Muse will begin to be integrated into Meta's smart glasses.

In terms of model development, we released the first closed-source cutting-edge model, Muse Wonder, which is at the forefront of the industry in several key metrics; we also released Muse Spark 1.2/1.3, Muse Code, and two open-source models, Muse Glimmer and Muse Spark Open. In the future, we will continue to push the boundaries of intelligence and plan to distill cutting-edge models into smaller, more efficient product models.

Regarding computing power and efficiency, we are building computing power that can serve over 1 billion people in the coming years. Although Muse is currently a compute-intensive application, we expect cost efficiency to improve by more than tenfold next year, enabling profitability at scale. We anticipate that in a year, we will be able to deliver the same intelligence in smaller models at a fraction of the current cost of Muse Wonder.

In terms of departmental adjustments, we will change the release rhythm of VR devices, prioritizing the delivery of personal superintelligence. Reality Labs' operational losses are expected to decrease to about $15 billion next year and trend towards profitability thereafter.

CFO Susan Li then added information about capital expenditures and funding allocation. She noted that in the coming years, Meta's capital expenditures will continue to increase, primarily to support core business and emerging AI initiatives. Although the time to realize returns on AI investments is longer than for core business, it is already generating significant incremental revenue for core advertising through improved recommendations and monetization efficiency.

The following is the Q&A session with analysts:

Q: What are the limiting factors for user growth as Muse moves forward? What is Meta's vision for its enterprise business?

Mark Zuckerberg: The biggest limitation for Muse right now is the computational intensity, which is why it hasn't been widely launched outside of the U.S. and Canada; we are working hard to improve computational efficiency and build sufficient computing power to support a future scale of over a billion users.

In terms of enterprise business, our recent focus has been on the model API and Muse Code. In the future, we will build a Muse agent network that allows consumers' Muse to negotiate and transact directly with enterprise agents.

Q: What friction exists in Agentic Commerce currently? How is the company responding? Additionally, what considerations are there regarding the new class of shares announced tonight for capital structure and future AI investments?

Mark Zuckerberg: Most companies are very excited to support this because they see it not only as a way to increase sales and improve existing customer experiences but also as a potential source of new customers. The speed of adoption will vary among different businesses, and we will cooperate on reasonable terms for all parties. In the long run, making this work well is valuable for everyone, and we are currently on a good trajectory.

Susan Li: Agents will become an important part of the evolution of reality, and the market will ultimately shift towards more permissioned and structured interactions—this includes certified APIs, approved agent partnerships, or direct communication between consumer personal agents and enterprise agents. We are working with partners like Sierra, Walmart, Stripe, and Shopify to develop the Personal Agent Protocol, which defines secure and reliable interaction methods. Many experiences in the near term will still be AI-assisted rather than fully autonomous, but as trusted integrations develop, more businesses will actively participate.

Regarding the new class of shares, we created this new class of non-voting capital stock (Class C) to distribute stock dividends to all existing shareholders, forming publicly tradable non-voting shares that can be used for financing, employee compensation, acquisitions, and other general corporate purposes, while retaining the voting rights of existing Class A and Class B shareholders.

This move maintains the founder-led structure, expands financing tools, and gives us the flexibility to invest under the strong momentum driven by AI. There are currently no specific financing plans, but establishing an appropriate capital structure framework is prudent.

Q: What is the cost structure per user for Muse? How will the cost curve decline in the future? What is the company doing to reduce costs?

Susan Li: We don't have specific numbers to share, but as Zuckerberg mentioned, the current computing costs are quite high. Ensuring a focus on computational efficiency is one of the key vectors that can both scale Muse and develop it into an ideal business.

In the long run, as LLMs scale, we expect to accomplish more tasks with fewer tokens and rely on the AI models themselves to help improve the efficiency of the software used to run the models. There is a lot of work on the efficiency roadmap that can help us bend the cost curve, and the arc of AI technological advancement will also bring greater savings.

Q: Does the commercialization pace of Muse need to be faster than previous new products due to significantly higher associated costs? Additionally, does the data showing 25% of users participating in shopping and an average of over $200 include travel purchases?

Mark Zuckerberg: We will start monetizing earlier than usual for two main reasons: first, both the costs and the opportunity scale are large, and we must make progress to scale; second, the extensive usage scenarios themselves are commercial. That data does indeed include travel. We will establish a model that captures a certain percentage of the value created for people. It is very important to make these numbers work so that we can continue to build computational capacity to serve everyone at scale.

The trend we see is that, year-on-year, in a year we will be able to deliver equivalent intelligence in smaller models at a fraction of the cost of the current Muse Wonder (or any current cutting-edge model) intelligence level. If this represents a baseline level sufficient to support product-market fit, ongoing model development will drive efficiency improvements.

Of course, a year from now, most people will not be satisfied with the current intelligence level (as intelligence itself is also advancing), so we need to simultaneously advance revenue, expansion, model development, and efficiency to fund computational construction and reach billions of people as quickly as possible. Not charging everyone right from the start is because we want to first build a large-scale Muse network.

Q: What is the layout for Muse's open-source and closed-source models? How will you balance cutting-edge intelligence and efficiency in the future?

Mark Zuckerberg: We released the first closed-source cutting-edge model, Muse Wonder, which is at the forefront of several key dimensions of personal superintelligence, validating the importance of self-trained models to deliver the best experience. At the same time, we also released Muse Spark 1.2 and 1.3, Muse Code, as well as two open-source models with open weights, Muse Glimmer and Muse Spark Open, along with Muse Voice.

Looking ahead, as the field advances rapidly, we plan to push the intelligence frontier while distilling these models into smaller, more efficient product models optimized for personal agents. We are already researching larger models after Muse Wonder, which will be very impressive and powerful; we are also developing new smaller models that will be launched soon, providing Muse Wonder-level intelligence in a more efficient manner.

My view is that we will be able to build superintelligence through scaling, so we are methodically moving up this ladder.

Q: Is computational power currently the main limitation for scaling Muse? How is the company improving computational efficiency? What is the basis for the expectation of a tenfold improvement in cost efficiency next year?

Mark Zuckerberg: Yes, one of the biggest limitations right now is computational intensity, which is why we haven't launched globally (currently only in the U.S. and Canada). Part of the reason is to optimize and refine connectors for each location, and more importantly, to ensure efficient progress. We are building computational capacity that can serve over a billion people in the coming years.

Currently, Muse operations are very computation-intensive, but we expect its cost efficiency to improve by more than tenfold next year, which will allow it to become profitable once revenue reaches scale. If computational efficiency continues to improve rapidly in the coming years (which we believe is highly likely), then Muse will be highly profitable for many years.

Q: Regarding capital expenditures, the guidance for 2026 has been raised to $135 billion to $145 billion. How will capital expenditures look in the coming years? How will these investments support core business and AI?

Susan Li: Capital expenditures for the third quarter (including principal payments on finance leases) were $34.9 billion, driven by investments in servers, data centers, and network infrastructure.

We are now adjusting the full-year capital expenditure guidance for 2026 to a range of $135 billion to $145 billion (previously $130 billion to $145 billion).

Looking beyond 2026, we are still formulating next year's budget and capacity plans, but we expect to continue investing actively to support core business and emerging AI initiatives. Current capacity and planned online capacity give us the flexibility to allocate resources between model training, core business growth, and new product development and support.

Our AI investments have already brought significant incremental revenue to the core advertising business through stronger engagement, better recommendations, and more efficient monetization. While the time for these investments to generate returns is longer than for the core business, the breadth of monetization opportunities and internal efficiencies give us confidence in long-term potential. We expect to maintain strong overall revenue growth throughout 2027.