Today, Meta released its fourth-quarter financial results. Following the earnings release, Meta CEO Mark Zuckerberg, CFO Susan Li, and other executives held an analyst conference call to answer questions regarding the business. Details: META Q4 Revenue Up 24% Year-Over-Year; Shares Rise Over 7% After Hours The following is a transcript of the conference call:
Morgan Stanley analyst Brian Nowak: My first question is for Mark. Mark, looking long-term at the company's various investments, personal AI services, and the recently announced Meta Compute project, could you share your perspective on how these investment opportunities will contribute to revenue growth over the next three, five, and ten years? How will the return on invested capital (ROIC) look? Susan, my second question focuses on the short term. For 2026, I know you have many optimizations and improvements planned for the recommendation algorithm and monetization efficiency. Could you outline what management views as the top 2-3 primary drivers for the company's revenue changes in 2026?
Mark Zuckerberg: I'll start with your first question. But before answering the questions, I must clarify that my responses to many queries during this call may not be extremely detailed or entirely satisfactory, as Meta is currently in an interesting phase—we are rebuilding the company's AI strategy, a process that has been underway for about six months, and I am generally pleased with the overall progress. In the coming months, we will launch a series of early-stage models, products, and businesses, and I will share more details then. So, I want to start by sharing our business outlook at a macro level. However, please be aware that many details will not be highly specific. As these related products and businesses roll out, I believe everything will become very exciting. Regarding business themes. We are currently focused on several major business opportunities, one of which is optimizing core products and accelerating the development of our existing business. I've mentioned before that I firmly believe integrating recommendation systems with large language models (LLMs) will significantly enhance the quality of both the user experience and the advertising experience. Going forward, we anticipate improvements in content generation capabilities and content quality, coupled with enhancements to the recommendation system, which we expect will collectively accelerate the quality and effectiveness of our core business, substantially improving the product experience for both platform users and our business customers. I believe this will create a compounding effect. Furthermore, I am confident many new commercial opportunities will emerge. We have been working on Meta AI for some time now. You can observe how similar products are monetized within the industry. As we scale this technology to the desired depth and breadth, we believe various monetization opportunities, such as subscriptions and advertising, will gradually materialize. Additionally, I am very optimistic about the opportunities in shopping and e-commerce, which I briefly mentioned in previous briefings. With the release of new models, both the initial ones and future iterations, I believe you will see significant improvements in our models' capabilities and performance. We will also develop complementary products to help business users and direct-to-consumer businesses better leverage our platform. It's worth noting that we haven't previously discussed the Manus acquisition in our briefings. This serves as a good example. Manus has a large base of subscribed enterprise users who use various tools to accelerate business outcomes. Integrating these functionalities into our advertising and business management platforms allows us to offer a more integrated solution for the millions of businesses using our platform. This not only fosters continued customer acquisition for existing products but also opens up new business lines for us. If we execute well over the next year, I believe our roadmap will become clearer and even more exciting.
Susan Li: Regarding your second question, our revenue forecast for the first quarter of 2026 indeed encompasses a range of possible outcomes. Overall, it reflects our expectation for strong quarterly growth. This range already factors in the potential for accelerated growth, based on the robust demand we observed at the end of Q4 2025, which is expected to continue into early 2026. I should also note that we anticipate foreign exchange factors will contribute approximately 4 percentage points positively to the year-over-year growth, which is about 3 percentage points more than the impact in Q4 2025. In general, we see advertisers responding positively to improvements in ad effectiveness, driving strong conversion growth. Throughout 2025, we made significant investments in multiple areas, including enhancements to ad ranking and delivery systems, more effective allocation of ad inventory, new features and ad products (like Advantage+), and more accurate measurement. These efforts have greatly contributed to the positive performance of our advertising business.
Goldman Sachs analyst Eric Sheridan: Management previously mentioned internal compute constraints, indicating the company lacks sufficient compute capacity to achieve its platform and product goals. I'd like to understand management's current view on the alignment between internal compute demand and the established product roadmap? Has this situation changed recently? My second question is, as the advertising business continues to expand, calculated from a year-over-year growth perspective (in USD), has the increase in compute capacity positively impacted business progression? In other words, as investors, how should we understand the relationship between the company investing more in compute and the monetization of outcomes?
Susan Li: Regarding your first question. Indeed, we still face compute constraints. Our teams worked extensively in 2025 to enhance our infrastructure capabilities. However, the demand for compute resources has, at times, outpaced our supply. Therefore, we expect a significant increase in available compute capacity in 2026 as cloud capacity expands. That said, the additional capacity from our self-built infrastructure will come online later this year. Prior to that, for most of 2026, our compute capacity will remain somewhat constrained. Nevertheless, I believe we have minimized the impact of these constraints on the business as much as possible internally. Thus, I expect this situation will persist for a portion of 2026. Furthermore, we are improving infrastructure efficiency through various methods, including optimizing workloads, increasing infrastructure utilization, diversifying chip supply, and investing in core technologies like content and ad ranking to continuously enhance efficiency. As for your second question regarding advertising business expansion, I cannot provide precise data at this time. One way we enhance ad performance is by achieving synergy between large models and lighter models used for ad inference. Typically, we do not directly use large model architectures (like GEM) in inference due to their size and complexity leading to high costs. Our common approach is to leverage large models to transfer knowledge to smaller, lightweight models used at runtime, thereby improving performance. However, I believe there is still room for large models to benefit from increased compute. As we allocate more compute to these models, we anticipate gaining further performance improvements in the future.
Bernstein Research analyst Mark Shmulik: My first question is for Mark. In the briefing, you mentioned expecting significant changes in the company's work methods and processes this year. My question is, if by year-end, noticeable results from new products and initiatives aren't apparent, do you think that's a possibility? Or should we exercise more patience? My second question is for Susan. I see the operating income (OI) guidance you provided; is the company's operating income expected to grow faster this year compared to last year? Assuming that in a few months, management believes capturing the AI opportunity requires more resources, but the macro environment might be softer. How will the company balance this high level of investment with core business growth?
Mark Zuckerberg: I think your first question is essentially asking when we expect the impact of these new products to become apparent. Indeed, we will be rolling out a series of new products throughout the year. It's important to understand that we are not just launching products; we are building the future. I have always believed that AI technology will enable many new experiences. I mentioned some in the earlier briefing, such as personal AI and the integration of LLMs with recommendation systems. I view this as a relatively long-term research project expected to yield returns over an extended period. But we are already seeing optimization effects in the recommendation system itself as more AI outcomes and progress are integrated, including improvements in content quality, new presentation formats, and functional enhancements. Of course, there are many other new experiments. In summary, we expect to launch these new products throughout the year. Some products may require several iterations to truly achieve product-market fit. Personally, I believe we have sufficient time; we will start launching new products early this year. By year-end, we expect to see some products achieving good results, with optimizations in work efficiency becoming evident. I think it's difficult to predict exactly what the situation will be at that point. Another notable point is the increasingly important role of AI agents, which has profound implications. We are beginning to see significant productivity improvements among employees using agent technology. The gap between those who use it effectively and those who don't is substantial. I believe this will create deep dynamic effects across the industry and the broader economy. Agent technology will effectively enhance company operational productivity and efficiency. I also hope everyone can leverage these tools to accomplish more than before. We need to ensure Meta becomes a company that delivers profound impact, where users can utilize these agent tools anywhere. Only Meta has the capability to accomplish such a task. I believe if we effectively leverage these tools, we will gradually see accelerated growth in related outcomes over time. As for a specific timeline, it's indeed difficult to predict accurately, and I won't speculate on a particular quarter here. But the trend is very clear: it will happen. I am very excited about it, and as I said, it will be fascinating. AI technology allows us to build more interesting things, which is core to our purpose.
Susan Li: Mark, regarding your second question, I want to clarify a point. In your question, you mentioned that our operating income growth in 2026 would be higher than in 2025. That characterization isn't accurate; I hope my response clarifies the relationship. In our guidance for the new year, we forecast that operating income in 2026 will be higher than the operating income in 2025. Here we are comparing absolute amounts, not year-over-year growth rates. For context, I'll provide some background. We are starting 2026 with strong overall revenue. Of course, 2026 is only a few weeks old; the macro environment is generally healthy, making it difficult to extrapolate current trends for the full year, especially with many variables in flux. We intend to leverage the current business strength to reinvest revenue into what we see as very attractive opportunities in AI infrastructure and talent. As capabilities are optimized and enhanced, it's challenging to assess the specifics of different investment opportunities throughout the year. Certainly, the hiring market remains competitive. But we aim to invest as proactively as possible. In the past, we've shared our principle of guiding investments based on operating profit growth. Based on our current expectations, we anticipate the company's operating income in 2026 will be higher than in 2025.
JP Morgan analyst Doug Anmuth: Mark, could you share more details on the progress of the MSL team over the past few months? Also, what are your views on the path to achieving frontier models this year? Susan, you mentioned expecting operating income growth in 2026. Will the company also achieve positive free cash flow this year? Looking ahead, for data center construction and compute development, is management considering finding joint venture partners?
Mark Zuckerberg: Regarding the progress, I've already mentioned it earlier, and I don't have much to add. I want to reiterate that when answering questions, some responses might seem less specific. We are now in the sixth month of building the MSL team. I am very satisfied with the team's quality; I believe we have one of the top talent teams in the industry, and some early data shows positive signals. But it's important to understand this is a long-term endeavor, not just about launching a single model or product. We will continue to release more models and different products in the future. I hope our work will demonstrate its value. We all need to recognize this will be an ongoing journey. The first results we release will primarily indicate our future trajectory. In summary, I am very optimistic about the company's prospects, but I don't have other specific details to share at this time.
Susan Li: Regarding the first part of your question. We are making significant investments in infrastructure capacity this year to support the company's future AI technology development. We believe that the cash flow generated by the business this year will provide sufficient capacity to support these investments. Simultaneously, as we build infrastructure capacity, considering potential different scenarios for capacity needs in the coming years, we continue to explore various paths to provide long-term flexibility and options for future capacity requirements. Therefore, we don't have any new announcements to make at this time. We are evaluating opportunities at different points in time to meet our capacity needs.
Bank of America Merrill Lynch analyst Justin Post: I believe the company will have very substantial infrastructure capacity in the future. How is management considering expanding opportunities beyond advertising, such as subscription services or cloud model licensing? Although you are developing many interesting things, and I know there won't be specific product announcements today, could management discuss businesses beyond advertising? Also, Susan, even excluding FX impacts, the acceleration in the advertising business is interesting. Is management observing an acceleration in overall e-commerce activity? Where is this revenue primarily coming from? Is the entire internet ecosystem accelerating? I'd like to hear your views.
Mark Zuckerberg: We are indeed looking at areas beyond advertising, but based on the data, advertising will remain the most important driver of our business growth for the next few years. While pursuing new businesses, we maintain a balance: experimenting with new things while heavily investing in the existing business to ensure all our AI efforts ultimately enhance the quality and performance of our core apps and business. However, even if these new businesses grow rapidly, it will take some time for them to reach the scale and impact of the advertising business. During this period, we will also focus on creating more value for businesses and improving the overall quality of our ad delivery.
Susan Li: Regarding your second question. We saw healthy year-over-year growth across most verticals in Q4, except for political ads. Due to the US election last year, political ad share declined this year. Online commerce advertising was the largest contributor to year-over-year growth, followed by professional services and the tech sector. The growth in online commerce ads was strong, relatively consistent with Q3 levels, with similar trends across regions and advertiser sizes. From holiday shopping demand, through Cyber Five, and into year-end, we observed very strong related ad demand. In the professional services sector, ad growth was broad and steady, boosted by the full rollout of Advantage+ lead ad campaigns early in Q4. The tech sector also maintained strong growth for us. Overall, our advertising business performance is very healthy, with positive growth across sectors.
Barclays analyst Ross Sandler: Mark, you mentioned bringing Horizon Worlds to mobile. But we haven't heard much about the Horizon Worlds team in recent calls. So we're glad to see action here. It seems management plans to combine AI with development on Horizon Worlds, which could become a new form of gaming or communication in the future. Could you elaborate on your plans in this area?
Mark Zuckerberg: In past calls, I've mentioned a core idea: people always want to express themselves and experience the world in the richest forms possible. I also touched on this in today's briefing. Initially, our primary mode of expression was text, right? That was the best way then. Later, we all got phones with cameras, making much content visual—photos. Mobile networks were weaker then, buffering videos. As network issues resolved, most content became video. But our core belief is that video isn't the endpoint. Video will be around for a long time; it will continue to grow, not disappear. Like photos and text, they continue to grow as the market develops. But I don't believe video is the ultimate form. I think more interactive, immersive content forms will emerge and eventually appear in users' feeds. Imagine users creating a virtual world or game with simple prompts and sharing it with people they care about; you could see it in your feed while scrolling and jump right in to participate. Such content could have 3D versions and 2D versions, with Horizon being well-suited for immersive 3D. In the future, users might enter any video they see, interact within it, experience the content, and participate in more meaningful ways. I believe our investments in VR software, the Horizon platform, and other related areas will align perfectly with the pace of AI technology development, ultimately enabling these immersive experiences to reach hundreds of millions, even billions, of users via mobile devices. In short, this is a direction I'm very excited about, but it's just one manifestation of our future vision. I think there will be many different types of interactive and immersive content, and the Horizon platform will be a very interesting case study; I look forward to its development.
Citi analyst Ronald Josey: Susan, I have a follow-up question on changes to the ad recommendation models. Clearly, integrating models like GEM, Andromeda, Lattice offers many benefits. Could you detail the company's model roadmap? At what stage is the company regarding progress on recommendation model changes? Some suggest there are limiting factors, like waiting for model updates. Please discuss this further to help us understand the future direction.
Susan Li: I'm not entirely sure if your question leans more towards advertising or user engagement, but I'll try to address both. Regarding core user engagement. We implemented several ranking optimizations on both Facebook and Instagram in Q4, which contributed to additional user engagement growth. However, it wasn't a single feature driving most growth; rather, multiple optimizations to the recommendation system collectively helped us better predict content each user finds interesting. I've shared examples from both platforms before. We see significant room for improvement in 2026 and expect these optimizations to further drive engagement on both platforms. First, we plan to continue scaling models, increasing the volume of training data, including longer content interaction data, to further enhance overall recommendation quality. We will also validate combining ad signals with content recommendations, steadily working towards the long-term goal of building a shared platform for both content and ad recommendations. Second, we'll make the recommendation system more adaptive. It will adjust recommended content based on real-time user behavior during a session, making content recommendations more aligned with immediate interests. Finally, we will more deeply integrate large language models into the existing recommendation system, as they can understand content at a deeper level. This is particularly useful for newly posted content with little interaction data, helping to surface it better. On the advertising side, we've made significant model advancements involving Andromeda, Lattice, and GEM. I'll focus on GEM. In Q4, we expanded GEM model coverage to Facebook Reels; it now covers all major surfaces on Facebook and Instagram. Concurrently, we doubled the size of the GPU cluster used to train GEM. Looking ahead to 2026, we expect to significantly scale GEM model training, using larger cluster data, increasing model complexity, expanding training data volume, and leveraging the new sequence learning architecture deployed in Q4. We will further optimize how knowledge learned by the GEM base model is transferred for use in other models. Overall, many of our models and components have substantial room for improvement. This is the first time we've found a model architecture that can scale as efficiently as LLMs. I hope this allows us to dramatically scale recommendation models while maintaining a good return on investment.
Wells Fargo analyst Ken Gawrelski: Mark, how critical is it for Meta to possess an industry-leading general-purpose model? Or do you believe having a model that excels in a specific application scenario is sufficient? For instance, Anthropic's excellent performance in code generation. I'd like to hear your thoughts on this. Also, Susan, you discussed model improvements for 2026, including fine-tuning core models, with optimizations in user engagement and ad relevance. For this capital allocation, are you observing signs of diminishing marginal returns? Looking at 2026, what other opportunities do you see?
Mark Zuckerberg: I think the core of this question is: How important is it to have a general-purpose large model? In my view, Meta is fundamentally a deeply technical company. Some might perceive us as an application or product experience company, but what truly enables us to build these applications and experiences is the underlying technology we develop and control ourselves. This allows us to integrate various technologies to design the user experience as we envision, rather than being constrained by what other companies in the ecosystem have already built or limited to innovating within their permitted boundaries. Therefore, I consider this a very fundamental issue. My judgment is that frontier AI technology, for various reasons—some competitive, some related to safety—will not always be available to everyone via API. So, if you aspire to be a globally significant company that shapes future product forms, it is crucial to have the ability to build the required capabilities autonomously and create the experiences you desire. This is critical from a business perspective, and equally so from a creativity and mission standpoint. We need the genuine ability to design and build the experiences we believe we should provide for our users. Thus, I do believe this matter is very important. Otherwise, we wouldn't be investing so much effort in it. We maintain a high level of focus on this area.
Susan Li: Regarding your second question. Interestingly, about a year ago on a call, I mentioned that during our 2025 budgeting process, we made a series of investment decisions related to ad effectiveness and content engagement improvements. Overall, these investments proved effective, and we were very satisfied with the decision-making process employed, which involved ranking investments based on expected return on investment (ROI) to ensure capital was allocated to projects with positive ROI. We also established robust measurement systems and tracked the performance of these investments throughout the year. Currently, we have just completed the 2026 budgeting process and approved a set of investments aligned with the expected direction. We anticipate these allocations will contribute to continued solid revenue growth in 2026. However, it's important to note that we expect both full-year reported revenue growth and constant currency revenue growth to be lower than the Q1 level, primarily for several reasons: First, based on current exchange rates, we expect the FX benefit to gradually diminish later in the year. Second, as the year progresses, we will be comping against higher prior-year periods, which benefited partly from 2025 ad effectiveness investments and a favorable macro environment. Third, we anticipate beginning the phased rollout of the new "low personalization ads" option in the EU later in Q1, which may create some headwind. But overall, similar to 2025, we remain confident in our investment decision-making process. We believe we successfully identified and supported a set of high-ROI investment opportunities, incorporating them into the budget to drive key business investments. These are crucial factors for long-term growth.
Evercore ISI analyst Mark Mahaney: I'd like to ask about the latest developments with Meta AI. What changes is management observing in terms of user engagement and usage? My second question concerns share repurchases. Susan, I noticed you didn't seem to repurchase shares this quarter. Looking back, it's been nearly a year since the company last repurchased shares. You discussed capital allocation earlier in the year, suggesting no immediate plans to resume buybacks. Could you clarify the company's specific thinking?
Susan Li: I can answer both questions. First, on Meta AI. Meta AI is now available in over 200 markets worldwide. The markets with the largest daily active user bases for Meta AI generally align with the regions where our other apps are most popular; however, users in different regions primarily access Meta AI through different channels. For example, in India, Indonesia, etc., usage is mainly driven by WhatsApp; whereas in the US, Facebook is the primary entry point. Overall, we believe Meta AI has significant room for growth in better helping users accomplish the tasks they come to our platforms to do daily. We believe doing this well will continuously expand how and how deeply users engage with our products. Therefore, our focus is on making Meta AI the most personalized AI assistant, while leveraging the vast amount of information, trends, and content on our platforms to provide differentiated, personalized perspectives. We have deep expertise in creating highly personalized user experiences and are now bringing that capability to Meta AI to provide answers tailored to each user's interests and preferences. Regarding your second question on share repurchases. The scale of share repurchases fluctuates over time. There are various reasons for this, including evaluating whether certain areas have higher short-term capital needs. Currently, we believe the highest priority for capital use is investing resources to ensure we maintain a leading position in AI. This is the top priority for capital allocation. Of course, we maintain flexibility in our investments, continuously evaluating share repurchases against other potential uses of cash.
Comments