Baidu, Inc. (BIDU) Earnings Call Transcript
August 18, 2026
Earnings Call Speaker Segments
Hello, everyone, and thank you for being on the line for Baidu's call about our second-quarter 2026 results. Baidu's earnings release came out earlier today, and you can find a copy on our website as well as through Newswire services. On the call with us 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 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. Under the Private Securities Litigation Reform Act of 1995, these forward-looking statements come with risks and uncertainties that might cause actual results to differ materially from our current expectations. For a detailed discussion of those risks and uncertainties, please see our latest annual report and other filings with the SEC and the Hong Kong Stock Exchange. Except where applicable law requires it, Baidu assumes no obligation to update any forward-looking statements. In our earnings press release and on this call, we discuss certain unaudited non-GAAP financial measures. A reconciliation of the unaudited non-GAAP measures to the unaudited most directly comparable GAAP measures appears in our press release, which is available on our IR website at ir.baidu.com. Just a note before we start — this call is being captured on tape. Also, you'll be able to catch a webcast of today's discussion over on Baidu's IR site. At this point, I'll hand things off to Robin, our CEO.
Hi, everyone. During the second quarter, total revenue from Baidu General Business came in at RMB 25.2 billion, and Baidu Core AI-powered Business kept accounting for 50% of that figure, which underscores how central AI remains to what we do. AI Cloud Infra posted yet another quarter of robust expansion, with overall revenue up 50% compared with a year earlier, once more growing faster than the wider market. Inside AI Cloud Infra, GPU Cloud revenue came close to quadrupling versus the prior-year period, rising 283% and speeding up considerably from the already impressive 184% growth recorded during the previous quarter. Now that AI-powered business sits at the heart of how our revenue is composed, our attention is on laying a sturdier groundwork for its coming growth phase throughout our entire AI stack, spanning chips and cloud infrastructure through to models and applications. We keep bolstering the capabilities needed to sustain ongoing innovation, fuel our future expansion and shore up our long-term competitiveness. Let me move on now to the main business highlights for this quarter, beginning with our in-house AI chips, Kunlunxin. During the second quarter, Kunlunxin carried on showing solid business momentum, with demand staying strong and widening across sectors. An expanding set of customers is taking up its chips for a broadening array of AI workloads, which shows growing market appreciation for Kunlunxin's stability, efficiency and versatility at scale. Kunlunxin kept reinforcing its software ecosystem, expanding compatibility with top models and frameworks while making deployment simpler across enterprise settings. Building on the support it offers for ERNIE and other leading foundation models in China, Kunlunxin widened its coverage further in the second quarter to take in newer releases of major Chinese foundation models, including Kimi K3, GLM5.2, Minimax M3 and Hunyuan 3. It additionally boosted inference throughput and overall compute efficiency, sharpening its capacity to handle varied, demanding AI workloads at scale. Across more than 10 years, Kunlunxin has managed to develop and commercialize 3 generations of AI chips. With that track record behind it, it kept pushing forward a well-defined product road map, covering the newest M100 tuned for large-scale inference plus the forthcoming M300s. That road map shows how deeply Kunlunxin grasps evolving AI technology, workloads and the compute they demand, putting it in position to enable the next wave of AI innovation. As we keep enhancing our AI infrastructure capabilities, we expect Kunlunxin to take on a more and more significant role inside our full stack AI architecture and to sharpen our capacity to provide high-performance, dependable and cost-efficient AI computing at scale. With demand for AI computing in China still climbing, we expect our in-house AI chips and full stack capabilities to grow ever more valuable. This backs the future expansion of our AI businesses and strengthens our long-term competitiveness in AI. Leveraging our strength at the infrastructure layer, AI Cloud Infra turned in another quarter of robust growth. In the second quarter, AI Cloud Infra revenue rose 50% versus a year earlier, continuing to grow faster than the wider industry. A number of factors came together to fuel this sustained growth momentum. To begin with, AI Cloud Infra kept gaining from robust demand for AI computing. On the demand side, things stayed strong for both training and inference workloads, even as compute availability continues to be tight throughout the market. Next, our established major customers — among them top players in online gaming, e-commerce, and lifestyle content — keep raising what they use and what they spend with us. At the same time, our total client base expanded quickly, with new customers coming from businesses of many different sizes. Third, demand stayed wide-ranging across verticals such as Internet, embedded AI, autonomous driving, smartphones, financial services, and others. Looking at that mix, Internet and autonomous driving maintained solid growth, and embedded AI revenue was up roughly sixfold from a year earlier in Q2. Given these patterns, we expect AI Cloud Infra revenue growth to stay strong during the second half, with room for even faster expansion. Notably, growth in AI Cloud Infra came alongside swift profit gains and widening margins compared with a year ago, which shows the business keeps getting healthier and higher in quality. Inside AI Cloud Infra, GPU Cloud revenue growth sped up dramatically, reaching 283% versus a year ago, on top of an already elevated 184% base from the prior quarter. That momentum mirrors solid underlying appetite for scalable AI compute in the public cloud. Our business mix kept moving toward higher-quality revenue streams, with GPU Cloud making up a rising portion of AI Cloud Infra revenue. Because its margin profile is more appealing, this shift is helping create a healthier revenue mix and boosting the long-term profitability of our cloud business; on MaaS, our Qianfan MaaS platform provides one of the most extensive model libraries, spanning Baidu's ERNIE family plus practically every leading model in China. A core goal for Qianfan is making model inference at scale more dependable and cost effective for customers. Drawing on our deep know-how in AI infrastructure and engineering, we further improved model serving via ongoing inference optimization, which delivered higher throughput and better service stability while cutting latency and inference costs. During Q2, revenue from external customers' token usage on Qianfan rose more than ninefold compared with a year earlier, mainly fueled by fast growth in daily average token consumption among those customers. Now on to foundation models. Pushing ERNIE forward and improving our overall model capabilities stays essential to our next stage of AI-driven growth. Our dedication to foundation model innovation has not wavered. As we mentioned in earlier quarters, we restructured our model teams into two groups, giving them clearer mandates and sharper focus across foundation models and applications. More recently, we brought in a new wave of leading AI talent to work on foundation models, which further shows our resolve to compete and innovate at the cutting edge of AI; we think these steps will help ERNIE keep evolving and reinforce the groundwork for future innovation across both models and AI applications. Now let's turn to AI applications, where we keep improving product capabilities and broadening real-world use cases. I'll start with digital humans. With digital human tech still evolving, we're seeing better results at reduced expense, which opens up more and more applications—everything from live streaming in e-commerce and digital human clips to real-time interactive digital humans, plus the video podcast we just rolled out. Thanks to these strides, there's a much wider scope for deploying digital humans throughout various sectors. Clients are increasingly acknowledging what our digital human solutions can do. During Q2, we kept bringing on new customers, among them top firms in a variety of industries, and those already with us significantly increased how much they used our offerings. Certain clients began with a trial, and once they witnessed the outcomes our digital human technology produced, they ramped up their adoption. Take a prominent Chinese Internet company: following merely a single quarter of usage, it scaled its digital human live streaming rollout to roughly 2.5x what it had been before. At the same time, we're pushing forward with taking our digital human offerings worldwide. Ever since we introduced our overseas digital human platform in the prior quarter, momentum has been encouraging, driven by its distinctive features, and it's yielding strong outcomes for merchants and creators abroad. With demand continuing to grow across additional industries and geographies, we think digital humans hold considerable potential for growth over the long term. Now shifting to Miaoda, our vibe coding platform. Following the release of Miaoda 3.0 in the last quarter, users are able to create stand-alone mobile apps for Android as well as iOS through natural language.
Tasks that once demanded a whole professional development crew, a lengthy construction schedule, and substantial funding are now far easier to accomplish using Miaoda — you can even handle them right from a phone. Engagement with Miaoda is getting deeper. A growing number of users do more than try it once; they return to Miaoda to continue building, refining, and polishing their apps over time, which signals stronger stickiness. Miaoda's monthly active users in June rose 67% compared with March. Adoption is also spreading across industries — including tech and education, health care, manufacturing, financial services, and logistics — demonstrating that Miaoda functions in many different business scenarios and holds broader commercialization potential. AI is additionally being leveraged by us to assist businesses in resolving intricate operational challenges. A notable illustration is Famou Agent, capable of autonomously investigating possible approaches in order to identify optimal ones. Following the release of Famou Agent 2.0 during the prior quarter, we continued efforts toward enhancing its ease of use as well as broadening the range of scenarios it can address. Growing interest from major enterprises has been attracted by Famou Agent, and initial commercial uptake began emerging this quarter. It is wonderful to observe Famou Agent moving past productivity improvements to assist organizations in refining operations and producing genuine, concrete business value. We believe its potential will continue expanding as its abilities further develop. General-purpose agents represent another important area of focus for our AI applications. At the start of this year, DuMate came out — it's our all-purpose agent built to help people get things done day to day, working smoothly on both computers and phones. Then during Q2, we rolled out a version aimed at enterprises, and we kept building out DuMate's in-house Baidu skills along with its purpose-built toolkits, which let it handle a wider set of tasks and more complex ones too. On top of that, the AI apps we offer consumers as flagships — Baidu Wenku and Baidu Drive — are still going all-in on AI, adding fresh AI features while refining the ones they already had. And in this quarter, we shipped an upgrade to GenFlow that weaves AI further into the routines users follow every day. Come June, the share of daily active users engaging with AI on Baidu Wenku and Baidu Drive rose 27.4% compared with the same month a year earlier, a sign that more people are picking up their AI-driven capabilities. Now, looking at the AI search side of things, we keep making gains on two fronts: how good the answers our AI produces actually are, and how the whole experience feels for the person using it. More and more, what people get back is dependable, cleanly organized, and shown in a way that works well. On top of that, hallucinations stayed at a low level, and our models got better at judging whether content measures up, which cut down on the number of weak answers showing up. Those gains together meant a smoother experience and happier users. We also went further in connecting AI search with ERNIE Assistant, so searching no longer stops at a single answer — it turns into a back-and-forth exchange that flows naturally and handles follow-up questions and other needs users bring up. During June, ERNIE Assistant saw its daily active users rise 83% compared to the same month last year, and the average number of conversation rounds each day increased by more than threefold—signs that more people are adopting the product and engaging more deeply with this evolving search offering. Now shifting to AI in the physical world, I'd like to talk about Apollo Go, which is our autonomous ride-hailing business. Over this quarter, we continued advancing its expansion across the globe while also achieving further improvements in safety, operational performance, and the rider experience. Within this quarter, Hong Kong marked an important milestone for Apollo Go. We secured Hong Kong's first-ever permits allowing fully driverless testing in June, and testing began on Airport Island in July. That achievement meant Apollo Go became the first autonomous ride-hailing service provider anywhere in the world to run fully driverless testing in a robotaxi market that is right-hand drive with left-hand traffic. When it comes to urban mobility, Hong Kong ranks among the most sophisticated markets globally, with an operating environment that is complex and standards that are rigorous for technology as well as operations. Hitting this milestone in Hong Kong strongly validates how mature and adaptable our technology and our operational capabilities have become. What we have learned in Hong Kong is already enabling us to move ahead more efficiently in London. There, in July, Apollo Go started open road testing alongside Uber and Lyft as partners. Taken together, what we have achieved in these two markets shows that our technology can generalize across differing operating environments, which gives us more confidence that over time we can expand into a growing number of high-value robotaxi markets that are right-hand drive with left-hand traffic. Progress also came in a number of other international markets. In Dubai, we kicked off fully driverless commercial operations in July, and we now run at the largest scale of any fully driverless autonomous ride-hailing service in that city, with rides bookable via both the Apollo Go app and the Uber app; in Switzerland, we started open-road testing with PostBus as our partner. On top of that, we reached a memorandum of understanding with Kazakhstan's Turlov Private Holding Limited, aimed at jointly investigating autonomous ride-hailing offerings within that nation. Taking everything into account, Apollo Go racked up close to 1 million operational rides with no driver at the wheel throughout Q2. As of June 2026, the total number of rides Apollo Go had provided to members of the public exceeded 23 million. A number of domestic cities saw operational tweaks brought about by regulatory factors, which put a temporary dent in ride volume over the quarter. Throughout this stretch, we conducted a methodical examination intended to further bolster the robustness of our autonomous driving systems along with the rigor of our operational procedures. Once August arrived, service in the impacted cities had begun coming back online, and from a sturdier position. Meanwhile, our footprint keeps growing across additional U.S. cities, and we believe trip counts will pick up speed again in upcoming quarters as we gradually scale up and keep pushing into new areas. During Q2, we also kept pushing standards higher when it comes to safety as well as how riders feel about using the service. As of the end of June, our vehicles operating with no one behind the wheel had logged roughly one airbag deployment for every 14.4 million kilometers on average — a figure that highlights safety results better than anyone else in the industry. On top of that, we made our suggested pickup and drop-off locations smarter, cutting down how far riders have to walk and steering clear of spots that aren't safe to stop. We also kept sharpening how our systems perceive the world and plan movement, so trips come out smoother and more reliable. Those gains raise the bar for how we operate even higher, and we plan to keep pushing past it as Apollo Go becomes more deeply woven into city transit networks, turning it into a handier and more dependable piece of daily getting around. As we turn our attention to the back half, the goals we've set for Apollo Go are straightforward: keep raising the bar on safety and how well we operate, push further into markets worldwide, grow both the size of our fleet and the number of rides taken, and move additional cities to the point where unit economics break even. We're confident that making headway on each of these fronts will only cement Apollo Go's position at the forefront of autonomous ride hailing and build a sturdier base for expanding those operations in a way that's both safe and sustainable over the long haul. In sum, what we achieved throughout our entire AI stack during this quarter confirms once more that Baidu has become an AI-first company, and it has reinforced the groundwork for the growth chapter ahead of us. Beyond that, we're pushing our AI businesses into markets around the world, and the early traction we're witnessing makes us optimistic — in areas like AI applications as well as robotaxi. Armed with this firmer footing, we're confident we're positioned well to seize a wider set of opportunities spanning markets as time goes on. Having said that, I'll hand things over to Henry so he can walk everyone through the financial results.
Thank you, Robin, and greetings to all of you. We felt good about the ongoing traction achieved by Baidu Core AI-powered Business during this quarter. The top line from Baidu Core AI-powered Business came in at RMB 12.5 billion, and it kept making up half of Baidu General Business revenue. AI Cloud Infra revenue rose roughly 50% compared with a year earlier, while GPU Cloud revenue climbed 283% versus the prior-year period in Q2. These results confirm that AI-powered business serves as a major engine behind Baidu's growth over the long term. Beyond that, we moved ahead with a significant capital market effort aimed at widening our access to capital markets and building long-term shareholder value. Back in July, our Board passed a motion to pursue the voluntary conversion of our Hong Kong listing to dual primary status. Since that time, we filed our application and got acknowledgment from the Hong Kong Stock Exchange. We intend as well to hold an Extraordinary General Meeting on August 26 to request shareholder approval for certain related matters. At present, we anticipate the conversion taking effect before this year ends, assuming shareholders and the Hong Kong Stock Exchange give their approval. After it takes effect, we think the dual primary listing will widen our investor base, improve the liquidity of our shares and deliver more flexibility when it comes to accessing capital in both markets. Looking ahead, AI stays at the heart of Baidu's long-term growth and competitiveness. We'll keep investing with conviction and discipline in the capabilities and businesses where we identify the strongest long-term opportunities. Now allow me to take you through the specifics of our second quarter 2026 financial results. Baidu's total revenue stood at RMB 31.3 billion, down 2% quarter-over-quarter and down 4% year-over-year. Baidu General Business revenue was RMB 25.2 billion, down 3% quarter-over-quarter and down 4% year-over-year. iQIYI revenue was RMB 6.3 billion, up 1% quarter-over-quarter and down 5% year-over-year. Cost of revenues came to RMB 19.1 billion, down 3% quarter-over-quarter, mainly because of lower costs tied to the AI Cloud business, partly offset by higher traffic acquisition costs, and up 4% year-over-year, mainly because of higher costs tied to the AI Cloud business. Operating expenses were RMB 9.2 billion, down 1% quarter-over-quarter and down 17% year-over-year. That year-over-year drop stemmed mainly from lower expected credit losses, channel spending expenses and R&D personnel-related expenses. Operating income was RMB 3.0 billion and operating margin was 10%. Non-GAAP operating income was RMB 3.8 billion and non-GAAP operating margin was 12%. Total other income net was RMB 184 million, versus RMB 626 million the previous quarter and RMB 4.9 billion in the same period last year. That year-over-year decline came mainly from a smaller fair value gain on long-term investments and a larger net foreign exchange loss caused by exchange rate fluctuation between Renminbi and U.S. dollar. Income tax expense was RMB 1.0 billion, versus RMB 528 million the previous quarter and RMB 881 million in the same period last year. Net income attributable to Baidu was RMB 2.3 billion. Net margin for Baidu was 7% and diluted earnings per ADS was RMB 5.74. On a non-GAAP basis, net income attributable to Baidu came to RMB 2.6 billion. Baidu's non-GAAP net margin stood at 8%, while non-GAAP diluted earnings per ADS reached RMB 7.22. Our definition of total cash and investments covers cash, cash equivalents, restricted cash, short-term investments, net long-term time deposits, held-to-maturity investments, and adjusted long-term investments. Total cash and investments amounted to RMB 283.1 billion as of June 30, 2026. Cash flow from operations was RMB 3.4 billion. As of June 30, 2026, Baidu General Business employed roughly 27,000 people. With that, operator, please open the call for questions.
With multitrillion parameter models now appearing so fast and setting new records on benchmarks, where does Baidu see ERNIE's competitive position going forward? Now that a senior foundation model talent has recently joined, which technical and product goals matter most for ERNIE? And what should investors anticipate from the next phase of its development?
Robin here. To begin with an industry view: foundation models keep advancing quickly—about every few months, some different model moves ahead in a given capability. That tells us the field is still very dynamic, and who wins is anything but decided. In a market of this kind, we think lasting competitiveness boils down in the end to continuous tech investment, an approach driven by applications, and patience. Baidu has long been a firm that puts its faith in technology and is ready to back it over the long haul. What we have lived through again and again proves that real technological innovation demands persistence and patience. Right now, a number of Baidu's significant AI assets—Kunlunxin and Apollo Go among them—exist because we kept investing for over a decade. They have turned into a major wellspring of our differentiated competitiveness. And the results they deliver plus their commercial value are winning ever broader recognition. That makes us extremely proud. In the same way, ERNIE has consistently formed a key piece of Baidu's AI strategy and its full stack AI capabilities. Among companies in China, we were early to put money into foundation models. Mistakes happened along the way, yet our resolve to keep ERNIE competitive has never wavered. Moving forward, we will keep supplying whatever resources are required to fuel ERNIE's continued progress. Toward that end, we have further refined our organization and lately recruited top AI talent. We feel sure we can speed up AI's iteration and return ERNIE to the top tier of foundation models. Going forward, an application-driven approach will remain ours. The capabilities foundation models cover are extremely wide, and at any moment no single model can be best across every dimension. So we will concentrate on the capabilities that count most for Baidu's applications and make ERNIE the strongest in those. That stretches across AI search, digital human, Miaoda, Famou and general-purpose agents such as DuMate. Such applications are essential to ERNIE's steady enhancement. Take AI search: when we sharpen ERNIE's capacity to grasp user intent and judge content quality, those gains go straight into search and feed. That way we spot outcomes fast, see what still needs fixing, and feed the relevant data back into model training, which sharpens our model at understanding user intent and assessing content quality. This loop, in our view, is a crucial route for ERNIE's growth—one that turns technological progress more directly into stronger product experiences and genuine user and commercial value, and in the end benefits a wider set of users and businesses.
I'd like to ask about cloud. Baidu AI Cloud Infra's top line has kept up impressive expansion. Could the management team walk us through what's driving that growth, plus your expectations for how revenue will trend across the coming quarters? And beyond that, how should we frame the margin opportunity over the long run as the business scales up?
Alicia, it's Dou here. For Q2, AI Cloud Infra revenue was up 50% versus the same period a year earlier. That's still a solid pace, and in my view it sits ahead of where the industry is growing. So if you look further out, our AI Cloud Infra has now posted fast growth across many quarters in a row, and we've kept beating the industry's pace throughout. The highlight this quarter was GPU Cloud, where revenue climbed 283% compared with a year ago — that's the fourth consecutive quarter of triple-digit expansion, and it's speeding up even more, from 184% back in Q1. Going forward, multiple drivers underpin our expectation of continued growth. Right now, demand for AI compute within China is extremely robust. And as AI gets woven more deeply into practical applications and business workflows — especially with inference scaling up quickly — we anticipate that demand will keep rising. At the same time, our customer roster is growing fast as well, with new clients of varying sizes coming on board for our AI Cloud Infra, and our existing major customers continuing to raise both their usage and their spend. On top of that, demand is widening across sectors and scenarios — internet, gaming, embodied AI, autonomous driving, smartphones, financial services, and more. Actually, and this matters most, we've established and keep reinforcing a differentiated full stack AI architecture that reaches across chips, cloud infrastructure, models, and applications, with competitive products at each layer. Especially at the application layer, we acted early to assemble a set of agents and AI applications, and offerings like Famou Agent, DuMate, Miaoda, and Baidu Yijing are picking up traction, which bolsters our capacity to seize an ever more varied set of AI opportunities. So given today's demand patterns, our customer pipelines, and these differentiated strengths, we're confident AI Cloud Infra will sustain strong growth through the second half, with room to accelerate even further. On the profitability front that you brought up, we're happy to see ongoing gains alongside the fast revenue growth. AI Cloud Infra profit and margins each rose year-over-year during Q2. Looking forward, we believe a number of factors ought to drive additional margin expansion. First, GPU Cloud is expanding much more quickly than AI Cloud Infra broadly, and it keeps accounting for a big portion of the mix. Its margin profile is also more appealing than that of traditional CPU Cloud, and there's additional room to improve margins as it scales, helped by ongoing optimization of product and customer mix, improved resource utilization, and stronger operating efficiency. So with GPU Cloud's contribution rising, that mix shift ought to keep pushing overall margins upward. Second, turning to MaaS: revenue from external customers' token calls on Qianfan is rising extremely quickly. MaaS is still a fairly small portion of our AI Cloud Infra revenue at present, yet the early momentum looks quite promising. As usage skills and per-unit inference costs continue to decline, we think that over the longer run, MaaS-related businesses can unlock greater profit potential and grow into a more and more significant margin contributor. Lastly, our full stack AI capabilities plus self-developed chips give us end-to-end cost advantages that ought to aid margin expansion as well. So all things considered, we believe there remains substantial room for AI Cloud Infra margins to rise over the long term. Thank you.
I'd like to ask about margins. AI-driven operations are now bringing in 50% of revenues, and CapEx keeps climbing, so how should we view the path of Baidu's operating margin? And in what way will leadership reconcile ongoing AI spending with staying profitable?
Thanks, Miranda. Henry here. During this quarter, AI-powered operations once again represented 50% of Baidu General Business revenue, which reinforces how central AI has become to what we do. Inside that AI-powered business, AI Cloud Infrastructure kept up swift top-line expansion, with profits rising at a fast clip too and margins up year-over-year. Within AI Cloud Infra, our GPU Cloud operation, which generally comes with a stronger margin profile, kept growing its share of revenue. With this positive mix shift persisting, alongside robust market demand and the cost edge derived from our in-house chips and full stack AI capabilities, we feel there remains substantial room for AI Cloud Infrastructure margins to widen over the long run. As operations grow in scale, we likewise anticipate improved resource utilization and heightened operational leverage to lend additional support to that margin expansion. We also perceive appealing long-term profitability prospects in our AI applications. A large number of these applications are sticky and subscription-based in nature, with the capacity to produce progressively appealing margins over time as they scale. As adoption grows and monetization advances, we expect them to turn into a more significant contributor to overall profitability. At the same time, I believe we remain within an AI investment cycle, and our dedication to that investment does not waver. We invest with conviction, yet equally important, we spend wisely and keep a close eye on the ROI. Our investments stem from clear demand among both customers and our internal business, enabling much of where we will put money to work quickly and start contributing to revenue fairly soon. At the same time, we keep bolstering our supply chain management capabilities, which we believe will increasingly assist us in enhancing capital efficiency as we scale. Taken together, these strengths grant us solid visibility into returns, and we are confident in our capacity to boost investment efficiency over time. That said, various AI investments unfold across differing time lines, and fully delivering their value will take us longer. Some of them will require more time. We are presently in a critical investment phase, and we plan to continue investing decisively in the areas that matter most to our long-term competitive position while upholding the same discipline around ROIC, operating efficiency and cash flow. As our AI business scales further and monetization matures, we believe these investments will increasingly convert into some and more sustainable profit growth. Thank you...
Can you bring us up to date on where things stand with the proposed listing of Kunlunxin, plus the important milestones that lie ahead? What do you expect to fuel its growth going forward? Also, how does the management team see its commercial potential over the long term and its strategic part in the Baidu AI ecosystem?
All right. I'll handle that one; Dou speaking. Kunlunxin's listing is a process that remains underway, and once there is more we can share, we'll bring the market up to date. Looking at the business side, our confidence stays strong regarding Kunlunxin's growth over the long term and its commercial potential, and there are a few reasons why. First, throughout the industry, the need for AI compute keeps expanding, in training as well as inference. As model abilities continue to get better and additional applications shift into practical deployment, particularly with agents getting more advanced and reaching a broader set of use cases. So inference, in particular, is what we see accelerating. So our view is that this trend will persist, which opens a structural growth opportunity over the long term for the AI chip sector. Secondly, there is considerable room to grow in the domestic market, while supply is probably going to stay tight for a period of time. So given that situation, customers increasingly want domestic AI chips that are high-performing, dependable and cost-efficient. So our view is that this generates major opportunities for chip suppliers that have robust technical strengths and can deliver at scale. After investing for more than 10 years, Kunlunxin has established dependable strengths in chip performance, hardware-software integration, compatibility with mainstream models and frameworks, and large-scale deployment, which has earned it growing recognition among customers; those factors place Kunlunxin well in this market and let it seize the commercial opportunities emerging from China's expanding AI computing requirements. Thirdly, inside Baidu's AI ecosystem, Kunlunxin plays a key role in the infrastructure layer of our full stack AI architecture, covering chips, cloud infrastructure, models and applications. Tight coordination among these layers makes end-to-end optimization possible, which lets us provide better performance, reliability and cost efficiency; so this underpins the long-term deployment of AI Cloud Infra and our other AI businesses, while making Baidu's full stack AI capabilities even more competitive. Going forward, we anticipate that Kunlunxin will continue to hold a meaningful place in our AI infrastructure, seizing a wider set of commercial opportunities and meeting a broader set of market needs. Thank you.
Can you take us through the anticipated schedule for the Hong Kong dual primary listing conversion and possible Stock Connect inclusion? In addition, what is the strategic reasoning behind it? And in what ways might it influence Baidu's investor base, share liquidity, and valuation as time goes on?
Thanks; Henry here. I'll begin with the schedule. The conversion to a dual primary listing was approved by our Board in July. We also submitted our application to the Hong Kong Stock Exchange and got its acknowledgment. Then comes our extraordinary general meeting, set for August 26. At that meeting, we will request shareholder approval for certain matters needed to prepare for conversion. After that, provided we receive clearance from the Hong Kong Stock Exchange and satisfy the other relevant requirements, we expect the conversion to take effect sometime in the current year. On the Southbound Stock Connect front, we are also actively preparing for the possibility of being added once that conversion is complete, and we hope our shares will be admitted as soon as they can be. Of course, whether that happens still hinges on the eligibility criteria that apply, the review of relevant procedures, and whatever the exchange in question decides. Turning to the rationale: the point of a dual primary listing is essentially to expand the pool of investors we can reach, improve how easily our shares trade, and give us more room to maneuver when tapping capital in Hong Kong as well as the U.S. Beyond that, it gives a wider set of investors—especially those based in Asia—a clearer picture of Baidu and an easier way to take part in what an AI-first company is worth. Further down the road, should we secure Stock Connect inclusion, our expectation is that this would notably broaden involvement from investors located in Mainland China in particular, which in turn ought to help build an increasingly varied shareholder base as time goes on. As we move further along, we'll gladly share updates with all of you.
Can the management team bring us up to date on where things stand with AI search — the product features, how users are experiencing it, and efforts to make money from it? In Q2, we noticed online marketing revenue was still under strain. What drove that mainly? Also, how does management anticipate the business will perform in the second half?
Thomas, Julius here. I'll handle your question. Over the last several quarters, I'd say, our attention to the AI search transformations has been lifting the quality of our AI answers by greatly enhancing how users experience them. Being accurate and authoritative has always been our fundamental advantage, and we've continued to strengthen that as the AI transformation progresses. These days our AI search understands better what people are seeking. Answers come across as more dependable, organized more effectively, and shown in formats that are richer. At the same time, rates of hallucination stay low, and our models are now improving at distinguishing good content from bad. So we're encountering more high-quality answers and fewer weak ones. Users have reacted rather positively to these changes, and we're witnessing consistent gains in satisfaction among users, their readiness to search, and retention. During this quarter, we went further in combining AI search with the ERNIE assistant, converting one of the search answers into multi-round conversations that are more coherent and interactive, which handle follow-up questions and wider user needs better. We're also persisting in bolstering the to-use multistep planning and executions of complex tasks to assist users in accomplishing more. Not long ago, the ERNIE Assistant's Task Agent came out on top in two prominent third-party agent benchmarks: the Pinch Bench V2, a worldwide benchmark centered on real-world complex tax complexion, and the SuperCLUE XCloud evaluation covering leading domestic agent products. I believe these outcomes can help cement the ERNIE Assistant's foremost capabilities in to-use multistep planning and task executions. Even so, competition within this industry stays extremely fierce, and as new product forms such as AI chatbots keep gaining traction, the methods by which users find and consume information keep changing, and the battle for users' time and attention has grown even more intense. At the same time, we've kept advancing the AI search transformations while intentionally refraining from monetizing the AI search, and both of those have weighed on our advertising businesses in the near term. Since these dynamics will probably continue, we anticipate our advertising business will stay under pressure in the second half. When it comes to monetization, what matters most to us at present is still getting the products and the user experience right. As model capabilities, the user experiences, and the task complexions keep getting better, we think additional monetization opportunities that fit naturally into the AI experiences will arise down the road. Thank you, Thomas.
I'd like to ask about robotaxi, if I may. With China having recently rolled out new robotaxi regulations, what's management's take on how that regulatory landscape is developing? How should we frame Apollo Go's relative priorities, plus how fast it's expanding both at home and abroad? And it would be helpful if management could address what Apollo Go has accomplished on the overseas commercialization front.
Robin here, and I'll take that. Things are moving very fast across the worldwide robotaxi sector. Previously, the question everyone cared about was whether robotaxis could provide rides that were safe and comfortable. These days, the question has grown into whether they can run dependably at scale and integrate with the wider transportation ecosystem; along with that shift, regulators in major markets worldwide are revising and honing their rules for robotaxis. Take China: a first-ever mandatory national standard covering safety requirements for Level 3 and Level 4 automated driving systems was just released, and for the L4 provisions in that standard, Apollo Go supplied a wealth of technical and operational know-how. Our number one priority has always been safety, and around the world we hold a safety record that leads the industry. We'll keep holding ourselves to demanding standards on safety and operations. More broadly, regulatory frameworks that are clearer and more systematic will lift operational standards industry-wide, foster public confidence and create a sturdier base for robotaxis to grow in an orderly way over the long term. Given all this, we stay bullish about Apollo Go's expansion globally. Domestic and international markets aren't something we treat as an either/or decision. We're very open and adaptable; we evaluate every city according to its regulatory framework, mobility demand, right pricing, road conditions, and commercial viability, then set how fast we enter and expand there. With proven technology and operational experience behind us, we're prepared to move fast and scale efficiently in any city where regulations and market conditions permit. Wherever we go, across any national border, our aim is to go deep and establish a solid presence. You can see this in what we've achieved city by city; in Dubai, Apollo Go has launched fully driverless commercial operations and keeps scaling up, and among robotaxi service providers there we now run the biggest operation. In London, together with partners such as Uber and Lyft, we're pushing forward testing and development. In Hong Kong, we were the first robotaxi service anywhere in the world to run fully driverless testing in a right-hand drive, left-hand traffic market; in Shenzhen, rides are climbing very fast, which makes it one of our biggest markets. As the fleet grows and our operating model matures, we expect vehicle and operating costs to keep declining, while scale delivers further efficiency gains. Previously, Apollo Go reached unit economics breakeven in a market where taxi fares are relatively low. Down the road, in markets abroad where fares run higher, our inexpensive cars plus an operating approach that's already been validated could produce unit economics that are even more robust. Beyond the U.S. and China, the worldwide market is bigger as well than China's home market. That makes the opportunity we can go after fairly sizeable. Going forward, with the strengths we hold in technology, cost and operations backing us, we're confident about moving additional cities to breakeven on unit economics. Thank you.
Everyone, that brings today's call to a close. We appreciate you joining us, and 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.