Baidu, Inc. (BIDU) Earnings Call Transcript
November 18, 2025
Earnings Call Speaker Segments
Hello, everyone, and thank you for being here on the Baidu Third Quarter 2025 Earnings Conference Call. We published our earnings release earlier today; you can find copies on our website as well as via Newswire Services. On the line with us are Co-Founder and CEO Robin Li; Julius Rong Luo, our EVP overseeing Baidu Mobile Ecosystem Group, MEG; Dou Shen, our EVP overseeing Baidu AI Cloud Group ACG; and Henry Haijian He, our CFO. After we finish our prepared remarks, we'll take your questions in a Q&A session. Please note that during today's discussion, we will make forward-looking statements under the safe harbor provisions of the U.S. Credit Securities Litigation Reform Act of 1995. There are risks and uncertainties tied to these forward-looking statements, and because of them, actual outcomes might end up substantially different from our present expectations. If you'd like a thorough look at those risks and uncertainties, you can turn to the annual report we issued most recently, along with the documents we file with the SEC and the Hong Kong Stock Exchange. Unless the law requires it, Baidu assumes no duty to revise any forward-looking statements. During our earnings press release and on this call, we also go over some unaudited non-GAAP financial measures. A reconciliation showing how those unaudited non-GAAP figures line up with the unaudited GAAP measures that are most directly comparable appears in our press release, which you can find on our IR site, ir.baidu.com. One quick note: this conference is being recorded. Also, you'll be able to catch a webcast of today's call on Baidu's investor relations site. At this point, I'll hand things over to Robin, our CEO.
Hello, everyone. In Q3, Baidu Core generated total revenue of RMB 24.7 billion. AI Cloud brought in RMB 6.2 billion, representing a 21% increase compared to the same period last year, and that value growth trajectory continued. Apollo Go accelerated its expansion significantly. We delivered over 3 million fully driverless operational rides in the third quarter. That translates to 212% growth year-over-year, versus 148% during the previous quarter. This quarter demonstrated the transformative value AI is generating across our entire business. Whether it's enterprise services, consumer-facing products, or smart mobility, our AI capabilities are delivering proven, tangible impact at scale. Let's start with the enterprise segment, where our AI Cloud business continues to scale with healthy momentum and produces measurable business results. During the third quarter, AI Cloud continued along its robust upward path. Within that business, the segments tied most closely to artificial intelligence posted the quickest gains. Specifically, revenue derived from subscriptions to AI accelerator infrastructure surged 128% compared with the same period a year earlier, and it has become the primary force behind AI Cloud's growth. This demonstrates a favorable move toward a revenue model that is more recurring and sounder in structure, alongside powerful appetite for our AI offerings and solutions. Our ability to satisfy this rising need stems from having deployed early and strategically across Baidu's entire AI architecture stack — spanning infrastructure, framework, models and applications — which enables us to serve enterprises at every phase of their AI journey. When it comes to the infrastructure layer, the AI infrastructure we've built is among the most sophisticated anywhere in China. It draws on a varied combination of high-performance computing resources, both from within China and abroad, and these include an AI computing architecture we developed ourselves. By innovating technically on an ongoing basis, we're able to push performance and efficiency higher while bringing inference costs down steadily. On top of that, our resource management capabilities lead the industry, and they substantially increase both utilization and scalability. Thanks to these strengths, enterprise clients get AI infrastructure that is dependable, can scale up, and is extremely cost effective. Turning to the model layer, we offer our own early foundation model, and it keeps iterating at a rapid pace. At Baidu World 2025, held not long ago, we introduced ERNIE 5.0. It's our first native omni-model, a foundation model that performs exceptionally well in omni-model understanding, creative writing and instruction following. Beyond being at the forefront of our AI capabilities, ERNIE also functions as a core foundation supporting a great deal of the AI-powered innovation happening throughout our business lines. When it comes to the application layer, we provide a variety of AI solutions designed around the needs of enterprise customers. Here are a few illustrations. To start with, there's Famou, also called FM agent — a self-evolving agent we introduced not long ago that drove notable gains in enterprise efficiency. Powered by ERNIE, it independently investigates a vast number of options and keeps refining its approaches in order to pinpoint optimal solutions for extremely intricate, ever-shifting challenges in the real world. FM agent has already been put to work in sectors such as transportation, energy, logistics and ports, streamlining sophisticated operations that conventional methods find difficult to manage. This functionality matters a great deal in China, since we operate across a wide range of industries with countless situations that call for better efficiency. Being able to substantially raise efficiency across so many different applications creates a significant benefit for society. Take the Daniel Wu English coach as another illustration — it's a digital employee powered by ERNIE that we built for Yashi Education, and it carries the lightness of that famous actor. Users can practice English conversation one-on-one in real time, at any hour and from any place, which directly addresses a central challenge. Yashi's users require frequent, on-demand speaking practice, and expanding that through human instructors is difficult. The digital employee delivers round-the-clock availability plus an immersive, engaging learning experience. Our AI offerings aren't just for businesses—they help people get more done as well. Baidu Wenku and Baidu Drive, the two largest consumer AI apps we offer, have been given new life through AI. Their combined MAU has climbed to close to 300 million. This August, the two apps together introduced a general-purpose AI agent platform, which coordinates hundreds of specialized agents so users can complete complicated end-to-end tasks through straightforward natural language exchanges. Traction since the launch has been robust, demonstrating how AI can substantially enhance individual productivity on a large scale; when it comes to the physical world, autonomous driving showcases AI's revolutionary impact, creating fresh opportunities in mobility, safety and efficiency; during Q3, Apollo Go accelerated its growth markedly, reaching new peaks. During this quarter, we provided the public with over 3 million rides that were fully driverless and operational. That represents an impressive 212% increase compared with the same period a year earlier, while the previous quarter had seen growth of 148%. In October, the weekly average of fully driverless operational rides exceeded 250,000 — among the highest figures achieved for real-world Robotaxi operations anywhere in the world. As of now, our fleets have accumulated more than 240 million autonomous kilometers, and over 140 million of those were fully driverless, all achieved while keeping an excellent safety record. Reaching this pace of rapid expansion together with exceptional safety results strongly confirms that our technology and operational abilities have matured. Our ten-year dedication to autonomous driving is now producing results at large-scale operations, which fills us with pride. Operating at this size calls for a lot of things to come together: cutting-edge tech, a safety track record that's strict and broadly acknowledged, proof that the business works, serious hands-on know-how, and the capacity to grow quickly. All those years of steady investment haven't just handed us the advantage of getting there first — even more significant, they've produced an all-around set of capabilities that makes Apollo Go, without question, the top player worldwide in this space. Armed with that, Apollo Go has moved into a period of swift worldwide growth. By October, Apollo Go's reach around the globe had grown to 22 cities, compared with 16 in the prior quarter. During October, Apollo Go launched in Switzerland via a strategic alliance with PostBus, the foremost public transport provider in that nation. So together, we're planning to roll out autonomous ride-hailing in Eastern Switzerland. That's an important move for growing our presence in Europe, and it's one more milestone on our worldwide journey. Over in the Middle East, we recently landed one of the first permits for fully driverless commercial operations in Abu Dhabi, and we've strengthened ties with partners there. In Dubai, back in September, at the fourth Dubai World Congress for self-driving transport, Apollo Go received exclusive clearance to run self-driving trials on public roads. RT6 gave demo rides to attendees from around the world — government officials, business leaders, media and investors among them — which showed how mature our technology is on an international stage and demonstrated our global leadership. In Hong Kong, the market for Robotaxis with right-hand drive where Apollo Go has clearly built the dominant position, we recently broadened the zones for open road testing to cover Kolon and Kung Tong District, which makes our standing even stronger in that strategically vital market. These achievements across Europe, the Middle East and Asia confirm that our technology can adapt and that we can collaborate successfully with top local operators under varying regulatory conditions. Going forward, we intend to enter additional markets that offer significant commercial promise and chances for partnership, while keeping our steadfast commitment to safety and operational excellence as we strive to bring smart mobility to a broad audience. Within our mobile ecosystem, agents and digital humans are AI-native monetization breakthroughs that are reshaping our online marketing operations, delivering considerable value to advertisers via greater engagement, improved lead conversion and better ROI. Through intelligent multi-round conversations, our agents assist advertisers in clearly understanding user intent and swiftly identifying the most pertinent high-quality sales leads; this guarantees that advertisers get leads that are more accurate and qualified than what traditional methods provide. Leveraging this strength, agents have grown into various formats; tech-based, voice-enabled and visually-embodied digital humans, with each one crafted to serve distinct scenarios and interaction requirements.
That kind of flexibility lets advertisers pick whichever format works best for their particular needs, which means they can cover more scenarios and convert more efficiently. Because of this, agents have picked up serious momentum across a wide range of industries, from healthcare to business services to lifestyle services. During September, roughly 33,000 advertisers were driving ad spend every day through our agents. Digital humans likewise built strong momentum. Driven by ERNIE, our digital humans deliver round-the-clock AI live streaming for advertisers at minimal expense, which brings professional live streaming within reach across additional scenarios and industries. The technology keeps advancing, offering more realism, more natural interaction, and real-time engagement with viewers. That produces results that beat human hosts in numerous instances, which makes our digital humans more and more appealing to advertisers. Adoption has moved past merchants into fields like healthcare, automotive, and legal services. We're watching current clients raise their budgets while new clients sign on quickly. During September, digital humans streaming live on our platform nearly tripled compared to a year earlier, which highlights swift uptake across industries and expanding monetization potential. These innovations already bring in meaningful revenue with rapid growth rates; during Q3, revenue from agents and digital humans combined hit RMB 2.8 billion, climbing 262% versus a year ago, which confirms the robust market demand for our AI native monetization approaches. Going forward, we see considerable room to scale these innovations even more, extending adoption into additional verticals and deepening penetration among advertisers we already serve. Let me now walk through the key highlights for each business. Within our AI Cloud business, our client portfolio kept improving during Q3, showing deeper collaboration across the board. Top enterprise clients spent more and broadened usage past AI infrastructure. Mid-tier enterprise clients produced healthy growth, with subscription-based revenue and client count both climbing. A number of key verticals showed strong momentum. In embodied AI, our client base grew to 35 from 20 in the prior quarter, covering almost every major industry player in China. The automotive vertical likewise posted strong growth, with revenue nearly doubling compared to a year ago. Beyond that, during this quarter, we formed new collaborations with leading players, among them Neolix, a top provider of autonomous delivery vehicles in China. Taken together, these results confirm broadening adoption and strong recognition of Baidu AI Cloud. To meet rapidly growing demand, we made a strategic upgrade to our MaaS platform Qianfan, making it agent-centric. Qianfan is now positioned to offer not just leading model services backed by a constantly enriched model library, but also cutting-edge agent development capabilities and best-in-class agent infrastructure. Through integrating high-quality proprietary and third-party capabilities and tools, Qianfan makes agent creation seamless and empowers enterprises to speed up AI native application development. At the application level, we are driving productivity gains, both internally and externally. Internally, our developers make wide use of Comate, our AI coding assistant for developers. During September, AI accounted for over 50% of new code generation under developer oversight, substantially improving our overall engineering and R&D productivity. Comate exemplifies our belief that AI should free humans from repetitive tasks and deliver immediate efficiency gains. When it comes to our outward-facing efforts, Miaoda—our platform that requires no coding—is how we're making AI available to everyone. Following steady improvements to what it can do, we introduced an international edition of Miaoda called MeDo in November, which delivers strong no-code functionality to users around the world. Our goal in eliminating obstacles such as the need for specialized training is to let more individuals everywhere innovate and build with AI. Turning to intelligent driving, Apollo Go delivered 3.1 million operational rides with no driver at all in Q3, which is a 212% increase compared with the same period a year earlier. By November 2025, the total number of rides given to the public had exceeded 17 million. Regarding how we're growing geographically, Apollo Go brought on 6 additional cities, which raised its worldwide presence to 22 cities as of October 2025. Within Chinese Mainland, Apollo Go has now reached 100% fully driverless operations across a number of cities, among them Beijing, Shanghai, Shenzhen, Chengdu, Chongqing, Wuhan, Haikou, Sanya and others; these aren't test areas—they represent actual services that are already available to the public, demonstrating how mature both our technology and our operations have become. When it comes to our asset-light model and partnerships inside China, we also advanced well during this quarter. This asset-light strategy lets us grow our autonomous driving offerings via collaborations and makes expansion quicker and more efficient in terms of capital. After rolling out fully autonomous vehicle rental services together with CAR Inc, Apollo Go is now making cross-city travel possible in Hainan province using fully driverless rental cars, giving customers a distinctive experience that traditional car rental offerings usually can't provide, especially for tourism and leisure trips. Beyond our collaboration with Hello Ride, we reached scaled fully driverless operations in 2 cities in China, which further confirms that the asset-light model is workable. Going forward, we'll keep growing quickly while putting safety first, speeding up uptake of autonomous ride-hailing services in wider markets; within our mobile ecosystem, the AI-driven transformation of Baidu Search kept making headway in Q3. By the end of October, about 70% of mobile search result pages included AI-generated content. Our view is that this is an ideal and sustainable level. During this quarter, we focused on improving the quality of multimodal content in AI search results while also growing our content ecosystem overall. AI generated multimodal content grew quickly in both how much of it there is and how good it is. Specifically, with daily AIGC video generation consistently at the scale of millions, our overall AIGC video content keeps expanding fast, and daily distribution inside Baidu App is also rising strongly. As the supply of content gets better, users get richer, more relevant and more engaging search results, and user metrics keep improving. In September, Baidu App MAU hit 708 million, a 1% rise versus a year earlier. The daily average time spent per user in Q3 went up 2.3% year-over-year. We're also bringing our AI search capabilities to outside partners via the Baidu AI search API, which allows integration of our industry-leading search technology that provides better accuracy, authority and comprehensiveness. Major companies including Samsung, Xiaomi and Honor have already taken up the API. This strategic move extends our technology's reach past our own ecosystem, opening up new business models and generating wider value throughout the industry. With our full stack AI capabilities serving as the foundation, every business group at Baidu has made quick strides, with AI fueling both the creation of new products and expansion of business. Viewed through an AI-native lens, what we offer spans across those groups, covering a wide array of AI-driven operations that stretch from AI Cloud Infra to AI Applications like Baidu Wenku and Baidu Drive, and onward to AI native marketing services, which include agents and digital humans — every one of these is demonstrating robust momentum for growth. When it comes to the physical world, Apollo Go, the biggest AI application we have, keeps scaling up at a fast pace, and that is merely a handful of illustrations, highlighting how widespread the growth is among our AI-powered businesses and the significant business impact our AI capabilities are already providing at scale. As we look forward, we will keep broadening our revenue streams powered by AI and reinforce our standing so we can seize the long-term opportunities that lie ahead; we have confidence that our AI capabilities will deliver still more transformative value throughout our portfolio in the years ahead. Having said that, I'll now hand the call over to Henry so he can walk through the financial results.
Thanks, Robin, and hi everybody. Robin touched on our AI-driven operations a moment ago, so I want to go deeper. Following continuous input from investors, and so that valuation drivers tied to our present portfolio come through more clearly, this quarter we're rolling out something new — an AI native view that runs across business groups to monitor AI-empowered assets throughout the company; under this fresh lens, our operations are arranged by what our products and services actually are, which helps investors grasp the core valuation drivers spanning our wide-ranging offerings. Moving ahead, business updates will come through this AI native view regularly, and at the same time we'll keep reporting results using the reporting methods already in place, so investors get complementary angles for judging what our portfolio is worth. Through this AI native view, we hold a rich array of AI in power assets. This quarter we're putting a spotlight on 3 categories. AI Cloud Infra, AI applications and AI native marketing services. Let's begin with AI Cloud Infra — that's the AI infrastructure and platform services we provide to businesses and government entities. During the third quarter, this segment brought in RMB 4.2 billion, which represents a 33% increase compared to the same period last year. Our AI accelerator infrastructure ranks among the most sophisticated in China, enabling training and inference that are both highly efficient and cost-effective for a wide range of enterprise workloads. Within AI Cloud Infra, the portion of revenue coming from subscription-based AI accelerator infrastructure jumped 128% year-over-year. Moving on to number two, AI Applications. This category covers products that are either AI-native or AI-enhanced, designed for specific use cases serving both consumers and companies — our flagship software offerings like Baidu Wenku, Baidu Drive, and digital employee are examples. AI is transforming the way applications create value, opening up much more potent capabilities that address real-world scenarios with greater effectiveness. Our portfolio is now broad and best-in-class, covering consumers as well as businesses. A large share of our AI offerings run on subscription arrangements that keep users locked in, which generates dependable, high-caliber revenue. During Q3, our AI Applications segment generated RMB 2.6 billion in revenue. Third, our AI-native marketing offerings — agents and digital humans among them — continue to expand at a rapid clip; this represents a second engine of growth alongside our established operations; demand is strong for these novel products among buyers who want AI-native solutions built around performance. More and more, customers are willing to spend extra for advanced AI capabilities that deliver concrete improvements in productivity, marketing efficiency and ROI. During the third quarter, AI-native marketing services generated RMB 2.8 billion in revenue, which represents a strong 262% gain compared to the same period last year and accounts for 18% of online marketing revenue at Baidu Core. I'd now like to walk you through the specifics of our Q3 financial performance. We recorded total revenues of RMB 31.2 billion, which is a 7% decline year-over-year. Baidu Core brought in RMB 24.7 billion in revenue, also down 7% from a year earlier. Online marketing revenue at Baidu Core stood at RMB 15.3 billion, falling 18% year-over-year. Non-online marketing revenue for Baidu Core reached RMB 9.3 billion, rising 21% compared to the prior-year period. Fueled by momentum in the AI Cloud business, which sits within Baidu Core's non-online marketing revenue, AI Cloud revenue amounted to RMB 6.2 billion, marking a 21% year-over-year increase. iQIYI contributed RMB 6.7 billion in revenue, a decrease of 8% year-over-year. Revenue costs came to RMB 18.3 billion, up 12% compared to a year earlier, largely driven by higher expenses tied to the AI Cloud business along with content costs. If you set aside impairment of long-lived assets, operating expenses stood at RMB 11.8 billion, a 5% increase versus the prior year. Meanwhile, Baidu Core's operating expenses reached RMB 10.4 billion, also up 5% from a year ago. At Baidu Core, SG&A expenses totaled RMB 5.7 billion, rising 14% year-over-year, mainly because of greater expected credit losses and more spending on channels. SG&A represented 23% of Baidu Core's revenue this quarter, versus 19% during the same quarter last year. During the quarter, Baidu Core's spending on research and development totaled RMB 4.8 billion, representing a 3% decline compared with the prior year. As a share of Baidu Core's quarterly revenue, R&D accounted for 19%, essentially flat versus the same period last year. We recorded RMB 16.2 billion in impairment of long-lived assets, reflecting an impairment loss on the Core asset group connected to how quickly we've been advancing our high-performance computing capabilities. We chose to proactively conduct a comprehensive review of our asset base and recorded write-downs on assets that include — though aren't limited to — legacy infrastructure that can no longer meet the computing efficiency requirements of today. As a result, we now hold a healthier, more thoroughly optimized set of assets that does a better job of supporting the future growth of our AI native business going forward. Operating loss came to RMB 15.1 billion. Baidu Core posted an operating loss of RMB 15.0 billion, which works out to a 61% operating loss margin for that segment. If you take out the impairment of long-lived assets, we had operating income of RMB 1.1 billion, while Baidu Core delivered operating income of RMB 1.2 billion. On a non-GAAP basis, operating income came in at RMB 2.2 billion. Baidu Core's non-GAAP operating income was also RMB 2.2 billion, giving that segment a non-GAAP operating margin of 9%. Turning to other items, total other income, net, was RMB 1.9 billion, down from RMB 2.7 billion in the prior-year period. We recorded an income tax benefit of RMB 1.8 billion, compared with income tax expense of RMB 814 million in last year's comparable period. Net loss attributable to Baidu was RMB 11.2 billion, and diluted loss per ADS was RMB 33.88. For Baidu Core, the net loss attributable to it came to RMB 11.1 billion, with a net loss margin of 45%. If you take out the impairment of long-lived assets, net income attributable to Baidu would be RMB 2.6 billion, while net income attributable to Baidu Core would be RMB 2.7 billion. On a non-GAAP basis, net income attributable to Baidu was RMB 3.8 billion. Non-GAAP diluted earnings per ADS came in at RMB 11.12. Non-GAAP net income attributable to Baidu Core was RMB 3.8 billion, and Baidu Core's non-GAAP net margin was 16%. When we refer to total cash and investments, that means cash, cash equivalents, restricted cash, short-term investments, net long-term time deposits and held-to-maturity investments and adjusted long-term investments. By the end of September 30, 2025, our cash investments came to RMB 296.4 billion overall, while cash and investments on a total basis, leaving out iQIYI, stood at RMB 290.4 billion. From operations, cash flow reached RMB 1.3 billion, and if you set iQIYI aside, that operating cash flow figure was RMB 1.5 billion. Baidu Core had approximately 31,000 employees as of that same date, September 30, 2025. With that said, operator, please go ahead and open the line for questions.
I'd like to ask about ERNIE 5.0, which was just introduced at Baidu World. How is this new model going to push the next phase of applications, things like digital humans? And what are the main priorities when it comes to earnings, upcoming versions, and what sets it apart?
Alicia, Robin here. During the last couple of years, I've said many times that our approach is application-driven when it comes to earnings iteration. Only a few days ago at Baidu World, we introduced ERNIE 5.0, which is our first native omni-model foundation model. For omni-model comprehension, creative writing, and following instructions—areas that are hugely important to both our current and upcoming products—it has reached top-tier, world-class performance. Across the ERNIE line, things keep getting better, starting with ERNIE 4.5 and ERNIE X1 back in March and going all the way to ERNIE 5.0 this November. A good example is digital humans. When they run on ERNIE, their conversations come out smooth, fitting the context, and full of expression. Those strengths come from how good ERNIE is with language. Beyond language, our model also drives the visuals—making them realistic—along with appearance, motion, and even subtle micro expressions, and all of it stays aligned with what's being said. Once these pieces combine, our digital humans perform at a truly exceptional level and are genuinely persuasive, able to drive user engagement along with purchasing decisions. ERNIE additionally powers FM agent, our self-evolving agent that greatly boosts enterprise efficiency; in sectors such as manufacturing, energy, finance, transportation, and logistics, it has shown itself to be very effective. In the same way, our AI search and cloud business gain from ERNIE's capabilities as well. Even though ERNIE has produced outstanding results for these applications, we believe there is considerable room to improve. Our aim is for digital humans to beat real humans in sales across every type of live streaming e-commerce and in numerous product categories. We also want FM agents to keep finding increasingly better answers in more complex situations across every industry. Another goal: AI-created content should align with what users care about more effectively than content made by KOLs. And we'd like ERNIE-based agents to be capable of judging which piece of content is higher quality on given subjects, plus many other things along those lines. In all these areas, no current model does well — nowhere near it. So fixing that problem is what we're after. This application-first strategy actually reflects how strongly we believe in where the value of AI will end up living. Right now, a lot of economic worth is concentrated down in the infrastructure tier, but if the AI ecosystem becomes healthier, then the biggest gains ought to come from applications — the places where what we build genuinely makes a difference for users, advertisers and enterprises. As I look forward, my view is that when it comes to any single capability, no foundation model is going to outshine every other one. Our ongoing priority will be getting ERNIE to be at its best in the areas that count most for our portfolio. Baidu has never stopped being a company that puts deep faith in technology, and we'll keep putting money behind our convictions in the places where tech can generate value you can actually measure. That's why remaining tightly connected to applications keeps AI development on a path it can sustain.
I'd like to ask about the Cloud segment. In Q3, we saw that Cloud growth slowed a bit. Are we seeing any changes in Cloud demand? And should we expect it to speed up again in the quarters ahead? What's your outlook for next year? And what main drivers should support durable growth for our cloud business?
This is Dou. Thanks, Lincoln. Based on our performance so far this year, our Cloud business is growing much faster than the industry average. Looking at quarter-to-quarter, you can see some ups and downs, but the bigger picture is still a strong one, and our confidence in that growth path ahead hasn't wavered. On the demand side, companies are putting AI to work throughout their operations, which fuels robust, wide-ranging appetite for cloud services built around AI. Inside AI cloud, the piece most directly connected to AI workloads is scaling the < >. Customers turn to our cloud for model training, and more and more for inference as well. AI Cloud Infra revenue came in at RMB 4.2 billion in Q3, a 33% year-over-year increase that beat overall cloud growth. Meanwhile, revenue from subscription-based AI accelerator infrastructure rose 128% year-over-year, speeding up from roughly 50% in the prior quarter. The numbers are solid — they show robust demand fueled by AI as well as an improved revenue mix. Our full-stack AI capabilities are what drive this momentum. When it comes to infrastructure, our high-performance AI infrastructure — particularly the self-developed AI computing architecture — keeps gaining traction thanks to better performance, greater efficiency and lower cost. A lot of customers begin with AI Infrastructure and later branch out into other offerings. On top of that, our Qianfan MaaS platform now centers on agents; it offers a broader model library, built-in tools and stronger backing for intricate agent workflows, so Qianfan delivers top-tier agent infrastructure that lets enterprises build and deploy AI agents at scale with ease. When it comes to the application layer, we offer apps that are ready to be used in actual business situations. Investment in AI infrastructure isn't the only driver behind our cloud growth — applications matter enormously to us. We have a broad lineup of AI products and solutions that's expanding rapidly, such as digital employee, Yijian, Miaoda, FM agent, and others. Therefore, we have strong conviction that over the long run, AI applications will generate considerable value for our cloud businesses. In summary, taking our cloud business on an annualized basis, we're confident that our full-stack AI capabilities, together with robust demand for AI-centric cloud services, will make healthy, scalable, and sustainable growth possible going forward. Thank you.
As Baidu's app transforms into an AI-driven application and web search turns into a foundational component of AI chatbots, the distinction between search and chatbots is becoming less clear. From what you've observed, how are users' behaviors shifting? And what's your plan for competing moving forward?
Alex, I'll address your questions. AI chatbots are actually evolving at a very rapid pace. That makes it essential to remain adaptable and provide various products tailored to different use cases. Here at Baidu, we tap into chatbot abilities via 2 complementary options. The first is the ERNIE assistant, the chatbot embedded within the Baidu App; it enables multi-round conversations, calling, and thanks to its tight integration with search. Because lots of users reach the ERNIE assistant straight from search, it can use query context and interaction history to give answers that are more relevant and personalized; it also links to a set of tools via MPT, letting users transition smoothly from finding information to getting tasks done. The ERNIE assistant is expanding rapidly in our app. You can observe that conversation logs are up about fivefold compared to a year ago, and DAU has exceeded 12 million, showing very robust month-over-month growth. We anticipate this pattern will keep going in the next few quarters. At the same time, we provide the ERNIE bot as a separate chatbot app. Although it shares core capabilities with ERNIE assistants, the ERNIE bot adopts an experimental and innovative style, with a short-term emphasis on boosting retention and a long-term goal to compete at the leading edge of the chatbot space. For instance, it offers some advanced multimodel features like AI images or comic-style generations, which have been particularly well-liked among younger users. Looking forward, we think chatbots aren't just the final form of AI applications; the future of AI interactions will be multimodel, real-time, generative, and interactive. For example, at the latest Baidu World, we demonstrated the upgraded < > digital human, which can interact instantly via real-time voices and video, similar to live conversations. As many of you might remember, we even experienced a minor technical glitch during the live demo, which actually demonstrated that it was genuinely real-time and not pre-recorded. Once fixed, the digital human responded very vividly and produced dynamic back-and-forth conversations that feel almost human; so we'll keep integrating these advanced capabilities into search, making it smarter, more personalized, and able to complete tasks. This ongoing innovation is how we plan to seize the long-term opportunities in AI and reinforce our competitive strengths. Thank you, Alex.
We also value the extra information you provided about AI-driven operations. Might the team elaborate on what kind of expansion we can expect ahead, plus how profitable Baidu's newly AI-empowered operations are? In what ways will those segments assist in speeding up our total top-line expansion later on?
Gary, thanks. Henry here. I'll give you some context regarding these fresh AI-native perspectives. From what investors have told us, there's a demand for clearer insight into our fast-expanding AI units; these perspectives sort our holdings according to what the products fundamentally are, which gives investors better sight of the value drivers beneath; both this AI-native lens and our current reporting approaches will continue side by side, delivering views of how our business performs that complement each other. Looking through this fresh lens, I'd say our collection of AI-empowered holdings is quite substantial. Allow me to point out a few notable things. To begin with, regarding the AI Cloud Infra, this segment covers our market-leading AI infrastructure, which takes in a self-developed AI computing architecture, cloud infrastructure and a top-tier MaaS platform. With AI uptake speeding up, appetite for sturdy infrastructure keeps climbing, and our distinctive strengths leave us favorably placed. We're securing durable, long-term revenue and anticipate margins to rise alongside greater utilization. Next, on the AI applications side, this segment covers marquee offerings such as our Baidu Wenku and Baidu Drive. Across these offerings, AI has markedly boosted what they can do. Ours ranks among China's widest-ranging AI application collections, with the bulk of these apps built on subscriptions, which adds to revenue and margins of superior quality. Third, concerning AI native marketing services, spanning agents and digital humans, this shows how AI delivers bigger efficiency gains and fuels the second growth curve within our advertising operations via better engagement, conversation and ROI; during this quarter, AI-native marketing services accounted for 18% of Baidu Core's online marketing revenue, climbing from 4% in the same period last year, and we anticipate that share will keep growing as more customers adopt it. These outcome-focused AI offerings are being taken up by clients, who are ready to spend for concrete improvements in productivity and marketing efficiency. Crucially, these AI-empowered operations strengthen each other throughout Baidu's ecosystem; and as AI becomes more deeply woven into products. Thus, we anticipate these operations expanding at a quicker pace. So as we look to what's ahead, our confidence in their revenue and profitability potential holds firm, and we think this will underpin a more robust growth path for Baidu in the long run. Thanks, Gary.
My question concerns the Robotaxi segment. Apollo Go's expansion has really picked up speed during this year. With that in mind, could the management team share an update on what's ahead for Apollo Go in the coming year and further out, covering your plans for going global as well? Also, what do unit economics look like market by market? And how does management assess the profit potential of the Robotaxi business over the long run?
Robin here. As you may recall, we began our Robotaxi efforts back in 2013. That makes this our 13th year. At present, Apollo Go ranks among the biggest robotaxi service operators anywhere in the world. By November, cumulative rides delivered had surpassed 17 million, a milestone only a handful of companies worldwide can claim. Within China, our market leadership is unquestioned. Across the first 3 quarters of this year, our ride volumes came in more than 15x above those of our closest domestic competitors, based on data that has been publicly disclosed. Every one of those rides is fully driverless, which showcases operational scale and technological excellence that nobody else matches. Scale is hugely important. What lets us hold a leading position in autonomous driving technology worldwide is precisely that scale. We've run into plenty of problems and corner cases that others haven't witnessed. That let us train our models to cope with such situations, making them smarter and smarter. In my view, robotaxi has hit a tipping point, in China and in the U.S. alike. Enough people now have had the chance to take driverless rides, and word of mouth has generated positive feedback on social media, which I believe will push regulators to open up or relax the relevant rules. Looking at 2026 and afterward, we'll keep scaling our operations, at home and abroad. More cars will be added in the cities where we already operate. We'll move into additional cities. We'll rack up more fully driverless mileage and keep refining our technology using the operational data collected on the road. And yes, additional data is needed to train our models. Superior models make the cars safer and faster. Technological innovation and operational efficiency will let us keep pushing the cost per mile downward. At this point, positive unit economics have been reached in a few cities. As we scale, we're hoping more cities will turn positive during 2026. Beyond that, we're expanding via flexible business models, among them asset-light ones — we are very — we stand ready to enter any city quickly once regulatory and market conditions permit. By October, Apollo Go's global footprint had reached 22 cities, with meaningful progress across Europe, the Middle East and Hong Kong. We're confident that UE will keep getting better as we scale. To sum up, for 2026 and beyond, we anticipate strong growth in 3 areas: rapid growth in ride volumes and fee size, geographic expansion in new markets — into new markets and accelerated adoption of new business models. We believe Apollo Go is well positioned for continued global expansion and long-term profitability. Thank you.
I'd like to ask how monetization of AI search is coming along? And what kind of response are you getting from both advertisers and users? Is it possible for AI native marketing services to make up for the traditional ad business? And when we look ahead, how should core advertising profitability be viewed?
Thomas, Julius here. Back in October, close to 70% — that's 7 to 0 — of mobile search result pages carried content that is AI generated along with multi-model first content. That format is fairly distinctive to us, and we're either the first or possibly the only company doing it. We anticipate that level staying more or less steady, since we've already largely addressed the query types in which AI delivered a meaningful boost to user experiences. What we're concentrating on now has moved toward lifting quality, especially rich media content such as images and videos, and this quarter we're observing quite evident gains in content quality, which feed straight into improved user experiences. We can also observe that retention among users is greater, and those users who encounter AI search results are kicking off 6% more queries while giving us more of their time. That signals to us that real value is being found by users in AI search, and they're engaging more deeply. Turning to monetization, we're testing actively and early results look somewhat encouraging. To begin with, we're testing MCP within the commercial modules of AI search. As an illustration, during the recent Double 11 shopping festival our e-commerce MCP module hit a daily GMV peak of almost RMB 6 million. That's a very early stage, yet the results are pretty encouraging. Next, agents for advertisers are bringing in more than RMB 25 million in daily revenue, and we anticipate growth here as we add more agents into the earning assistance too. And third, testing has begun on digital human live streaming that includes real-time interaction capabilities, as we try to explore fresh ways of building engaging commercial experiences. Looking forward, significant monetization potential for AI search is something we see, and active testing will continue. That said, our priority in the near term is still user experiences rather than immediate monetization. This AI transformation is essential for long-term competitiveness and will unavoidably place near-term pressure on revenue as well as margins. So in our view this trade-off is the correct one for capturing the large opportunities that lie ahead. Thank you.
I actually have several things to ask. To start, here's an easy one. Can you walk us through the asset impairment recorded this quarter and the reasoning behind it? Then, what are your CapEx intentions for the coming year? Also, how ought we to view the margin path as AI revenues expand? Finally, please give us an update regarding shareholder returns after the buyback program now in place runs out.
Thank you, Xiong. To begin with your initial question about asset impairment, the context here is that we're stepping up spending on the newest AI computing technologies with no reluctance at all. As part of that push, we carried out a thorough examination of our infrastructure portfolio. Certain assets we already hold fail to satisfy the computing efficiency standards of today. Therefore, we chose to take some impairments on our own initiative. Following this one-time impairment, both our asset base and portfolio profile sit in a far healthier spot and line up better with sophisticated AI computing demand and higher-value application scenarios ahead. Next, regarding capital expenditures, we're keeping investment at an elevated level. Here's a single illustration. From the time Baidu introduced ERNIE in March of 2023, our AI investment has totaled considerably more than RMB 100 billion. In the future, we'll keep raising how intensely we invest in AI. We do anticipate greater operational leverage as our AI business grows; we're carrying this out across 3 fronts. To start, the asset review along with the impairments has given us a leaner, more efficient asset base. Next, we invest in a disciplined manner to guarantee capital efficiency. And naturally, third, we're boosting utilization of our AI infrastructure, for instance by dynamically allocating capacity between internal products and external cloud services. Consequently, thanks to these initiatives, we think Q3 marks a low point for margins. As we look toward next year, we'll work to lift our non-GAAP operational income and margins once these benefits begin to come through. Now for your final point about shareholder returns, through the plan and program authorized in 2023, we've repurchased shares valued at USD 2.3 billion already. At present we're examining the future buyback mechanism; we recognize that it matters to offer greater certainty and clarity so that buyback programs become less volatile in the future. We're likewise actively looking into diversified return mechanisms, such as potentially establishing a dividend policy. Taken together, these efforts are meant to bring more consistent value to our shareholders. Thank you.
Everyone, that brings today’s call to a close. We appreciate your participation. You’re welcome to leave the line now.
Read the full transcript via the API
You're viewing the first half of this call. Get the complete Baidu, Inc. transcript - plus 131,000+ transcripts from 7,000+ companies, speaker segments and full-text search - through the EarningsAPI REST API or hosted MCP server.
Get an API key View API docs →For developers and AI pipelines
Programmatic access to Baidu, Inc. earnings transcripts and 131,000+ others is available through the
EarningsAPI REST API and the hosted MCP server.
Quarterly plans from $162 - full transcripts, speaker segments, full-text search,
and the /api/v1/transcripts/recent polling endpoint for ETL pipelines.