Baidu, Inc. (BIDU) Earnings Call Transcript & Summary
May 18, 2026
What were the key takeaways from Baidu, Inc.'s May 18, 2026 earnings call?
In Q1 2026, Baidu, Inc. (BIDU:US) reported total revenue of RMB 26.0 billion, up 2% year-over-year, marking a return to positive growth after several quarters of decline. The company's core AI-powered business generated RMB 13.6 billion in revenue, a significant 49% increase year-over-year, and now constitutes 52% of total revenue. Management expressed confidence in continued AI growth, citing strong momentum in AI Cloud Infrastructure and applications, while maintaining a positive outlook for the fiscal year ahead.
What topics did Baidu, Inc. cover?
- Return to Revenue Growth: Baidu reported total revenue of RMB 26.0 billion, up 2% year-over-year, indicating a return to positive growth after previous declines. CEO Robin Li stated, "This is an important milestone as AI-powered business has now become the majority of our revenue mix."
- AI Cloud Infrastructure Momentum: Revenue from AI Cloud Infrastructure grew 79% year-over-year, with GPU cloud revenue accelerating to 184% growth. This segment is becoming a key driver of Baidu's overall business, as highlighted by Li's comment on AI's role as the primary growth driver.
- AI-Powered Business Dominance: The core AI-powered business accounted for over half of total revenue for the first time, reaching 52%. This shift underscores Baidu's strategic pivot towards AI, as stated by CFO Henry Hu, "AI has now become the core driver and the majority of our business."
- Apollo Go Expansion: Baidu's robotaxi service, Apollo Go, maintained strong momentum with 3.2 million fully driverless rides in Q1. Management emphasized the service's potential for broader monetization opportunities beyond traditional ride-hailing.
- Foundation Model Advancements: The launch of ERNIE 5.1 showcased Baidu's commitment to enhancing its AI capabilities, achieving top rankings in various leaderboards. Li noted, "We will continue to iterate ERNIE in line with the needs of our key applications," indicating ongoing investment in AI models.
What were Baidu, Inc.'s May 18, 2026 results?
- Total Revenue: RMB 26.0 billion (up 2% YoY, returning to positive growth)
- Core AI Revenue: RMB 13.6 billion (up 49% YoY, now 52% of total revenue)
- AI Cloud Revenue: RMB 11.3 billion (significant growth driven by AI Cloud Infrastructure)
- Operating Income: RMB 3.2 billion (operating margin at 10%)
- Net Income: RMB 3.4 billion (net margin at 11%)
- Operating Cash Flow: RMB 2.7 billion (positive for the third consecutive quarter)
Baidu's strong Q1 results reflect a successful pivot towards AI-driven business models, with significant growth in AI Cloud and applications. The company's commitment to innovation and strategic investments positions it well for future growth. Investors should monitor the execution of international expansion plans and the competitive dynamics in the AI landscape as potential catalysts and risks.
Earnings Call Speaker Segments
Good day, everyone, and thanks for joining us for Baidu's first quarter 2026 earnings conference call. Our earnings release went out earlier today, and copies are available on our website and through newswire services. With us on the call are Robin Li, our Co-Founder and CEO; Julius Rong Luo, our EVP overseeing Baidu Mobile Ecosystem Group, MEG; Dou Shen, our EVP overseeing Baidu AI Cloud Group, ACG; and Henry Haijian Hu, our CFO. Following our prepared remarks, there will be a Q&A session. I'd like to note that today's discussion includes forward-looking statements made under the safe harbor provisions of the U.S. Credit Securities Litigation Reform Act of 1995. Such forward-looking statements carry risks and uncertainties that may cause 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 our other filings with the SEC and Hong Kong Stock Exchange. Baidu assumes no obligation to update any forward-looking statements, except as required under applicable rule. Our earnings press release and this call include discussions of certain unaudited non-GAAP financial measures. The press release contains a reconciliation of the unaudited non-GAAP measures to the audited most directly comparable GAAP measures and is available on our IR website at ir.com. Please be aware that this conference is being recorded. A webcast of this conference call will also be available on Baidu's IR website. I'll now hand the call over to our CEO, Robin.
In the first quarter, we brought in total revenue of RMB 26.0 billion, which was up 2% compared to the same period last year — that marks a return to positive growth. Our core AI power business generated RMB 13.6 billion in revenue, rising 49% year-over-year, and for the very first time it made up over half of Baidu's general business revenue, hitting 52%. That's a significant milestone, since AI-powered business now represents the majority of our revenue mix. AI cloud infra showed outstanding momentum in Q1, with overall revenue climbing 79% year-over-year. Inside AI Cloud infra, GPU cloud revenue kept up its strong trajectory from last quarter's 143% growth, speeding up even more to 184% year-over-year. Apollo Go turned in a strong quarter as well. During Q1, we delivered 3.2 million rides with no driver at all, and our total rides kept expanding at a triple-digit rate compared with the same period a year earlier — a reflection of how our operations continue to scale. Viewed alongside our other results, this makes one thing plain: AI now serves as Baidu's main engine of growth, and it reinforces where we stand as a company that puts AI first. With adoption of AI still speeding up, practical uses are widening, which generates fresh and ever more varied demand for what AI can do. We feel sure we can capture these chances as they present themselves, and we expect AI to continue powering Baidu's next stage of expansion. Let me now walk you through the quarter's most important points, starting with AI Cloud Infra. We keep seeing demand climb for both training and inference workloads as AI adoption accelerates throughout various sectors, and inference is scaling up particularly quickly, accounting for an expanding portion of total demand. For our AI Cloud Infra, Q1 brought substantially faster growth, with revenue increasing well beyond the wider market's pace. Higher-quality revenue sources now represent a larger part of our mix, and GPU Cloud—usually associated with better margins—has grown into a significant piece of our overall AI Cloud Infra revenue, underscoring continuous gains in the health of the business. Baidu's full stack AI capabilities provide a distinctive edge that drives this momentum, something only a handful of firms worldwide can genuinely assert. Because we control proprietary technology at each level—everything from the foundational infrastructure up through the applications—we could keep our compute supply both stable and dependable, while simultaneously fine-tuning the whole stack from one end to the other. That let us keep raising performance, keep bringing expenses down, and keep giving customers very attractive value for what they pay. As AI use cases keep spreading, this benefit of owning the complete stack grows more and more obvious, which lets us go after a wider and more varied set of opportunities. When it comes to the infrastructure tier, Kunlunxin—our AI chips that we developed in-house—gives us a clear edge. Demand for Kunlunxin has been robust and keeps expanding, as more and more customers in a wide range of industries pick it up for an ever-widening variety of AI workloads. This shows the market increasingly acknowledging that Kunlunxin delivers stability, efficiency, compatibility, and versatility. It also stands among the earliest domestically produced AI chips to reach large-scale commercial deployment within one AI computing cluster exceeding 30,000 accelerators, with cluster performance and stability that lead the industry. Kunlunxin operates on a complete software stack, offering compatibility across a variety of models and frameworks, along with dependable usability for enterprise environments. To date, tuning and verification have covered workloads across numerous models, including the latest ERNIE versions and other mainstream foundation models, while inference support was most recently extended to DeepSeek V4 GLM 5.1 and MiniMax M2.7. As a key piece of our AI infrastructure, Kunlunxin adds further strength to our infrastructure layer's foundation, letting Baidu AI Cloud help customers deploy AI with more efficiency, reliability and cost effectiveness, which lifts the overall competitiveness of our cloud products. Those benefits are showing up as robust client momentum on the infrastructure side. Baidu AI Cloud has turned into a trusted infrastructure partner for an expanding group of major companies spanning a wide set of industries — Internet, gaming, embodied AI, autonomous driving, smartphones, financial services and others; during this quarter, several notable new clients came on board, among them leading model companies. Names like Unity, Honor, Opel and Vivo are also part of our client base. Meanwhile, top-tier clients already with us keep broadening their work with us and increasing their usage, fueling healthy expansion throughout our client base; when it comes to MaaS, as open cloud won adoption industry-wide, we acted fast to grow the model library, our < > Mass platform. Beyond ERNIE, < > today backs a widening lineup of in-demand models, among them popular offerings from < > AI, MiniMax, < > and DeepSeek, which keeps our model library thorough and current. In March, average daily token consumption by external customers climbed to almost 7x what it was a year earlier, and our MaaS revenue likewise grew quickly; we think the MaaS platform holds substantial untapped potential as the ecosystem surrounding agents and AI applications keeps developing. On foundation models, we just introduced ERNIE 5.1, which offers stronger tax capabilities, a more compact size and improved reasoning relative to the prior version; we also progressed in key areas such as cold generation, agent capabilities and deep search. Lately, on the LM Arena, ERNIE 5.1 holds first place among Chinese models on the text leaderboard. ERNIE 5.1 likewise led the LM Arena search leaderboard among Chinese models, coming in fourth worldwide, which makes it the sole Chinese model to show up on that leaderboard too. Going forward, we stay firmly dedicated to pushing ERNIE ahead via an application-driven approach, continually iterating off real-world needs so we keep earning a place at the forefront of AI capabilities. Let me now shift to AI applications. We have long held that AI's true value is ultimately delivered through applications, and we were early and persistent in assembling a comprehensive portfolio that serves enterprises and individual users alike. This quarter, we keep witnessing encouraging progress in several high-potential directions. Let me call out a few examples. First, we have our AI agent focused on day-to-day productivity, which was recently presented at Baidu Create. It's built to carry out intricate multi-step workflows spanning apps and files on its own, managing lengthy tasks all the way through. It works on both PC and mobile, letting people kick off tasks whenever and wherever, while running nonstop in the background like a 24/7 AI assistant. All users have to do is say what they want, then check back for the outcome. What sets it apart is how tightly it connects with Baidu's proprietary skills, such as AI search, among others. As we keep broadening its ecosystem, we're confident it will handle an increasingly wide variety of office workflows and tough real-world tests, assisting users in finishing the end to end more efficiently. Now on to digital humans. Our hyperrealistic digital human tech kept progressing, with better performance and growing readiness for large-scale rollout. On the cost side, we recorded roughly an 80% reduction across the last 2 quarters, which brought down the barrier to adoption and made our digital humans cheaper and easier to access for a wider set of clients. At the same time, we're bringing our digital human capabilities to the global stage. At the latest Baidu Create event, we rolled out a digital human platform for overseas markets, which lets merchants and creators generate digital human content fast — everything from e-commerce live streaming to digital human videos and beyond.
For our digital humans to genuinely serve markets worldwide, localization was engineered into them from the very beginning. They handle 24 languages — Spanish, French, and Thai among them — with scripts and presentation approaches tailored to fit each culture and connect with local viewers. That enables merchants to keep digital human live streams running 24/7 that come across as truly native, opening up fresh gains in efficiency and conversion across the globe. Our partner network, expanding both inside China and abroad, features < >, TikTok and < >, and multiple partners are taking their work with us to a deeper level. Turning to Miaoda, our vibe coding platform. With Miaoda, anybody can turn ideas into reality without writing even one line of code, and that value is gaining wider recognition; during March, Miaoda's monthly active users climbed roughly 70% versus the prior quarter, and our domestic paying user rate hit about 3x what it was at the close of last year. At Baidu Create, we rolled out Miaoda 3.0, bringing an enterprise edition plus a mobile app, which allows wider uptake among individuals and enterprises alike and more adaptable use across different times and scenarios. Worth noting, Miaoda can now generate stand-alone mobile applications, widening further what people are able to build with it. Take Famou agent as another illustration. This self-evolving agent was created to tackle intricate operational problems spanning industries and to help enterprises achieve substantial productivity improvements; when Famou agent 2.0 debuted at Baidu Create, we broadened how accessible it is. Whereas prior iterations mostly served developers and technical teams, Famou agent 2.0 brings down the entry barrier by letting domain experts engage the agent straight through natural language, with no coding skills needed; for instance, at Qingdao Port — among the top ports globally, with a highly advanced scheduling system and profoundly complicated operational logic — Famou agent is assisting in driving an already sophisticated system to even greater efficiency; in a setting where thousands of interdependent variables require real-time coordination, Famou agent independently probes the solution space to pinpoint optimal decisions spanning birth scheduling, equipment allocation and cargo prioritization, and even against an already highly optimized baseline, Famou agent keeps delivering incremental efficiency gains, boosting overall operational performance further still. Moving on to AI Search. During Q1, we kept pushing forward with our AI Search transformation, concentrating especially on lifting user satisfaction and the broader search experience. By continually upgrading model capabilities, we made further gains in how our search results are planned, structured and generated, which allows stronger evaluation of content quality, wider reach for high-quality information and a notable drop in low-quality content. In parallel, ERNIE assistant maintained robust user engagement, fueled by continuous refinements to how it interacts. During March, early assistant's daily active users almost doubled versus a year earlier, and daily average conversation rounds more than tripled across that same span. Next day, retention likewise rose significantly, showing users stick around more. As we look forward, we'll keep tightening the connection between AI Search and the ERNIE assistant, which will improve things even more when people look for information, make sense of content, and get tasks done. Outside the digital realm, AI is remaking the physical world in dramatic ways, and robotaxi represents the strongest setting for this shift. During Q1, Apollo Go, which is our robotaxi operation, kept up robust momentum: fully driverless vehicles kept increasing, and our safety performance stayed the best in the industry; we also kept pushing our international growth, registering solid progress in major markets abroad. In Europe, we remain on schedule to start open road testing in Switzerland, and our initial cars have reached London, getting ready for testing with Uber and Lyft that should kick off shortly. In the Middle East, Apollo Go's fully driverless services are now operating in several zones of Dubai; additionally, in late March we formally introduced the Apollo Go app, which makes us the first and only autonomous ride hailing service there to have its own stand-alone app. Past conventional ride hailing, we're testing new use cases that widen Apollo Go's commercial reach. In Hainan, among China's most visited tourist spots, we worked with < > on a rental model using our fully driverless vehicles, positioned right at the arrival level of < > airport. That way Apollo Go is there waiting the moment visitors walk out of the terminal. We think robotaxi can create value far past everyday commuting, unlocking fresh monetization possibilities along the way. As Apollo Go kept expanding fully driverless operations, we've run into a wider and more intricate set of real-world situations, among them system and operational complexities that surface only at greater scale. Whenever those situations come up, we deal with them carefully and leverage them to keep making our operations stronger. More generally, Apollo Go has advanced far past technology demonstration and small-scale pilots. At this scale, we're tackling a new frontier focused on how robotaxi services can fit more naturally into public transportation, city operations and daily life. Such experiences are assisting us in developing the expertise and knowledge required for Apollo Go to coexist more smoothly with the wider transportation ecosystem over time and, in the end, to turn into a more convenient and trusted service for those we serve. To wrap up, our AI-powered businesses generated strong momentum across every area in Q1. AI has now turned into the core driver and the bulk of our business, and we expect that role to do nothing but grow from here as we scale AI across an ever more diversified portfolio and stretch our reach into global markets—from AI applications to autonomous ride hailing, we see substantial opportunities emerging on many fronts. We're confident in our capacity to seize them. With that, I'll hand the call over to Henry to walk through the financial results.
Thanks, Robin, and good day to all of you. During the first quarter, we kept moving forward on our main priorities: strengthening that closure, expanding our core AI-driven business, and making operations more efficient. I want to call out some results from the quarter. Baidu General business total revenue rose 2% compared with a year earlier, which brought us back to positive growth following multiple quarters of contraction. Non-GAAP operating income for Baidu General business climbed 39% versus the prior quarter, reaching RMB 4.0 billion. For a third quarter in a row, Baidu's operating cash flow stayed positive, at RMB 2.7 billion, which shows our operating efficiency and overall business health kept getting better. We also hit a significant milestone: for the first time, our core AI-powered business represented over half of Baidu General business revenue. Baidu Core AI-powered business revenue surpassed RMB 13 billion, a 49% increase year-over-year. Inside that, AI Cloud Infra's growth far exceeded the wider market, and our AI applications portfolio kept thriving on many fronts. Together, AI Cloud Infra and AI applications lifted our total AI cloud revenue to RMB 11.3 billion in Q1 2026. Outside cloud, Apollo Go strengthened further its standing as a worldwide leader in autonomous ride-hailing and kept broadening its operations. Taken together, these results show a business turning both more AI-centric and more financially sound; let me now take you through the specifics of our first quarter 2026 financial results. Baidu's total revenue came to RMB 32.1 billion, a 2% decline quarter-over-quarter. Baidu General business revenue was RMB 26.0 billion, up 2% year-over-year and unchanged quarter-over-quarter, with the rise in others mainly due to expansion of the AI cloud business. IT revenue was RMB 6.2 billion, down 8% quarter-over-quarter. Cost of revenues totaled RMB 19.6 billion, up 7% quarter-over-quarter, largely from higher costs tied to the AI cloud business, partly offset by lower content costs and traffic acquisition costs. Operating expenses came to RMB 9.3 billion, down 28% quarter-over-quarter, mainly from lower expected credit losses and personnel-related expenses. Operating income was RMB 3.2 billion, with an operating margin of 10%. Non-GAAP operating income was RMB 3.8 billion, with a non-GAAP operating margin of 12%. Total other income net was RMB 626 million, versus RMB 1.2 billion in the previous quarter. Income tax expense was RMB 528 million, versus RMB 1.0 billion in the previous quarter. Net income attributable to Baidu was RMB 3.4 billion, Baidu's net margin was 11%, and diluted earnings per ADS was RMB 8.76. Non-GAAP net income attributable to Baidu was RMB 4.3 billion. Baidu's non-GAAP net margin was 14%, and non-GAAP diluted earnings per ADS was RMB 12.06. We define total cash and investments as cash, cash equivalents, restricted cash, short-term investments, net long-term time deposits and held to mature the investments and adjusted long-term investments as of March 31, 2026. Total cash and investments stood at RMB 279.3 billion. Operating cash flow was RMB 2.7 billion. As of March 31, 2026, Baidu General business employed roughly 28,000 people. With that said, operator, we're ready to take questions on the call now.
Congratulations on the impressive acceleration in AI Cloud Infra revenue this quarter — could you share more color on the key drivers behind this revenue momentum? Also, do you have sufficient compute capacity to support future growth, and how should we think about the margin profile of AI cloud compared with a more traditional, for example, CPU cloud, and the long-term margin trajectory?
Thank you, Alex; this is Dou. We've observed remarkably strong enterprise demand for AI infrastructure, showing particularly strong momentum, which is a pretty healthy signal; it tells us customers have moved beyond training models, and they're now running AI across more parts of their business at an accelerating pace. Closely tied to this, our MaaS platform is gaining strong traction. It is one of the few MaaS platforms in China. As Robin just mentioned, beyond ERNIE we've quickly grown our model library to include the most in-demand models such as DeepSeek, Gemini, MiniMax and others. And we're seeing continued growth in token consumption from external customers. Importantly, supporting new models quickly isn't a simple plug and play process; it demands higher throughput inference and efficient model serving capabilities, so we can run these models reliably at scale and serve more token demand with less compute. Demand is broad-based across verticals, including air night, autonomous driving, embody AI, gaming, advanced manufacturing and more; it's not just existing customers spending more. We keep winning new ones, too, including industries that historically wouldn't have users of AI and cloud computing, like retail and IP-based consumer brands. The addressable market is still expanding. And with demand remaining strong and supply relatively tight, we're actively expanding capacity and improving resource efficiency to better support growing customer needs. Our confidence in capturing this demand comes from our differentiated full-stack AI capabilities, which provide 2 tangible advantages. First, efficiency. Owning and optimizing across the full stack lets us deliver highly competitive price performance for our customers. And second, our proprietary have earned strong recognition in real-world deployment; on our margins, the key driver is business mix. GPU cloud usually carries better margin profiles than traditional CPU cloud for a few reasons. First, GPU cloud is technically more complex with much higher barriers to entry. Baidu was actually one of the earliest cloud providers in China to build GPU cloud at scale, and we remain at the forefront. Second, demand remains very strong, while high-quality supply is relatively tight, customers prioritize stability and availability, not just cost. Certainly, our AI chips and full-stack capabilities give us more room to optimize costs, and continued improvement in our customer mix further supports margin expansion. So as GPU cloud takes a larger and larger share of our total cloud infrastructure revenue, we believe the blended margins for cloud improved structurally. This is a lasting pattern that keeps going, so we have confidence in the path toward long-term profitability for our cloud business.
Alicis, thanks for the kind words about the quarter. Turning to your question on our foundation model: where does Baidu see ERNIE sitting in a field that keeps getting more crowded? And what do we plan to spend, and where will future iterations head?
Alicia, Robin here. Things are shifting fast across the model space, with lots of releases happening from companies in China and around the world. Our view is that capabilities will keep climbing quickly, and having robust foundation model work done in-house stays critical. So our commitment to backing ERNIE won't waver. At the same time, our long-held belief is that value from these models shows up through applications. That's the reason we've always let applications steer our approach. Every new version of ERNIE comes out of actual product requirements and business use cases. Just recently, ERNIE 5.1 arrived, and it posted top-tier scores on LM Arenas, Text and Search leaderboard, which shows we keep making headway on text ability, reasoning and search. Down the road, we'll keep updating ERNIE to match what our main applications need — AI search, digital humans, < > Famou agent; those rank among the areas where we see the most value created. Our aim is to deliver the strongest capabilities precisely where they count. Take one example: we'll go on refining how well ERNIE grasps what users want and judges content quality. That way AI Search can serve up results that are more precise, better in quality and smarter. We've also boosted ERNIE's text and multi-model abilities so our digital humans feel more lifelike and hyper realistic, and do a better job of holding user attention and generating sales during e-commerce live streaming. We'll upgrade coding abilities to give vibe coding stronger support, so users can create applications using natural language. Since coding is turning into — is becoming an ever more essential skill these days, that's an area getting more of our attention. And we'll keep sharpening how ERNIE spots better and better answers in messy real-world situations, which helps companies across many industries capture bigger efficiency gains. To serve — to serve this direction better, we've restructured our model teams and will keep reshaping our organization when necessary. We're sure ERNIE will grow stronger in every one of these areas. Beyond ERNIE, we also offer a set of smaller, quicker and more efficient models, plus combinations tuned for particular scenarios. Distinct use cases demand distinct things when it comes to capability cost, latency and deployment efficiency. Our aim is always to get the best result for whatever the application is. Looking further out, we think the fourth potential AI execution is still a long way from everyday reality. As additional AI use cases appear, the worth of our application-driven approach will only grow more obvious, and ERNIE will keep gaining capability and value as that happens.
Great job on the really impressive cloud traction. I'd like to hear your perspective regarding profitability. Given that AI Cloud Infrastructure sales keep climbing quickly and AI-driven offerings now make up more than 50% of overall revenue, what's the right way to view Baidu's operating margin over the long haul, and what will mainly push margins higher moving forward?
Thanks, Wei. Henry here. During Q1, as you saw, the AI-powered portion of our Baidu Cloud business—covering primarily operations outside classic online marketing—surpassed half of our overall revenue for the very first time. That marks a significant milestone, showing both the rising role of AI and a broader revenue foundation. A lot of these rapidly expanding operations remain in scale-up mode and keep growing. Since they now represent a bigger slice of what we earn, we anticipate that they will add to top-line growth as well as to margin gains, handing us several levers for lasting earnings enhancement eventually. Right now, we're putting money into the most strategic chances with confidence. ROI on those investments matters greatly to us. And I think what we're constructing at present will define our margin profile for upcoming years and build lasting competitive edges. Let me go through the main businesses where we observe this unfolding. To begin with, as you pointed out, AI Cloud Infrastructure. GPU Cloud carries structurally better margins compared with conventional CPU Cloud, thanks to stronger demand, a tighter supply chain, greater technical hurdles and pricing power; thus, once it takes up more of our mix, we anticipate it will serve as a major lever for margin gains and widening. Next, AI applications. That's an inherently high-margin operation, powered by sticky, subscription-based models plus operating leverage as time passes. Third, robotaxi. Our unit economics have steadily gotten better ever since reaching breakeven in Wuhan City. We remain in the investment stage, yet the route to profitability is growing clearer as we expand. Beyond that, at the company level, a few extra levers deserve mention now. First, we keep pushing cost optimization and operational efficiency throughout the whole organization. And second, we're applying AI broadly to boost internal productivity too. Then, not least, third, regarding infrastructure, we keep raising server utilization rates, which feed straight into margins eventually. So to sum up as I said, our revenue mix is shifting toward faster-growing, higher-margin operations, our full stack AI capability generates cost efficiencies, and company-wide productivity improvements compound as time goes on; we think the medium- to long-term margin path looks compelling for us, and we consider it sustainable. Thank you.
Congrats again on the great AI Cloud Infra figures. I'd like to ask about robotaxi. Can management update us on robotaxi operations abroad? And for Robotaxi, how should we think about operating scale and the split of revenue between domestic and overseas markets? How should we think about comparing margin profiles? Longer term, how does Baidu see its role in the robotaxi ecosystem — operator, technology provider, or platform?
This is Robin. Starting with scale: apollo Go is still a global leader. By April, cumulative rides had passed 22 million. Within the global economy, China stands out for how accessible its markets are, which explains why the size of our business at home currently far exceeds what we've achieved abroad. That said, an increasing number of countries worldwide are creating room for robotaxi services as regulations grow more favorable. It brings us real satisfaction that the operational know-how we've built domestically positions us strongly to grow internationally. Over a remarkably brief window, we've accomplished a great deal. Our push to expand abroad only started gaining momentum a handful of quarters back, yet we now have a presence in major markets spanning Europe, the Middle East and Asia. Such speed demonstrates how readily both our technology and our operations can scale across varying market conditions. We feel good about growing abroad, and that comes from the big operational strengths we've already demonstrated in China. After running fully driverless services in real conditions for years, we've built up serious know-how around tricky roads, operational hurdles and < > that only show up once you reach a certain size. You can't put that together quickly; what we've learned keeps refining our algorithms and operational standards, which makes our robotaxi operation stronger with every mile driven. That means when we go into new markets, we carry all of that experience along and can move at a quicker pace. One illustration is how we went from Hong Kong to London. For us, Hong Kong has been a key right and dry global taxi market. Looking back at the last twelve months, the hands-on knowledge we gained there has been quite useful as we moved into London, which is yet another significant market where vehicles are driven on the right-hand side. Now, turning to the question of making money. In its biggest operating city within China, Apollo Go has already reached UE breakeven, even with fares set at extremely low levels. When we look at scaling up around the world, the way prices are set becomes far more favorable, and we're of the view that our business abroad could generate significantly better profitability once those operations keep growing; additionally, the total international market beyond the U.S. and China is actually larger than China's own domestic market. As for where we ultimately fit in the broader ecosystem over the long haul, I'd say it's too soon to make that determination. The robotaxi sector hasn't settled yet — its value chain and ways of doing business are both still forming. Our current priority is to keep growing, strengthen the technological and operational edge we hold, and stay number one worldwide. With that foundation in place, we'll have the strategic room to shape our role as the ecosystem develops, and by doing so, lock in value for the long haul.
Congrats on the strong results. With AI spending rising throughout the sector, what's the right way to view Baidu's CapEx for 2026? Also, how does management weigh capital allocation between AI investments and returns to shareholders? Lastly, any update on the possibility of a dual primary listing in Hong Kong?
Thanks, Thomas. It's Henry here. Regarding the CapEx point you raised, our guiding principle is to keep up strategic investment intensity while staying financially disciplined. AI is still Baidu's biggest long-term opportunity, and we'll keep funding foundation models so we stay competitive at the frontier, as well as across our broader full-stack AI capabilities. Financially, we're able to sustain that level of investment. Thomas, as you've seen, our operating profit and operating cash flow are still healthy, and our total cash position is healthy too. Operating cash flow for Baidu stayed positive at RMB 2.7 billion in Q1 this quarter. That marks the third quarter in a row since it turned positive in Q3 last year. At the same time, we're tapping a range of financing channels and instruments, such as operating leases, financial leases, and other low-cost bank borrowings, to fund our AI investments while keeping a sound cash position, a healthy balance sheet, and a long-term financing structure. Also, as you noted, shareholder returns remain a priority for us and stay as they are. Last quarter, we rolled out a new buyback program and introduced our first dividend policy. We'll keep balancing long-term AI investment against shareholder returns, and we stay committed to creating sustainable value for our shareholders. On your last point about the Hong Kong dual primary listing, as you said, we always assess initiatives that could help unlock long-term value for the company, including capital market ones; we think Baidu's underlying value is substantial, and we'll be flexible and proactive in exploring ways to surface it. We'll take into account market conditions, regulatory requirements, and shareholder interest, and we'll update the market when there's meaningful development.
Nice work on those impressive numbers. My question concerns the AI agent. With us racing into the agentic AI period, how should we view Baidu's approach to AI applications and agents? What does management think about ways to make money, say token-based pricing, a subscription model, or a project-based one?
Miranda, Robin here. During my talk at the Baidu Create conference one week ago, I noted that across the last 3 years, the largest AI moments came from breakthroughs in models; this year, though, things look different. For the first time, an agent — an AI application — has grabbed global attention. That change confirms a long-held conviction of ours: within the AI era, value is ultimately realized at the application layer. AI applications remain a strategic priority for us at all times. We have assembled a portfolio around user needs and business scenarios, covering AI native products plus AI transformations of our current offerings, serving consumers, companies, and industry verticals alike. Inside that portfolio, a few directions strike us as especially valuable. Number one is search, our biggest consumer-facing product and one we keep transforming through AI. AI Search is where we're setting the global pace, bringing the latest AI features to our user base of hundreds of millions, and enabling search to deliver answers that are smarter, better organized, and useful across a wide range of needs, all at scale. Digital humans comes in at number two. Our efforts keep driving toward elevated quality, reduced expense, and the ability to scale further. With expenses coming down, merchants find digital humans much easier to adopt, and they can be deployed across a broader footprint. When it comes to live streaming for e-commerce, these digital humans are showing growing effectiveness at boosting engagement and conversion, with results that in numerous instances match or surpass what a human host achieves. Moving on to item three: Miaoda, the product we offer for vibe coding. With ongoing improvements to what it can do, Miaoda now handles a wider variety of use cases and workflows. This makes it easier for both individual users and companies to create AI applications simply by using everyday language. Item four covers Famou agents, which target business use cases. These agents assist organizations in working through intricate, ever-changing conditions out in the real world, and they keep advancing so they can find superior approaches and deliver significant boosts in efficiency. On top of that, as these agents take on scenarios of growing complexity spanning an expanding set of sectors, their performance continues to improve. When you look at these paths, the first one obviously leans much more heavily on advertising, while the remaining three rely more on subscriptions, usage, or token-based pricing. As for search, there's fairly broad agreement that it's a substantial market for AI native applications. And when it comes to Miaoda, the vibe coding product, you can find comparable offerings outside of China too. But digital humans and the Famou agent aren't areas where there's consensus — those reflect our own conviction. We believe both hold enormous market potential, yet not many companies are devoting sufficient resources and effort to this kind of work. On top of that, we're also exploring new forms of AI native products. DuMate serves as one illustration of this trend. It's an agent built for general purposes. It takes AI into the everyday workloads and tasks that individuals handle; it is < >; it retains a great deal of information about you; therefore we expect it will be a fairly sticky application. Regarding monetization, the sector remains at a quite early phase worldwide, and the ways of making money are still developing. At present, pricing based on tokens is the more typical approach. What people are basically doing is paying for foundation models; however as time goes on, AI applications and agents are going to get better at finishing real cash work the way a human being would. We believe money-making will broaden out and move in the direction of outcomes. In practice, that translates to people eventually shelling out for gains in productivity, hours reclaimed, superior experiences, and concrete results. So in the time ahead, folks will buy agents or apps, and that market should end up substantially larger than token.
Nice work on the strong numbers. I'd like to ask whether the management team could offer its perspective regarding how growth may trend and how rivalry is shaping up among makers of AI chips made at home in China? In that market, then, where does Kunlunxin sit? And what shifts in demand have we actually been observing lately?
Thanks for asking; Dou here. In my view, the market for AI chips developed domestically in China remains at an early stage; even so, it is advancing quickly. A structural change is underway in demand for AI compute, moving from < > toward a mix in which inference makes up a rising share. With agentic applications, use cases tailored to specific industries and fresh categories of applications appearing one after another, inference accounts for an expanding slice of the overall picture. The open client illustrates this well, fueling a surge in inference demand marked by greater frequency, more immediacy and more endeavors. Turning to supply, near-term hurdles persist for homegrown AI chips, tied to capacity as well as how mature the supply chain is. < > demand is climbing more rapidly than supply. Yet looking further out, China's semiconductor sector is progressing swiftly, buoyed by a robust base in manufacturing and supply chains; our expectation is that domestic supply capabilities will continue to strengthen as time goes on. Should that unfold, competition will hinge increasingly not merely on shipping chips, but on whether those chips can run dependably and efficiently across endeavors real-world workloads. In certain frontier training scenarios, domestic chips < > are still trailing the most advanced products worldwide. Inference represents a space in which homegrown chips can be highly pertinent and able to compete. As inference keeps expanding, that edge grows more pertinent; from an enterprise standpoint, peak chip performance alone is not the whole story. Of greater importance are stability at scale, compatibility with mainstream models and frameworks migration cost and friction support for our large-scale cluster deployment and, in the end, cost efficiency; thus, in our view, the market will progressively concentrate around those players capable of meeting all of these dimensions. On every one of these brands, Kunlunxin holds a strong position. What's more, Kunlunxin is far from being merely a stand-alone chip product. It forms an essential piece of Baidu's full-stack AI capabilities. Extending from infrastructure through to applications, we are able to keep optimizing across the intersect, boosting model efficiency, lowering inference costs and providing AI infrastructure that is more cost effective, stable and simpler to deploy; customer demand for Kunlunxin is strong and rising, with adoption widening throughout industries. So our belief is that Kunlunxin is well positioned to seize the opportunities that lie ahead.
We don't have any more questions at this point, so that brings today's conference to an end. We appreciate everyone for joining in. You're welcome to disconnect at this time.
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.