Good day, and welcome to the Cheetah Mobile Fourth Quarter 2024 Earnings Conference Call. All participants will be in listen-only mode. [Operator Instructions] After today’s presentation, there will be an opportunity to ask questions. [Operator Instructions] Please note, today's event is being recorded. I would now like to turn the conference over to Helen Jing Zhu, Investor Relations for Cheetah Mobile. Please go ahead.

Thank you, operator. Welcome to Cheetah Mobile's fourth quarter 2024 earnings conference call. With us today are our company's Chairman and CEO, Mr. Fu Sheng; and our company's Director and CFO, Mr. Thomas Ren. Following management's prepared remarks, we will conduct a Q&A section. Please note that the CEO's script will be presented by an AI agent. Before we begin, I refer you to the safe harbor statements in our earnings release, which also applies to our earnings conference call today, as we will make forward-looking statements. At this time, I would like to turn the conference call over to our CEO and Chairman, Mr. Fu Sheng. Please go ahead.

Hello, everyone. Thank you for joining us today. In Q4, our total revenue accelerated robustly, growing 42% year-over-year and 23% quarter-over-quarter. We continue to make progress by consistently reducing non-GAAP operating losses quarter-over-quarter, and we expect our first quarter's non-GAAP operating losses to further decrease both year-over-year and quarter-over-quarter. For the full year 2024, total revenue increased by 21% to RMB807 million, and our non-GAAP net loss attributable to shareholders narrowed year-over-year. This performance highlights our commitment to high-quality growth and positions us strongly for a continued business turnaround in 2025. Our growth is driven not only by our enterprise-facing AI and robotic businesses, but also by our legacy Internet business, which continues to deliver strong year-over-year revenue growth and margin expansion. Our apps for PC and mobile users continue to address pain points effectively, delivering excellent product experience even in a mature market. We believe that this positive trajectory will continue into 2025, benefiting from the momentum of LMS. Cheetah entered 2025 with tremendous opportunities for our robotic business. LMS are playing a key role in transforming the service robotics industry, expanding its total addressable market, and laying the groundwork for service robots to evolve into general purpose physical AI. At [indiscernible], we are taking a measured, strategic approach to growing our robotics business, focusing on product types that can generate seasonal revenue and profits, aiming to become a top three service robot provider globally within three years. Our wheelbase robots are designed to serve real-world business environment, enabling a data flywheel that will further drive product innovation and LLM advancements. At the same time, we continue to enhance our general purpose robotic capabilities by leveraging the most advanced LLMs available in the market, such as DeepSeek, ChatGPT, and et cetera, alongside our self-developed LLMS capabilities, such as voice…

Thank you, Fu Sheng. Hello, everyone, on the call. Please note that, unless stated otherwise, all money amounts are in R&D terms. In Q4, we once again accelerated year-over-year revenue growth and achieved quarter-over-quarter reductions in non-GAAP operating losses. Also, this represented the first quarter that we reduced our non-GAAP operating losses on a year-over-year basis since Q4 2022. Our strong performance in both top line and bottom line demonstrates Cheetah's continued business improvement since Q1 2024. In Q4, total revenue increased by 42% year-over-year, and 23% quarter-over-quarter to RMB237 million. Revenues from our Internet business increased by 49% year-over-year and 19% quarter-over-quarter, accounting for 68% of total revenue. This segment remains a strong cash cow, supporting our initiatives in AI and robotics. Revenues from AI and others increased by 29% year-over-year and 33% quarter-over-quarter. Within this segment, our robotics business contributed about one-third of AI and other revenues, or about 10% of total revenues. Notably, in the quarter, revenue from our Robotics business outgrew other business in this segment, including our advertising agency business and multi-cloud management services. On the profitability side, we continue to manage our costs and expenses while strategically investing in top R&D talent and leading sales professionals to drive our robotics business forward. As of December 31, 2024, we had 935 employees, an increase from 845 in 2023. About 40% of our employees are R&D talent, and about 30% are in sales. From a business perspective, we continue to optimize our costs and expenses and focus our resources on AI and robotics. On a corporate level, our non-GAAP gross profits increased by 74% year-over-year and 32% quarter-over-quarter to RMB172 million in Q4. Non-GAAP gross margins increased to 73% in the quarter from 59% in the same period last year and 68% in the previous quarter. Non-GAAP operating loss reduced by about RMB18 million quarter-over-quarter and about RMB7 million year-over-year to RMB42 million. As we enter the anniversary of consolidating Beijing OrionStar and our investments in AI and robotic business, we expect our operating losses to continue reducing on both year-over-year and quarter-over-quarter basis. As segment, operating margin for our Internet business increased to 16% in Q4, up from 9% in the year-ago quarter and 10% in the previous quarter. Another highlight of the quarter is our strong cash flow generation. We generated about $39 million in operating cash flow, closing Q4 with a net cash position of about $250 million. Additionally, we hold about $100 million in long-term investments on our balance sheet. We are confident in keeping our balance sheet strong, in particular maintaining our strong net cash position going forward. Before opening the call for questions, I want to emphasize our clear objective, to drive revenue growth while achieving breakeven and generating operating profit as soon as possible. Furthermore, we continue to carefully manage our net cash position, as we have recognized that earnings and cash position are the two key metrics that we measure our progress, both operationally and financially.

Everyone, for today's conference call, now we will start the Q&A session. Everyone, for today's call, management will answer the question in Chinese. I think all the analysts on the line will ask their question in Chinese, and we'll have an AI agent who will translate the analysts' questions and also management's comments into English in another line. Please note that translation is for convenience purposes only in the case of any discrepancies or management statement in Chinese as well. If you are unable to hear the Chinese translation, I think a transcript in English will be available on our IR website within seven working days. Operator, now we are ready to take questions. Thank you so much.

[Operator Instructions]

[Technical Difficulty] And can these products achieve coordinated development with Cheetah's robot business?

Well, the term ag is quite popular. We also believe that AI is on the verge of a breakout in various fields, which is our judgment. AI agents within enterprises will surely merge in large numbers and become a mainstream way of applying AI within enterprises. We can better improve the efficiency of enterprises. Our company has also taken a lot of actions in this regard internally, and we also have some external products. Recently, we mentioned our so-called AI tutoring product, which can actually realize internal enterprise training completely through AI. We also have some large partners or rather some of them have become our customers. Regarding the coordinated development with the robot business you just asked about, I think the robot business ultimately depends on the software's intelligent capabilities to determine whether the robot business can truly meet user expectations and reach the stage of commercialization or large scale commercial application. I think that in the past year or so, Cheetah's efforts in AI and large language models can be well applied to the robot field because ultimately their architectures are very similar. For example, as stated in our speech, we will release an AgentOS, a robot operating system. It is actually based entirely on the current large language models and the agent system, raising the intelligent level of robots to a new height. Without our previous experience in applying agents within the enterprise, including helping our customers with AI related work as well as our understanding of various model capabilities and even training models ourselves, it would not be easy to develop and perfect this system. Although everyone is talking about these things these days, to actually make it useful, there are still quite a few detailed technical points to be achieved. I think this is an embodiment of the combination of our efforts in software agents and the robot business that is the [indiscernible] robot operating platform we are going to release.

Thank you. The third question, DeepSeek has promoted the development of the global large language model industry this year. Do you think AI already has the conditions to achieve large scale implementation at a lower cost or do we still need to wait for some key turning points?

Well, okay. I think the emergence of DeepSeek is indeed of great significance to the popularization of the entire AI industry. However, what I want to say is that the most crucial aspect of DeepSeek is not just its low cost, but its top tier performance in the inference model within the industry. And this top tier performance is completely open, sourced and free, which enables every manufacturer to have a large language model with independent intellectual property rights. So I have an opinion. I also mentioned this in an interview not long ago. I said that perhaps in the future, there may be no so-called large language model companies or there will be very few of them and all companies will become application companies. That is how to enable users to truly solve their very practical problems will become a very crucial point this year. Due to the improvement in the logical capabilities of the inference model, we believe that the right time has arrived. So, we think that for the entire AI industry, whether it's software applications or hardware applications, the conditions for achieving large scale implementation, as you asked, have been met. Of course, it will still take some time. This time is more about specific requirements and engineering implementation rather than the time for technical preconditions. So, in this regard, we are very optimistic and confident. Therefore, both our robot products and Internet products will focus on this technological mainline of AI and make every effort. Thank you.

The next question. With the rapid development of large language model technology, the market space for wheeled robots is expanding rapidly. Which types of major companies do you think are likely to enter this track, especially companies like EV manufacturers? Is it possible for them to enter the field of wheeled robots by leveraging their accumulation in hardware, manufacturing and intelligent technology? Do you think Cheetah's accumulation and advantages are sufficient to build a deep enough moat to resist potential competition from major companies? How do you view the changes in the future market competition pattern?

Well, currently, I think that for a rather long period, perhaps in the next two or three years, in my opinion, these major companies may not enter this field, for example, compared with the electric vehicle field. Although wheeled robots have a promising future today, the market is still too small compared to that of electric vehicles and the competition in the electric vehicle market itself is also extremely fierce. So, I think that within the foreseeable two or three years, not many major companies should enter this market because this market is still in the early stage with a lot of investments. In fact, last year, our major investment was in AI and robots. Perhaps in the long-term development, major companies may enter. I think it comes down to focus, that is, we need to seize this window period and go all out to build our own iterative system in the wheeled robot market, especially in this area. I don't think a single technical point can become a moat because with the open source of technology and the availability of various information today, it's difficult for technology to be a complete moat. Instead, it's about truly obtaining user recognition through your products and then having the ability to rapidly iterate according to user needs based on these recognition. I think if we can seize this opportunity window in the next two or three years and establish a complete and rapid product iterative system from sales to R&D, that will be our real moat. And even when major companies enter at that time, there are still many detailed points that are difficult to replicate all at once. Let me give you an example. Some people ask me that if electric vehicle manufacturers already have self-driving technology, won't it be quick for them to enter the robot field? In fact, I said the problems are different. Why? Because the chips installed in the car may cost tens of thousands of yuan, while the selling price of an entire robot is just over CNY10,000. I need to achieve indoor self-driving on a chip with much less computing power compared to them. In fact, the technical difficulty here does not lie in the visual framework, but rather in whether you can use such limited computing power to truly achieve reliable and autonomous indoor navigation, right? In fact, the technical points are different. An inappropriate example is that just because a PC manufacturer makes a small mobile phone doesn't necessarily mean it can do a good job. So I think in this era of rapid technological development, it still depends on your iteration speed and your focus in this field. Thank you.

With the increase in the number of high-quality models with reasonable cost, have you noticed that small and medium sized companies are shifting from model training to model inference in terms of computing power usage? At the same time, which enterprises do you think are still continuously training their own large language models and maintaining a high investment in computing power? What impact do these trends have on Cheetah's future business development?

Yes. Regarding the question you just asked, I've always thought that small and medium sized companies don't need to do model training. In fact, large language models will ultimately become an infrastructure. This infrastructure capability will be accessible to everyone in the future. My view, which hasn't changed much from two years ago until now, is that in the future, large companies will train their own models because they think that without their own large language models, they may not have a secure boundary, right? Otherwise, they might be held back by others. Besides, open source models will continue to rise. And as it seems now, right, I've always believed that the performance of open source models won't be inferior to so-called closed models. And this trend, I think, will become more and more obvious, because when computing power is not abundant, it will instead lead to resource constrained innovation with more attempts in algorithms and other aspects. So ultimately, a kind of balance will be achieved. I think it's correct that many enterprises are now shifting from model training to inference because in the future, the computing power of each enterprise will become a standard feature just like human resources. There will be a lot of work completed with computing power instead of relying solely on human labor. As for the impact on Cheetah's business, first, it's definitely changing our R&D model. We're already making extensive use of AI in aspects such as programming and design rather than the completely manual methods we used before. So, our entire reliance on so-called professional and highly skilled personnel is decreasing. At the end of last year, we also established a R&D center in Shanghai with the help of AI young people recruited in second and third. Tier cities like this can quickly develop strong problem solving abilities. As for the impact on ourselves, I think there are several aspects. As I mentioned, in the robot field will make AI agents our core focus. In fact, robots today are a very broad concept, right? There are humanoid robots, cleaning robots, heavy load robots, robotic arms, et cetera. But our advantage lies in truly interacting with users and intelligently completing tasks. This is reflected in the AI based OS we released. Second, essentially, all these Internet applications will become AI enabled and intelligent. We're now intensively transforming some of our traditional software to better present it to users in an AI based way. So I think [Technical Difficulty]

[Technical Difficulty] regarding the progress of large language models such as DeepSeek-V3 and R1, how do you view the impact of the introduction of these models on the competitive landscape of AI applications? How will Cheetah utilize these technologies to drive the company's development in the next few quarters, especially to enhance product competitiveness?

More or less, I don't know if you guys noticed, the day before yesterday, DeepSeek upgraded its V3 version again. We've just tested this latest version. In fact, its coding ability is on a par with that of the top-notch cloud-based coding. And in many aspects, its performance is similar to that of those closed up stores and even paid large language models. So in terms of the impact on the competitive landscape, I think in the foreseeable future, the innovation of DeepSeek will continue. The competition in the performance of large language models will become even more intense. Eventually, only a few manufacturers will remain in the competition for large language models and the vast majority of companies will shift towards the development of AI applications. Also, I think this is a significant boost to the AI startup ecosystem in China because previously, if you were starting an AI business in China, there was indeed a problem. When developing applications, you couldn't use those top tier models. But now, the experience with this model is almost the same as that of top tier models. So, I think this will have very positive factors for this wave of AI startups in China. How will we use this technology to drive the company's development in the next few quarters? As I mentioned just now, we've been working in the robot field for several years. We've always attached great importance to the interaction aspect. We've developed a voice interactive robot, which can listen to your commands and perform some actions. But frankly, at that time, due to the industry-wide technological feeling, it was difficult to achieve a satisfactory voice interaction experience in a public, noisy and crowded environment. We could only achieve satisfaction in some very specific vertical fields. For example, when…

The rapid development of the AI industry has brought new opportunities to the future space, development path and business model of Cheetah's robot business. Compared with the last quarter, have Cheetah adjusted its robot business strategy?

Well, yes. Regarding the question you just asked, I think it's been quite clearly stated. Let me summarize. Currently, the robotics industry is booming, right? But it seems that everyone's focus is on humanoid robots. My consistent view is that humanoid robots face not only software technical difficulties, but also high hardware technical complexity. So there is still a long way to go before they can be commercially applied on a large scale. Today, humanoid robots mainly serve customers like research institutions. We've always firmly believed that for robots to provide human like services, the mechanical complexity should not be too high, and they should be fully intelligent in understanding the environment and user needs. In this regard, with the maturity of large language model technology, we think we've reached a crucial point. So we don't think there has been any significant adjustment. We are still following the established path of robot development. However, we believe that the market for robots in voice interaction such as reception, escalation, promotion and translation should experience relatively large scale growth compared to last year or previous years. We've made some internal adjustments in commercialization and channels, but I won't go into detail. Overall, we believe that today, artificial intelligence and robotics have indeed entered a track where they can develop rapidly.

In Cheetah's robot business, how do you think real world data feedback can be used to accelerate technological iteration? What role will the data flywheel play in promoting the expansion of Cheetah's robot business to a wider range of industry applications?

Yes, based on my previous answers, let me explain. In fact, our robots today focus on two main aspects of technology, aside from what I've already mentioned about language. One is in voice interaction. Let me give you an example. In many public places, the ambient noise can be around 60 to 70 decibels or even higher. The microphone is actually quite sensitive to noise, and there are also issues like spatial echoes. Through our in-depth work over the past few years, we've achieved significant improvements in voice capabilities through data driven iteration. For instance, our current in-house voice recognition model in such environments has a much higher recognition rate compared to some well-known models outside. This is one aspect of data iteration. After we launched the AI enabled OS 4, it can respond to more complex scenarios involving voice interaction, visual interaction and multi person situations. However, there will still be some bad cases and relevant data in these responses. So, when we introduce our products to the market, I believe that with the continuous collection of data, we will surely improve the user experience, making it closer to interacting with a real person. It will be better at discerning whether you're talking to it or chatting with someone else and whether your intention is to ask the robot to do something or not. These experiences, I think, will be enhanced. The second aspect is indoor navigation for our robots. As I mentioned before, it's limited by computing power. In the past, we mainly relied on SLAM based engineering algorithms. Some attentive friends may know that in the past, in restaurants, we had to stick QR codes on the ceiling. Without such physical markers, the robot was prone to getting lost or even bumping into people. But now with the developmental vision technology and the continuous increase of data, we no longer need to stick QR codes on the ceiling. Moreover, with the continuous growth of data, we will also enhance the robot's self-positioning and path planning capabilities in space. So, in this regard, I think the data flywheel is a great help to us and as we continuously improve the user experience and attract more users, I believe our robots can maintain a sufficient lead in terms of intelligent experience. Thank you.

With the rapid development of AI products such as [indiscernible], like AI assistance, Copilot, and AI Search, is it possible that the emergence of AI agents will have a significant impact on the existing application ecosystem? Some investors believe that AI agents may replace many traditional apps. In this context, how do you view the integration of AI agents with existing applications? In particular, how do you think AI agents will complement super apps like WeChat?

Regarding whether the emergence of AI agents will have a significant impact on the application ecosystem, I think the emergence of AI agents will definitely have a major impact on the existing app technology roadmap. As for whether it will have a significant impact on the existing ecosystem, it depends a lot on the response of these major players in the ecosystem. Whoever responds quickly may be able to withstand the change, while those who respond slowly may experience some changes. It's a bit similar to the situation when mobile Internet emerged. Could traditional Internet companies get a share in the mobile Internet market or could they further develop in the mobile Internet field? We've seen different companies have different developments later, but I think this is not pre-destined. It depends on everyone's investment in AI and the speed of their actions in product height in AI. Later, for example, with the emergence of short video platforms like Douyin, could those video websites not have made something similar at the beginning? I don't think so. Because we invested in and sold the product to ByteDance which later became TikTok. Essentially, at that time, people didn't have high expectations, now that they couldn't make it. This time, regarding AI agents, I think there is a broad consensus, whether among entrepreneurs or large companies. So it's hard for me to predict how things will change because it sometimes depends on the actions of each company. But for your second question, whether AI agents will replace many traditional apps, I think this technology will definitely replace traditional apps. When users get used to solving problems with just one sentence, they won't use the traditional click-based way anymore because natural language requires no learning. In the past, when using any app, we had to…

Could you review the major achievements of Cheetah's robot business in 2024 and share the plans and goals for this business area in 2025?

I know, in fact, we've disclosed some of our achievements in AI and agents in 2024, right? Well, for us, one of the achievements in 2024 was that we expanded into some overseas markets. We now believe that the demand for service robots in overseas markets is clearly on the rise. Of course, I have to say that the overseas market is different from our previous APP business. It takes time to build channels, develop agents, and recruit salespeople. In 2024, we witnessed real market-driven growth in the robot business. Even in markets like Europe, which I previously thought were slow changing due to labor shortages, we can clearly see this growing demand for 2025. As I mentioned, one of our major goals this year is to achieve differentiated competition from many robot manufacturers. I think we won't just focus on the single so-called delivery market, because as we all know, the domestic market is highly competitive, especially in terms of price wars. This year, the differentiation we aim to achieve is centered around intelligence and interaction as a development strategy. We want to truly lead in product performance and user experience in this market. And this market, I think, can officially take off with the development of large language models this year. Also, we take a long-term view on robots. We hope that within three years, we can become one of the top three robot companies globally and we also hope that robot revenue can account for half of the company's total revenue. This is our plan for the next three years or more. Thank you.

In the past, the market generally believed that due to the lack of sufficient general purpose capabilities, wheeled robots had limited market space and average monetization capabilities, which led to restricted R&D investment and formed a vicious cycle. Do you agree with this view? In your opinion, how can Cheetah break through these limitations and open up the market space for wheeled robots? How far are we from this turning point?

Well, I don't quite agree with the first part of your conclusion. The limited market space of wheeled robots is not because they are wheeled, but rather due to the limited intelligence level of robots in the past. In the past, when you saw many restaurants and hotels using them, there was a large amount of deployment work required and their tasks were very single. They couldn't handle complex solution level problems which restricted their practical application, right? So the insufficient intelligence level was the limiting factor, enabling them to only complete very fixed tasks, such as point-to-point delivery, right? Therefore, I think the biggest problem with humanoid robots before, or in fact, with the entire robotics industry excluding industrial robots, as industrial robots have reached a sufficient level of automation through decades of development to meet many industrial production line requirements was the lack of intelligence. This insufficient intelligence level led to very limited task completion capabilities. However, this year with the addition of the AgentOS we just mentioned, which brings new multimodal interaction capabilities, task key planning capabilities, I believe the intelligence level of robots can be significantly enhanced. Once the intelligence level is improved, the task they can complete will be, let's say, not on par with humans, but approaching human-like capabilities in some aspects. Let me give you a detailed example to help you understand. In the past, restaurant robots were limited. They could only move from one table to another, and waiters still had to be trained for this simple task, and that was about it. They had no idea about the customer's reactions. But with our system now, first of all, waiters don't need to be trained to tell the robot to deliver something to a certain table. It's as simple as telling a…

Okay, so thank you so much for joining the conference call. This is Helen from Cheetah Mobile’s IR team. This line is hosted by an AI agent to translate management Q&A section. So thank you for your patience and if there is any misunderstanding or for translation, I think, or transcript for the earnings call will become available on our IR website as soon as possible. And if you have any further questions, please reach out to our IR team. Thank you so much. Bye.

The conference has now concluded. Thank you for attending today's presentation. You may now disconnect your lines.