Baidu, Inc. (BIDU) Earnings Call Transcript
February 26, 2026
Earnings Call Speaker Segments
Good day, everyone, and thanks for joining us for Baidu's Fourth Quarter and Fiscal Year 2025 Earnings Conference Call. Earlier today, Baidu's earnings release went out, and a copy can be found on our website and through newswire services. Joining us on this call are Robin Li, our Co-Founder and CEO; Julius Rong Luo, our EVP in charge of Baidu Mobile Ecosystem Group, MEG; Dou Shen, our EVP in charge of Baidu AI Cloud Group, ACG; and Henry Haijian Hu, our CFO. Following our prepared remarks, we'll open things up for a Q&A session. I'd like to note that today's discussion will include forward-looking statements made under the safe harbor provisions of the U.S. Private Securities Litigation Reform Act of 1995. Such forward-looking statements carry risks and uncertainties that could lead actual results to differ materially from what we currently expect. For a detailed discussion of these risks and uncertainties, please see our latest annual report and other filings with the SEC and Hong Kong Stock Exchange. Except where applicable law requires it, Baidu assumes no obligation to update any forward-looking statements. Our earnings press release and this call include discussions of certain unaudited non-GAAP financial measures. A reconciliation of the unaudited non-GAAP measures to the unaudited most directly comparable GAAP measures is included in our press release, which is available on our IR website at ir.baidu.com. Please be aware that this conference is being recorded. A webcast of this conference call will also be accessible on Baidu's IR website. I'll now hand the call over to our CEO, Robin.
Hi, everyone. For the fourth quarter, total revenue at Baidu General Business came in at RMB 26.1 billion. Our core AI-powered business brought in more than RMB 11 billion, which works out to 43% of Baidu General Business revenue. Over in AI Cloud Infra, subscription revenue tied to AI accelerator infrastructure jumped 143% compared with a year earlier, building on even faster momentum than the 128% seen in Q3. At the same time, Apollo Go kept up its strong pace, providing 3.4 million fully driverless operational rides during the quarter. Overall rides rose by more than 200% versus the prior year. The year 2025 represented our third year working on Gen AI and proved decisive, as AI turned into the new heart of what we offer. During 2025, we advanced considerably in bringing AI to scale throughout our operations, boosting AI cloud expansion, growing robotaxi activity while unit economics improved, and embedding AI more deeply within our mobile ecosystem. Viewed through an AI native perspective, momentum kept building in 2025 across our core AI-powered businesses. AI Cloud Infra drew strong interest thanks to its highly efficient, cost-effective training and inference capabilities. AI Cloud Infra revenue came to roughly RMB 20 billion in 2025, a 34% increase from a year earlier, ahead of the industry's pace. Our lineup of AI applications ranks among the broadest anywhere, pairing AI-empowered flagship products with AI native offerings that open up completely new use cases; for all of 2025, AI application revenue topped RMB 10 billion. Apollo Go hit a major milestone. During 2025 by itself, we provided more than 10 million fully driverless operational rides. Cumulatively to date, we have given the public a total of over 20 million rides. Thanks to faster expansion around the world, Apollo Go now operates in 26 cities globally, strengthening our position at the forefront of autonomous ride-hailing. Finally, our AI native marketing services, which include digital humans and agents, kept growing strongly, with revenue climbing 110% from a year earlier. Taken together, these figures show AI contributing more and more to Baidu's value creation and our capacity to turn AI capabilities into commercial impact at scale. Now I'll walk through the quarter's key highlights, beginning with our proprietary AI chips. During the quarter, we announced plans to spin off and separately list Kunlunxin. Having invested steadfastly for over a decade in self-developed AI chips, we take pride in seeing the market increasingly appreciate their worth and demonstrated performance. This milestone confirms our long-term strategic vision and opens up fresh opportunities for value creation. Our AI chips rest on a proprietary architecture that was developed in-house from day 1. They provide stable, high-performance AI computing at scale, with wide compatibility across various models and frameworks. That lets customers deploy more quickly while spending less on integration. What sets our AI chips apart is a demonstrated record of large-scale, real-world deployments alongside leading enterprises spanning many industries, including financial services, telecommunications, energy and Internet sectors. Customers pick our chips for dependable performance, stable supply at scale, outstanding software compatibility and strong efficiency, particularly for inference workloads. Going forward, we foresee major opportunities for both Baidu and Kunlunxin as demand for AI infrastructure keeps speeding up. Now let's move on to the AI cloud infrastructure side of things. When it comes to infrastructure in China, ours ranks among the most sophisticated, and it runs on a varied combination of high-performance computing resources, both domestic and international. During Q4, revenue from subscriptions to AI accelerator infrastructure climbed 143% compared to the same period a year earlier, and for the entirety of 2025 we hit triple-digit growth. A key point here is that we observed a continuing move toward a revenue model that is more recurring and structurally sounder. This strong expansion was driven by enterprise AI adoption that is scaling up quickly. When customers embed AI into their core operations, the distinctive worth of our full stack end-to-end AI architecture grows ever more apparent. Because we own and fine-tune across all 4 layers, we secured lasting edges in stability and cost effectiveness, which better serves what enterprises require for AI deployment. Those edges are now converting into concrete market traction, propelling faster uptake of our AI Cloud Infra; during Q4, we widened our client base even further. Top enterprise clients are taking their partnerships with us deeper, which pushes up both usage and spending. We likewise enjoyed a solid growth contribution from clients in the mid-tier. We keep reinforcing our footprint across a range of industries, such as Internet services, gaming, autonomous driving and embedded AI, which highlights how versatile our infrastructure is. Embodied AI especially demonstrated striking momentum. For this vertical, revenue doubled from Q3 to Q4. We brought on board a fresh set of leading humanoid robotics companies, which locks in our standing as the preferred cloud service provider for China's rapidly expanding embodied AI sector. Now I'll discuss our foundation model progress, a vital component of our AI capabilities; we stay wholly dedicated to pushing forward our proprietary foundation model, ERNIE. After ERNIE 5.0 was unveiled last quarter, we rolled out an updated version in January. As we push ERNIE ahead, a clear application-driven approach keeps guiding us, so that ERNIE is at its strongest precisely where it counts most for our portfolio. To carry out this approach with greater effectiveness, we recently reorganized our model development organization into 2 dedicated teams. One team pushes forward ERNIE state-of-the-art foundation model capabilities, preserving our technological edge in this rapidly evolving space. The other team customizes models for particular business needs, lowering costs, enhancing response latency and fine-tuning model size and efficiency to guarantee our technologies are not merely cutting edge, but easily scalable throughout our businesses. Tight collaboration between both teams keeps our technologies rooted in real-world needs while our applications gain from ongoing technological advancement. Now let's turn to AI applications. This is where we think AI's greatest value will eventually be found; we are breaking new ground with AI applications to tackle complicated real-world problems for individuals and enterprises alike. Allow me to walk through our progress across several key areas, beginning with AI-powered search; during Q4, we carried on with our AI search transformation, chasing one of the most comprehensive and ambitious transformations anywhere in the world. Our emphasis stays on constantly enhancing the quality of AI search results while broadening what users can get done right inside search. This quarter, for instance, we rolled out AI-generated infographics into our search results, drawing on text-based information where suitable to make key insights instantly clear and easy to absorb. Additionally, we've brought additional MCP features into several important areas, like e-commerce, health care, and local services. As a result, things like shopping, booking, and health care consultation can now be finished without friction inside the search experience. Around Chinese New Year, we acted fast to adopt the newest AI agent breakthrough by putting OpenClaw—an open source agent framework that recently gained popularity—straight into the Baidu app with one-click access, so our users could instantly take advantage of cutting-edge agentic AI capabilities, with an MAU of roughly 700 million. That gives close to half of China's population an easy way to reach OpenClaw. As for ERNIE Assistant, the AI chatbot built into our platform, we improved the user experience by adding wider multimodal capabilities. Users have responded well to this enhancement, which pushed ERNIE Assistant's MAU past 200 million in December. We're also expanding our AI search API. Q4 saw faster uptake, with call volume climbing more than 110% compared with the prior quarter. Thanks to industry-leading authority, comprehensiveness, and newly introduced multilingual abilities, our AI search API is now creating wider opportunities internationally. Turning to digital humans, they are a powerful kind of AI application. By merging visual presence, voice, and real-time interaction, they deliver experiences that are more engaging and more effective.
During December 2025, digital humans streaming live on our platform jumped by almost 200% compared to the same period a year earlier. Our digital human tech is now reaching beyond Baidu's own properties to strengthen the wider sector. We've formed partnerships with major players such as Jingdong, Zuoyebang, and TikTok, which confirms how well and how efficiently our digital humans perform. When it comes to technology, we see our hyperrealistic digital human as the next generation of capabilities. During this quarter, the expense of production dropped to about one-third of what it was in the prior quarter, delivering top-tier cost performance and setting up this technology for wider uptake. Miaoda, our vibe coding platform, marks another step forward — it lets people with no coding background create applications using everyday language, covering things like WeChat Mini Programs, websites, mini games, and much more. After MeDo, the international edition of Miaoda, launched in Q4, users around the world had built more than 1 million AI applications by early February, none of whom wrote even one line of code. Going forward, we anticipate substantial opportunities to open up still bigger possibilities in AI application development. Finally, we're applying AI to tackle operational issues and generate efficiency improvements throughout various industries. Yijian, our sophisticated visual intelligence platform, serves as one illustration. Through intelligent visual analysis, Yijian lets enterprises automate checks for operational compliance and safety. Well-known brands spanning coffee chains, quick service restaurants, and fine dining now rely on Yijian to maintain high operational standards across their thousands of sites. FM Agent, our self-evolving agent built to address complicated operational challenges, offers another illustration. It autonomously reasons across data, rules, and real-world constraints, simulating innumerable scenarios to pinpoint optimal solutions; we've observed robust validation both internally, via our own cloud resource optimization, and externally in sectors such as manufacturing, energy, finance, and logistics, where better efficiency is a shared priority. Organizationally, we recently set up the Personal Super Intelligence Business Group, or PSIG. PSIG brings together Baidu Wenku and Baidu Drive, our 2 flagship consumer-facing AI applications. Even prior to this organizational integration, the 2 teams had already worked together at the product level to produce innovations such as Free Canvas and GenFlow. This new group allows for still deeper collaboration moving forward as we speed up the release of new applications to cultivate a strong growth curve powered by application layer innovation. Turning to physical AI. Apollo Go stands as our biggest AI application in the physical world. For Apollo Go, 2025 was a year of accelerated scaling, in which we strengthened our leadership in operational scale and made significant progress in global expansion. We keep expanding fully driverless operations rapidly, providing 3.4 million fully driverless operational rides in Q4, with weekly rides reaching a peak of more than 300,000. Total rides increased by over 200% year-over-year. Cumulative rides given to the public have exceeded 20 million as of February 2026, solidly securing our standing as the world's leading autonomous ride-hailing service provider; we carried momentum across key international markets into 2026. Over in the U.K., our partnerships with Uber and Lyft moved ahead, and we're proceeding with plans to trial autonomous vehicles in London, with testing anticipated to kick off during the first half of 2026. That marks a significant milestone for Apollo Go's international growth, taking our right-hand drive robotaxi expertise from Hong Kong into yet another strategically vital market; in Switzerland, we kicked off testing in St. Gallen after entering that market during the prior quarter. In the Middle East, we made headway in Abu Dhabi as well as Dubai. In Abu Dhabi, this January we rolled out fully autonomous ride-hailing services on Yas Island together with AutoGo; in Dubai, we obtained the city's first fully driverless testing permit issued by the Roads and Transport Authority. We additionally revealed the next stage of our worldwide partnership with Uber, which will bring our fully autonomous ride-hailing services to Dubai through the Uber platform. These milestones are critical ones that speed up our momentum throughout the Emirates. In Asia, we stepped into a new market, South Korea, beginning with the Seoul metropolitan area, which broadens our presence even further across the Asian region. At the same time, in Hong Kong, we extended our open road testing to Tsuen Wan and began cross-district testing linking Airport Island and Tung Chung, which moves us nearer to commercial readiness in that location. As of February 2026, Apollo Go's worldwide footprint hit 26 cities, showing how scalable our autonomous driving technology is across varied regulatory and operational settings. Going forward, our focus is on speeding up expansion into additional cities around the globe while we keep enhancing operational excellence and unit economics. The experience we keep gaining in varied markets gives us confidence that we can scale even further, and we anticipate that additional cities will reach positive unit economics in time. Safety underpins this expansion and stays our number one priority, the foundation for all that we do. Our autonomous ride-hailing service ranks as the safest anywhere, with our fully driverless vehicles seeing an airbag deployment accident just once per over 12 million kilometers. As we grow, we'll keep raising safety standards and make sure reliability is sustained. In the end, our mission is to leverage AI to reinvent mobility, rendering it fundamentally safer, more affordable and more comfortable, and enhancing the way millions of people move, work and live. To sum up, now that AI is solidly woven throughout our portfolio, we believe we're well positioned to generate sustainable value and help shape what comes next in the AI era. With that, I'll hand the call over to Henry so he can walk through the financial results.
Appreciate that, Robin, and hello everyone. We're making progress on the priorities that count the most for us. Over the last few quarters, we've enhanced what we disclose, giving you greater visibility, and continued advancing on operational efficiency gains; this quarter brought a significant step toward unlocking value from our strategic bets in AI chips through the planned Kunlunxin spin-off and a separate listing — an achievement that brings us real joy. On top of that, we introduced a new USD 5 billion share buyback program and, for the very first time, established a dividend policy. On top of that, we've honed our strategic emphasis on AI applications with high potential by setting up a PSIG business group that brings together Baidu Wenku and Baidu Drive. All of this shows how steadily we execute and how we keep concentrating on building shareholder value. Turning to Q4 results, momentum was positive. Total revenue for Baidu General Business climbed 6% compared with the prior quarter, while non-GAAP operating profit grew 28% sequentially to RMB 2.8 billion. Operating cash flow for Baidu became positive in Q3 and stayed positive in Q4, producing RMB 3.9 billion combined over the two quarters; when it comes to our core AI-powered business, Q4 revenue topped RMB 11 billion, which represents 43% of Baidu General Business revenue. We're witnessing robust momentum in a number of areas. AI Cloud Infra keeps winning market traction and growing faster than the industry average. Our portfolio of AI applications is expanding quickly, with strong uptake among enterprises. If you put AI Cloud Infra and AI applications together, our cloud revenue hit RMB 30 billion for the whole of 2025. Meanwhile, Apollo Go is cementing its position as a global frontrunner in autonomous ride-hailing, holding one of the industry's biggest footprints along with the most powerful growth momentum. AI native marketing services, too, are growing quickly. These figures reflect the headway we've achieved, and from where we sit, this is merely the beginning. A robust slate of initiatives lies ahead of us, and we trust in our ability to generate durable shareholder value. Now, allow me to walk you through the details of our Q4 and full year 2025 financial results. Total revenues in Q4 reached RMB 32.7 billion, representing a 5% increase quarter over quarter, driven primarily by expansion in Baidu core AI-powered business. Looking at the complete 2025 fiscal year, we brought in RMB 129.1 billion in total revenues, which represents a 3% decrease from the previous year. The main factor behind that drop was a reduction in our legacy business, though this was partially balanced out by gains in Baidu core AI-powered business. During Q4, our cost of revenues came to RMB 18.3 billion, which was unchanged from the quarter before. Across all of 2025, cost of revenues reached RMB 72.4 billion, marking a 10% increase year-over-year, primarily driven by elevated expenses associated with Baidu core AI-powered business. Operating expenses for Q4 totaled RMB 13.0 billion, rising 10% from the previous quarter, largely due to increased expected credit losses along with one-time employee severance costs intended to boost efficiency. For the entire 2025 year, operating expenses amounted to RMB 46.3 billion, reflecting a 1% rise year-over-year. During 2025, we recorded a RMB 16.2 billion impairment tied to long-lived assets, and the loss stemmed from writing down the core asset group. Operating income for Q4 landed at RMB 1.5 billion, which equates to a 5% operating margin. Across the full year 2025, we posted an operating loss of RMB 5.8 billion, with an operating loss margin of 5%. Excluding the long-lived asset impairment, operating income for 2025 would have been RMB 10.4 billion. On a non-GAAP basis, Q4 operating income was RMB 3.0 billion, giving a non-GAAP operating margin of 9%. For the entirety of 2025, non-GAAP operating income totaled RMB 15.0 billion, and the non-GAAP operating margin stood at 12%. Now looking at Q4, total other income net came in at RMB 1.2 billion, compared with RMB 1.9 billion during the previous quarter. On the tax side, we recorded RMB 1.0 billion of income tax expense, whereas the same quarter last year showed an income tax benefit of RMB 1.8 billion. Turning to 2025, total other income net reached RMB 12.5 billion, versus RMB 7.4 billion a year earlier. Income tax expense totaled RMB 1.3 billion, compared with RMB 4.4 billion in the prior-year period. For Q4, net income attributable to Baidu stood at RMB 1.8 billion, with a net margin for Baidu of 5% and diluted earnings per ADS of RMB 3.71. Non-GAAP net income attributable to Baidu was RMB 3.9 billion, non-GAAP net margin for Baidu came in at 12%, and non-GAAP diluted earnings per ADS was RMB 10.62. Looking at 2025, Baidu's attributable net income came in at RMB 5.6 billion, with a 4% net margin for the company and RMB 11.78 in diluted earnings per ADS. Once you set aside the effect of long-lived asset impairment, the figure for net income attributable to Baidu reaches RMB 19.4 billion. On a non-GAAP basis, net income attributable to Baidu totaled RMB 18.9 billion, the non-GAAP net margin for Baidu stood at 15%, and non-GAAP diluted earnings per ADS amounted to RMB 53.41. When we refer to total cash and investments, we mean cash, cash equivalents, restricted cash short-term investments, net, long-term time deposits, held-to-maturity investments, and adjusted long-term investments. At the close of December 31, 2025, our cash and investments together came to RMB 294.1 billion. Cash flow from operations during Q4 reached RMB 2.6 billion. Looking at all of 2025, operating cash flow came in at negative RMB 3.0 billion, though it held a positive position over the last 2 quarters in a row. Turning to headcount, Baidu General business employed about 29,000 people as of December 31, 2025. With that, operator, we're ready to take questions.
Alicis, regarding your question about the model — we've indeed observed quite frequent iteration on models lately. How does the management team assess the competitive environment right now? Also, Baidu just rolled out an updated ERNIE 5.0 and made certain organizational changes; could management talk about the strategic reasoning driving those actions, and how the company views the connection between model advancement and the application side within your overall AI approach?
Alicis, it's Robin here; we have indeed witnessed a lot of model launches lately. Competition in this market is intense and the pace is rapid. Yet through all of that rivalry, our longstanding conviction is that applications carry more weight than models, since value from models is ultimately delivered via applications. That's the reason we consistently adopt an application-driven strategy with ERNIE. The most valuable and promising use cases steer how we improve the model. This has held true through each iteration of ERNIE. As I said earlier, we recently put out the upgraded version of ERNIE 5.0. Simultaneously, we have been proactively implementing organizational changes so we remain agile in this rapidly shifting market. We split our model team into distinct focus areas. One team keeps advancing frontier capabilities at the foundation model level in order to sustain technical leadership. ERNIE holds obvious strengths in a number of key areas, including creative writing, omnimodel understanding and instruction following. We feel confident that we will continue enhancing ERNIE's performance across key application scenarios. At the same time, these high-value application scenarios keep supplying ERNIE with real data and feedback, which propels model iteration and makes ERNIE progressively better. The other team operates far closer to specific business needs and application scenarios, concentrating on cutting costs, boosting speed and raising efficiency, or drawing on the best models available for particular use cases, all with the aim of helping businesses make better use of AI according to what they actually need. We understand that model capabilities span widely and application scenarios can be extremely varied. And no lone model can excel everywhere. So we make full use of ERNIE wherever its strengths are obvious, and we remain open to adopting other models in places where those fit better. The objective is always to secure the best application results. To wrap up, we will stick with our application-driven strategy, drawing on genuine application needs to continually iterate and optimize our models, while also continuing to refine the applications themselves so as to produce better and better outcomes, and in the end, generate tangible value for users and businesses.
I'd like to ask about Baidu AI Cloud. We saw that revenue for Baidu AI Cloud put up very strong growth across the whole of 2025. Could you go into more detail and give us a sense of what the main driver was behind that robust top-line increase? And what's the right way to frame the outlook for AI cloud revenue growth in 2026?
Thanks, Alex. It's Dou here. In 2025, revenue for our AI cloud, which covers what we get from AI Cloud Infra plus AI applications, came to RMB 30 billion. The portion from AI Cloud Infra was up 34% compared with a year earlier, growing faster than the wider market. Inside AI Cloud Infra, subscription revenue tied to AI accelerator infrastructure rose 143% versus the prior year in Q4 and is now the main engine of growth, showing real momentum. We have a lot of confidence that we can keep up strong momentum in 2026. What's driving us is faster adoption of AI among enterprises. Demand is rising for both training and inference workloads, and we think appetite for AI computing will continue to expand, which opens up meaningful opportunities ahead. A key differentiator for us in seizing that opportunity is Baidu's full stack end-to-end AI architecture. At the base of this architecture sits our industry-leading AI infrastructure, which strikes an excellent balance among performance, efficiency and cost. A diverse mix of chips powers our AI infra. We've developed deep expertise in heterogeneous computing and unified scheduling, and that lets us manage computing resources efficiently across different chip vendors while delivering industry-leading performance and efficiency. At the same time, our proprietary chip capabilities give us a significant competitive edge. As Robin said earlier, the Kunlunxin AI chips we developed ourselves offer strong performance, compatibility and cost efficiency; they've been deployed at scale with leading enterprise customers in financial services, telecom, energy and Internet sectors, and feedback from the market has been very positive. Kunlunxin is a key component of the computing power on our own cloud platform and plays an important role in our overall AI infra. As AI demand grows, the advantages of our AI infra will become increasingly evident. Beyond the AI infra we just covered, we keep evolving our best-in-class agent infra so enterprises can rapidly build and deploy AI agents at scale. We continue to bring in the latest, most cutting-edge capabilities. For instance, we recently launched simplified open cloud deployment on Baidu AI Cloud, which makes the process more streamlined so that even users with no coding experience can quickly deploy their own open cloud agents. Then as we look toward 2026, with enterprise AI deployments going deeper, we're confident our cloud business will keep growing faster than the industry. We expect AI Cloud Infra to maintain strong momentum, with AI accelerator infrastructure remaining a core driver, pushing our overall cloud business toward a more sustainable and high-quality growth mode.
This quarter, our AI-driven operations kept up quite robust expansion. How does leadership assess where these AI-centric businesses stand right now in their development? At what point do we anticipate that portion surpassing, let's say, half of Baidu General Business? Also, what's going to be the primary catalyst ahead for the AI-centric operations?
All right. I'll begin by explaining our perspective on the main AI-centric operations. These encompass AI cloud infrastructure, AI applications such as Baidu Wenku and Baidu Drive, plus our robotaxi operation, Apollo Go, along with AI-native marketing offerings, which include agents and digital humans. The AI-centric operations structure our company based on what our products and services fundamentally are, with AI enhancing each one to generate real customer value and business results. During Q4, revenue from AI-centric operations surpassed RMB 11 billion. That represents roughly 43% of Baidu General Business revenue. This proportion has climbed swiftly across recent quarters, and AI-centric operations are turning into the primary engine behind our total revenue expansion. Every one of our AI-centric operations possesses distinct strategic positioning and competitive edge. To begin, AI Cloud Infra. We observe companies taking AI from trial stages into full deployment, and our comprehensive, end-to-end AI strengths deliver robust results at aggressive pricing. In 2025, AI Cloud Infra revenue outpaced the sector norm, with subscription-based AI accelerator infrastructure revenue surging dramatically during Q4. Next comes AI applications. We have long held that AI's greatest worth will ultimately reside at the application layer, and we assembled one of China's broadest AI application lineups. As AI abilities keep advancing and fresh use cases arise, we anticipate considerable room for growth in this segment; and the third one is our robotaxi operation, Apollo Go. Apollo Go is growing quickly while also extending its reach abroad. We hold the top position worldwide for operational scale, safety track record, efficiency, and cost structure. And fourth are AI native marketing offerings such as agents and digital humans; these boosted interaction and conversion rates, and we're witnessing robust uptake in the market with substantial upside remaining; so considering the medium to long term, as enterprise AI adoption goes deeper, the ability to monetize AI applications strengthens and physical AI uses like autonomous driving keep broadening, we remain confident about the growth path of our AI-centric operations. These AI-centric operations don't stand alone. They consistently strengthen one another via our comprehensive strengths; and given what we can currently see, we expect our main AI-centric operations to account for most of Baidu General Business within the foreseeable horizon.
Might the management team walk us through the approach you take when deciding where capital goes — things like returning money to shareholders, funding organic growth, and possible strategic openings? And on top of that, could management share its view on where Kunlunxin fits strategically over the long run inside the Baidu Group?
Henry here. My guess is a lot of you have already picked up on the wave of actions we've rolled out lately. Among them: stepping up what we disclose, running things more efficiently, reshaping how our organization is structured, pushing forward with the planned Kunlunxin spin-off and separate listing, and also unveiling both our fresh share repurchase program and our inaugural dividend policy. More recently, we've been restructuring the PSIG, the Personal Super Intelligent Group business group, bringing Baidu Wenku together with Baidu Drive. Taken as a whole, these steps show a consistent framework for execution, which speaks to how our management execution has gotten better and our continued dedication to building shareholder value. Let me use our new share repurchase program to illustrate. Giving shareholders clear, sustainable returns is something we care about deeply. Thus, back in February — in the recent period — the Board gave the green light to a fresh USD 5 billion share repurchase program, and our intention is to carry it out routinely, in a highly disciplined and transparent fashion. We're additionally rolling out Baidu's first-ever dividend policy. Our view is that pairing this policy with a buyback program of considerable size will make our shareholder return profile even stronger and draw in a wider set of investors, which in turn makes our investor base more diverse. As we've said before, the planned spin-off and separate listing of Kunlunxin serves as yet another solid illustration. The listing process is moving along quite well for us. Kunlunxin came about through more than a decade of investment and stands as an essential infrastructure piece within our full-stack AI capabilities; our belief is that this spin-off and separate listing will earn robust recognition from the market and free up meaningful value for Baidu as a group. So as we look forward, our conviction is firm that this company holds enormous value, and we'll keep freeing it up via a range of initiatives. Delivering returns to our shareholders that are sustainable and consistent remains our pledge. So additional initiatives are coming in due course. So keep watching us. Thank you.
I'd like to ask about robotaxi. To start, congrats on growing robotaxi into additional nations, and in particular my home city of Hong Kong. Could you tell us your plan abroad for 2026? What would you say are your main edges in those markets? Also, given that Waymo was just valued at $126 billion, what is management's view on how to unlock Apollo Go's value? Might a spin-off be something you'd explore?
Gary, like I said during the prior quarter, my view is that robotaxi has hit a global tipping point. Because we keep delivering safe autonomous drives and because of positive word of mouth, we observe a growing number of nations and regions building a supportive environment for robotaxi operations. Our expectation is that the sector speeds up in 2026. In this space, Apollo Go is plainly a worldwide leader. More than 20 million cumulative rides have been finished by us. During the busiest stretch, our weekly fully driverless rides were above 300,000. As of now, Apollo Go fleets have rolled up over 300 million autonomous kilometers, among them more than 190 million fully driverless autonomous kilometers, with a superb safety record; and we keep pushing forward our industry-leading technology so that rides become safer and more pleasant. We are likewise speeding up international growth to seize worldwide opportunities. As of today, our worldwide reach covers 26 cities on various continents, spanning robotaxi markets with both left-hand and right-hand drive. Our autonomous driving system performs dependably in any location, across differing traffic patterns and differing urban environments. It's worth noting that hardly any players have gone into the right-hand drive robotaxi market, yet we have already set up a presence and are progressing quickly. On top of that, we hold a basic cost edge. RT6 is the first purpose-built production vehicle anywhere in the world engineered from scratch for Level 4 autonomous driving. Priced below USD 30,000 per vehicle, RT6 delivers the industry's finest cost structure, and when paired with our top operational efficiency, that lets us reach the lowest cost per mile anywhere in the world while keeping superior safety. In Wuhan, during late 2024, we became the first to reach UE breakeven. And as you're aware, ride-hailing prices in most major cities are above those in Wuhan. To speed up global growth, we are drawing on a range of strategic partnerships. For instance, together with Uber and Lyft we are working in London toward a launch this year, and in Dubai with Uber as well. Such partnerships produce quicker, more efficient market growth. We regard Apollo Go as a strategic growth engine possessing considerable long-term potential. A great many major cities lack enough human drivers. Additional supply through robotaxi service not only gives safer rides, but additionally sparks ride-hailing demand, and so adds tax revenue for the government; it further frees precious land from parking spaces and offers extra monetization opportunities for those real estate assets. Our attention is on 3 areas: first, aggressively scale up safe and comfortable operations by deploying more vehicles. Second, continuously improving unit economics with the goal of achieving UE breakeven in more cities this year. Third, expanding with flexible business models, both domestically and internationally. On the strategic side, we're going to stay flexible and weigh whichever route delivers the greatest long-term returns for shareholders; and naturally, execution and sustainable growth are always where our attention sits. In our view, the autonomous ride-hailing industry overall is still undervalued. Down the road, we anticipate that valuations throughout the sector will come to reflect more fully what this technology can transform, and that opens up substantial upside for Apollo Go.
Wishing you a joyful year filled with hope. My question concerns competition. Recently, we've noticed that rivalry in consumer-facing AI has been heating up, particularly around Chinese New Year. What's your take on how the competitive landscape looks right now? In this market, where do you place Baidu's AI2C offerings — things like the early assistance, the differentiation, and the positioning? And finally, what's the thinking on the route toward monetization?
This is Julius speaking. Competition within the AI2C product space is fierce; throughout the recent Chinese New Year period, we observed certain rivals pursuing quite aggressive market tactics to grow their user bases quickly. That said, given how fast technology and products keep evolving, our conviction remains that the core strategy ought to stay anchored in genuine user needs; we're deeply dedicated to continually upgrading the capabilities of our current products and services via AI innovation so that we serve our users better. Among our flagship consumer-facing offerings, such as the Baidu app today, we've created an ERNIE Assistant to bolster our service capabilities throughout the whole user journey — from information thinking all the way to delivering solutions and getting tasks done; when it comes to information thinking, we've considerably strengthened how users assess information via ERNIE Assistant. As an illustration, we've boosted answer accuracy and relevance using RAG, and ERNIE Assistant keeps error rates low with very few hallucinations, bringing highly trustworthy content to users. We've likewise embedded multilingual AI capabilities, such as API functions, which let users assess a broad range of information sources while conversing, boosting how rich and usable the information is; this proves especially valuable in scenarios such as travel planning. And in December, ERNIE Assistant's MAU went beyond 200 million, with conversation rounds and engagement rising quite rapidly. Regarding past complexion, we're integrating MCP agents to link users with tools and real-world services. During this quarter alone, we're introducing close to 100 service capabilities, particularly across health care, travel, education and e-commerce. As an example, via the Baidu Health MCP integrated into ERNIE Assistant, users can assess a variety of health care capabilities, extending from online through to offline services; in e-commerce, our MCP module experienced extremely robust GMV growth quarter-over-quarter. At the same time, with the stand-alone ERNIE app we're pursuing a different route, positioning it as a platform for innovation and experimentation. Our earlier multi-model AI features have secured solid traction among young audiences. And more lately, we've introduced AI capabilities centered on workplace productivity, drawing on ERNIE's capacity to manage complex tasks in professional settings. We're noticing encouraging early signals across these productivity scenarios. We adopt a measured stance toward monetizing the AI tools and products, putting product excellence and user experiences first. Monetization will come naturally once the products mature. Thank you for your question.
My question is mostly about the AI investment. How should we think about AI-related CapEx over the coming 12 to 24 months? What's the right way to think about the return profile on these AI investments and the expected effect on ROIC over time? And broadly, where do you see additional efficiency opportunities that could support margin and cash flow improvements down the road?
This is Henry. First, regarding CapEx and AI investment, since our launch in early March 2023, we've put over RMB 100 billion into AI. Looking ahead, we'll keep up that same investment density. Second, we're very mindful about returns and we understand investors' focus on return on invested capital. That's why we've worked to improve our financial performance, and we've delivered solid results on key metrics across the past few quarters. For instance, in Q4, Baidu's gross profit rose double digits sequentially, and Baidu's non-GAAP operating income was up roughly 35% quarter-over-quarter. We also did better on margin profile, with both gross margin and operating margin rising sequentially. Importantly, Baidu's operating cash flow turned positive in Q3 and stayed positive in Q4. Across the second half, operating cash flow came to nearly RMB 4 billion. Baidu's free cash flow also turned positive in Q4. Third, we've identified and explored other ways to support our financial needs, such as operational and financing leasing, plus we can tap low-cost interest banking borrowing. For instance, certain bank borrowings and leasing facilities carry interest rates as low as under 2%. These approaches help us keep a healthy long-term financing structure while continuing our AI investments and supporting our business growth. So to sum up, we'll keep maintaining our AI investment density, while balancing investor focus on profitability and return time lines. We believe that even with substantial AI investment, our operating cash flow will stay positive going forward as well. Thank you.
Thank you. Everyone, that brings today's call to a close. We appreciate your joining us. You're free to hang up 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.