Meta (Minutes including small meetings): "Every year" is a key year, and this year is not just about Meta AI

The following is the summary of Meta's Q4 2024 earnings conference call (including the Callback session). For a review of the financial report, see Meta: Why is the market not panicking this time with Zuck excitedly spending money?

1. Overview of Financial Indicators

(1) In Q4 2024, Meta achieved total revenue of $48.4 billion, a year-on-year increase of 21% (also a 21% increase when calculated at constant exchange rates). During the same period, the company's total expenses were $25 billion, a year-on-year increase of 5%.

(2) Operating costs increased by 15% year-on-year, mainly due to increased infrastructure spending;

(3) R&D expenses increased by 16% year-on-year, driven mainly by employee compensation and infrastructure investment, although partially offset by a decrease in restructuring costs;

(4) Sales and marketing expenses remained basically flat year-on-year;

(5) General and administrative expenses (G&A) decreased by 67% year-on-year, mainly due to a reduction in legal-related expenses of $1.55 billion.

(6) By the end of 2024, the company's total global employee count exceeded 74,000, a year-on-year increase of 10%, with the main growth coming from hiring in core areas such as commercialization, infrastructure, generative AI, Reality Labs, and regulatory compliance.

(7) In Q4 2024, the company's operating profit reached $23.4 billion, with an operating profit margin of 48%. The company's tax rate for the quarter was 12%, resulting in a net profit of $20.8 billion and earnings per share (EPS) of $8.02.

(8) Capital expenditures (including principal payments on finance leases) were $14.8 billion, mainly for investments in servers, data centers, and network infrastructure;

(9) Free cash flow was $13.2 billion;

(10) The company paid $1.3 billion in dividends to shareholders, with cash and cash equivalents at the end of the year totaling $77.8 billion and total debt of $28.8 billion.

1. Meta Family of Apps

As of December 2024, at least 3.3 billion people use at least one app from our family of apps daily.

In Q4 2024, total revenue from the family of apps was $47.3 billion, a year-on-year increase of 21%; advertising revenue was $46.8 billion, also a year-on-year increase of 21% (a 21% increase when calculated at constant exchange rates).

In advertising revenue, the e-commerce vertical was the largest contributor to year-on-year growth. By user region, the strongest growth in advertising revenue was in other regions, which grew by 27%, followed by the Asia-Pacific region and Europe, which grew by 23% and 22%, respectively. North America grew by 18%.

In the fourth quarter, the total number of ad impressions across our services grew by 6%, and the average price per ad increased by 14%. The growth in ad impressions was mainly driven by the Asia-Pacific region. The increase in ad prices was due to higher demand from advertisers, partly because of improved ad effectivenessHowever, this growth was partially offset by the increase in display volume in low monetization regions.

Other revenue from Apps reached $519 million, a year-on-year increase of 55%, mainly due to the growth in business message revenue from the WhatsApp business platform.

In the fourth quarter of 2024, total spending on Apps was $19 billion, accounting for 76% of the company's total spending, a year-on-year increase of 5%, primarily due to rising infrastructure costs and employee compensation, although this was partially offset by a decrease in legal-related expenses. The operating profit for Apps reached $28.3 billion, with an operating profit margin of 60%.

2. Reality Labs

In the fourth quarter, Reality Labs revenue was $1.1 billion, primarily contributed by hardware sales, a year-on-year increase of 1%;

Reality Labs expenses reached $6 billion, a year-on-year increase of 6%, mainly due to rising infrastructure costs and employee compensation, although this was partially offset by a decrease in restructuring costs;

Reality Labs reported an operating loss of $5 billion.

II. Management Report

1. AI

1.1 Meta AI

It is expected that this year AI assistants will cover 1 billion users, and Meta AI will become a leader in this field. Meta AI is currently the most widely used AI assistant globally, and typically when a service reaches such scale, it forms a long-term competitive advantage. This year, we have exciting product plans focused on personalized experiences.

We believe that users do not want to use a generic AI assistant but prefer one that can be personalized. In the future, there will not be just one single universal AI. Instead, users will have the autonomy to decide how their AI assistant operates and presents itself. This will become one of the most transformative products we have ever created, and we will also launch some exciting innovations this year that we believe users will love.

1.2 Llama Model

Meanwhile, I also believe that the Llama open-source model will become one of the most advanced and widely used AI models this year. Currently, the training progress of Llama 4 is going smoothly, and Llama 4 Mini is undergoing pre-training, with our inference models and large models also performing well.

The goal of Llama 3 is to enable open-source models to compete with closed models, while the goal of Llama 4 is to lead the industry. Llama 4 will have native multimodal capabilities, be an all-purpose model, and possess agency capabilities, which will unlock many new application scenarios. In the coming months, we will share more about our plans for Llama 4.

1.3 AI Engineer

This year, we will also be able to build an AI engineer whose coding and problem-solving abilities will reach the level of a mid-level engineer. This will be a historic milestone, potentially becoming one of the most significant innovations and also creating a huge market opportunity. Any company that launches this system first will have a significant advantage in AI research and technological evolution, which is why I believe this year will be a key year that determines the future1.4 AI Smart Glasses

Our Ray-Ban Meta AI Smart Glasses are currently receiving positive market feedback. This year, we will clarify the future development path of AI glasses as a product category. In the consumer electronics field, many breakthrough products often sell between 5 million to 10 million units by their third generation. This year will determine whether we can develop along the path of hundreds of millions or even billions of units, making glasses the next generation computing platform, or if this will be a market that requires more time to refine.

Overall, we see that users recognize smart glasses not only as an ideal carrier for AI platforms but also possessing fashion attributes. Of course, these are all large-scale investments, especially as we plan to invest hundreds of billions of dollars in AI infrastructure in the long term.

1.5 Data Centers

Last week, I announced our plan to launch nearly 1GW of data center computing power this year, and we are building a data center of 2GW or even larger scale. If this data center were placed in Manhattan, its footprint would be quite substantial. At the same time, we will actively grow revenue by promoting the commercialization of AI to support these investments. To this end, we have developed plans to accelerate these revenue growth initiatives in the coming years. This is also the main direction for our new hires, and the level of execution will determine our financial trajectory in the coming years.

2. Basic Apps

In terms of our app ecosystem, this year will also be a year where key trends for multiple products become clearer. For example, there is still uncertainty about the future of TikTok, but regardless, I expect Instagram Reels and Facebook Reels to continue to grow. Meanwhile, Threads will continue to move towards becoming a leading global discussion platform, ultimately reaching 1 billion users in the coming years. Currently, Threads has over 320 million monthly active users, with more than 1 million new registered users added daily.

Additionally, I expect WhatsApp's market share in the U.S. to continue to grow, moving towards becoming the leading messaging app in the U.S., just as it is in other parts of the world. Currently, WhatsApp has over 100 million monthly active users in the U.S..

Facebook currently has over 3 billion monthly active users, and our focus is on enhancing its cultural influence. This year, I am also very much looking forward to bringing Facebook back to its early core social experience.

Finally, this year is also a key year for our metaverse strategy. Users of Quest and Horizon are steadily increasing. This year, we will see the results of many long-term investments, which will make the metaverse visually more stunning and more attractive. By the end of this year, we expect to have a clearer view of Horizon’s future development path3. Government Relations

This year is also an important year for us to reshape our relationship with the government. The new U.S. government is more friendly towards technological innovation, supports the global competitiveness of domestic tech companies, and is willing to defend our values and interests. I am optimistic about the potential progress and innovative opportunities brought by this change.

In summary, 2025 will be a big year and one of the most exciting and vibrant years I have seen in this industry. Artificial intelligence, smart glasses, large-scale infrastructure construction, accelerated business growth, and the future development of social media—there are so many things we need to drive forward. I firmly believe that we will create excellent products that shape the future of human connection.

3. Business Guidance

Revenue drivers: first is user volume; second is monetization rate.

3.1 User Volume

First, the daily active users of Facebook, Instagram, and WhatsApp continue to grow year-over-year, both globally and in the U.S. In the fourth quarter, the global video watch time on Instagram grew by double digits year-over-year, and we also saw strong growth on Facebook in the U.S., with video watch time increasing by more than double digits year-over-year.

We expect that by continuously optimizing our ranking system, we will continue to drive video growth in 2025. We are also undertaking some product innovations to ensure that our platform can achieve long-term success. Creators are one of our core focuses.

On Instagram, we continue to prioritize original posts and content to help small creators get discovered. We also want to provide creators with a space to experiment with content. Therefore, we launched a new feature in the fourth quarter that allows creators to share a short video first with people who do not follow them. This enables them to test content before deciding to share it with fans, understand which content performs best, and also helps them reach a brand new audience.

Creative tools are another area we are investing in. In the coming weeks, we will launch a standalone app called "Edits," providing a full suite of creative tools to help creators make great short videos on their phones.

Another focus is making it easier for people to connect through content. Short videos are already being shared over 4.5 billion times a day, and we have introduced more features that combine Instagram's social and entertainment functionalities.

In the U.S., we recently launched a new feature in short videos that showcases content liked or commented on by friends. We have seen very positive initial results and plan to roll it out globally in the coming months.

On Threads, we made significant progress in 2024, and our goal this year is to make Threads a platform where people can follow the content they care about. We will implement multiple updates to the recommendation system, prioritizing more recent posts, showcasing content from top creators, and ensuring users see more content from accounts they follow. We will also continue to improve the custom feed, allowing users to create personalized feeds based on their interestsFinally, the usage of Meta AI continues to grow, with over 700 million monthly active users. We are rolling out updates that enable Meta AI to provide more personalized and relevant responses by remembering users' previous query details and considering their interactions on Facebook and Instagram, thereby better understanding their interests and preferences.

3.2 Monetization Efficiency

Next, let's look at the second factor driving revenue performance, which is improving monetization efficiency. The first part of this work is optimizing the balance between ad placements and organic engagement. We continue to increase ad supply on surfaces with low monetization, such as video, while optimizing our ad supply on each service to deliver ads at the most relevant times and places.

For example, we continue to better personalize the timing of ad displays, including introducing ad supply in deeper positions within users' feeds to find the optimal timing for users and revenue. This allows ad supply to grow efficiently.

In the long term, we also see opportunities for increased ad display volume on unmonetized surfaces, such as Threads, where we will begin testing ads this quarter. We expect the rollout of ads on Threads to be a gradual process and do not anticipate it to become a significant driver of overall ad display volume or revenue growth by 2025.

The second part of improving monetization efficiency is enhancing marketing effectiveness. Our ongoing improvements to the ad ranking system are a key driver of this work.

In the second half of 2024, we will launch an innovative machine learning system called "Andromeda" in collaboration with NVIDIA (ad retrieval system). This more efficient system increases the model complexity we use for ad retrieval by 10,000 times. Ad retrieval is the part of the ranking process where we narrow down tens of millions of ads to a few thousand displayed to users. The increase in model complexity allows us to run more complex predictive models, better personalizing ad displays.

This has resulted in an 8% increase in ad accuracy for users in our testing targets. Andromeda's ability to efficiently handle a large volume of ads also prepares us for the future, as advertisers will use our generative AI tools to create and test more ads.

Another way we provide value to advertisers is through the automation of campaigns via Advantage+. The scale of Advantage+ shopping campaigns continues to expand, with annualized revenue exceeding $20 billion and a 70% year-over-year growth in the fourth quarter. Given the strong performance and interest we see in Advantage+ shopping and other end-to-end solutions, we are testing a new streamlined campaign creation process that allows advertisers to no longer choose between manual or Advantage+ sales or application campaigns.

In the new setup, all campaigns optimized for sales, applications, or lead goals will enable Advantage+ from the start. This will allow more advertisers to leverage the performance offered by Advantage+, while still being able to further customize certain aspects of the campaign when neededWe plan to promote to more advertisers in the coming months and fully launch by the end of the year.

Advantage+ creative is also an area where we see growth momentum. There are now over 4 million advertisers using at least one of our generative artificial intelligence advertising creative tools, up from just 1 million six months ago. The first video generation tool we launched in October has seen significant early adoption, and hundreds of thousands of advertisers are using the image animation tool each month.

3.4, Business Outlook

Our top priority remains reinvesting capital back into the business, with infrastructure and talent being our priorities. First, we expect computing power to be at the core of many opportunities we pursue as we advance the capabilities of Llama, drive the use of generative artificial intelligence products and features across the platform, and push core advertising and natural engagement programs. We are meeting the growing capacity demands of these services by scaling our infrastructure and improving workload efficiency.

Another way we achieve efficiency gains is by extending the lifespan of servers and related network equipment. We expect to be able to use non-AI and AI servers for a longer time before replacement, which we estimate to be about 5.5 years. This will save on annual capital expenditures and the resulting depreciation costs, which are already included in our guidance.

Finally, we achieve cost efficiency by deploying our custom MTIA chips in areas where we can optimize chips to fit our unique workloads . In 2024, we will begin deploying MTIA to our advertising and natural content recommendations. We expect to further expand the adoption of MTIA in these use cases in 2025 and extend our custom chip work to ranking and recommendation training workloads next year.

From a hiring perspective, our focus remains on increasing technical talent to support our strategic priorities. In the fourth quarter, nearly 90% of our annual employee growth was in R&D functions. The remaining growth was primarily in revenue costs as we increased infrastructure staff to support our data center operations.

In 2025, we expect employee growth to continue to be driven mainly by technical positions in priority plans such as infrastructure, monetization, reality labs, generative artificial intelligence, and regulatory and compliance. We expect employee growth in business units to be relatively limited. To achieve our ambitions in these areas, we need to continue executing quickly. We support this by building tools to help our engineering teams improve productivity. Over the past two years of focusing on efficiency, we have made significant improvements in internal processes and development tools, introducing new tools like AI-driven coding assistants to help our engineers write code faster.

Looking ahead, we expect continued advancements in Llama's coding capabilities to provide greater leverage for our engineers as we focus on expanding its capabilities to not only assist our engineers in writing and reviewing code but also to start generating code changes to automate tool updates and improve the quality of our codebaseFinally, we expect our strong financial position will enable us to support these investments while continuing to return capital to shareholders through stock buybacks and dividends.

(1) We expect total revenue in the first quarter to be between $39.5 billion and $41.8 billion. This reflects an annual growth of 8% to 15%, or a growth of 11% to 18% on a constant currency basis, as our guidance assumes a negative impact of approximately 3% from foreign exchange on annual total revenue growth based on current exchange rates. This also reflects the impact of the leap day in the first quarter of 2024.

While we do not provide revenue guidance for the full year of 2025, we expect our investments in core business this year will give us the opportunity to continue achieving strong revenue growth throughout 2025.

(2) Next, let's look at the spending outlook. We expect total expenses for the full year of 2025 to be between $114 billion and $119 billion. We anticipate that the largest driver of expense growth in 2025 will be infrastructure depreciation, followed by employee compensation, as we will be increasing technical talent in the priority areas I mentioned earlier.

(3) Next, let's look at the capital expenditure outlook. We expect capital expenditures for the full year of 2025 to be between $60 billion and $65 billion. We expect the growth in capital expenditures in 2025 to be driven by investments supporting our generative AI efforts and increased investment in core business. In 2025, most of our capital expenditures will continue to be directed towards the core business.

(4) We expect the full-year tax rate for 2025 to be between 12% and 15%. Additionally, we continue to closely monitor the active regulatory environment, including legal and regulatory headwinds in the EU and the US, which could have a significant impact on our business and financial results.

IV. Analyst Q&A

Q: I know there will be many product announcements this year, but can you share some examples that illustrate your vision for new potential application scenarios and products? How will these products enhance user experience and create value for advertisers? Particularly regarding the development direction of Llama 4 and Meta AI in 2025?

A: Regarding your first question, I have already mentioned some points in my opening remarks. Our current focus is on Meta AI, which we aim to develop into a highly intelligent and personalized assistant that can be used across our various applications and can also be accessed through a standalone website.

We believe that the quality of Meta AI will continue to improve, with significant progress made over the past year. Additionally, we are actively exploring how to better integrate it into our services, allowing more users to discover and use it. As a result, Meta AI is currently being used by hundreds of millions of users, which is also a result of our ongoing efforts to optimize product usability.

As for Llama 4, I have previously provided some technical details that may be interesting to tech enthusiasts, as we have never discussed these topics before. However, I won't reveal too many details about the products set to be released this year, leaving some surprises. Nevertheless, we will be innovating extensively around Meta AI, and Llama 4 will also bring a series of exciting updatesLlama4 will not be a single product release, but rather a phased rollout of multiple different models, similar to Llama 3.

In addition, I am very excited about our progress in the AI engineering direction. Although it will not become an external product in the short term, we are actively advancing AI research and applying it to internal product development. This technology may bring significant market opportunities in the future, but this year it is mainly used to enhance the capabilities of our existing products, driving greater changes in AI in 2026 and beyond.

Q: Regarding custom chips, what experiences does Meta have in the application of self-developed chips and third-party chips in computational efficiency and ranking models? What are the main reasons for adopting custom chips in the future?

A: First of all, we will continue to procure third-party chips from industry-leading suppliers and maintain long-term partnerships. At the same time, we are also actively developing self-developed chips to address specific computational needs, especially for tasks that cannot be optimized with standard chips available on the market.

Our self-developed chips can optimize the entire computing architecture, achieving higher performance in computational efficiency, unit cost, and power consumption. This is mainly because Meta's workloads have unique requirements in memory, network bandwidth, and computational resource allocation, and self-developed chips can be precisely optimized according to these needs.

Currently, our MTIA (Meta Training and Inference Accelerator) is primarily used for core ranking and recommendation inference tasks and will officially be put into use in the first half of 2024. This year, we will continue to expand the application scope of MTIA, not only to meet new computing power demands but also to gradually replace some GPU servers that have reached the end of their lifespan.

Looking ahead, we hope to enable MTIA to support some core AI training tasks by 2025 and gradually expand to generative AI-related application scenarios.

Q: How has Meta's view on open-source strategy evolved? How do you see the competitive landscape of open-source strategy in the AI field? From the perspective of cost control and capital returns, can the open-source model bring long-term advantages to Meta?

A: In terms of open-source strategy, our thinking is similar to the OpenCompute project. Back then, we were not the first company to establish a data center hardware system, so we decided to open-source it to promote the adoption of our solutions across the industry and encourage ecosystem innovation. Ultimately, this standardization made the supply chain more efficient and reduced costs.

In the AI field, we believe that the widespread application of Llama will drive more chip suppliers, API developers, and development platforms to optimize it, thereby reducing usage costs and improving overall performance. As the industry's adoption rate increases, we can also benefit from these optimization results.

Additionally, we have noticed the recent rise of open-source competitors like DeepSeek in China. A global open-source standard may emerge, and we hope that this standard can be led by the United StatesTherefore, we will continue to promote the development of the AI ecosystem, allowing global developers to use the AI technology developed by Meta. Recent industry dynamics have further strengthened our confidence in the open-source strategy, which we believe is the right direction.

Q: In terms of the development trend of smart glasses, does Meta believe that the best hardware form for AI assistants is smart glasses? Or is it just a scenario supplement to the in-app AI experience?

A: Yes, I have always believed that smart glasses are the ideal form of AI devices because they allow AI assistants to "see" your perspective and "hear" what you hear, thus providing a more contextually aware interactive experience. In the future, smart glasses will become an important computing platform.

When smartphones became the primary computing platform, traditional PCs were not eliminated. Similarly, smartphones will still exist in the future, but more and more users will turn to AI glasses. This is an irreversible trend, and I expect that in the next decade, all glasses on the market will have AI capabilities, and even some people who originally did not wear glasses will start using AI glasses.

Last year, I believed that smart glasses would only become mainstream products when they had holographic display technology. But now, I think the value brought by AI itself may be as important as holographic AR. Therefore, the development of AI glasses may happen faster than we anticipated.

The success of Ray-Ban Meta smart glasses indicates that there is market demand for such products. Although we are still unclear about the long-term development trajectory, 2025 will be a key year for exploring the potential of the smart glasses market.

Q: Regarding the advertising business, Meta's advertising revenue growth in recent quarters has mainly been driven by an increase in ad prices. What will be the main driving factors for future advertising growth? Will the trend of price growth continue?

A: In the long term, we believe that the growth of advertising revenue can still be driven by both ad pricing and ad display volume. In terms of pricing, growth may be influenced by various factors, including the supply of ad placements, bidding dynamics, and changes in ad display positions. For example, the monetization efficiency of video ads is usually lower than that of other ad types, which can also affect the overall pricing trend. Additionally, the macroeconomic environment is an important factor.

However, we expect that Meta can still enhance the return on investment (ROI) for advertisers through continuous optimization of ad performance. Our technological innovations in ad monetization, including optimizing delivery algorithms and improving ad targeting accuracy, will contribute to the long-term growth of ad prices.

It is important to note that ad pricing is a composite metric influenced by multiple variables, such as advertisers' bidding strategies and ad conversion costs. Therefore, although prices may fluctuate, we generally see that ad conversion costs remain at a healthy level. As Meta continues to improve ad conversion rates, we expect future CPM (cost per thousand impressions) to continue to show an upward trend.

Q: Mark, you mentioned that political changes in the United States may make American companies more competitive overseas. But domestically in the U.S., how do you view the impact of these changes on user usage and advertiser adoption?You recently canceled the fact-checking mechanism. Do you think this will affect the platform's content? Could this potentially attract more users? At the same time, will this have an impact on the advertising business?**

Susan, regarding Meta AI, there is excitement about its application scenarios, but there is also concern about its commercialization path. How are you considering the monetization model for Meta AI? Is it possible to adopt a CPC (cost-per-click) advertising model in the future?

A: Regarding the questions about fact-checking and content policy, our goal has always been to create the best service for users. I have always supported freedom of speech, while we also recognize that people do not want to see misinformation, so we need to establish an effective system to provide more contextual information.

From practical experience, we have found that the Community Notes system is more effective than the previous fact-checking mechanism. I am willing to admit that when other companies develop better methods, we should also learn and optimize our own systems.

The recent policy adjustments do not mean we have lost focus on the quality of platform content. In fact, we believe that the Community Notes system (similar to the model on platform X) is more effective than the previous fact-checking mechanism and can provide a better experience for users. Therefore, I believe these adjustments will ultimately make our products better.

As for the advertising business, we currently do not see a significant impact on advertising spending due to the content policy adjustments. Advertiser demand remains strong, especially as AI-powered advertising tools help businesses optimize the value of their advertising spending more effectively. Therefore, our commitment to brand safety remains unchanged, and we will continue to invest in brand safety tools to meet advertisers' needs.

Regarding the commercialization of Meta AI, our initial focus is on creating an excellent user experience, and all our efforts are currently concentrated on improving product quality.

In the future, the monetization methods for Meta AI may include: paid recommendations and premium subscription services, but for now, our main goal remains product development rather than monetization.

Q: Mark, regarding the open-source issue, new models like DeepSeek are leveraging Llama for faster and lower-cost training. What impact does this have on Meta? Will it affect Meta's investment direction in the coming years?

Regarding the capital expenditure (CapEx) for 2025, you expect to invest $60-65 billion. Has there been any change in the investment structure compared to last year? Is the server still the largest expenditure item? Has there been any adjustment in the investment allocation for data centers and network equipment? Particularly in terms of allocation for AI training and inference computing?

A: Regarding DeepSeek, we are still evaluating their innovations and plan to absorb some of their technological advantages. Regardless of where competitors come from, advancements in the entire AI field are often a process of mutual learning, and each new technology release pushes the entire industry forward.

Currently, it is difficult to determine the long-term impact of these developments on infrastructure investment and capital expenditure. The allocation of AI computing resources is undergoing structural changes: In the past, computing power was mainly concentrated on pre-trainingNow, more and more computing resources are shifting towards inference, which refers to the computational demands of AI in real-time applications.

This trend suggests that future AI computing may not necessarily require more pre-training resources, but the load for inference computing could significantly increase. This presents a long-term advantage for companies supported by strong business models, as we can continuously invest in infrastructure to provide higher quality AI services, while some competitors who cannot bear the high computing costs may struggle to sustain themselves.

Additionally, our AI computing infrastructure is not only used for Meta AI but is also widely applied in information flow recommendations (Feeds) and advertising products. Therefore, we need to ensure that we have sufficient computing capacity to support the demands of billions of users.

Regarding capital expenditures in 2025, servers remain the largest investment direction and are expected to grow significantly, mainly due to:

(1) Increased AI computing capacity: Supporting generative AI (GenAI) and the expansion of core AI businesses.

(2) Maintaining core businesses: Expanding infrastructure to meet user growth and updating outdated servers.

(3) Investment in data centers will also increase: Building large-scale AI training clusters; core construction phase of high power density data centers.

(4) Investment in network equipment: Higher capacity networks to meet the traffic growth of AI and non-AI computing; construction of fiber optic infrastructure to optimize cross-regional AI training data transmission.

Overall, Meta's infrastructure investments in basic AI, non-AI businesses, and generative AI will all increase, with a continued focus on supporting core businesses.

We are still in the early stages of AI computing, and it remains uncertain whether computing costs will significantly decrease in the future. Therefore, we are currently maintaining large-scale investments to ensure long-term competitiveness.

Q: Meta recently emphasized a return to "OG Facebook" (the original Facebook experience). Can you elaborate on how to expand the application scenarios in this direction? Are video and marketplace a focus?

At the same time, Meta AI currently has over 600 million monthly active users. How do you view the evolution of user experience? What are users mainly doing right now?

A: Facebook remains an important part of many users' daily lives. I believe that Facebook still has great potential to unleash and can rise again in terms of cultural influence.

Currently, we are exploring some new product directions, but they will not immediately impact short-term business growth. This process requires trade-offs, and we may make sacrifices in certain areas to focus on long-term product innovation.

I will share more details in the next 6-12 months. Overall, we hope Facebook can return to the early community interaction model and regain its vitalityRegarding Meta AI, we are currently in the user behavior learning phase. From the data of various applications: WhatsApp is the platform with the highest usage of Meta AI, with main application scenarios including information inquiry, learning support, and emotional communication. Facebook is the second largest platform for Meta AI usage, where users primarily experience AI recommendation features deeply integrated through the Feed.

Overall trends show that users are gradually applying Meta AI to: information search and organization; social interaction and communication; leisure and entertainment (such as humorous exchanges); content creation and editing (like writing assistance).

By 2025, the core goal of Meta AI is to optimize personalized experiences, including: enhancing AI memory capabilities to remember key information from users' private chats for more accurate responses. Improving the quality of content recommendations to enhance the value of Facebook and Instagram.

Five, here is the content from the Meta 4Q24 Callback meeting:

Q: Susan, could you please provide an update on the business AI (artificial intelligence) initiatives for WhatsApp? Share some key lessons learned so far. Additionally, could you discuss the rollout of products in non-English speaking countries?

A: We are still in the early stages of piloting business AI, with a relatively small number of pilot companies, but we are gradually expanding the testing scope. Currently, our focus is on supporting businesses that use WhatsApp and Messenger click-to-message ads. For WhatsApp, we have launched English services to more small businesses in India. Additionally, we have expanded to several countries in the Asia-Pacific region (including Indonesia and Malaysia) and are gradually expanding Spanish business AI in Latin America (including Mexico). In terms of Messenger, we have conducted English testing in several countries in the Asia-Pacific region (including the Philippines, Singapore, and Malaysia) and are also expanding to Spanish-speaking countries in Latin America.

We are in the early stages, and as the testing scope expands, we have received encouraging feedback. Businesses tell us that AI saves them time, while consumers report that they receive responses to inquiries more promptly. Our newly established business AI product team is dedicated to accelerating the scaling efforts of business AI, and we are making comprehensive investments in technology, marketing, and the partner ecosystem.

We focus on helping businesses leverage AI to drive ad conversions, which is a use case we want to expand. We are exploring how to activate AI and chat features to support business use cases, helping to generate or filter potential leads. Our ultimate goal is to create AI products that businesses can use across various use cases to deepen customer relationships and make the entire transaction process more efficient.

However, there is still much work to be done on our roadmap, with many goals we want to achieve. We need to execute many things, including enabling businesses to access business data and systems more broadly so that AI can respond appropriately and take action. We also need to build richer in-thread experiencesI believe this will take time, and we are working hard to advance it. This is a very important priority for us in 2025, and it is also another area where we expect to benefit from the continued progress of Llama, which will further improve response quality and enable new features.

Q: Does the total expense guidance include severance costs related to the recent 5% layoffs? How significant are these costs? When you talk about employee growth in fiscal year 2025, can it be reasonably assumed that the high single-digit percentage growth rate seen in the past two quarters is what you are implying?

Additionally, I would like to ask again about DeepSeek. From what we have learned about DeepSeek over the past week or even the past couple of months, there have been some fantastic efficiency improvements and cost savings. Do you expect to leverage these results in the future, thereby either significantly increasing the ROAI (Return on AI Investment) of all your investments or providing you with a way to reduce AI inputs? Can it be assumed that the future cost structure will be positively impacted?

A: Regarding total expense guidance: The total expense guidance does indeed include severance costs related to the 5% layoffs. The total expense guidance covers everything in our current outlook, including savings generated from extending server lifespans. Without these savings, the total expense guidance would be higher. We have not quantified the specific amount of severance costs, so we will not list them separately.

Regarding employee growth: This is not a 5% layoff; the gaps in positions will be filled. We do not expect the company's employee size to shrink by 5% as a result. As for the employee growth rate in 2025, we have not quantified it, but we have discussed which areas are growing and where we are very focused on improving efficiency.

Regarding DeepSeek: I am essentially reiterating what Mark said earlier. We all agree that DeepSeek has made good research progress in model design and efficiency. As you mentioned, we are studying the results they have released and considering where it makes sense for us to make adaptive adjustments. At the same time, as Mark mentioned, we are still committed to investing in the infrastructure for training and inference, partly because we are still unclear about what we actually need and how broad our inference use cases will be, and partly because we believe this is a real competitive advantage for us.

Therefore, while I agree with your point that we can conduct some work more efficiently in the future, I must reiterate my previous comment that we are still in the early stages of capital expenditure investment, and we are not clear on how the overall landscape will evolve.

Q: I don't think I have taken it for granted, Susan, that you may have indicated over the past year that advertising growth rates might be somewhat constrained by capacity limitations. So, can it be assumed that as you set the capital expenditure budget for this year, the constraints on the development of the advertising business have been somewhat lifted, thereby enabling the growth of the advertising business?Additionally, you have mentioned that in certain regions of the world, there is a stronger cultural tendency for users and businesses to communicate through platforms like WhatsApp and Messenger. So, is this behavior something we need to promote in developed markets where such communication is not yet widespread, to make business AI more practical? Thank you.

A: Regarding capacity constraints: We expect to significantly alleviate capacity constraints in the first half of this year as more data center space comes online and we work to improve workload efficiency to free up more capacity. At the same time, we are also focusing on other investments to enhance advertising performance more broadly.

Regarding cultural differences and business behavior: I think you are right. You may have noticed my previous comments about the regions where business AI has been launched, which reflect markets where we know business messaging is already active. We also know we need to invest more aggressively to figure out how to operate in markets that have not yet taken this path for various historical reasons. Therefore, I see this as a challenge to establish broader business messaging, but at the same time, it is an area where we believe AI can help reduce the cost of business-to-business conversations to some extent, thereby facilitating dialogue and interaction between businesses and users in developed markets in a scalable way.

Q: In the days leading up to TikTok's suspension (even though the actual suspension lasted only about 14 hours), did you observe any changes in advertiser behavior and/or their spending on the platform?

Another question about gross margins for 2025. I know you do not provide gross margin guidance, but given the significant increase in capital expenditures planned for this year and the extension of server lifespans to five and a half years, do you think these two factors could offset each other? How will this year's gross margin be affected? Can they offset each other?

A: Regarding TikTok's suspension: As you pointed out, the actual suspension time for TikTok was very brief. Therefore, it did not have a material impact on our revenue or user engagement metrics for the first quarter of 2025. Moreover, I don't think we have more specific information to share on this matter. Our focus remains on building the best services for our community and driving growth through these efforts.

Regarding the gross margin question: The main drivers of cost revenue growth in 2025 will be infrastructure costs, followed by Reality Labs.

Q: I would like to ask a question about Reality Labs. Susan, could you elaborate on the main categories of spending you anticipate for Reality Labs in 2025? Do you think the scale of spending for Reality Labs has already peaked?

A: Overall, we believe that investments in Reality Labs are primarily distributed across two categories:

(1) Metaverse, including virtual reality (VR), mixed reality (MR), and social ecosystems;(2) Wearable devices, including our augmented reality (AR) and artificial intelligence glasses work. We expect that by 2025, about half of RL's investments will be allocated to wearable device programs, with the other half going to the metaverse.

What I want to say is that in these two areas, we are still making significant product bets, and there are a range of complex challenges that we need to unlock in order to bring scalable consumer products to market, which is our true focus. Therefore, it remains one of our core investment priorities. We expect RL's operating losses to increase in 2025, just like in 2024.

In terms of growth, we anticipate that wearable devices will be the main factor driving the expansion of RL's operating losses in 2025, reflected in both cost and revenue as well as operating expenses. This mainly stems from our efforts to further accelerate the adoption of AI glasses products.

Q: Susan, you mentioned that there might be some gray areas between different categories of capital expenditures, such as older generation GPUs being used for core functions. But if you go back to the definitions you mentioned, namely Gen-AI, core AI infrastructure, and non-AI infrastructure, could you elaborate further on the definitions of these categories?

Also, I may have missed this detail in the conference call, but does the five-and-a-half-year lifespan for servers apply to all these categories?

A: Regarding how to roughly think about these categories, generative AI (Gen AI) capabilities mainly refer to the capabilities we are currently using to train the next generation of Llama models. Therefore, it is primarily about training capabilities, but we expect and hope that there will be significant demand for inference capabilities in the future.

The core AI we mentioned actually refers to the capabilities that power ranking and recommendation engines, which support effective content experiences as well as advertising experiences. Essentially, it is about determining which ads and organic content any given user might find most interesting and useful in their experience.

Non-AI refers to things that do not fall into these two categories. For example, the capacity used to provide you with video streaming falls into the non-AI category. So, these three categories are defined this way. And as you suggested, it is not a very clear allocation. The extended lifespan applies to all categories.

Q: Susan, I have two questions. One is about Dan's earlier response regarding the definition of expenditure categories. I would like to know if you could help us understand the geographic distribution of capital investments to better understand the ranking and advertising experience in different regions? Because I am trying to look at the average pricing in various regions, and I know there is a lot of marketing involved. But could you provide more information about the regions where you have already launched these AI programs and what we are seeing? That would be very helpful.

The second question is about Reality Labs; how do you measure return on investment (ROI)? I know there are many new ideas that are still in the early stages. So as the CFO, how do you test whether these expenditures are effective or ineffective? Thank you.A: Regarding the geographical distribution of capital expenditures: In terms of core AI use cases, such as improving next-generation advertising architecture models and for Andromeda (Meta's advertising retrieval system), this is unrelated to regional distribution. Generally speaking, the features we release to improve advertising performance, such as ranking and recommendation systems, delivery backends, etc., are typically rolled out globally, except during the testing phase.

The generative AI program itself, particularly business AI, we have previously discussed , is actually being tested in markets that have already launched commercial messaging and have a thriving click-to-message advertising ecosystem. Therefore, this tends to be more concentrated in regions like Southeast Asia and Latin American countries. However, much of the foundational work we are doing in advertising growth, such as ensuring we launch advertising features targeted at different verticals, does not have any specific geographical limitations.

Regarding the ROI of RL: I believe Mark has articulated the vision for the Reality Lab, which is truly committed to building the next-generation computing platform. If we can successfully achieve this, it will position us very importantly in shaping the next-generation computing platform and building an ecosystem that is very complementary to today's Family of Apps. I also believe that building a business on top of this will be a natural extension of our existing advertising business, but you can imagine that it will also unlock new businesses around digital goods and so on.

In the short term, I think one factor driving our investment in 2025 is our emphasis on wearable devices. This has prompted us to accelerate the timeline for the next-generation computing platform. Therefore, we do expect to continue making significant investments in the wearable device category to help accelerate this process. Mark has already set a target of 5 to 10 million units for the next-generation product. Ultimately, this will put you on a path toward hundreds of millions of devices and become a scaled consumer product. But this is a field where we are developing the market, which requires our investment.

Q: I just want to ask one question. As the confirmation hearing for RFK Jr. (Robert F. Kennedy Jr.) is underway, we have been asked about pharmaceutical advertising in digital companies. Is there any way we can think about the overall risk size in the pharmaceutical sector? Of course, I know there are many things we don't know. But if pharmaceutical companies change their advertising strategies, do we have any way to frame our thinking around this situation? Thank you.

A: I don't think this is something we need to share specifically. However, I just want to say that overall, we have a very diverse advertising business across geographical regions and verticals. Therefore, this will not affect our outlook for 2025.

Q: I think this is a more philosophical question and expands on what Mark mentioned at the beginning of the call. He talked about the goal of reaching a billion users. When you think about the AI landscape, much of what exists in the digital and internet space is winner-takes-all. How do you think this will evolve?Once again, this is related to what has happened with DeepSeek over the past five weeks and the news announced by Alibaba today. How do you think we should think about this? Or how are you thinking about it? How will market share be distributed in this landscape and the development of AI? Is this different from past digital and internet cycles?

A: This is a very interesting question. I think the part I am most confident in discussing is that I believe there are many reasonable ways to build businesses based on leading AI technology. These businesses are, in some cases, extensions of our existing business.

Therefore, whether it is generative AI creating engaging content for our ecosystem in many new and simpler ways, which is clearly beneficial for the experience of our Family of Apps, or leveraging these technologies to help advertisers run highly customized advertising campaigns at scale, making our ads better. Whether it is building a richer commercial messaging ecosystem or Meta AI, which is our consumer-facing experience version.

I believe we have many avenues to truly build adjacent businesses based on AI technology that are intuitively related to our existing business and have significant opportunity scale.

I think it may be difficult to comment on how other parts of the entire landscape will develop. But I am very optimistic about our opportunity to truly build great businesses in this landscape.

Q: May I ask two questions? First, Clara (a new team member) is a huge addition for us. Can you discuss some areas where Clara and her team may have a significant impact in the short to mid-term in the enterprise and SMB sectors? Do you think there is an opportunity to monetize Llama through APIs, subscriptions, or more directly through certain cloud services?

Additionally, I would like to ask about the disclosure of the Advantage+ shopping revenue run rate. As you promote the simplified campaign creation as the default option to a broader range of advertisers, this metric accounts for about 10% of total revenue. So, does this performance improvement apply to all types of advertisers, or only to certain specific categories? Basically, I am trying to figure out to what extent the performance improvements you see through Advantage+ can be further scaled. Thank you very much.

A: Regarding Advantage+ shopping: I believe you are asking about the new simplified campaign setup process we started testing in the fourth quarter. This process combines Advantage+ shopping and Advantage+ app campaigns with the manual setup process. In the new design, there is now only one campaign creation process. Advertisers automatically benefit from the performance advantages of Advantage+. Previously, they had to choose between setting up a manual solution or an automated solution. They still have access to the features that were previously available if they prefer to manually adjust settings. But overall, we generally believe that the new simplified process will help advertisers leverage the performance improvements achieved through Advantage+. We also plan to expand these tests to more advertisers in the first quarter and roll them out more broadly in the coming monthsOverall, we are optimistic about this area because in our research, advertisers have seen a significant improvement in return on ad spend after adopting Advantage+ Shopping campaigns. Therefore, we believe that helping them achieve this more easily will enhance their performance.

Regarding Clara joining the team: We are very pleased to have Clara join our team. Her initial focus will be on general business agency, and we believe this will leverage our strengths in AI and messaging platforms like WhatsApp and Messenger. I think I mentioned before that the product team she leads is expected to accelerate our efforts to scale business AI. We are investing in technology, marketing, and ensuring we have a rich partner ecosystem. Therefore, we are very excited to have her join our team.

Q: Can you provide more details about the first quarter revenue guidance? Given the strong performance we saw in the fourth quarter, the guidance for the first quarter seems a bit conservative. It looks like you had a more challenging comparison base in the first quarter of last year, as well as foreign exchange (FX) factors. So, can it be said that these factors are the reason the guidance isn't a bit higher? Thank you.

A: Yes, the first quarter outlook includes multiple outcomes. Overall, it reflects our expectations for continued strong growth in currency revenue. However, the biggest factor leading to a slowdown on a reported basis is the exchange rate factor, which we expect to impact by 3 percentage points in the first quarter, while in the fourth quarter, the currency factor's impact on revenue was roughly neutral, mainly due to the strengthening of the dollar against the euro.

In addition to currency factors, we also face a high comparison base from the first quarter of last year, which includes the benefits of an extra day from the leap year in 2024. Furthermore, since we have just concluded the U.S. elections, political ad spending is also unlikely to be a tailwind for year-over-year growth as it was in the fourth quarter.

Therefore, our outlook reflects all these impacts, but overall, ad demand remains strong. We continue to see opportunities for strong revenue growth throughout 2025, and we are excited about the investments we are making in our core business. I think that’s all for now, thank you all for joining, and I wish everyone a Happy Lunar New Year.

Disclosure and Statement: Dolphin Investment Research Disclaimer and General Disclosure