Prepared remarks
Good day, and welcome to the Cheetah Mobile First Quarter 2026 Earnings Call. Please note this event is being recorded. I would now like to turn the conference over to Cheetah Mobile, Investor Relations. Helen, please go ahead.
Thank you, operator. Welcome to Cheetah Mobile's First Quarter 2026 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 the Q&A section. Please note that parts of the management presentation will be presented by an AI agent. Before we begin, I refer you to the safe harbor statement in our earnings release, which also applies to our conference call today as we will make forward-looking statements. At this time, I would now like to turn the conference call over to our Chairman and CEO, Mr. Fu Sheng. Please go ahead.
2026 remains an important transition year for Cheetah Mobile. We are continuing to evolve from a traditional Internet company into a company focused on AI enabled applications for AI agents and robotics. More importantly, we believe we are gradually moving from capability building into early stage commercial validation. Our focus is not only on developing AI capabilities but on turning these capabilities into practical products for real business scenarios, helping customers deliver better ROI. Starting from this quarter, we are separating our robotics and others business into an independent reportable segment. In the first quarter, revenue from robotics and others increased 176% year over year to RMB 51 million, approaching 20% of total revenue. At the same time, adjusted operating loss from this segment narrowed by 57% year over year. Customer demand remained strong, and we expect robotics and others revenue to grow strongly in 2026. In Q2, our robotics and other revenue will continue growing on both a year over year and quarter over quarter basis. Today, our robotics business mainly focuses on commercial scenarios with real customer demand and clear long term value, including reception, guided tours and intelligent service applications. Our smart personal mobility is another important step for us. This product extends our robotics and AI capabilities into personal mobility and health care related scenarios. More importantly, it further validates that our robotic platform can expand beyond commercial service robots into broader consumer applications. We are encouraged to see recognition from leading industry partners. During the second quarter, we started initial product shipments to a top global designer and manufacturer of mobility products as well as to the leading elderly mobility scooter manufacturer in China. We are seeing encouraging early market feedback and initial commercial traction. Moving to our agents, we are seeing strong customer adoption. We worked closely with Google Cloud and AWS, helping enterprises serving international markets access AI models and use multi cloud environments more efficiently. In the first quarter of 2026, revenue from our cloud and AI infrastructure services, as part of global enterprise services revenue, increased 68% year over year, contributing 18% of total revenue. Daily token usage has increased more than 20 times since January 2026, exceeding RMB 400 million in May. We expect this revenue growth to continue. We also kept building EasyClaw; it is still early, but we believe it will help customers deploy AI agents and boost productivity. The two fast growing businesses, robotics and others as well as cloud and AI infrastructure, already accounted for 38% of our first quarter revenue, and we expect their revenue growth and contribution to continue increasing in the coming quarter and to exceed more than 50% of our total revenue in the second half of this year. During the quarter, revenue from our advertising agency business within the Global Enterprise Services segment was affected by policy changes from certain overseas advertising platforms. We believe this revenue decline was primarily driven by external factors rather than changes in customer demand. This was the primary reason for the company's widening year over year operating loss in the first quarter. Our Internet services business continues to provide important profit and cash flow support for the company. In the first quarter of 2026, our Internet service business generated approximately RMB 15 million in adjusted operating profit. Profit and cash were supported while agency revenue was hit by policy changes, which impacts our financial results in the near term. Due to a stronger base for growth, moreover, our USD 186 million also supports our AI agents and robotics. Thank you.
Thank you, Fu Sheng. Hello, everyone, and thank you for joining us. Unless otherwise stated, all financial figures are presented in RMB. During the first quarter of 2026, we continued focusing on operating discipline, improving revenue quality and maintaining financial flexibility as we invest in AI and robotics initiatives. Total revenue remained relatively stable year-over-year at RMB 259 million during the quarter, while Internet service revenue declined due to continued weakness in online advertising. The quality of our revenue mix continued improving. Within the Internet Services segment, revenue from Internet value-added services continued to grow steadily and increased 8.2% year-over-year, contributing 72.8% of segment revenue given a larger portion of internet value-added services. Our Internet service revenue is becoming increasingly predictable. More importantly, the Internet service business remained profitable and continued generating stable cash, which provides an important financial foundation for our long-term AI and robotics investment. Turning to our robotics and other segments: starting from this quarter, we began recording the robotics and others business as a separate segment to present the operating progress of this business. Historical results previously reported in AI and others are now presented as robotics and others as well as global enterprise services. During the first quarter, revenue from robotics and others increased significantly year-over-year, with revenue increasing 175.9% year-over-year to RMB 51.2 million, accounting for 19.8% of total revenue. Adjusted operating loss from this segment narrowed by 57.1% year-over-year, reflecting continued improvement in operating efficiency and commercial execution. Turning to Global Enterprise Services: this business remains strategically important to the company. In addition to profitability contribution, it provides us with valuable enterprise customer relationships, overseas operating experience and real-world deployment scenarios for AI-related services. During the quarter, revenue from the advertising agency business was affected by policy changes from overseas advertising platforms, which impacted year-over-year segment revenue performance. However, revenue from our cloud and AI infrastructure services business increased by 68.3%, supported by increasing advertiser demand for AI-related cloud and token management services. Moving to profitability: operating loss was RMB 28.3 million during the quarter compared with RMB 26.5 million in the same period last year. The increase mainly reflected lower profitability from Internet and Global Enterprise Services businesses following revenue declines in online advertising and advertising agency services, as well as our continued investments in AI and robotics initiatives. More importantly, the Internet service and Global Enterprise Services businesses remained profitable during the quarter. The Internet service business generated approximately RMB 15.2 million in adjusted operating profit, while our Global Enterprise Services generated approximately RMB 13.8 million in adjusted operating profit. We also maintained a strong balance sheet. As of March 31, 2026, we had approximately $186 million in cash and cash equivalents as well as over $100 million in long-term investments. We believe our financial position provides sufficient flexibility to continue investing in AI and robotics with a disciplined and sustainable approach. Looking ahead, our financial priorities remain consistent: a) maintaining operating discipline; b) improving revenue quality and operating efficiency; c) supporting long-term investments while preserving financial flexibility. Overall, we believe the company continues moving toward a more sustainable and balanced operating structure as our AI and robotics businesses gradually scale. Thank you. We are now ready to take your questions.
Questions and answers
The first question comes from Thomas Chong with Jefferies. Please go ahead.
Thanks for management for taking my question. Recently, we can see that the market is paying more and more attention to robot AI, arguing that the real value of robots is not only hardware but also data and software. Cheetah has long operated commercial robots. From your perspective, during this operation, do you need to collect or simulate dynamic data to improve robot performance? Thank you.
Okay. Let me answer. Thanks, Thomas, for your question. I think you also pointed out a very important issue in the robotics industry, which is the issue of insufficient training data today. The rapid development of AI has given us very high expectations for the robotics industry, believing that today's AI capabilities have improved and robots should soon be able to achieve various behavioral capabilities. But in fact, I don't think so, because the development of AI agents, including the development of large language models, is actually built on the development of the Internet over two or three decades. The Internet essentially forms the basic training data of large language models. It is a very high-quality data set, and one of the various problems in the robotics industry today is the lack of data. Many approaches are being tried today with many manufacturers trying to use training data, including data migration and simulation training. However, there is a very serious problem: the physical world is much more complex than the laboratory environment and the simulator environment. So today, whether it's data migration, collection or truly migrating to different ontologies, this adaptability will be a huge challenge. Let me give you an example. Tesla's FSD is already very good. But in fact, some older versions of Tesla's own cars cannot install the latest FSD. So indeed, data is a very big problem. I also very much agree with what you said: the data continuously generated in the real deployment environment is actually very important for the robotics industry from our own experience. Let me give you two examples. One aspect is our voice interaction capability in different environments, which is actually closely related to our long-term exploration in various scenarios: different noises, different environments, multiple people, and so on. We have made optimizations and training on that data. Therefore, the interaction effect of our interaction robots, including reception robots, is leading in the industry today. We have a reputation in the industry for that. Another example is mechanical mobility: a very simple robot can navigate indoors from point A to point B. It is similar to a small low-speed driverless vehicle. How to use cheap chips and sensors to achieve automatic obstacle avoidance in different environments — in fact, all of these can only be achieved based on massive amounts of data. We recently launched a smart wheelchair, which we just mentioned; we started mass production in May, and now it seems that in overseas markets, especially in Europe, the sales momentum is quite good. For a traditional wheelchair product like this to achieve obstacle avoidance and assisted driving, many manufacturers, including some start-ups, want to achieve this kind of assisted driving capability, but to create a prototype and truly achieve good performance in many environments actually requires quite a lot of effort. This is related to the fact that we have deployed many robots in many environments over the years, regardless of surface conditions such as carpets or floors. We have also addressed reflections from walls and so on, all of which have been accumulated over time, with continuous algorithm optimization based on actual scenarios. Therefore, our wheelchair can truly achieve lower cost, highly assisted driving capability. It has also received positive feedback. So at this stage, the value chain is definitely in this regard. But I want to say the first reason I think it is not simply the model layer is because although the model competition is fierce, the gap between models is not too wide, and it is not easy to widen. Today, for example, the models of China and the United States — we think there is probably a gap of about half a year. And this gap fluctuates. Among large manufacturers, I think the gap is a bit like ebb and flow. Of course, today's models are also in the early stage. In the future, with the continuous increase in production of inference chips and training chips, training costs will gradually decrease. So I think the model layer will be an infrastructure, but in the long run, it will not be monopolized. With continuous improvement in model capabilities, many models, even if they are not top models but are adapted to specific daily tasks, have actually achieved very good results. For example, some open-source models in China this year have seen a significant increase in the amount of calls. The core reason is that they offer great cost effectiveness and have achieved high completion rates in some tasks. Therefore, I even think that in the future, various specialized models will continue to emerge, although this will take time. The second infrastructure layer we do not fully participate in, but we also see that because we have our own cloud business and we have token clients consuming here, the growth is also very fast. So I think this is a state of mismatch between supply and demand at this stage. But eventually, the infrastructure will also enter an economy of scale. For applications, today AI can actually reshape almost all applications. So there are huge opportunities in the application layer today, whether it is the industry we are working in like robots — we have been doing it for a long time and are still very firmly optimistic — and as the capabilities of the models continue to improve, the application of robots is wider; it may be a bigger industry than the automotive industry. There are also many opportunities at the software level, which I will not expand on here. Even today, when we look at some large model companies, their valuations are very high or excellent. Some have truly delved deep into a certain application, such as programming, or image generation like Stable Diffusion; the rise of specialized application models is actually an application. Its application is focused. Its agent has been well developed, including OpenAI-related developments earlier this year. We have also developed products like EasyClaw. So I think there is still large room and opportunities in the application layer. Thank you.
The next question comes from the line. Please go ahead.
I have a question about the robotics industry. There is a lot of discussion about the future of technology—some say the core is product operation, others say product deployment. What do you think is the core competitive barrier for robots in the future? Which capabilities are the most difficult to replicate?
From my understanding of the robotics industry today, I believe that in the short term or within the next two to five years, the possibility of a particularly versatile robot appearing is very low. This is limited by both the so-called model capabilities and the entire hardware industry chain. The update on the hardware industry chain is relatively slow, and it involves some of the most basic physics and materials as well as the underlying logic of physical laws and materials. So I believe today that the core skill barriers in the integrated industry in the future still lie in true scenario operation capabilities. In terms of client network, if we can have enough scenarios and a good client network so that our products can really be used in those scenarios, we can accumulate our own unique experience and data. The first question has been answered, which is that we can optimize based on that data. That optimization enables the product to provide better cost effectiveness and truly meet users' needs. The machinery industry is costly. But when it comes to business implementation, clients don't care whether you are a robot, a machine or a human. What they care more about is cost effectiveness, ROI, input and output. This has been very significantly reflected in our operations in recent years. So whether it is in the media, where you've seen many amazing things before, you will find that in a real scenario, very few things translate directly. Without going through actual scenarios, the operation of robots in the physical environment—whether actions or work—their complexity is actually much higher than that of autonomous driving of cars. Given this very high complexity, I think forming vertical and penetrating operational points and customer networks is much more important than building a generalized machine and model, because today I don't think generalized models and machines can quickly deliver the ROI required in these vertical scenarios.
The next question comes from Nancy Lu with JPMorgan. Please go ahead.
We see that recently basic model capabilities converging and API costs continuing to decline are driving the acceleration of commoditization of the underlying model, but enterprises generally adopt a multimodal strategy and no longer rely on a single model supplier. The shift has been from model performance to model application. I would like to ask: in the future enterprise AI market, where is the irreplaceable scarce capability and for future enterprise-level AI products, where is the ultimate moat?
Thank you, Lu. I think this is a very broad question. The ultimate moat of enterprise-level AI products should come from a deep understanding of user needs and a deep understanding of the industry, and then forming an extremely high-level organizational capability. The points you mentioned today are realistic in terms of model capabilities: one thing rises and another falls, and cost effectiveness is increasingly emphasized. The essence today is enabling enterprises to save a lot of costs previously spent on noncommercial insights and allowing them to truly focus on understanding user needs. So the real moat comes from keen insight into user needs, quickly launching new products and services, and improving your products and services. We often talk about using AI to reconstruct the internal organizational processes of an enterprise to quickly and efficiently operate and launch products and services. For example, if you pay attention, we have launched various products and services in the past year, much more than in the past, but our investment in R&D cost has decreased significantly from the perspective of cost, although there is still room for improvement. So when you can launch products and services quickly, where is your real moat? It comes from user demand: you can really find user demand and respond quickly. By the way, we have also launched some corresponding services and courses for building AI organizations for enterprises and shared some of our experiences with our clients. Now some large clients have started to sign contracts and operations have begun. The essence of business competition lies in efficiency and insight into user demand, and I believe AI products can accelerate the arrival of these two points.
The next question comes from Qiong Yang with Guoyuan Securities. Please go ahead.
Hello, you just mentioned our company is investing in enterprise AI projects. We would like to know: currently a large number of enterprise projects still rely on customized development and manual services compared to the standardized interaction model of traditional large products. The LLM moat will remain a mixed model of software and services for a long time. What's a key change in this process?
I think the core reason why there is still such a large amount of customization and manual services today is that AI is still in its early stage. Although we are seeing many media moments, most people's understanding of AI and its use is still insufficient. I think only a few people today can really make good use of AI. So this is a generation gap. Today's AI projects in traditional enterprises need customized development and manual services. Traditional SaaS has been developing for many years and has condensed many things in code, making it more standardized. I think as people increasingly understand AI and staff become more proficient in AI application, the proportion of the purely customized service model will continue to decline. Our company has already reached a point where all employees are using AI to write code and some of our internal systems are being created directly by business departments using AI rather than relying solely on a SaaS vendor and the service department. So the most critical change in this process is, on one hand, model capabilities are constantly increasing. This year we feel that the business departments are writing internal software and services, and model capabilities have improved a lot compared to last year. Many solutions that were demos last year are now usable internally. The model capabilities will continue to increase. Another important change is organizational: our organizational structure today is still based on the traditional industrial-software model. With the emergence of AI-native organizations, the traditional standardized SaaS model will be challenged. What we provide to our customers today is no longer the traditional type of service, but more training for employees and assessment of AI capabilities to help them transform their organizations. I think this change is the most critical: companies need to change their organizational structures and employee roles to take advantage of AI.
Please go ahead.
I'd like to ask: in terms of commercialization, wheeled robots and robotic arms are still the most widely deployed and most mature. What's your opinion on the development structure of robots in the coming years?
I have made my view on humanoid robots quite clear in the media. I think humanoid robots will not be able to replace humans in most commercial applications in the next three to five years, whether in factories, the service industry, or households. The difficulty of developing humanoid robots is extremely high. Wheeled robots and robotic arms—like xArm from UFACTORY—have been steadily growing in recent years and showed good growth in Q1 this year. Wheeled robots are also doing well because practicality, cost effectiveness and indoor navigation technology are already in a mature stage. Therefore, I believe we will see rapid growth in these areas. Robots should evolve from specialized vertical models that continuously gather data and improve performance, and only then might they gradually integrate toward a more general form. As for bipedal robots, I don't think they are needed in most scenarios. There's no need to add such cost and complexity, including reliability concerns. What we care most about in making robots is the commercial landing that can really be accepted and paid for by the market, not just project demos or integrated showcases. I think wheeled robots will gradually be matched with product fees in the future and will remain a main form for a long time.
The next question comes from Guangtao Jiang from Bohai Securities. Please go ahead.
Domestic service and worldwide service markets are considered to be the largest markets for robots in the long term, but at the same time they are also the most complex in demand and present the highest challenges. In the past quarter, you also launched your own intelligent wheelchair. What is your commercialization plan in the next two to three years?
Yes. Home robots are a broad concept. If you really talk about home robots, the only real breakthrough so far has been the sweeping robot. If you consider robots that can do more household tasks like an adult, I think the first reason we made an intelligent wheelchair is that in our view it is a robot: previously our robots were used for delivery, and intelligent wheelchairs can also be understood as delivering people. So I think the first type of family application is mobility—moving from A to B. The second is to add functions on this mobility, such as floor sweeping for a sweeping robot. What we see now is companionship: helping users with home control via voice, integrating companionship into robots, and assisting the elderly. These are all part of the same direction. Our wheelchair products have such functions, including companionship features for the elderly, and will soon be launched on our HTP. I think the ability to do full housework is not likely within two to three years because we also have our own robotic arm company, and while robotic arms are used in many scenarios, interacting with the physical world is extremely complex. It's not only about being able to perform certain actions but ensuring stability and success rates over time. Even picking up a cup in a kitchen environment is not a 100% success for any company. If the success rate is 99%, we still accumulate broken cups. That negative impact is significant. If robots enter households, there are issues like falling, bumping into things or people, and long-term reliability concerns. Ensuring quality over several years without malfunction is a tough challenge for many robotics companies today. Therefore, I believe that for home robots we should be pragmatic. Our view is that robots should be able to truly provide companionship for the family and assist the elderly and people with disabilities in mobility. I think this is a great breakthrough direction.
Operator, please check if there are any further questions. If not, we can conclude the meeting.
Thank you. Seeing there are no further questions, this concludes both our question-and-answer session and today's conference. Thank you for attending today's presentation. You may now disconnect.
Thank you. Bye-bye.
Thank you.
The conference has now concluded. We thank you for attending today's presentation. You may now disconnect your lines.