Meta conference call: AI significantly enhances user engagement, capital expenditures will continue to "soar" next year, focusing on both talent and computing power, equipping AI glasses is a trend

The financial report released after the U.S. stock market on Wednesday showed that Meta's revenue in the second quarter was $47.52 billion, exceeding analysts' expectations of $44.83 billion, with advertising revenue of $46.5 billion also surpassing expectations. The Reality Labs division reported a loss of $4.5 billion, which was better than market expectations. The company also raised the lower limit of its capital expenditures for 2025 from $64 billion to $66 billion, and its stock price surged by as much as 10% in after-hours trading.

In a later earnings call, Meta founder, chairman, and CEO Mark Zuckerberg elaborated on his vision for superintelligence, stating that "superintelligence is now just around the corner," and that Meta now has all the conditions to achieve this goal, which will "soon reshape all of our systems."

To realize this grand vision, Meta will continue to increase its investment in computing power and talent resources.

Zuckerberg revealed that the company has established a new "Meta Superintelligence Lab," led by newly recruited AI elites, aimed at developing the next generation of cutting-edge models; additionally, the company is also building multiple gigawatt-level computing clusters, with the goal of providing "personal superintelligence" to billions of users worldwide, thereby reshaping its product line.

Chief Financial Officer Susan Li clearly stated that the total expense growth rate for 2026 will be higher than that of 2025, while capital expenditures will see another "similarly significant dollar growth year."

AI monetization shows "promising results," strong growth in advertising, and increased user engagement

The financial report indicated that artificial intelligence has become the core engine driving Meta's current business growth. In the second quarter, the company's operating profit margin reached 43%, demonstrating strong profitability. Zuckerberg emphasized in the meeting that the outstanding performance this quarter is mainly attributed to the efficiency dividends released by AI in the advertising system.

Zuckerberg introduced that the new AI-driven advertising recommendation model has produced significant effects on Meta's platforms, boosting Instagram's ad conversion rate by about 5% and Facebook's by 3%. Meanwhile, the usage of generative AI creative tools continues to expand, particularly favored by small advertisers with limited budgets.

In terms of user engagement, the advancements in the AI recommendation system have also played a crucial role. Susan Li stated that this quarter alone has led to a 5% increase in Facebook's user time spent and a 6% increase for Instagram, with users spending over 20% more time on videos year-on-year.

Product activity has also improved. Susan Li added that Meta expects further improvements, as it will adjust content recommendations based on user interactions during specific activity periods to ensure these recommendations are "most relevant to what they are interested in at that time," which will mean higher engagement and advertising revenue.

Susan Li stated in the meeting: "In the second half of the year, we will focus on further enhancing the freshness of original content so that the right audience can discover creators' original content as soon as it is published." Currently, over two-thirds of the recommended content on Instagram in the U.S. region is original In addition to its core advertising and recommendation business, Meta is also expanding its AI capabilities into new commercial areas. Zuckerberg stated that business messaging has shown market potential in several test countries, and the active users of its AI assistant Meta AI have exceeded 1 billion, penetrating broader scenarios from platforms like WhatsApp to Facebook feeds.

"Superintelligence is just around the corner," continuing to increase capital expenditure

Meta has narrowed its full-year capital expenditure forecast for 2025 to between $66 billion and $72 billion, with the median being adjusted upward from previous estimates, and it expects capital expenditures to continue to grow in 2026.

Susan Li emphasized that developing leading AI infrastructure will be a core advantage in developing the best AI models and product experiences, and therefore, significant investments will be made in 2026 to support this work.

According to Susan Li, 2026 will be a year of "similarly significant" growth in capital expenditure, primarily for expanding the servers, networks, and data centers needed for generative AI. She explained that the biggest driver of cost growth comes from infrastructure costs, including sharply accelerated depreciation expenses due to massive capital investments, as well as increased costs for data center operations, energy, and maintenance.

In the future, those without AI glasses will be at a disadvantage

Zuckerberg stated at the conference that glasses will become the ideal foundational form for AI because "you can let artificial intelligence see what you see all day, hear what you hear, and talk to you."

He mentioned that adding displays to these glasses will unlock more value, such as providing a wider holographic view like Meta's next-generation Orion AR glasses, or equipping smaller displays in everyday AI glasses.

Zuckerberg also emphasized: "If you don't have AI-equipped glasses in the future, or some way to interact with artificial intelligence, your cognitive abilities will likely be at a significant disadvantage compared to others."

The talent war has begun, creating a "small but elite" top team

Susan Li emphasized that the company's main focus remains on reinvesting capital back into the business, with infrastructure and talent being top priorities.

Susan Li stated that the way to increase staff continues to target the company's highest priority areas, and it is expected that talent increases in all priority areas will continue to drive overall personnel growth from this year to 2026, while personnel growth in other functional departments remains limited. The AI field will particularly focus on recruiting industry-leading talent to build the Meta Superintelligence Lab to accelerate our AI model development and product planning.

Zuckerberg emphasized that superintelligence research requires "small, talent-dense teams," which differs from the configuration of Meta's other machine learning systems. The company is assembling a world-class team of AI researchers, infrastructure, and data engineers, led by top industry talent, with Nat Friedman leading AI product and application research, and Shengjia Zhao serving as chief scientist Mark Zuckerberg stated that the reason for attracting many top talents is that Meta possesses "all the elements needed to build leading models and deliver them to billions of people."

The meeting also indicated that employee compensation will become the second largest driver of expense growth in 2026, primarily due to investments in technical talent, including confirming the annual salary expenditures for AI talent recruited this year. The company had over 75,900 employees at the end of the second quarter and continues to hire in key areas such as AI and infrastructure.

The Dogewave-level computing cluster will be rolled out successively

Next is the infrastructure aspect.

Susan Li stated that having sufficient computing power is expected to be central to realizing many of the biggest opportunities in the coming years, and the company will continue to see very attractive returns from investments in AI capabilities for core advertising and organic engagement programs, with expectations for significant investments in this area in 2026.

The scale of the computing infrastructure that Meta is building is impressive. In addition to the upcoming Prometheus cluster, the company is also building the Hyperion cluster, which can scale to 5 gigawatts in a few years, as well as multiple Titan clusters. Zuckerberg mentioned that the company is making these investments because it "firmly believes that superintelligence will improve every aspect of what we do."

Infrastructure costs will become the largest driver of expense growth in 2026, primarily due to a sharp acceleration in depreciation expenses and higher operating costs, including energy, leasing, maintenance, and other expenses. The company also expects to increase spending on cloud services and network-related costs.

Susan Li stated that the company is exploring ways to collaborate with financial partners to develop data centers to support large-scale infrastructure projects. Although no completed transactions have been announced, the company believes this will attract external financing for large data center projects while providing flexibility as infrastructure demands change.

Below is the Meta earnings call (translated with AI tools):

Meeting Date: July 30, 2025

Company Name: Meta

Meeting Nature: Q2 2025 Earnings Call

Meeting Host: Krista (call operator)

Meta Participants:

• Kenneth Dorell - Director of Investor Relations
• Mark Zuckerberg - Founder, Chairman, and CEO
• Susan Li - Chief Financial Officer

Host Opening:

Good afternoon, everyone. I am Krista, and I will be the host for today's meeting. Welcome to the Meta Q2 earnings call. To prevent background noise, all lines have been set to mute. There will be a Q&A session after the speeches, and this meeting will be recorded. Now, please welcome Kenneth Dorell, Meta's Director of Investor Relations, to begin his remarks.

Kenneth Dorell:

Thank you. Good afternoon, everyone, and welcome to the Meta Q2 2025 earnings call. Joining me today are CEO Mark Zuckerberg and CFO Susan Li Our remarks today will contain forward-looking statements that are based on assumptions as of today. Actual results may differ significantly due to various factors, including those described in today’s earnings press release and the 10-Q quarterly report filed with the SEC. We do not undertake any obligation to update any forward-looking statements.

In this meeting, we will present GAAP and certain non-GAAP financial metrics. The reconciliation of GAAP to non-GAAP metrics is included in today’s earnings press release. The earnings press release and accompanying investor presentation can be found on our website investor.atmeta.com.

Now, please welcome Mark to speak.

Mark Zuckerberg, Founder, Chairman, and CEO:

Okay. Thank you, Ken. Thank you all for joining us today. We’ve had another strong quarter, with over 3.4 billion people using at least one of our apps every day, and engagement across the board remains strong. Our business continues to perform exceptionally well, allowing us to invest heavily in AI projects. Over the past few months, we’ve begun to see signs of AI systems self-improving. The improvements are still slow, but undeniable. Developing superintelligence, which we define as AI that surpasses human intelligence in all aspects, we believe is now within reach.

Meta’s vision is to bring personal superintelligence to everyone, enabling people to direct it towards what they value in their own lives. We believe this has the potential to unlock an exciting new era of personal empowerment. There has been much written about all the economic and scientific advancements that superintelligence can bring, and I am extremely optimistic about this. However, I believe that if history serves as a guide, the more important role of superintelligence will be how it empowers people to become more creative, develop culture and community, connect with one another, and lead more fulfilling lives. To build this future, we have established the Meta Superintelligence Lab, which includes our infrastructure, product, and fairness teams, as well as a new lab focused on developing the next generation of models. We have made good progress on Llama 4.1 and 4.2. Meanwhile, we are also developing next-generation models that will push the frontier of technology in about a year.

We are building an elite, talent-dense team. Alexandr Wang leads the entire team. Nat Friedman leads our AI product and application research. Shengjia Zhao is the chief scientist for this new work. They are all exceptionally talented leaders, and I am excited to work closely with them and the world-class AI researchers, infrastructure, and data engineers we are assembling. I have spent a lot of time this quarter building this team. The reason so many people are excited to join is that Meta has all the elements needed to build leading models and deliver them to billions of people.

Those who join us will have access to unparalleled computing resources, as we are building several multi-gigawatt clusters. Our Prometheus cluster will go live next year, and we believe it will become the world’s first cluster exceeding one gigawatt. We are also building Hyperion, which will be able to scale to 5 gigawatts in a few years We have multiple Titan clusters in development. We are making all these investments because we firmly believe that superintelligence will improve every aspect of what we do. From a business perspective, I mentioned last quarter that we are pursuing five fundamental opportunities: improved advertising, more engaging experiences, business messaging, Meta AI, and AI devices. I can elaborate on each one. In terms of advertising, the strong performance this quarter is primarily due to AI unlocking greater efficiency and revenue in our advertising system.

This quarter, we expanded our new AI-driven ad recommendation model to new interfaces and improved performance by using more signals and longer context. The model drove approximately 5% growth in ad conversions on Instagram and 3% growth on Facebook. We have also made good progress in AI ad creative, with a significant portion of ad revenue now coming from campaigns using our generative AI capabilities. This is particularly valuable for small advertisers with limited budgets. Agencies will continue to play an important role in helping large brands strategically apply these tools.

The second opportunity is more engaging experiences. AI has significantly enhanced our ability to show users content that is interesting and useful to them. Advances in our recommendation systems have greatly improved quality, resulting in a 5% increase in Facebook user engagement time and a 6% increase on Instagram just this quarter. The content itself also has significant improvement potential. We are seeing early progress with the release of AI video editing tools in Meta AI and new editing applications, but there is still much work to be done in this area.

The third opportunity is business messaging. I have previously mentioned that I believe every business will soon have business AI, just as they have email addresses, social media accounts, and websites. We are starting to see some product-market fit in several test countries, and we are integrating these business AIs into Facebook and Instagram ads, as well as directly into e-commerce websites.

The fourth opportunity is Meta AI, which has a considerable reach, with over 1 billion monthly active users. Our current focus is on deepening the experience to make Meta AI the leading personal AI. As we continue to improve the models, we are seeing increased user engagement. Our next-generation models will continue to play an important role in this regard.

The fifth opportunity is AI devices. Our Ray-Ban Meta glasses continue to gain strong momentum, with accelerating sales growth.

We have also partnered with Oakley to launch the new high-performance AI glasses, the Meta Houston series. These glasses feature longer battery life and higher resolution cameras, designed specifically for sports. The proportion of users utilizing Meta AI is increasing, and we are seeing improved AI retention rates among new users, which is a good sign for continued usage. I believe AI glasses will become a primary way we integrate superintelligence into our daily lives. Therefore, it is essential to have these different styles and products to attract different users in various environments

Finally, we see people continuing to spend more time in the Quest ecosystem, with the community steadily growing. Last month, we launched the MetaQuest 3S Xbox version, and we saw interest in cloud gaming reach record levels. Beyond gaming, we continue to see broader use cases, with media and web browsing contributing significantly to engagement. We will share more relevant information, especially regarding the work of Reality Labs at the Connect conference on September 17th.

I encourage everyone to stay tuned. Overall, it has been a busy quarter with strong business performance, and we have made real progress in gathering the talent and computing power needed to build personal superintelligence for everyone. I am very grateful to our team for their hard work in delivering all these results, and I thank everyone for joining us on this journey. Now, I will hand it over to Susan.

Susan Li, Chief Financial Officer:

Thank you, Mark. Good afternoon, everyone. Let's first look at the consolidated financial results. All comparative data is year-over-year unless otherwise noted.

Total revenue for the second quarter was $47.5 billion, growing 22% on both a reported basis and a constant currency basis. Total expenses for the second quarter were $27.1 billion, an increase of 12% year-over-year. In terms of specific items, the cost of revenue increased by 16%, primarily due to rising infrastructure costs and payments to partners, although this was partially offset by gains from the previously announced extension of server useful lives. Research and development expenses increased by 23%, mainly due to higher employee compensation and infrastructure costs.

Marketing and sales expenses grew by 9%, primarily due to increased professional services related to ongoing platform integrity work and higher marketing costs, although this was partially offset by a decrease in employee compensation. General and administrative expenses decreased by 27%, mainly due to lower legal-related costs. At the end of the second quarter, we had over 75,900 employees, a decrease of 1% quarter-over-quarter, as the vast majority of employees affected by performance-related layoffs earlier this year are no longer included in our total employee count. This decline was partially offset by ongoing hiring in key areas such as monetization, infrastructure, Reality Labs, artificial intelligence, and regulatory compliance.

Operating income for the second quarter was $20.4 billion, with an operating margin of 43%. The tax rate for the quarter was 11%, reflecting excess tax benefits from stock-based compensation due to the increase in stock price compared to prior periods. Net income was $18.3 billion, or $7.14 per share. Capital expenditures (including principal payments on finance leases) were $17 billion, primarily for investments in servers, data centers, and network infrastructure.

Free cash flow was $8.5 billion. We repurchased $9.8 billion of Class A common stock and paid $1.3 billion in dividends to shareholders. We also made $15.1 billion in private equity investments during the second quarter, including a minority investment in Scale AI and other investment activities. At the end of the quarter, we held $47.1 billion in cash and cash equivalents and $28.8 billion in debt.

Now turning to segment performance. I will start with the Family of Apps segment. Our Family of Apps community continues to grow, and we estimate that in June, over 3.4 billion people used at least one of our Family of Apps products daily In the second quarter, total revenue from the application series was $47.1 billion, a year-on-year increase of 22%.

In the second quarter, advertising revenue from the application series was $46.6 billion, an increase of 21%, and a 22% increase on a constant currency basis. Among the advertising revenue, the online commerce vertical was the largest contributor to the year-on-year growth. By user region, advertising revenue growth was strongest in Europe and other parts of the world, at 24% and 23%, respectively. North America and the Asia-Pacific region grew by 21% and 18%, respectively.

In the second quarter, the total number of ad impressions across our services grew by 11%, primarily driven by the Asia-Pacific region. Growth in impressions accelerated across all regions, mainly due to increased user engagement on Facebook and Instagram, followed by ad loading optimizations on Facebook. The average ad price increased by 9%, benefiting from increased advertiser demand, primarily driven by improved ad performance. Due to the accelerated growth in impressions in the second quarter, pricing growth slightly slowed compared to the first quarter.

Other revenue from the application series was $583 million, a 50% increase, primarily driven by growth in WhatsApp paid messaging revenue and Meta Verified subscriptions. We continue to invest most of our resources in the development and operation of the application series. In the second quarter, spending on the application series was $22.2 billion, accounting for 82% of total spending. Spending on the application series grew by 14%, mainly due to increases in employee compensation and infrastructure costs, although this was partially offset by a decrease in legal-related costs. Operating income from the application series was $25 billion, with an operating margin of 53%.

In the Reality Labs division, second-quarter revenue was $370 million, a year-on-year increase of 5%, primarily due to increased sales of AI glasses, although this was partially offset by a decline in Quest sales. Reality Labs spending was $4.9 billion, a year-on-year increase of 1%, mainly due to increased non-employee-related technology development costs. Reality Labs reported an operating loss of $4.5 billion.

Now let's talk about the business outlook. There are two main factors driving our revenue performance. One is our ability to provide engaging experiences for the community, and the other is our long-term efficiency in effectively monetizing this engagement. In the first aspect, daily active users on Facebook, Instagram, and WhatsApp continue to grow as we further improve our recommendation systems and product experiences.

We continue to see strong momentum in video interactions. In the second quarter, Instagram video time grew by over 20% year-on-year globally. Facebook also performed strongly, especially in the United States, where video time also grew by over 20% year-on-year. These increases are attributed to our ongoing optimization of ranking systems to better identify and display the most relevant content.

As we further scale our models to better align recommendations with users' interests in conversations, we expect to achieve more improvements throughout the year. Another focus of our recommendation efforts is to promote original content. On Instagram, more than two-thirds of recommended content in the U.S. now comes from original posts. In the second half of the year, we will focus on further enhancing the freshness of original posts, enabling the right audiences to discover original content as soon as creators publish it

We have also made good progress in long-term ranking innovation, which is expected to provide the next round of improvements in the coming years. Research work on developing cross-platform foundational recommendation models continues to advance. We have also seen encouraging results from the use of large language models in the Threads recommendation system. The integration of large language models is now driving a significant increase in time spent related to ranking on Threads.

We are exploring how to extend the use of large language models in recommendation systems to other applications. We are leveraging Llama and several other backend processes, including handling error reports, to identify and resolve recurring issues more quickly and efficiently. This has led to a reduction of about 30% in the top error reports for Facebook feeds and notifications in the U.S. and Canada over the past 10 months. Our primary way of using Llama in applications is to support Meta AI, which is now available in over 200 countries and regions.

WhatsApp remains the biggest driver of queries, as people directly message Meta AI to complete tasks such as information gathering, homework tutoring, and image generation. Beyond WhatsApp, we see Meta AI becoming an increasingly valuable complement to our content discovery engine. The use of Meta AI on Facebook is expanding, with people using it to inquire about posts they see in their feeds and to search for content on our platform. We anticipate that another way Meta AI will assist content discovery is through automatic translation and voiceover, converting foreign language content into the local language of the audience.

We will share more related work later this year. Moving on to Reality Labs. Ray-Ban Meta sales accelerated in the second quarter, although production was increased earlier this year, demand for the most popular SKUs still exceeds supply. We are working to increase supply to better meet consumer demand later this year.

Now, turning to the second driver of revenue performance, which is improving monetization efficiency. The first part of this work involves optimizing ad levels in organic interactions. We continue to optimize ad supply on each platform to serve ads to users at the most relevant times and places. In the second quarter, we also began introducing ads in Threads feeds and WhatsApp's update tabs, which is a separate space from chatting with people.

As of May, advertisers worldwide can now serve video and image ads to Threads users in most countries, including the U.S. Although ad supply remains low, Threads is not expected to make a significant contribution to overall impression growth in the short term. However, we are optimistic about Threads' long-term opportunities as the community and interactions grow and monetization scales up. On WhatsApp, we are rolling out status and channel ads, as well as channel subscriptions in the update tab, to help businesses reach over 1.5 billion daily active users accessing that part of the app. We expect the introduction of status ads to be gradual this year and next, initially with low ad supply

We also expect that WhatsApp status ads will achieve lower average prices than Facebook or Instagram ads in the foreseeable future, partly because WhatsApp tends to monetize in lower-level markets, and the information available for targeting is more limited. Given this, we anticipate that status ads will not make a significant contribution to total impressions or revenue growth in the coming years. The second part of improving monetization efficiency is enhancing marketing performance. I will focus on three aspects of this work.

Improving our advertising system, advancing our advertising products, including building tools to assist in ad creation, and developing our advertising platform to drive results optimized for each business goal. First is our advertising system, where we are innovating at both the ad retrieval and ranking stages to provide users with more relevant ads. A significant part of this work involves our continued advancement of the modeling innovations introduced earlier while expanding their adoption on our platform. The Andromeda model architecture, which we began introducing in the second half of 2024, supports the ad retrieval stage of our advertising system, allowing us to select the most relevant ads from tens of millions of potential candidates.

In the second quarter, we optimized Andromeda to select more relevant and personalized ad candidates while expanding its reach to Facebook Reels. These improvements increased the conversion rates for Facebook mobile news feeds and Reels by nearly 4%. Our new generative ad recommendation system, GEM, supports the ranking stage of the advertising system, which is the process after ad retrieval, where we determine which ads to show users from the candidates suggested by the retrieval engine.

In the second quarter, we enhanced GEM's performance by further expanding training capabilities and adding organic content and ad interaction data from Instagram. We also adopted new advanced sequence modeling techniques, doubling the length of the event sequences used, allowing our system to consider users' longer content or ad interaction histories, thereby providing better ad choices. The combination of these improvements increased Instagram's ad conversion rates by about 5%, and Facebook news feeds and Reels by 3%.

Finally, we expanded the coverage of the Lattice model architecture in the second quarter. We began deploying Lattice for post-ad ranking work for the first time in 2023, enabling us to run significantly larger models that generalize learning across targets and platforms, replacing numerous small ad models historically optimized for individual targets and platforms. In April, we began deploying Lattice to other early-stage ad ranking models. This not only improved capacity and engineering efficiency but also enhanced performance. The recent Lattice deployment increased ad conversion rates for Facebook news feeds and Reels by nearly 4% in the second quarter.

Next is the advertising product. Our Advantage+ AI solution suite is showing strong growth momentum. In the second quarter, we completed the rollout of a simplified campaign creation process for Advantage+ sales and application promotion activities, making it easier for advertisers to realize the performance advantages of Advantage+ by enabling this feature at the start Since expanding availability, we have seen an increase in advertisers' adoption of sales and app promotion activities, and we are working to complete the launch of lead generation activities in the coming months.

The adoption of generative AI creative tools in the Advantage+ creative suite continues to expand. Currently, nearly 2 million advertisers are using our video generation features, image animation, and video extension capabilities. Our text generation tools have also performed well and are continuously adding new features. In the second quarter, we began testing AI translation capabilities, allowing advertisers to automatically translate ad headlines into 10 different languages. Although still in the early stages, we have seen encouraging performance improvements in pre-release testing.

We also continue to see strong adoption of image extension features among small and medium-sized advertisers, demonstrating how these tools help resource-constrained businesses develop creative content. For large advertisers, we expect agencies to continue to be valuable partners in applying these new tools to enhance performance.

In addition to Advantage+, we have also seen good momentum in business messaging, particularly in the United States, where click-to-message revenue grew by over 40% year-over-year in the second quarter. The strong growth in the U.S. benefited from an increase in the adoption of website-to-message ads, which guide users to first visit a business's website for more information before choosing to start a chat with the business in our messaging app.

Finally, we continue to improve our advertising platform to optimize results based on each business's goals and measurement methods. In the second quarter, we completed the global rollout of incremental attribution capabilities, which is the only product on the market that can optimize and report incremental conversions—those that would not occur without the user seeing the ad. We also globally launched cross-channel advertising in the second quarter, allowing advertisers to optimize in-store and online incremental sales through a single campaign. In testing, advertisers using cross-channel ads saw a 15% reduction in median total purchase costs compared to those optimizing only for the website.

Next, I would like to discuss our capital allocation approach. Our primary focus remains on reinvesting capital back into the business, with infrastructure and talent as our top priorities. First is recruitment. Our personnel increase strategy continues to target the company's highest priority areas. We expect talent increases in all priority areas to continue driving overall headcount growth from this year through 2026, while personnel growth in other functional areas remains constrained. In the field of artificial intelligence, we place particular emphasis on recruiting industry-leading talent to build the Meta Super Intelligence Lab to accelerate our AI model development and product planning.

Second is infrastructure. We expect having ample computing power will be central to realizing many of our biggest opportunities in the coming years. **We continue to see very attractive returns from our investments in AI capabilities for core advertising and organic engagement programs, and we anticipate making significant investments in this area through 2026. We also expect that developing leading AI infrastructure will be a core advantage in developing the best AI models and product experiences, so we anticipate significantly increasing investments to support this work by 2026 **

Turning to our financial outlook. We expect total revenue in the third quarter of 2025 to be between $47.5 billion and $50.5 billion. Our guidance is based on current exchange rates, with foreign exchange having a tailwind effect of approximately 1% year-over-year on total revenue.

While we are not providing a revenue outlook for the fourth quarter, we expect the year-over-year growth rate for the fourth quarter of 2025 to be slower than that of the third quarter, as we will be facing a comparison against a stronger growth period in the fourth quarter of 2024.

Now turning to expense outlook. We expect total expenses for the full year of 2025 to be in the range of $114 billion to $118 billion, narrowing from the previous outlook of $113 billion to $118 billion, with a year-over-year growth rate of 20% to 24%. Although our planning for next year is still in the early stages, several factors are expected to exert significant upward pressure on our total expense growth rate in 2026. The largest single driver of growth will be infrastructure costs, due to a sharp acceleration in depreciation expenses and higher operating costs as we continue to scale our infrastructure. In addition to infrastructure, we expect the second largest driver of growth to be employee compensation, as we increase technical talent in priority areas and confirm the full-year compensation costs for hiring employees in 2025. We expect these factors to lead to a year-over-year expense growth rate in 2026 that exceeds the expense growth rate in 2025.

Now turning to capital expenditure outlook. We currently expect capital expenditures for 2025 (including principal payments on finance leases) to be in the range of $66 billion to $72 billion, narrowing from the previous outlook of $64 billion to $72 billion, with a median year-over-year increase of approximately $30 billion. Although the infrastructure planning process remains highly dynamic, we currently expect 2026 to be another year of significant dollar growth in capital expenditures as we continue to actively seek opportunities to bring additional capacity online to meet the demands of our AI efforts and business operations.

Regarding taxes. With the enactment of the new U.S. tax law, we expect U.S. federal cash taxes for the remainder of this year and future years to decrease. The provisions of the bill have various implementation options, which we are currently evaluating. While we estimate that the tax rate for 2025 will be higher than the second quarter tax rate, we cannot currently quantify the extent. Additionally, we continue to monitor the active regulatory environment, including the increasing legal and regulatory resistance in the EU, which could have a significant impact on our business and financial performance. For example, we continue to engage with the European Commission regarding our lower personalization advertising product (LPA), which we launched in November 2024 based on feedback from the European Commission regarding the Digital Markets Act. As the Commission provides further feedback on the LPA, we cannot rule out the possibility that it may seek to impose further modifications, which could significantly worsen the experience for users and advertisers. This could have a significant negative impact on our European revenue as early as later this quarter. We have appealed the European Commission's decision on the Digital Markets Act, but any modifications to our model may be imposed during the appeal process

In summary, this has been another strong quarter for our business, and our investments in infrastructure and technical talent continue to improve core advertising performance and engagement on our platform. We expect that the significant investments we are making now will enable us to continue to leverage advancements in AI to expand these gains and unlock a range of new opportunities in the coming years.

Okay, Krista, let's start taking questions.

Host: Thank you. We will now begin the Q&A session. The first question comes from Eric Sheridan at Goldman Sachs.

Eric Sheridan: Thank you very much for taking my question. Mark, what are the key learnings from your AI business over the past three to six months? How have these learnings influenced adjustments in your talent acquisition and computing capacity strategies, as well as the strategic evolution mentioned in your recent blog post? Susan, based on Mark's comments about expanding talent and computing capacity, could you elaborate on how these two components are driving your expectations for operating and capital expenditures over the next 12 to 18 months? Thank you.

Mark Zuckerberg, Founder, Chairman, and CEO: Okay, I'll take that. From a macro perspective, people are asking how long it will take to achieve truly powerful AI or superintelligence. I have found that so far, at every stage, those more aggressive assumptions or the fastest expectations have often predicted actual outcomes most accurately. I think this is still the case this year. I have mentioned many such examples in past earnings calls. We see internal teams being able to use Llama 4 to build autonomous AI agents to improve Facebook's algorithms, enhancing quality and engagement, which is a profound achievement.

Currently, this is still at a small scale, so I'm not sure if this outcome itself has made a significant contribution to this quarter's earnings. But I think the trajectory of development in this area is very optimistic. I think one of the interesting challenges of running a business like this now is that the world is likely to undergo tremendous changes in the coming years.

On one hand, we can make many improvements to our existing core products. On the other hand, we maintain a principle throughout the company of taking superintelligence seriously. The basic idea is that we believe this will soon reshape all of our systems, not just the trajectory of one or two quarters, but over several years. This will change the operational assumptions across all aspects of the company.

We continuously observe the trajectory and pace of AI development, and I believe the progress remains at a rapid level. This influences our decision-making from various aspects, including the importance and value of having the absolute best elite talent team, ensuring we have leading computing clusters, providing researchers with more per capita computing resources for cutting-edge research and applying it to products for billions of users, and ensuring we build and drive these products across all business areas. I believe our company is among the best in the world in this regard; when we master a technology, we excel at applying it across all applications and advertising systems

This technology will not be idle. I believe no other company is as skilled as we are at bringing a technology to billions of users. We will actively advance in all these areas, but to some extent, this is a bet on the development trajectory we have observed, which are the signals we see, and we are working to interpret them.

Susan Li, Chief Financial Officer:

Eric, regarding the second part of your question, we have not actually initiated the budgeting process for 2026 yet. Given the situation for next year, there are clearly many variables in this very dynamic operating environment. However, we already have some visibility on certain aspects today, including a rough framework for our 2026 infrastructure plan, which directly impacts our spending expectations for next year.

We also have some visibility on the increase in compensation expenses generated by the AI talent we hired this year. Based on these two aspects, we have previewed the expected total spending growth for 2026 as well as the expected capital expenditures for 2026.

In terms of total spending, as I mentioned, we expect infrastructure to be the largest contributor to spending growth in 2026. This is primarily due to the sharp acceleration in depreciation expenses in 2026, largely driven by the incremental depreciation from assets purchased and put into use in 2026, as well as the full-year depreciation that will be recognized next year from the infrastructure deployed in 2025.

We also expect that, compared to previous years, the proportion of short-term assets in our capital expenditures for 2025 and 2026 will be larger. Another component of the infrastructure cost growth will come from higher operating expenses, including energy costs, leasing, maintenance, and operating expenses associated with maintaining that equipment. We also expect an increase in cloud service expenditures in 2026 to meet our capacity needs, as well as growth in network-related costs. Therefore, there are many aspects contributing to the infrastructure's contribution to the total spending figure for 2026.

Secondly, employee compensation is the second largest driver of spending growth in 2026. This primarily stems from our investments in technical talent, including the full-year compensation expenses for the AI talent hired this year.

In terms of capital expenditures, the main driver for the increase in our capital expenditures in 2026 will be the expansion of generative AI capacity, as we will build training capacity. This will drive higher spending next year on servers, networks, and data centers. We also expect to continue investing heavily in core AI in 2026. This is a very dynamic planning area, but we wanted to share some preliminary thoughts on the current situation.

A question from Brian Nowak of Morgan Stanley:

I have two questions. First, Mark, regarding the vision for smart labs and superintelligence. Compared to 12 months ago, can you describe any changes in the technical constraints or technical threshold factors that you are most focused on overcoming in the next 24 months to ensure you truly lead superintelligence over the next decade? The second question is for Susan or Mark; you have made many improvements to the core platform to enhance engagement, recommendations, etc. Can you share a few factors you are most looking forward to in the next 18 months that you believe will further enhance engagement on the core platform?

Mark Zuckerberg, Founder, Chairman, and CEO:

In terms of the research agenda and many areas we are focusing on, I believe that focusing on self-improvement is a very important area of research. There are clearly different paradigms of expansion, but I don't want to delve too much into the details of the research we are conducting. I think in developing superintelligence, to some extent, you cannot just learn from humans because what you are trying to build is fundamentally something smarter than humans.

Therefore, it needs to learn how to self-improve, and you need to develop a way for it to be able to self-improve. I think this is very fundamental and will have a very broad impact on how we build products, how we operate the company, the new things we can invent, the new discoveries we can make, and the broader society.

In terms of the overall work structure, I am more convinced that small, talent-dense teams are the best configuration for driving cutting-edge research. This is somewhat different from the setup of our other world-class machine learning systems. If you look at what we do in Instagram, Facebook, or the advertising system, we can very efficiently have hundreds or thousands of people essentially dedicated to improving these systems. We have very mature systems that allow individuals to run tests and be able to test many different things, where each researcher does not need to have the entire system in their head.

But I think for leading research in superintelligence, you really need a small team that can hold the entire system in their heads, which determines certain physical characteristics and operational dynamics of team size. Now let Susan talk about more practical matters.

Susan Li, Chief Financial Officer:

Brian, regarding the forward-looking roadmap for the core recommendation engine, we are currently focusing on several short-term matters.

First, we are focusing on making recommendations more adaptive to the content users are engaging with during conversations, making our pushed recommendations most relevant to what they are interested in at that moment. We are optimizing by matching content with the right audience more quickly after it is published, helping the best content from small creators stand out. We are also improving the system's ability to discover more diverse and niche interests for everyone through interest exploration and learning explicit user preferences. We also plan to further scale the model and incorporate more advanced technologies to improve the overall quality of recommendations.

But we also have many long-term investments in various areas, such as developing foundational models that support multi-service recommendations and integrating large language models more deeply into our recommendation systems. A focus of this work will be optimizing the system for efficiency, allowing us to continue to scale the usage capacity of the recommendation system without diminishing the return on investment we provide.

Host: Here is your next question, from Doug Anmuth at JP Morgan. Please go ahead with your question.

Doug Anmuth:

Thank you very much for answering these questions, one question for Mark and one for Susan. Mark, Meta has been a huge supporter of open-source AI. Has your thinking on this changed as you pursue superintelligence and seek greater returns on significant infrastructure investments? Susan, your comments on capital expenditures for 2026 suggest that there may be over $100 billion in spending next year. Do you still expect to fund all of this yourself, or might there be opportunities for partnerships? Thank you.

Susan Li, Chief Financial Officer: Regarding open source, I don't think our thoughts on this have changed significantly. We have always open-sourced some models, but open sourcing is not everything we do.

So I expect we will continue to produce and share leading open-source models. I also think there are several trends emerging. One is that our models are becoming so large that they are not practically usable for many others. So we are considering whether sharing these is effective or helpful, or if it mainly just helps competitors.

I think there is that concern, and obviously when you approach true superintelligence, I think there is a whole set of different safety concerns that we need to take very seriously, and I mentioned this in my remarks this morning. But I think the bottom line is that I expect we will continue our open-source work. I expect we will continue to be a leader in this area, and I also expect we will continue not to open source everything we do, which is a continuation of the work we have always been engaged in. I think this is indeed a huge investment.

We believe this will be good in the long run, but we take very seriously the issue of converting a large amount of capital into many gigawatts of computing power, and we believe this will help us generate leading research and high-quality products and operate our business. I am indeed looking for opportunities to fundamentally convert capital into the quality of products we can provide to people. But this is indeed a huge bet we are focused on, and we want to ensure that what we build helps create the best products we can offer to the billions of people using our services.

Susan Li, Chief Financial Officer:

Doug, regarding your second question about how we expect to fund the growing capital expenditures for next year, we certainly expect to fund a large portion of it ourselves, but we are also exploring ways to co-develop data centers with financial partners.

We do not have any completed deals to announce, but we generally believe there will be patterns here that attract significant external financing to support large-scale data center projects that leverage our ability to build world-class infrastructure while providing us with flexibility as our infrastructure needs change over time. So we are exploring many different paths.

Host: The next question comes from Justin Post at Bank of America. Please go ahead.

Justin Post:

Thank you. I would like to ask another question about infrastructure. Mark, your spending is now approaching some of the largest hyperscale cloud providers. Do you think all this capacity is primarily for internal use, or do you think there is a possibility of sharing it, or even coming up with a business model to utilize this capacity for external use? And Susan, when you consider the return on investment for these capital expenditures, I am sure you have internal models. I am sure you cannot share all of them But how do you view the return on investment, and are you optimistic about long-term returns?

Susan Li, Chief Financial Officer:

Thank you. Justin, I can answer these two questions first. Obviously, Mark, you should feel free to add at any time. Right now, we are focused on ensuring we have enough capacity for internal use cases, which includes all the core AI work we are doing to support organic content recommendation, supporting all ad ranking and recommendation work. Then, of course, ensuring we are building the training capacity we believe we need to construct cutting-edge AI models, and making sure we are prepared for the types of inference use cases we think we might face, because ultimately we are not only focused on developing cutting-edge models but also on how to scale to the types of consumer use cases we believe will be broadly useful and appealing to users. So, at the moment, we are not really considering external use cases for infrastructure, but that’s a good question.

Regarding your second question, which is mainly about the return on investment for capital expenditures, there are several aspects. In core AI, we continue to see strong returns on investment, our measurement capabilities are quite good, and we are very satisfied with the rigorous measurement and returns we are seeing. In generative AI, we are clearly still in the very early stages of the return curve, and we do not expect generative AI work to become a significant revenue driver this year or next, but we remain very optimistic about the monetization opportunities that will open up overall. Mark talked about the five pillars in his remarks, so I won’t repeat that here. We believe that within a medium to long-term time frame, these opportunities are very adjacent and intuitive to our business today, which is why they represent huge opportunities for us, and each opportunity will have a massive market. The last point I want to add is that we consider substitutability when building infrastructure. Clearly, there are many things you have to build in advance, such as data center shells, network infrastructure, etc. But we will order servers when and as needed, which will ultimately be the largest part of capital expenditures, and make the best decisions at those times to determine where to allocate capacity.

Operator: The next question comes from Mark Shmulik at Bernstein. Please go ahead.

Mark Shmulik:

Thank you for taking my question. Mark, as you pursue the vision of superintelligence, particularly for those of us on the outside, what metrics or KPIs are you tracking to determine if you are on track and making progress? Is this primarily focused on the five pillars you outlined above, or should we consider a broader scope? Susan, clearly AI is providing good returns on investment today, and all these investments are also building towards long-term goals. I wonder if there have been any changes or adjustments in how you view the relationship between revenue or core business performance and the pace of investment? Thank you.

Founder, Chairman, and CEO Mark Zuckerberg:

In terms of what to focus on, internally I will look at the quality of the team members, the quality of the models we are producing, the speed of improvement of our other AI systems in the company, and the extent to which the leading foundational models we build contribute to improving all other AI systems and everything we do around the company. ** Then I think you enter our standard product and business strategy, which is to transform the technology into new products, first expanding to billions of people, and then over time, we will monetize it. But I think there will be some lag. I think this has always been our way of working, whether we are building some new social products or something like Meta AI, or new products around this.

We will focus on achieving leading scale, building the highest quality products, focusing on this for a few years, and then once we are truly confident in that position, we will focus on advancing the business around it. Going back to the previous question, when you compare this business to some cloud businesses, we do have this lag; we focus on building research, then conducting research, and then advancing consumer products, which usually takes some time before we really push the business around it. I think this is a known attribute of our business and the cycle around it. But on the other hand, we believe that if you are building super intelligence, you should use all your GPUs to ensure you are serving your customers well. We believe that through direct generation, we can achieve higher returns than simply leasing or renting infrastructure to other companies.

Susan Li, Chief Financial Officer:

Regarding the second part of your question, we have previously stated that from a profitability perspective, our top priority is to drive long-term growth in consolidated operating profit. This will not be linear. In some years, we will achieve above-average profit growth, while in years when we make significant investments, I think we will see this affect the extent of operating profit growth we can achieve.

Currently, we see many attractive investment opportunities, and we believe these opportunities will prepare us for compelling profit growth for all investors in the coming years. Therefore, while pursuing these investments, we focus on limiting investments elsewhere. But we truly believe that now is the time for us to invest in the future of artificial intelligence, and I think this will not only open up new opportunities for us but also strengthen our core business.

Host:

Your next question comes from Ron Josey of Citigroup. Please go ahead.

Ron Josey:

Great. Thank you for taking my question. Mark, I want to ask you about Meta AI; I think you mentioned in the call about the overall increase in engagement, especially on WhatsApp, where we now have 1 billion users, and the focus now is on driving personalization.

So I want to learn more about how these next-generation models help drive adoption, especially with the behemoth that will be launched at some point. Then when people use AI on WhatsApp, the ideas around search and queries and potential monetization. Thank you.

Mark Zuckerberg, Founder, Chairman, and CEO:

Yes, I won't delve into the roadmap in this area, but basically, we do see that as we continue to improve the models and post-training behind Meta AI, engagement will increase.

When we switch to updated models, when we upgrade from Llama 4 to Llama 4.1, we expect these models to be quite general in nature. So you focus on specific areas, but overall, it will get better at many different things that people want to ask or want to do with it. I think with every version, whether it's the work we do weekly in training or when we launch significant releases of new generations or each generation, engagement will improve. So we focus on that. I won't go into detail about the specific research areas or features we plan to launch in the future, but obviously, I'm very excited about it. Thank you.

Host:

Our last question comes from Youssef H Squali of Truist Securities. Please go ahead.

Youssef H Squali:

Great. Thank you for taking my question. Mark, the Ray-Ban project has been your flagship product so far. How is the progress on the glasses development and the new computing platform you’ve talked about in the past? Is it progressing faster or slower than you expected? When you leverage Meta AI, do you think the glasses will eventually replace smartphones, or do you need a new AI-first form factor? And then Susan, simply put, how do you see FBC developing in the coming years? Is it reasonable to assume it will grow faster than revenue and operating expenses? How do you minimize shareholder dilution? Thank you.

Mark Zuckerberg, Founder, Chairman, and CEO:

Yes, I can talk about the glasses. I'm very excited about the progress we've made. I think the Ray-Ban Meta glasses, as well as my excitement about the Oakley Meta glasses and other products we have planned, are all very promising. This product category is clearly doing well.

I think it applies to many aspects. They are fashionable glasses, so people like to wear them as glasses. They have a range of interesting features. And then the use of Meta AI in them continues to grow. The percentage of people using it daily is increasing. All of this is good. I continue to believe that glasses will essentially become the ideal form factor for AI, because you can have AI see what you see all day, hear what you hear, and talk to you. Once you install a display in them, whether it's a wide holographic view like we showed in Orion or a small display that might fit to show some information.

This will also unlock a lot of value, allowing you to interact with AI systems in this multimodal way throughout the day. It can see the content around you. It can generate user interfaces for you, displaying a lot of information. Then it can show you information and provide assistance.

Personally, I feel like wearing glasses is like wearing contact lenses. I feel that if my vision is not corrected, I would be at some cognitive disadvantage in this world. I think in the future, if you don't have glasses equipped with AI or some way to interact with AI, I think it will be similar Compared to others you work with or compete against, we may be at a significant cognitive disadvantage.

So I think this is a fairly fundamental form. It has many different versions. Right now, we are building products that I think are stylish but not focused on display. I believe there is a whole set of different content about display that needs to be explored.

This is what we have been doing through Reality Labs for the past 5 to 10 years, basically researching all these different things. Ten years ago, I would say one of the amazing aspects of glasses is that they would be an ideal way to merge the physical and digital worlds. So I think the whole metaverse vision will ultimately become extremely important, and AI will accelerate this. Just if you had asked me five years ago whether we would first have holograms creating immersive experiences or superintelligence, I think most people would have thought you would get holograms first.

This is an interesting peculiarity of the tech industry, and I think we will ultimately have truly powerful AI first, but because we have been investing in this area, I believe we are several years ahead in building glasses. I think this is something we are very happy to continue investing heavily in because I believe it will be a very important part of the future.

Kenneth Dorell, Director of Investor Relations:

Youssef, we didn’t quite catch your second question. Could you repeat it?

Youssef H Squali:

Sure, with all this hiring, when you look at the stock-based compensation expenses in the coming years, I think we will see this part of the expense growing at a rate that may significantly outpace revenue growth. I wonder what you plan to do to minimize shareholder dilution. Mainly buybacks or other measures? Thank you.

Susan Li, Chief Financial Officer:

Thank you, Yusuf. This year, the impact of increased compensation costs from our AI hiring, including stock-based compensation, has already been reflected in the revised 2025 expense outlook. In my comments on the 2026 expense outlook, these are clearly important drivers of expense growth for 2026, as we will confirm the full-year compensation for the additional talent brought in. That said, we have incorporated these factors into our expense outlook. We are indeed very focused on the dilution issue. We generally believe that our strong financial position will enable us to support these investments while continuing to repurchase shares as part of a buyback program to offset equity compensation and provide quarterly cash dividend distributions to investors.

Kenneth Dorell, Director of Investor Relations:

Great. Thank you all for joining the meeting today. We look forward to speaking with you again soon. The meeting is adjourned