Meta conference call: The recommendation system is being rewritten by large models, and a future without smart glasses is "hard to imagine."

On January 28, Meta announced strong fourth-quarter results, with advertising driving revenue for Q4 last year and guidance for Q1 this year exceeding analysts' expectations. Additionally, the expected range for full-year capital expenditures was also above analysts' expectations, leading to a more than 11% increase in the company's stock price in after-hours trading. During the subsequent conference call, a clear signal was sent: while Meta is delivering strong performance in 2025, it is upgrading the logic of its next phase of growth from "a better advertising system" to "a product and infrastructure reconstruction centered on personal super intelligence." In the short term, revenue is still driven by advertising; in the medium to long term, recommendation systems, intelligent agents, content production methods, smart glasses, and computing power supply will collectively determine whether Meta can open up new growth curves. Zuckerberg clearly positioned AI smart glasses as the next generation of core computing terminals, comparing it to the key moment when smartphones replaced feature phones. The strategic focus of Reality Labs is shifting from the metaverse to AI wearable devices and self-developed models. He also previewed that in the coming months, Meta will begin to release new models and products. AI Glasses Will Reach the "Smartphone Moment" Zuckerberg has repeatedly mentioned the company's long-term judgment in the field of AI smart glasses, comparing it to the key moment when smartphones replaced feature phones. There are billions of people around the world who wear glasses or contact lenses to correct their vision. I believe we are at a stage similar to when smartphones first appeared, when everyone could see that it was only a matter of time before flip phones were replaced by smartphones. It’s hard to imagine that in a few years, most people will not be wearing AI glasses. He also revealed that sales of Meta's smart glasses have tripled over the past year, calling them "one of the fastest-growing consumer electronics products in history." The conference call indicated that Meta is shifting the investment focus of Reality Labs from its previous metaverse vision further towards AI wearable devices and self-developed AI models. The New Generation of AI Models and Products Will Be Launched in the "Coming Months" In response to investors' concerns about the return on AI investments, Zuckerberg emphasized that the results of the AI lab will soon be made available to users. This will be an important year for delivering personal super intelligence, accelerating business growth, building future infrastructure, and reshaping the way the company operates. He added that by 2025, Meta will have completed the reconstruction of its AI system, and will then begin to continuously release new models and products, pushing "personal super intelligence" into practical application. The "First Principle" of Growth: Recommendation Systems Are Being Rewritten by Large Models Meta breaks down growth into two main lines: enhancing interaction and improving monetization efficiency, with both sharing a common foundation in the upgrade of recommendation/ranking models. Interaction Side: Systematic Efficiency Improvement in Content Recommendation Instagram Reels viewing time in the U.S. increased by 30% year-on-year; Facebook video viewing time maintained double-digit growth. By "simplifying the ranking structure and supporting longer interaction history," Facebook's organic feed and video post views were boosted by 7% in Q4, with management emphasizing this as the largest single-quarter revenue-driving product optimization in two years. "Freshness/originality" is treated as a key variable: 75% of recommended content on Instagram in the U.S. comes from original sources; the number of Reels pushed on Facebook increased by over 25% month-on-month. The time spent on Threads increased by 20% after optimization in Q4, beginning to establish a more sustainable growth structure. Advertising side: From "large model training" to a "runtime inference" layered system Meta stated clearly: large models like GEM are not suitable for real-time inference due to high costs, therefore, the "knowledge of large models" must be transferred to lighter runtime models—this determines whether computational investment can be converted into ROI. Disclosed progress includes: The number of GPUs for GEM training has doubled; a new sequence learning architecture has been introduced (longer user behavior sequences, richer content information). Q4 results: Facebook ad clicks increased by 3.5%, Instagram conversion rate increased by over 1%; the new runtime model on Instagram brought a 3% increase in conversion rate. Lattice promotes model unification, combined with backend transformation, improving ad quality by 12%; more models will continue to be integrated by 2026, using "fewer but stronger" models for global operations. Core implication: Meta is turning the "recommendation system" into a scalable engineering system similar to LLMs—this not only determines user duration but also affects ad pricing and conversion rates. Monetization beyond advertising: but "advertising remains the largest certainty" Zuckerberg acknowledged: advertising will still be the most important growth driver in the coming years. However, Meta is also laying the groundwork for commercialization "beyond advertising," especially targeting the enterprise side. Two clear incremental paths: Enterprise messaging and WhatsApp commercialization: WhatsApp's paid messaging service annual revenue has surpassed $2 billion, and Click-to-message ads are accelerating growth, up 50% year-on-year in the U.S. Business AI is being promoted early in Mexico/Philippines: over 1 million conversations per week, with plans to expand the market and enable more tasks to be completed within WhatsApp. Threads and WhatsApp advertising catch-up: Threads advertising will expand to all remaining countries/regions (including the UK, EU, Brazil). WhatsApp advertising and status features will be fully promoted within the year, but short-term ad load will remain restrained, focusing first on optimizing effectiveness. Additionally, an important detail was mentioned during the call: Zuckerberg proactively pointed out the acquisition of Manus, and presented it as a case for "enterprise tool subscription/integration," aiming to integrate commonly used enterprise tools into the advertising and business management platform, enhancing the "integrated delivery capability" for small and medium-sized enterprises. This indicates that Meta is attempting to transition itself from a "traffic platform" to an "enterprise growth operating system." The Biggest Bet: Computing Power and Infrastructure, Capex Guidance Significantly Upgraded The most "asset-heavy" part of this conference call is Meta's long-term commitment to computing power: 2026 capital expenditure guidance: $115 billion–$135 billion (including principal of finance leases), clearly indicating a significant year-on-year increase, mainly for superintelligent laboratories and core business. Capacity bottlenecks still exist: demand is growing faster than supply; new capacity from self-built projects will not come online until later in 2026, and most of the time before that will remain tight. Chips and architecture: Andromeda can now run on NVIDIA, AMD, and MTIA; MTIA plans to expand from inference to training workloads. Establishing Meta Compute, emphasizing strategic advantages through engineering design, investment, and collaborative construction, viewing "silicon and energy" as part of long-term capabilities. Capital allocation is also very clear: management explicitly stated that the current priority is to invest resources in AI leadership, with buybacks remaining opportunistic rather than a priority (which echoes the lack of buybacks this quarter). Here is the full transcript of the conference call: Mark Zuckerberg, Founder, Chairman, and CEO: Alright, hello everyone. Thank you all for tuning in. By the end of 2025, we achieved strong performance, with at least 3.5 billion users using one of our applications every day. Daily active users for both Facebook and WhatsApp exceed 2 billion, and Instagram's daily active users are close to that number. Thanks to record demand during the holiday period and performance improvements brought by artificial intelligence, our business has also performed excellently. Currently, artificial intelligence is in a rapid development phase. I expect this wave to accelerate further in multiple ways in 2026. We are beginning to see agents truly come into play. This will enable us to develop entirely new products and fundamentally change the way we work. In 2025, we rebuilt the infrastructure for our AI projects. In the coming months, we will begin to release new models and products. I expect our first models to perform well, but more importantly, we will demonstrate our rapidly developing momentum. As we continue to release new models, I expect us to make steady progress this year, continually pushing the industry frontier. I am very excited about the products we are building. Our vision is to build personal superintelligence. We are beginning to see the potential of artificial intelligence, which can understand our personal backgrounds, including our experiences, interests, content, and relationships. The value of agents largely lies in their ability to understand unique contextual information. We believe that Meta will be able to provide a unique personalized experience. We are also working to integrate large models with Facebook, Instagram Threads, and the recommendation system of our advertising system. Our world-class recommendation system has significantly driven the growth of our applications and advertising business, but we believe that the current system is still quite primitive compared to what is about to be realized. Today, our system helps people stay connected with friends, understand the world, and find interesting content But soon, we will be able to understand users' unique personal goals and customize the information flow based on each individual's needs, showing them content that helps improve their lives. This will also have an impact on businesses. Today, our ads can help businesses accurately find target audiences interested in their products. The new smart shopping tools will enable users to find the most suitable specific products from merchants in our catalog. We are committed to making these experiences applicable to both our information flow and business messaging features, significantly enhancing WhatsApp's functionality over time. Additionally, more types of content will be launched in the future. People crave to express themselves and experience the world in the most immersive and interactive ways. We initially used text, then shifted to photos with the popularity of camera-equipped phones, and later to videos as mobile network speeds improved. In the near future, we will see more new media forms that are immersive and interactive, all thanks to advancements in artificial intelligence. Our information flow will become more interactive. Today, our app feels like an algorithm recommending content. Soon, when you open our app, you will find that artificial intelligence not only understands your needs but also pushes exciting content to you, even tailoring high-quality content for you. Smart glasses are the ultimate embodiment of this vision. They can see what you see, hear what you hear, converse with you, and assist you in your daily life, even displaying information or generating customized user interfaces directly in your line of sight. Last year, our smart glasses sales more than doubled, and we believe they are one of the fastest-growing consumer electronics products in history. Millions of people wear glasses or contact lenses to correct their vision, and I believe we are at a stage similar to the birth of smartphones; clearly, it is only a matter of time before all these feature phones turn into smart devices. It is hard to imagine that in a few years, most of the glasses people wear will not be AI glasses. For Reality Labs, we will invest most of our resources in the field of glasses and wearable devices in the future, while focusing on making Horizon a huge success on mobile and turning VR into a profitable ecosystem in the coming years. I expect Reality Labs' losses this year to be similar to last year, which is likely the peak of our losses, as we will gradually reduce losses while continuing to advance our vision. Looking ahead, we will continue to invest heavily in infrastructure to train leading models and provide personalized super intelligence for billions of people and businesses worldwide. I recently announced the establishment of Meta Compute because I believe that achieving maximum efficiency in the engineering design, investment, and collaborative construction of infrastructure will become our strategic advantage. Dina Powell McCormick has also joined us as President and Vice Chair, and she will lead our collaborations with governments, sovereign nations, and strategic capital partners to expand our long-term capabilities, including ensuring that we have a positive economic impact on the communities where we operate globally. A key component of Meta Compute is long-term investments in silicon and energy We will continue to work hand in hand with important partners while advancing our own chip research and development projects. We are building a system architecture to ensure the flexibility of system usage, and by optimizing technology and supply chains, we expect the cost per gigawatt to significantly decrease over time. Finally, I believe that 2026 will be the year when artificial intelligence begins to fundamentally change the way we work. In addressing this transformation, we have always adhered to a clear goal: to create the best platform for individuals to exert significant influence. To this end, we are investing in AI-native tools so that Meta employees can enhance their work efficiency. We are elevating the status of individual contributors and simplifying team structures. We are starting to see that projects that previously required large teams can now be completed independently by a single talented employee. I hope as many outstanding talents as possible choose Meta, where they can exert their maximum influence and provide personalized products for billions of users worldwide. If we do this, I believe we will accomplish more work, and it will be more enjoyable. Well, that’s all I wanted to say. This year will be significant for us in delivering personal superintelligence, accelerating business development, building future infrastructure, and shaping the operational model of the company. As always, I appreciate the hard work of our team and thank everyone for working together with us. Now, I would like to invite Susan. Susan Li, Chief Financial Officer: Thank you, Mark, and good afternoon, everyone. Let’s first look at the performance results of each segment. Unless otherwise noted, all comparisons are year-over-year data. Our family of apps continues to grow, with an estimated 3.5 billion people using at least one of our apps daily in December. Total revenue for the family of apps in the fourth quarter was $58.9 billion, a year-over-year increase of 25%. Advertising revenue for the family of apps in the fourth quarter was $58.1 billion, a year-over-year increase of 24%, and a 23% increase when adjusted for constant currency. In the fourth quarter, the total ad impressions across our services grew by 18%. Ad impressions showed healthy growth across all regions, primarily driven by user engagement and user growth, followed by ad loading optimization. The average price per ad increased by 6% year-over-year, benefiting from increased advertiser demand, which was mainly driven by improved ad effectiveness. Other revenue from the family of apps was $801 million, a year-over-year increase of 54%, primarily due to growth in WhatsApp paid messaging revenue and an increase in Meta verified subscription users. In our Reality Labs segment, fourth-quarter revenue was $955 million, a year-over-year decrease of 12%. As we mentioned in our last conference call, the reason for the year-over-year decline in Reality Labs revenue is that we delayed the launch of Quest 3S to the fourth quarter of 2024, and retail partners are procuring Quest headsets in the third quarter of 2025 in preparation for the holiday season, which has been accounted for in the third quarter. Next, let’s look at our consolidated performance. Total revenue for the fourth quarter was $59.9 billion, a year-over-year increase of 24%, and a 23% increase when adjusted for constant currency Total expenditure in the fourth quarter was $35.1 billion, a year-on-year increase of 40%. The year-on-year growth was mainly driven by employee compensation, legal fees, and infrastructure costs. The increase in employee compensation reflects the addition of technical talent this year, particularly AI engineer Tama. The rise in legal fees was due to the reversal of legal expenses in the fourth quarter of 2024 and the accrual of expenses for the fourth quarter of 2025. The increase in infrastructure costs was primarily driven by higher depreciation, cloud spending, and other operating expenses. At the end of the fourth quarter, our employee count exceeded 78,800, a year-on-year increase of 6%, mainly due to recruitment and growth in key areas such as profitability, infrastructure, Meta Superintelligence Labs, and regulatory compliance. Operating revenue for the fourth quarter was $24.7 billion, with an operating profit margin of 41%. Interest and other income for the fourth quarter was $609 million, mainly from unrealized gains on equity investments. Due to a settlement with tax authorities, the tax rate for this quarter was 10%, slightly lower than our previous expectation of 12% to 15%. Net profit was $22.8 billion, or $8.88 per share. Capital expenditures (including principal payments on finance leases) were $22.1 billion, primarily for investments in data centers, servers, and network infrastructure. Free cash flow was $14.1 billion. At the end of this quarter, the company held $81.6 billion in cash and cash equivalents, with liabilities of $58.7 billion. Now, let's look at business performance. Our revenue growth was driven by two main factors: first, our ability to provide interactive experiences for users, and second, our continued effectiveness in converting these interactions into revenue. First, we continuously enhance user engagement through ranking and product improvements. Instagram Reels performed strongly this quarter, with viewing time in the U.S. increasing by over 30% year-on-year. The increase in user engagement was due to several optimizations we made to improve recommendation quality, including simplifying the ranking architecture for more efficient model scaling. This allowed our system to consider longer interaction histories, thereby better identifying user interests. In the U.S., Facebook's video viewing time continued to show double-digit year-on-year growth, and our ranking and product optimization efforts on the news feed and video platforms also achieved significant results. The optimizations in the fourth quarter led to a 7% increase in the views of organic news feed and video posts on Facebook, marking the largest quarterly revenue growth from Facebook product releases in the past two years. We continue to enhance the freshness and originality of content recommendations. On Facebook, the number of Reels short videos pushed by our system on the same day increased by over 25% compared to the previous quarter. On Instagram, the proportion of original content in the U.S. increased by 10 percentage points in the fourth quarter, with 75% of recommended content coming from original posts. Threads also showed strong growth momentum, benefiting from improvements in the recommendation system. The optimizations in the fourth quarter led to a 20% increase in user retention time on Threads. Looking ahead to 2026, we see tremendous growth opportunities. This includes expanding the complexity and scale of the training data used in the model while continuously enhancing the system's responsiveness to users' real-time interests We are also committed to integrating large models to gain a deeper understanding of the content on our platform, thereby achieving more personalized recommendations. Another important area of investment this year is the development of next-generation recommendation systems. We have invested heavily in this area, including building a brand new model architecture based on LLM from scratch, leveraging the world knowledge and reasoning capabilities of LLM to more accurately infer user interests. In addition to improving the recommendation system, we also plan to utilize models developed by Meta Superintelligence Labs to create engaging and unique AI products. Currently, we have seen significant potential in AI video localization and dubbing. We currently support nine languages, with hundreds of millions of users watching AI-translated videos daily. This has significantly increased user engagement on Instagram, and we plan to launch more language support within this year. Additionally, our media creation tools have received strong feedback. Nearly 10% of the Reels videos watched by users daily come from our Edits app, which has nearly tripled compared to the previous quarter. In terms of Meta AI, the number of daily active users creating media content in the fourth quarter has tripled year-on-year. This year, we plan to enhance the performance of the underlying media generation models and launch new features to further improve the product experience. Another focus area for Meta AI is personalization. Early tests indicate that personalized responses can significantly enhance user engagement, and we plan to greatly enhance Meta AI's personalization capabilities this year. This complements our investment in content understanding, which will enable our system to gain deeper insights into each user's interests and preferences while identifying the most relevant content on the platform for responses. Next, let's discuss the second factor affecting revenue: improving monetization efficiency. The first step in this work is to optimize the ad placement ratio in organic traffic. Currently, our focus remains on adjusting the system to find the best times and places to serve ads. In some cases, this allows us to increase the overall ad volume while maintaining user experience. However, an increasingly important part of this work is how to find opportunities to improve conversion rates by determining when users are more willing to watch ads, without changing the ad volume. In fact, the cross-user and cross-session ad reallocation initiative we are implementing on Facebook in the second half of 2025 is expected to generate revenue growth nearly four times that of the growth in Facebook ad volume. We are also continuously advancing the introduction of ads into our latest services. On the Threads platform, we began expanding ads to all remaining countries/regions this month, including the UK, EU, and Brazil. On the WhatsApp platform, we expect to complete the full rollout of ads and status features within the year, with ad volume remaining low in the short term as we follow standard strategies to optimize ad formats and effectiveness before gradually increasing ad inventory. Next, we will focus on the second phase of improving monetization efficiency, which is to enhance the performance of businesses using our tools. We have seen significant results from investments in ad effectiveness for the full year of 2025, with year-on-year conversion rate growth accelerating in the fourth quarter We expect that a series of investments in 2026 will enable us to further enhance our performance as we continue to integrate artificial intelligence into all aspects of marketing and customer interaction channels. First is our advertising system, where we will continue to expand the complexity and scale of the models to better select which ads to display. In the fourth quarter, the number of GPUs we used to train the GEM model (for ad ranking) doubled. Additionally, we adopted a new sequence learning model architecture that can handle longer user behavior sequences and process richer information about each piece of content. In the fourth quarter, improvements in GEM and sequence learning technology jointly drove a 3.5% increase in Facebook ad clicks and over a 1% increase in Instagram conversion rates. This brand-new sequence learning architecture is significantly more efficient than our previous architecture, which will allow us to further scale the data volume, complexity, and computational power of future ranking models, thereby enhancing ad effectiveness. While scaling foundational ad models like GEM, we are also developing more advanced models for downstream ad inference at runtime. In the fourth quarter, we launched new runtime models in Instagram feeds, Stories, and Reels, resulting in a 3% increase in conversion rates in the fourth quarter. With the support of Lattice, our model unification efforts are also progressing. Following the successful integration of Facebook feed and video models in the first half of 2025, we integrated Facebook Stories and models from other platforms into the overall Facebook model in the fourth quarter. This initiative, along with a series of backend improvements, led to a 12% increase in ad quality. In 2026, we expect to integrate more models than in the past two years as we continue to improve the system to run fewer but more powerful models. Next is the advertising product area. We continue to invest in helping businesses leverage artificial intelligence to reduce the difficulty of setting up and optimizing advertising campaigns. In the fourth quarter, we began testing the MetaAI Business Assistant with advertisers, which can assist with tasks such as ad campaign optimization and account support. In the coming months, we will open this assistant to more advertisers to ensure that every business can experience and use it. In the coming months, we will open this service to more advertisers, allowing every business to have an AI assistant to interact with. This assistant can remember the business's goals and provide personalized performance improvement suggestions. We are also deploying AI in the advertising creative space to enhance performance. In the fourth quarter, the annualized revenue of our video generation tool reached $10 billion, with a growth rate nearly three times that of overall advertising revenue growth. Additionally, our incremental attribution feature has also achieved significant results, enabling real-time optimization of incremental conversions. The latest model we launched in the fourth quarter improved conversion rates by 24% compared to standard attribution models. This product has already reached billions in annual revenue just seven months after its launch. The last profitable area I want to introduce is business messaging services, where our entire solution portfolio has shown strong growth momentum In the fourth quarter, the revenue from click-to-message ads accelerated, with a year-on-year growth of over 50% in the United States, primarily due to the widespread application of message ads on our website. These ads guide users to the business's website for more information before initiating a chat. The paid messaging service within WhatsApp also continued to grow, with annual revenue surpassing $2 billion in the fourth quarter. Lastly, our business AI has made good early progress in Mexico and the Philippines, with over 1 million user interactions with business AI on our messaging platform each week. This year, we will expand business AI to more markets while enhancing its capabilities to not only answer questions about product inventory but also help users efficiently complete various tasks within WhatsApp. We often talk about how AI improves our products, but I would like to take this opportunity to briefly introduce how it is changing the way we work. Mark mentioned that we are committed to making Meta a platform where everyone can have a significant impact. One of our key focuses is to promote and develop our AI coding tools, where we are currently seeing strong growth. Since the beginning of 2025, the output per engineer has increased by 30%, with most of the growth coming from the adoption of agent coding, which saw significant growth in the fourth quarter. The growth of advanced users of AI coding tools is even more pronounced, with their output increasing by 80% year-on-year. We expect this growth momentum to accelerate over the next six months. Next, I would like to discuss our capital allocation strategy. In 2026, we have significant opportunities to enhance our core business. We plan to continue prioritizing investments in our core business to support these opportunities while preparing for a new product cycle driven by AI models in the coming years. Ensuring sufficient infrastructure capacity is at the core of these initiatives, and we are working to meet our chip needs by deploying various chips to optimally support our different workloads. To this end, in the fourth quarter, we expanded the Andromeda ad retrieval engine, which can now run on NVIDIA, AMD, and MTIA chips. This improvement, along with model innovations, has nearly tripled Andromeda's computational efficiency. In the first quarter, we will expand the MTIA project to support not only the inference workloads currently running but also our core ranking and recommendation training workloads. More broadly, while investing in infrastructure to meet business needs, we will continue to prioritize maintaining long-term flexibility to adapt to market changes. We are addressing challenges in various ways, including changing the way we build data centers, establishing strategic partnerships, signing cloud capacity contracts, and creating new ownership structures for some large data centers. We have a strong net cash balance and expect that by 2026, our business will continue to generate ample cash flow to support our infrastructure investments, which is already reflected in our expectations. Nevertheless, we will continue to seek opportunities to regularly supplement our strong operating cash flow through prudent, cost-effective external financing, which may ultimately allow us to maintain a positive net debt balance Next, let's look at the financial outlook. We expect total revenue in the first quarter of 2026 to be between $53.5 billion and $56.5 billion. Based on current exchange rates, our forecast assumes that foreign exchange factors will contribute approximately 4% to year-over-year total revenue growth. Next is the outlook for expenses and capital expenditures. We expect total expenses for the full year of 2026 to be between $162 billion and $169 billion. Most of the expense growth will be driven by infrastructure costs, which include third-party cloud spending, higher depreciation expenses, and increased infrastructure operating expenses. The second largest factor driving expense growth is employee compensation, primarily due to investments in technical talent. This includes hiring employees in 2026 to support our key areas, particularly artificial intelligence, as well as the full-year costs of employees hired in 2025. By business segment, we expect expense growth to be primarily driven by the Family of Apps, while operating losses in Reality Labs will remain roughly in line with 2025 levels. We expect capital expenditures in 2026 (including principal payments on finance leases) to be between $115 billion and $135 billion, with year-over-year growth primarily driven by increased investments in the Metaverse Super Intelligence Lab and core business. Despite significant growth in infrastructure investment, we expect operating income in 2026 to be higher than in 2025. Assuming no changes in the tax environment, we expect the full-year tax rate for 2026 to be between 13% and 16%. Additionally, we recently reached an agreement with the European Commission to further adjust our lower-personalization advertising services, which will begin to roll out gradually this quarter. However, we are still closely monitoring adverse legal and regulatory factors in the EU and the U.S. that could have a significant impact on our business and financial performance. For example, we continue to pay attention to issues related to teenagers, and we plan to conduct several trials in the U.S. this year, which could ultimately lead to significant losses. In summary, 2025 is expected to be another strong year for our company. The investments we are making to improve the business are continuing to drive strong growth, and this year we have an exciting roadmap to provide new experiences and services to our global community. As always, we thank our team for their hard work and steadfast commitment to our mission. Krista, let's start taking questions next. Operator: Thank you. We will now open the Q&A session. (Operator instructions) (Operator instructions) Your first question comes from Brian Nowak of Morgan Stanley. Please go ahead. Q1 Brian Nowak: Thank you for taking my question. I have one question for Mark and one for Susan. Mark, the first question is about long-term development. Considering your plans to scale up investments and seize opportunities in personal intelligence and the metaverse, could you briefly outline which long-term revenue or return on invested capital (ROIC) you value most through these investments over the next three to five years or even ten years? Susan, the second question is about the near term, specifically 2026. I believe your guidance business is the fastest-growing in the past five years I know you have made many improvements in recommendation and profitability efficiency, but could you help us analyze what the two or three main drivers of revenue growth will be in 2026? Mark Zuckerberg, Founder, Chairman, and CEO: Okay, I think I can answer the first question first. But I have to say that I feel my answers to many of your questions in this call may not be very satisfying because we are in an interesting period where we have been rebuilding our AI projects for six months, and I am satisfied with the current progress. However, in the coming months, we will gradually roll out the first models, products, and businesses around these topics. At that time, I will share more information on all these aspects. So, I am happy to provide an overview first. But I want to say in advance that this content may not be particularly detailed, but as we gradually roll out, I believe it will be exciting. I think, in terms of business, I mean—I don’t think I will present any new insights here, but—we are focusing on several important business opportunities. I think one of them is improving core products and accelerating the development of core businesses. I have talked before about connecting recommendation systems and LLMs, and I believe this will enhance both the organic traffic experience and the quality of ads. We will see advancements in media generation technology improving content quality, and combined with improvements in recommendation systems, we expect the quality and efficiency of our core business to improve overall, both for organic traffic users and enterprise users. So I think this will create a chain reaction. Then, I believe many new business opportunities will emerge. I mean, we have been studying meta AI, and I think you have already begun to see some monetization models for such products across the industry. When we develop this technology to the scale and depth we want, we believe there will be many opportunities in subscriptions, advertising, and various other aspects. Yes, I have a lot of excitement about shopping and e-commerce, which I mentioned in my earlier comments. With the rollout of models and the demonstration of capabilities, whether it’s the first models or subsequent models in the coming year, I believe these models will become more refined. We will be able to combine different products with these models to better serve different businesses—including those using our platform and those directly facing consumers. I think it’s worth mentioning that we haven’t seemed to mention the acquisition of Manus before. I think the acquisition of Manus is a great example that can help many businesses accelerate their growth. Currently, many businesses are already paying for Manus's tools, and we will integrate these tools into our advertising and business management platform. This way, we can provide a more integrated solution for millions of businesses that use and rely on our platform, which will be very powerful, not only helping them accelerate business growth with existing products but also further enhancing performance through new product lines. So, my answer is quite general; I believe if we do our job well this year, things will become clearer and more exciting Susan Li, Chief Financial Officer: Brian, regarding your second question, the revenue expectations for Q1 2026 clearly encompass a variety of possible outcomes. Overall, it reflects our expectations for strong growth in that quarter. This expected range reflects the prospects for accelerated growth, primarily driven by sustained strong demand at the end of Q4 and early 2026. Additionally, we anticipate that foreign exchange factors will contribute a positive 4 percentage points to year-over-year growth. This is 3 percentage points higher than the positive impact experienced during the dollar's strength in Q4 2025. But overall, we see advertisers responding positively to the improvements we have made in advertising effectiveness. They are driving strong growth in conversion rates. We made significant investments in 2025 in areas such as improving ad ranking and delivery systems, more effectively reallocating ad load, launching new features in advertising products like Advantage+, improving measurement methods, and many other excellent efforts that help drive sustained advertising performance. Operator: Your next question comes from Eric Sheridan at Goldman Sachs. Please go ahead. Q2 Eric Sheridan: Thank you very much for taking this question. If I may, I would like to ask two questions. You previously mentioned that the company has limited internal capacity and that computing power is insufficient to meet the goals you have set for your platform and products. I would like to understand how you currently view the alignment between your internal demand for computing power and your roadmap. The second question is, as the advertising business continues to scale, especially in terms of annual revenue growth, have we seen the full direct impact of the scaling of the business on the enhancement of computing power? Or how should investors view the directional relationship between the enhancement of computing power and the improvement of profitability? Thank you. Susan Li, Chief Financial Officer: Regarding your first question, we do still face capacity constraints. Our team did an excellent job upgrading infrastructure during 2025, but the demand for computing resources is growing even faster than our supply. Therefore, we expect that by 2026, as the cloud business expands, this year's computing capacity will significantly increase. However, until the additional capacity of our own facilities comes online later in 2026, our computing capacity may still be constrained for much of 2026. Nevertheless, I believe we have effectively mitigated the impact of computing capacity constraints on the business internally. I expect this situation to continue into 2026. We are continuously enhancing infrastructure efficiency from multiple angles, including optimizing workloads, increasing infrastructure utilization, diversifying chip supply, and investing in efficiency improvements in core technology research and development in areas like content and ad ranking. That addresses your first question. The second question is about how the advertising business is scaling. I think I cannot provide a very precise answer to that What I want to say is that one of the ways we enhance advertising effectiveness is by improving our large models and the lightweight models used for runtime advertising inference. We typically do not use large model architectures like GEM for inference because their scale and complexity lead to excessive costs. Therefore, the way we enhance the performance of these models is by leveraging them to transfer knowledge to smaller lightweight models used at runtime. However, I believe there is still room for improvement in our large models, which could benefit from more computational resources. As we increase the computational resources available for these models and the foundational models supporting various stages of ad ranking and recommendations, we expect performance to improve. Operator: The next question comes from Mark Shmulik at Bernstein. Please go ahead. Q3 Mark Shmulik: Mark, you mentioned that you expect some substantial changes in the way we work and operate this year. So, if by the end of the year we still do not see substantial progress or widespread adoption of some of the new products and initiatives you launched, would you be surprised? Or should we be more patient with the timeline? Susan, your previous guidance was — I still expect the growth rate this year to be faster than last year — assuming that a few months from now we realize we need more investment and resources to continue seizing the opportunities brought by artificial intelligence, but the macroeconomic situation may appear slightly weak. So, how closely related are investment levels to core performance? Thank you. Mark Zuckerberg, Founder, Chairman, and CEO: I think the first question is asking me when I expect the products to have an impact. I mean, we will be rolling out new products throughout this year. I think it’s important to note that we are not just launching a single feature, but we are building multiple features. Artificial intelligence will bring many brand new experiences. I have previously outlined some of these themes in my discussions about personal AI, LLMs, and recommendation systems. I believe this will be a relatively long-term research project that will yield returns over a long period. But as we integrate more improvements and advancements from AI research into the recommendation systems, we have already seen significant optimizations in the recommendation systems. Content will improve, there will be new forms, and the glasses will also be enhanced. We will try many different aspects, including some that we believe are entirely new and not just simple extensions of existing features. So, yes, I mean, I expect these features to be rolled out throughout this year. Sometimes, products need to go through several iterations to truly achieve market recognition and find the necessary market fit. But I believe we have enough time, and we started early in the year, so I expect that by the end of the year, whether in terms of content or work, we will achieve some success. We have discussed before that I think it’s difficult to accurately predict how the organization’s working patterns will change. But I believe that intelligent agents are now starting to really play a role, which is significant, and I think this will bring about substantial changes. We are already beginning to see that the productivity of those who adopt intelligent agents has significantly increased. There is a huge gap between those who can do this well and those who cannot. I believe this will have a profound impact on the entire industry and even the whole economy, affecting the productivity and efficiency with which we operate these companies. I hope we can leverage this to accomplish more than ever before. My main concern is to ensure that Meta becomes a great company that can make a significant impact. You can use these smart tools anywhere, but only by joining a company like Meta can you deliver products to billions of users, and there are not many companies like Meta. So I believe that if we can succeed—if we can harness these tools—then over time, we should see a significant increase in output. Now, it is indeed a bit difficult to accurately predict when this technology will become mainstream, right? I won't predict a specific quarter or anything like that, but the trend indicates that it is inevitable. For me, this is very exciting. As I said before, to be honest, it’s also quite interesting, right? I think being able to build a lot of things will make things more fun. And that’s exactly why we are here. Susan Li, Chief Financial Officer: Thank you, Mark. Regarding your second question, I want to confirm and clarify something. I remember you mentioned in your question that the revenue growth rate for 2026 would be higher than that of 2025. I want to make sure my explanation is very clear. We expect the revenue for 2026 to be higher than that of 2025. What is being compared here is the absolute amount, not the growth rate. For clarity, we achieved strong revenue growth right at the beginning of 2026. Of course, we are just a few weeks into this, and it is against the backdrop of a favorable macroeconomic situation. Therefore, it is obviously difficult to extrapolate the current trend to the entire year. Additionally, there are many variables in the current situation. We are indeed taking full advantage of the current business strengths and reinvesting a significant amount of revenue into what we believe are highly attractive opportunities in artificial intelligence infrastructure and talent investment. Since we are still weighing various capacity options, it is difficult to assess the specifics of all investment opportunities this year. Of course, the current hiring market remains very competitive, but we hope to invest actively within our means. We will continue to follow the framework proposed a few years ago, which is to continuously improve consolidated operating profit over time to guide these investments. According to our current plans, we expect the revenue for 2026 to exceed that of 2025. Operator: Your next question comes from Doug Anmuth of JP Morgan. Please go ahead. Q4: Doug Anmuth: Thank you very much for answering these questions. I have a question for Mark and one for Susan. Mark, could you elaborate on the progress of the MSL team over the past few months and your views on moving towards the frontier model this year? Susan, I know you expect revenue to grow in 2026. Do you also expect positive free cash flow? How should we view the current and future joint ventures in data center and computing facility construction? Thank you Mark Zuckerberg, Founder, Chairman, and CEO: I don't have much to add about the current progress. What I mean is, as I mentioned at the beginning, it might be a bit inappropriate to answer these questions right now. The MSL project has been underway for about six months, and I am very satisfied with the quality of the team. I believe we have the strongest R&D team in the industry, and some early indicators are showing positive trends. However, I think this is a long-term endeavor. Our purpose in doing this is not to launch a single model or product. We will develop many models and various products over time. I want to ensure that the work itself speaks for everything, and we also hope everyone understands that this is a journey. The first batch of products we launch is more about showcasing the trajectory we are on rather than the results of a specific moment. So, yes, I am optimistic, but there are no other particularly specific updates to share at this time. Susan Li, Chief Financial Officer: Doug, regarding the first part of your question, we have made significant investments in infrastructure this year to support our AI projects. We believe that with this year's business cash flow, we are fully capable of strongly supporting these investments. At the same time, we will continue to explore different paths to continuously improve our infrastructure to provide the long-term flexibility and optionality we need in the context of various capacity demands that may arise in the coming years. Therefore, we have no other news to announce at this time. You know, we are exploring all different opportunities to build and develop capabilities within the various timeframes we need. Operator: The next question comes from Justin Post at Bank of America. Please go ahead. Q5 Justin Post: Okay, I have two questions, one for Mark and one for Susan. It feels like your capacity is quite impressive. How do you plan to expand business opportunities beyond advertising? For example, subscription models, cloud licensing models, and the various interesting products you are developing. I'm not expecting you to announce any product launches, but what else can you do beyond advertising? Also, Susan, it's really interesting to see the acceleration in the advertising industry (including FX). I wonder if you have also observed the overall acceleration in e-commerce, and where you think the funding is primarily coming from? Is the entire internet ecosystem also accelerating? I would like to hear your thoughts. Mark Zuckerberg, Founder, Chairman, and CEO: Yes, our focus is not limited to advertising. I believe that based on the data, advertising will still be the most important driver of our business growth in the coming years. Therefore, while we are advancing related work, we are also striving to balance various innovative initiatives and heavily investing in AI to ensure that everything we do in the AI space enhances the quality and performance of our core applications and business. Of course, we will share more relevant information later. But it should be noted that all these initiatives, even if they can scale quickly, will take some time to truly have an impact on the scale of the advertising business At the same time, we are also very focused on providing more value to businesses and improving the quality of the applications we use for advertising. Susan Li, Chief Financial Officer: Justin, regarding your second question, all verticals achieved robust year-over-year growth in the fourth quarter, except for the political sector (because we missed last year's U.S. presidential election). Among them, the e-commerce sector was the largest contributor to year-over-year growth, followed by professional services and technology sectors. Therefore, our online commerce business saw strong year-over-year growth, remaining roughly flat compared to the third quarter, with growth momentum spanning advertisers of all regions and sizes. Overall, we observed sustained strong market demand from the pre-holiday shopping season through Cyber Monday and into the end of the year, which is very favorable for us. The professional services sector also achieved strong overall growth, primarily due to the significant contribution from lead generation ads resulting from our product improvements, including the full launch of the Advantage+ lead generation campaign at the beginning of the fourth quarter. The technology vertical also continued to maintain strong growth, with similar growth momentum across advertisers of all regions and sizes. In summary, I believe the overall growth is very healthy, with growth momentum coming from all aspects. Operator: Your next question comes from Ross Sandler of Barclays. Please go ahead. Q6 Ross Sandler: Yes, Mark, you mentioned the plan to port Horizon Worlds to mobile platforms. We haven't heard much from the Horizon Worlds team on the call before. So it's great to hear that this project is moving forward to mobile. It seems that the combination of artificial intelligence and the technology you are building in Horizon could open up many new possibilities for the gaming space or new ways of communication. So could you elaborate on what the specific plans are? Thank you. Mark Zuckerberg, Founder, Chairman, and CEO: Yes, let me talk about this fundamental theme. Over the years, I have mentioned a core point in some earnings calls, which is that people always crave to express themselves and experience the world in as rich a way as possible. So I talked about that today. When we first started, a lot of content was text-based, right? That was almost the limit of what we could do at the time. Later, with the proliferation of mobile phones and cameras, much of the content became visual. But with the popularity of photos, we went through a period of unstable mobile networks where every time we wanted to watch a video, it had to buffer. After that issue was resolved, video has now become the primary form of content. One of our core philosophies has always been that this is not the end, right? Video will exist for a long time, it will continue to evolve and grow, and it will not disappear, just like photos and text, which will continue to develop in many ways, even as the market itself continues to grow. But I believe that video is not the final form. I think we will see more interactive and immersive forms in the future, and these forms will appear in your feed. You can imagine, of course, there are many details to be worked out, but you can imagine that people can easily create a world or a game through prompts and share it with the people they care about You can see it in the information stream and participate in it immediately. Moreover, this technology has both 3D and 2D versions. I think Horizon is very suitable for creating immersive 3D experiences. But it is certain that in the future, any video can be clicked into for a more meaningful engagement, providing a richer experience. I believe our investments in virtual reality software, Horizon, and many other areas of the company will perfectly combine with advancements in artificial intelligence, bringing these experiences to hundreds of millions or even billions of users through mobile devices. In summary, this is what excites me a lot, but this is just one of many topics, and I think this topic will be very interesting. I believe many different types of interactive immersive content will emerge in the future. And Horizon is a very interesting example, and I look forward to seeing how it develops. Operator: The next question comes from Ron Josey at Citigroup. Please go ahead. Q7 Ron Josey: Great, thank you for answering this question. I would like to delve deeper into your previous comments about the transformation of ranking recommendation models with Susan. Given the achievements in GEM, Andromeda, Lattice, and model integration, there are clearly many favorable factors. So, could you help us understand the roadmap in more detail and the stage we are at in the transformation of ranking recommendation models? There is a viewpoint that we may be encountering some limiting factors, or we are waiting for updated models, but any insights would be very helpful for us to think about the next steps and future directions. Thank you. Susan Li, Chief Financial Officer: Yes, thank you for your question, Ron. I was just thinking whether your question is more focused on advertising or more on user interaction, but I will try to address both aspects. Regarding the core user interaction part, we launched a series of ranking optimization measures for Facebook and Instagram in the fourth quarter, which effectively enhanced user interaction. In fact, no single product release has been the main driver of growth. This is due to multiple optimizations we made to the recommendation system, which helped us predict each user's interests more accurately. I briefly introduced some of these optimizations and their specific applications on Facebook and Instagram. We believe that by 2026, there is still significant room for improvement in our recommendation system, which is expected to further drive user engagement growth in these two applications. First, we plan to continue expanding the scale of the models and increasing data usage, including incorporating longer content interaction histories to further enhance the overall quality of recommendations. At the same time, we will begin to validate the use of advertising signals and natural content recommendations, gradually building a more shared platform for natural content and advertising recommendations. Secondly, we will continue to optimize recommendations to better align with the interactive content users engage with during conversations, thereby providing recommendations that better match their current interests Finally, given that LLMs can understand content more deeply, we will focus on integrating them more thoroughly into the existing recommendation system. I believe this is especially useful for recently released content, as this type of content lacks sufficient interaction data to support recommendations. In terms of advertising, we have discussed the various works of models such as Andromeda, Lattice, and GEM in the advertising field. I will highlight GEM in the fourth quarter, as we have expanded GEM to cover Facebook Reels. Now, it covers all major platforms of Facebook and Instagram. We have also doubled the scale of the GPU cluster used to train GEM. By 2026, we expect to significantly expand the training scale of GEM to larger clusters, thereby enhancing the model's complexity, expanding training data, and leveraging our new sequence learning architecture that we began deploying in the fourth quarter. Additionally, we will further improve how to transfer the learning outcomes of the GEM base model to the runtime models we use. So I believe there is still a lot of room for improvement in many components of the tech stack. This is the first time we have found a recommendation model architecture that can scale as efficiently as LLMs. We hope this will allow us to significantly scale the ranking model while maintaining a substantial return on investment. Operator: The next question will come from Ken Gawrelski of Wells Fargo. Please go ahead. Q8 Ken Gawrelski: Thank you very much. First, for Mark, how important is it for Meta to have a leading general model? Or is a model that performs exceptionally well in specific use cases sufficient, like what Anthropic is currently demonstrating in coding? We would love to hear your thoughts on this. Secondly, I would like to follow up, Susan, regarding the information you currently have. You mentioned that you made some improvements to the model by 2026, fine-tuning core functionalities, including enhancing user engagement and ad relevance. Can you talk about whether these investments are showing signs of diminishing returns? In what areas do you see further development opportunities after 2026? Thank you. Mark Zuckerberg, Founder, Chairman, and CEO: I think the question is, how important is it for us to have a general model? In my view, Meta is like a deep tech company. So, people think of us because we are building these applications and experiences. But the key that enables us to build all these products is that we build and control the underlying technology, which allows us to integrate and design the experiences we want, rather than being limited to what other participants in the ecosystem are building or allowing us to build. Therefore, I think this is a very fundamental question. I suspect that for various reasons, including some competitive factors and some safety factors, cutting-edge AI technology cannot always be open to everyone through APIs. So, I believe that if you want to be one of the world-class companies shaping the future of these products, having the ability to create the experiences you want is crucial So I think, from a business perspective, this is very important. And from a creative and mission perspective, it is also crucial to be able to truly design and build the experiences that we believe should be created for people. Yes, I mean, I think this is very important. Otherwise, we wouldn't be so focused on it, and we are clearly placing a lot of emphasis on this now. Susan Li, Chief Financial Officer: Regarding your second question, interestingly, during this conference call a year ago, I seemed to have talked about a series of investments we made in the 2025 budgeting process targeting advertising effectiveness and organic traffic engagement plans. These investments have generally been effective, and we are very satisfied with the processes we adopted at that time, including prioritizing investments based on expected return on investment, ensuring we have a robust measurement system to fund projects with positive returns, and tracking their performance throughout the year. We have just completed the budgeting process for 2026 and have allocated funds for similar investment projects, and we expect these investments to enable us to continue to maintain strong revenue growth in 2026. Nevertheless, for various reasons, I expect that the full-year reported revenue and revenue growth at constant currency will be below the levels of the first quarter. First, based on current exchange rates, we expect the favorable exchange rate factors to fade later this year. Second, we will end the strong growth period that previously benefited from the 2025 advertising effectiveness investments and a strong macroeconomic environment later this year. Finally, we anticipate that the revised, less personalized advertising services launched in the EU may have some adverse effects, which will begin to roll out gradually at the end of the first quarter. However, similar to 2025, we are satisfied with our process of identifying attractive investment opportunities and incorporating them into the budget to support key initiatives for ranking and recommendation systems, as well as improving model capacity efficiency, all of which are key drivers of our growth. Operator: The next question comes from Mark Mahaney of Evercore, please go ahead. Q9 Mark Mahaney: Okay, I want to ask two questions. MetaAI, what is the latest progress regarding user engagement and usage? Have you just started to improve this feature? Additionally, Susan, regarding stock buybacks, I believe you haven't repurchased any shares this quarter. It seems you haven't repurchased shares for quite some time, probably about a year. Earlier this year, you talked about capital allocation, and it sounds like you have no plans to repurchase shares in the near future, can you confirm that? Thank you very much. Susan Li, Chief Financial Officer: Okay, I’m happy to answer both of those questions. The latest progress with MetaAI is that it now covers over 200 markets. The markets with the highest daily active users for MetaAI are largely consistent with the markets where our other applications are very popular. However, the main applications used by users in different regions vary. For example, in India and Indonesia, WhatsApp has the highest usage. In the United States, Facebook is a significant driver of user engagement. Overall, we see many opportunities to simplify the tasks users need to accomplish daily using our services If we can do this well, the way users engage with our products will continue to expand. Therefore, we are committed to making MetaAI the most personalized assistant, while leveraging the vast information, trends, and content on our platform to provide differentiated insights. We believe we have a very strong capability in building highly personalized experiences and will integrate these advantages into MetaAI so that we can tailor responses based on each individual's personal interests and preferences. Regarding your second question about stock buybacks, our level of stock repurchase can be adjusted at any time for various reasons, including whether we believe certain areas need funding more in the short term. Currently, we believe the company's top priority is to allocate resources to position itself as a leader in the field of artificial intelligence. Therefore, this is indeed the primary use of funds, but we will continue to remain opportunistic and assess the pros and cons of stock buybacks versus other uses of funds. Kenneth Dorell, Director of Investor Relations: Okay, I think we will wrap up today's program here. Thank you all for participating today. We look forward to communicating with you again soon. Operator: This concludes today's conference call. Thank you for your participation, and you may now disconnect. The event has ended