Meta conference call: "Before the arrival of 'super intelligence', continue to aggressively invest in infrastructure, even if there is short-term excess it can be resolved."

On Wednesday after the U.S. stock market closed, Meta released its third-quarter financial report, showing revenue of $51.24 billion, a year-on-year increase of 26%. However, due to a one-time non-cash tax expense of $15.93 billion, net profit plummeted from $15.69 billion in the same period last year to $2.71 billion, a decline of 83%. In the subsequent earnings call, Meta elaborated on the strength of its current core business and further introduced that the company is fully committed to sprinting towards the future of artificial intelligence, planning to significantly increase capital expenditures and total expenses by 2026 to establish what CEO Mark Zuckerberg described as a "leading-edge AI laboratory." When asked about the considerations for the accelerated growth in capital expenditures in 2026, Meta CEO Zuckerberg stated, "The company's consistent model is to build infrastructure based on aggressive assumptions, and it always finds that demand far exceeds expectations, especially in core businesses, where computing power can be efficiently utilized with high profitability. Therefore, continuing to accelerate investment is a reasonable choice." Zuckerberg emphasized that even in the event of short-term excess, the company has multiple ways to digest it: "First, to enhance the recommendation and advertising intelligence of the application family; second, to explore providing API or computing power services externally (though not yet implemented); in the worst-case scenario, it would just mean building a few years in advance, accompanied by depreciation costs, but in the long term, it can still be gradually utilized." Here are the key points from the call: There are differing expectations regarding when superintelligence will arrive; some believe it can be achieved within a few years, while others think it will take 5 to 7 years or even longer. Meta believes that in this uncertainty, the correct strategy is to actively build computing capacity in advance to prepare for the most optimistic scenarios. The company plans to meet these computing demands through building its own infrastructure and signing contracts with third-party cloud service providers. Zuckerberg divides the evolution of social media into three eras: the first is content from friends and family, the second adds creator content, and now, AI will bring about the third era of content. Within this framework, Meta's massive investment in AI is not only a new business but also a necessary requirement for the evolution of its core platform. The company's AI recommendation system continues to improve, driving increased user engagement. Facebook's usage time in the third quarter grew by 5% year-on-year, while Threads grew by 10%. Video performance was particularly outstanding, with Instagram video usage time increasing by over 30% year-on-year. Reels' annualized revenue has exceeded $50 billion. The user base of Meta AI is rapidly expanding, with monthly active users exceeding 1 billion. The annualized revenue generated through end-to-end AI-driven advertising tools has surpassed $60 billion. Instagram has reached an important milestone of 3 billion monthly active users, and Threads' daily active users recently surpassed 150 million, with the potential to become a leader in its category. The Vibes tool launched by the company is the next-generation AI creation tool, currently showing good retention rates and rapidly increasing usage each week. Since the launch of Vibes in Meta AI in September, the amount of media generated within the app has increased more than tenfold. The company's total expenditure growth rate in 2026 will significantly outpace that of 2025, driven mainly by infrastructure costs, including additional cloud spending and depreciation expenses. Meta has also raised its 2025 capital expenditure guidance to $70 billion to $72 billion, up from the previous range of $66 billion to $72 billion. In addition to capital expenditure pressures, employee compensation costs will become the second largest growth driver, primarily due to the full-year recognition of salaries for employees hired throughout 2025, especially for AI talent and advertising technology talent in priority areas. Here is the transcript of the conference call: Good afternoon, I’m Krista, today’s meeting host. Welcome everyone to Meta's third-quarter earnings call. All lines have been muted to prevent background noise. After the speakers' statements, we will enter the Q&A session. Mark Zuckerberg, Founder, Chairman, and CEO: Alright. Thank you all for joining today. We’ve had another strong quarter, with 3.5 billion people using at least one of our apps every day. Instagram has reached an important milestone, with monthly active users reaching 3 billion, and we are seeing good growth momentum in other apps as well, including Threads, which recently surpassed 150 million daily active users and is poised to become a leader in its category. I am very focused on building Meta into a leading frontier AI lab, creating personal superintelligence for everyone, and providing application experiences and computing devices that can improve the lives of billions of people around the world. Our approach to advancing open-source AI means that when Meta innovates, everyone can benefit. The Meta Superintelligence Lab has had a strong start. I believe we have built the lab with the highest talent density in the industry. We are focused on developing the next generation of models and products and look forward to sharing more progress in the coming months. We are also building the scale of computing power that we expect to reach industry-leading levels. There are different timelines for when superintelligence can be achieved. Some believe it can happen in a few years. Others think it will take 5, 7, or even longer. I believe the right strategy is to proactively build computing power in advance so that we can be prepared for the most optimistic scenarios. This way, if superintelligence arrives sooner, we will be ideally positioned for this generational paradigm shift and numerous significant opportunities. If it takes longer, then we will use the additional computing power to accelerate our core business, which can still profitably utilize far more computing power than we can invest, and we see strong demand for additional computing power both internally and externally. In the worst-case scenario, we are simply slowing down the construction of new infrastructure for a period while the business grows to match the scale we are building. There is significant upside potential for our existing applications and for AI to enable new products and businesses. Our AI recommendation systems on Facebook, Instagram, and Threads provide higher quality and more relevant content, which has led to a 5% increase in Facebook usage time in the third quarter and a 10% increase in Threads Video is a particularly bright field, with video watch time on Instagram increasing by over 30% compared to last year. As video continues to grow in our applications, Reels' annualized revenue has now exceeded $50 billion. With the increase in AI-generated content, the improvements in our recommendation systems will become even more effective. Social media has experienced two eras so far. The first era was when all content came from friends, family, and accounts you followed directly. The second era was when we added all creator content. Now, as AI makes it easier to create and remix content, we will add another massive content library on top of these foundations. The recommendation system that can understand all this content more deeply and show you the right content to help you achieve your goals will become increasingly valuable. Our advertising business continues to perform excellently, largely thanks to improvements in our AI ranking system. This quarter, we made significant progress in unifying different models into simpler, more general models, which drove better performance and efficiency. Now, the annualized revenue from our fully end-to-end AI-driven advertising tools has exceeded $60 billion. One way I think about our overall company is that there are three giant converters running Facebook, Instagram, and ad recommendations. We are improving these models by integrating new AI advancements and capabilities. At the same time, we are also working to merge these three major AI systems into a single unified AI system that will effectively run our family of applications and businesses, using increasingly sophisticated intelligence to improve the trillions of recommendations we make for people every day. I am also very excited about the new products we can build. Over 1 billion monthly active users are using Meta AI, and as we improve the underlying models, we see usage continuously increasing. I look forward to bringing cutting-edge models to Meta AI, as I believe there are significant opportunities there. The same goes for our business AI. Every day, people engage in over 1 billion active conversations with business accounts on our messaging platforms, ranging from product inquiries to customer support. Our business AI will enable millions of businesses to scale these conversations and increase sales at a low cost. The better our models, the better the results for all businesses. This quarter, we also launched Vibes, which is the next generation of our AI creation tools and content experiences. So far, retention looks good, and usage is rapidly increasing week by week. I look forward to accelerating the growth of Vibes in the coming months. More broadly, I see Vibes as an example of a new type of AI-supported content, and I believe there are many more opportunities to build many novel content types in the future. As our new models are ready, I look forward to starting to showcase some of the new products we are developing. **At the Connect conference, we launched the 2025 AI glasses series, which has received a great response so far. The new Ray-Ban Meta glasses and Oakley Meta Vanguards are selling well, as people love the improved battery life, camera resolution, new AI features, and excellent design ** We also have our new Meta Ray-Ban smart glasses, which are our first glasses equipped with a high-resolution display, along with the Meta neural wristband for interaction. They sold out in almost every store within 48 hours, and demonstration appointments are fully booked until the end of next month. Therefore, we will have to invest to increase manufacturing and sell more products. This is a field where we are clearly in a leading position and have tremendous opportunities. Looking at it from a broader perspective, if we can realize even a small portion of the potential of existing applications and new experiences, I believe the next few years will be the most exciting period in our history. We have a lot of work to do, but we are making substantial progress, delivering strong business results, building the talent density and infrastructure needed for the next era, and leading the way in AI devices that will define the next computing platform. I am proud of how our team has risen to the challenges, and I appreciate their dedication, hard work, and creativity. As always, thank you all for joining us on this journey. Now, I would like to invite Susan. Susan Li, Chief Financial Officer: Thank you, Zuckerberg, and good afternoon, everyone. Let’s start with segment performance. Unless otherwise noted, all comparisons are year-over-year. Our family of apps continues to grow its community, and we estimate that over 3.5 billion people use at least one app from our family every day in September. Total revenue for the family of apps in the third quarter was $50.8 billion, an increase of 26% year-over-year. Advertising revenue for the family of apps in the third quarter was $50.1 billion, up 26%, or 25% on a constant currency basis. In the third quarter, the total number of ad impressions served increased by 14%. Growth in impressions was healthy across all regions, driven by engagement and user growth, particularly in video services. The average price per ad increased by 10% year-over-year, benefiting from increased demand from advertisers, primarily driven by improvements in ad effectiveness. This was partially offset by growth in impressions, particularly from lower monetization regions and services. Other revenue from the family of apps was $690 million, an increase of 59%, driven by growth in WhatsApp paid messaging revenue and Meta Verified subscriptions. In our Reality Labs segment, third-quarter revenue was $470 million, an increase of 74%. The significant year-over-year growth in the third quarter was partly due to retail partners stocking Quest headsets ahead of the holiday season. We did not have similar gains in the third quarter last year, as our Quest 3S headset was launched in the fourth quarter of 2024. Additionally, strong revenue from AI glasses also contributed to the revenue growth in the third quarter. Now, turning to our consolidated performance. Total revenue for the third quarter was $51.2 billion, an increase of 26%, or 25% on a constant currency basis. Total expenses for the third quarter were $30.7 billion, an increase of 32% year-over-year. The year-over-year expense growth accelerated by 20 percentage points compared to the second quarter, primarily due to three factors. **First, legal-related expenses grew faster than in the second quarter, due to costs recorded in the third quarter, as well as the write-off of some provisions from the third quarter of last year Secondly, employee compensation growth has accelerated, driven by the recruitment of technical talent, particularly in AI. Finally, infrastructure cost growth has accelerated, due to increased operational costs associated with our expanded data center cluster, depreciation of additional capital expenditures, and third-party cloud spending. At the end of the third quarter, we had over 78,400 employees, an 8% year-over-year increase, driven by hiring in priority areas such as monetization, infrastructure, Reality Labs, Meta Super Intelligence Lab, and regulatory and compliance. The operating revenue for the third quarter was $20.5 billion, with an operating profit margin of 40%. Interest and other income for the third quarter was $1.1 billion, primarily driven by unrealized gains on our marketable equity securities. Our tax rate for the quarter was 87%, adversely affected by a one-time non-cash reduction in deferred tax assets that we no longer expect to utilize under the new U.S. tax law. Excluding this expense, our tax rate would have been 14%. Although the transition to the new U.S. tax law resulted in accounting expenses for the third quarter, we continue to expect significant cash tax savings for the remainder of this year and future years under the new law. The expenses for this quarter reflect the total expected impact of the transition to the new U.S. tax law. Net profit was $2.7 billion, or $1.05 per share. Excluding one-time tax expenses, our net profit and earnings per share would have been $18.6 billion and $7.25, respectively. Capital expenditures (including principal payments on finance leases) were $19.4 billion, driven by investments in servers, data centers, and network infrastructure. Free cash flow was $10.6 billion. We repurchased $3.2 billion of Class A common stock and paid $1.3 billion in dividends to shareholders. At the end of the quarter, we had $44.4 billion in cash and marketable securities and $28.8 billion in debt. Now, turning to business outlook. There are two main factors driving our revenue performance. Our ability to provide engaging experiences for communities and the effectiveness with which we monetize that engagement over time. Regarding the first point, daily active users on Facebook, Instagram, and WhatsApp continue to grow year-over-year. We continue to see improvements in products and recommendations driving additional engagement, with global usage time on Facebook and Instagram accelerating year-over-year in the third quarter. In the U.S., total usage time on Facebook and Instagram grew double digits year-over-year, thanks to sustained strength in video and healthy growth in non-video time on Facebook. The increase in engagement continues to be driven by ongoing improvements in product work and recommendation systems, as we optimize model architectures, implement advanced modeling techniques, and integrate more signals about people's interests. We also continue to focus on increasing the freshness of recommended content. On Facebook, the number of Reels shown by our system that were published on the same day is now double that of the beginning of the year. Looking ahead to 2026, we expect to advance our recommendation systems across multiple dimensions. On Instagram, a key focus is to develop our system to showcase content on a wider range of topics, catering to the diverse interests of everyone. This follows a similar approach we implemented on Facebook, which has yielded good results. We also expect to make significant progress in long-term ranking innovations by 2026. We have seen promising new results from our research work, namely the creation of foundational ranking models, and we anticipate that the new model innovations we are developing will enable us to significantly expand the scale of data and computing power used to train recommendation models by 2026, resulting in more relevant recommendations. Another focus for next year is to leverage large language models to improve content understanding. We expect this will allow our system to more accurately tag keywords and topics in videos and posts, enabling a deeper understanding of an individual's interests while retrieving matching content. Finally, we have made good progress with Meta AI and Threads. The number of users utilizing Meta AI across our application family continues to grow, and we are increasingly integrating first-party content into Meta AI results. In the U.S., most responses to deep queries on Facebook now display relevant Reels. We have also seen a lot of traction in media generation. People have created over 20 billion images using our products, and since the launch of Vibes in Meta AI in September, we have seen a tenfold increase in media generation within the app. On Threads, as we continue to improve recommendations, we are seeing strong growth in daily active users and engagement depth. The ranking optimizations we implemented in the third quarter alone drove a 10% increase in time spent on Threads. We are also continuing to roll out new features, including the launch of direct messaging in the third quarter, so now anyone on Threads can send messages to each other within the app. Now let's talk about the second driver of our revenue performance, which is improving monetization efficiency. The first part of this work is optimizing the ad levels in organic engagement. We continue to optimize the ad supply across the main interfaces of Facebook and Instagram to better serve ads to people at the most relevant times and places. In the long term, we have exciting advertising supply opportunities on Threads and WhatsApp Status. Ads are now being served in the global feed of Threads, and we are following our typical monetization playbook to optimize ad formats and effectiveness before increasing supply. In WhatsApp Status, we are continuing to gradually introduce ads and expect to complete the rollout next year. The second part of improving monetization efficiency is enhancing marketing effectiveness. Advancing our advertising system remains a key aspect of this work, and we are driving effectiveness improvements through continuous enhancements to larger-scale ad ranking models. For example, we continue to expand the adoption of Lattice (our unified model architecture) In the third quarter, we promoted Lattice to application advertising, which improved the conversion rate of this target by nearly 3%. Since the introduction of Lattice and other backend improvements in 2023, we have now reduced the number of ad ranking and recommendation models by about 100, as we merged smaller and more specialized models into larger models using the Lattice architecture to promote learning between the interface and the target. As we merge models, we continue to observe performance improvements and expect to drive additional gains when we merge another 200 models into a smaller number of high-capacity models over the next few years. In addition to advancing our foundational ad models, we are also innovating downstream runtime models for ad inference. For example, in the third quarter, we began piloting a new runtime ad ranking model that utilizes more computing power and data than our previous models to select more relevant ads. In testing, we saw that this new model improved the conversion rate on Instagram by over 2%. We also significantly improved the performance of Andromeda in the third quarter by merging retrieval and early ranking models into a single model, which enhanced the ad quality on the Facebook interface by 14%. In our ad products, we see continued momentum for Advantage+. In the third quarter, we completed the global rollout of a simplified campaign creation process for the Advantage+ lead generation campaigns. As a result, advertisers running sales applications or lead generation campaigns are now enabled with end-to-end automation from the start, allowing our system to optimize performance on our platform by automatically selecting criteria such as who to show ads to and where to display ads. We see Advantage+ continuing to drive performance improvements. Advertisers running lead generation campaigns with Advantage+ have an average cost per lead that is 14% lower than those not using it. I would say we believe there is still significant growth opportunity for the adoption of Advantage+. Many advertisers only use our end-to-end automation solution in part of their campaigns, so we can grow share there. To seize this opportunity, we are focused on driving continuous performance improvements and addressing some key use cases to increase adoption rates. We are also committed to expanding the adoption rate among advertisers using one of our single-step automation solutions. For example, advertisers who may only use part of it (such as Advantage+ audiences) by helping them understand the benefits of using multiple automation solutions simultaneously. So I would say Advantage+ is a continuous platform where we continue to expand the feature set available in Advantage+ while also extending the reach of that feature set to a broader group of advertisers. Zuckerberg mentioned that advertisers using these automation options currently have an annualized revenue of $60 billion. We again see room for continued growth. Q&A Session Q: I have two questions for Susan. First, the core for 2026, including models, ad ranking models, and more types of computing power is exciting, and the underlying infrastructure investment seems quite substantial. Can you share some early quantifiable signals from A/B testing that give you confidence in the return on investment for these capital expenditures? The second question is simpler: How significant is the headwind for Reality Labs' revenue in the fourth quarter?** Susan: The first question has multiple layers, and I will respond to them one by one. The growth of capital expenditures in 2026 will come from three main areas: MSL (Meta Supercluster Lab), core AI, and non-AI expenditures, with MSL experiencing the fastest growth in AI demand. We have established a resource investment roadmap across human resources and computing power in the 2025 budgeting process, which is expected to yield returns in 2026. Specifically, we continue to make progress in advertising effectiveness, ranking, and conversion optimization. A key metric is the value-weighted conversion rate—although advertisers have diverse optimization goals, this metric's year-on-year growth significantly outpaces the growth in impressions when we control for variables. Additionally, the new model architecture allows us to utilize data and computing power more efficiently to enhance advertising effectiveness, a trend expected to continue into 2026. We have currently initiated the 2026 budgeting process and see a large number of high-return investment opportunities, giving us confidence in driving revenue growth. Regarding Reality Labs, we have not quantified the specific potential contraction scale, but we expect Q4 revenue to be lower than the same period last year. The main reason is that the Quest 3S was launched in Q4 2023, while there are no new products this year. Last year's Q4 confirmed all holiday season Quest 3S sales (the product was released in October 2023), while this year's sales were partially confirmed in Q3 (due to retail partners stocking up early). Although AI glasses revenue is expected to achieve significant year-on-year growth, it is still offset by a decline in Quest headset sales. Q: I appreciate the strategy of preemptively laying out computing power for "super intelligence." Can you discuss the considerations when weighing the accelerated growth of capital expenditures in 2026 against the impacts on core business earnings and free cash flow? Is there a cash or net cash target on hand? Susan: The 2026 budget is still in the early stages, especially with a lot of dynamism in computing power planning. We are currently feeling a clear shortage of computing power—not only does the MSL team need more computing power, but there are also many high ROI uses in the core business that can be accommodated. Therefore, we are working to lay out in advance, not only to meet the 2026 demand but also to reserve flexibility and option value for 2027-2028. The budget is still being compiled, and there are no specific cash targets to share yet, but the strategic priorities are very clear: ensuring we have the computing power needed for success in the AI field. Mark Zuckerberg added: Our consistent model is to build infrastructure based on aggressive assumptions, and we always find that demand far exceeds expectations, especially in the core business, where computing power can be efficiently utilized with high margins. Therefore, accelerating investment is a reasonable choice. Even if there is a short-term surplus, we have multiple paths for digestion: First, it can be used to enhance recommendation and advertising intelligence for the application family; Second, we explore providing API or computing power services externally (though this has not yet been implemented); the worst-case scenario is just building a few years in advance, accompanied by depreciation costs, but in the long term, it can still be gradually utilized The key is that rather than missing high ROI opportunities due to capital constraints, it is better to actively accelerate layout. Of course, actual construction is also limited by operations, and we will further clarify plans in the coming months. Overall, this is a huge opportunity. Q: Zuckerberg, you mentioned "super intelligence." Can you relate it to consumer AI (like Meta AI)? How are current user interaction signals? How will the cutting-edge models derived from MSL change the practicality of Meta AI in the future? Zuckerberg: Currently, Meta AI has over 1 billion monthly active users. With the improved quality of the Llama 4 post-training model, usage continues to grow. Our view is that once we integrate the cutting-edge models developed by MSL (which have unique new capabilities), it will bring huge opportunities. Meta has the best industry record in scaling new products to billions of users. After inserting leading models, I expect Meta AI and related products (such as video creation tools, business AI, etc.) to experience explosive growth. At the same time, smarter models will continue to optimize core businesses—including feed recommendations and advertising effectiveness. Currently, we are still operating under "compute-constrained" conditions, and if we can invest more computing power, it will release significant value. Q: Susan, how visible will the improvements in advertising effectiveness and engagement be next year? Is the scale of improvement comparable to the past two years? Also, a question for Zuckerberg: Is the new work in the super intelligence lab anchored to an update of a cutting-edge model next year? Or should we focus on the progress of new products like Vibes? Susan: We are improving advertising effectiveness in two ways: first, by using large models for knowledge distillation to transfer capabilities to lightweight inference models; second, by continuously optimizing the inference architecture to expand computing power and complexity investment in a positive ROI manner. Even if we achieve only single-digit percentage increases in conversion rates each quarter, it can lead to significant revenue growth given the large base of advertisers. Zuckerberg: We do not have any specific timelines to announce regarding models or products, but I expect you will see both. We anticipate building novel models and novel products, and I will be happy to share more when we have content. Q: Zuckerberg, past technology cycles (such as mobile and social) have brought high profit margins. The initial investment in the AI cycle is huge; how do you view the user value of new tools and future profit margins? Will it be different from past cycles? Zuckerberg: It is too early to judge the profit margins of new products. Each product has different characteristics and requires time for validation. My goal has always been to maximize user value and overall profitability, rather than simply pursuing profit margins. We will prioritize creating the best products to create value for the most people. Q: Compared to other AI labs focusing on long-term goals like AGI, how does Meta balance cutting-edge research with immediate monetization? Is having billions of people using personal AI still a core direction? Are capabilities like Vibes or Sora equally important? Zuckerberg: Research will enable new technological capabilities to exist, and then these capabilities can be built into various different products. For example, the ability to reason more intelligently is very important in many aspects It is useful for assistants. It is also useful for business AI. It is useful for the AI agents we are building to help advertisers figure out what their campaigns will be. Ultimately, it will also impact how we rank and recommend people's information streams and make different decisions there. This is just one example. What I mean is that the ability to create very high-quality, good videos will definitely be useful for providing people with new creative tools. It will help increase the inventory of content that can be displayed on Instagram and Facebook, thus increasing engagement there. It should help advertisers create creatives that will help us monetize better. Each capability can unlock multiple product opportunities. The key to product development is identifying high-value use cases from the technology list and prioritizing their implementation. I firmly believe that achieving "the best" in each capability rather than just "good enough" is essential to capturing maximum value. We are focused on building novel capabilities (not simple copies) and leveraging Meta's scalable capabilities to bring them to billions of users. This will drive both new business and upgrades to existing business. Due to competitive considerations, I cannot disclose specific priorities, but the direction is clear. Q: Regarding Meta AI's products and monetization path—what are the most encouraging adoption signals currently? Is the monetization path clear? At which stage of "launch → deepen engagement → monetize" are we? Zuckerberg: The most encouraging aspect is that we have built products used by over 1 billion people, and improvements in models directly drive usage growth, indicating a clear path for enhancement. We have just completed the establishment of the Meta Super Intelligence Lab—bringing together top talent in the industry, focusing on training the next generation of models. Once ready, they will be quickly integrated into various product lines. In terms of monetization, beyond enhancing ads and recommendations, a greater opportunity lies in an end-to-end automated advertising system: advertisers only need to provide business goals and payment methods, and AI can automatically generate creatives, target audiences, and optimize placements. This will greatly lower the barriers to advertising and improve ROI. We are very optimistic about this. Q: Advantage+ has achieved an annualized revenue of $60 billion. Can you talk about the adoption rate among advertisers? How will automation further enhance advertisers' ROI after integrating ranking and recommendation upgrades like Andromeda, Gens, and Lattice? Susan: In Q3, we completed the global rollout of a simplified process for Advantage+ potential client activities, allowing advertisers to enable end-to-end automation from the outset, covering audience, placement, budget allocation, and more. Data shows that advertisers using Advantage+ have seen an average reduction of 14% in customer acquisition costs. However, there is still significant room for overall adoption—many advertisers only use it for certain campaigns or only adopt a single feature (like Advantage+ audience). We will drive comprehensive adoption by continuously improving effectiveness, optimizing key use cases, and educating users to combine multiple modules. Advantage+ is a continuously evolving platform; we are expanding both the feature set and the range of advertisers covered. The current $60 billion ARR is just the starting point Q: I would like to ask Zuckerberg whether wearable devices (such as Ray-Ban Meta) can recoup investments through hardware sales, or if they will rely on new revenues from AI services, etc.? What are the gating factors? Susan, how is the on-balance/off-balance financing for the AI plan arranged? Will the joint venture with Blue Owl be included in the 2026 capital expenditures? Will this model slow down future CapEx growth? Zuckerberg: Ray-Ban Meta and Oakley Meta are progressing well. Revenue will come from both hardware sales and AI services. Currently, users purchase mainly for fashion or audio features, but AI will become the core driver in the future, with huge commercial potential. At the same time, we are still investing in holographic AR prototypes like Orion, which are in the early stages, but the goal is to serve hundreds of millions or even billions of users—at that point, it will be highly profitable. Susan: The joint venture with Blue Owl is an innovative financing approach to address the uncertainty of future computing power demands. Previously, data center costs were included in capital expenditures; in the future, construction costs will no longer be counted as CapEx but will be borne by the joint venture. Meta will contribute in the form of "other investment cash flow" at a 20% ownership stake. This structure is not included in the guidance for 2026 capital expenditures, but it provides flexible expansion options for the future. Whether it slows down the growth of capital expenditures depends on the evolution of demand, and there is currently no conclusion. Q: Zuckerberg, when you hope to have leading-edge models next year, how do you think value will be distributed in the AI ecosystem? Is it at the platform level or on the side of scaled applications? Zuckerberg: I don't differentiate much between "platform" and "application"—there are opportunities for all parties in the AI ecosystem: hardware like NVIDIA, cloud service providers, and application developers. But history shows that truly sustainable businesses often combine technological capabilities with network effects and scalability. Meta's unique advantage is that not only can we build cutting-edge models, but we can also quickly integrate them into products used by billions of users and continuously enhance the experience through a unified intelligent system (such as cross-application recommendation pools). This will strengthen core business advertising and recommendations and will also give rise to new business models (advertising, e-commerce, subscriptions, etc.). Although we are still in the early stages, we have already seen clear returns in our core business, which makes us confident that we should increase investment and avoid underinvestment