Baidu Stock

Hello and thank you for standing by for Baidu's Second Quarter 2026 Earnings Conference Call. [Operator Instructions] Today's conference is being recorded.

[Operator Instructions] I would now like to turn the meeting over to your host for today's conference, Juan Lin, Baidu's Director of Investor Relations.

Hello, everyone, and welcome to Baidu's Second Quarter 2026 Earnings Conference Call. Baidu's earnings release was distributed earlier today, and you can find a copy on our website as well as on Newswire services.

On the call today, we have 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 He, our CFO. After our prepared remarks, we will hold a Q&A session.

Please note that the discussion today will contain forward-looking statements made under the safe harbor provisions of the U.S. Private Securities Litigation Reform Act of 1995. Forward-looking statements are subject to risks and uncertainties that may cause actual results to differ materially from our current expectations. For detailed discussions of these risks and uncertainties, please refer to our latest annual report and other filings with the SEC and the Hong Kong Stock Exchange. Baidu does not undertake any obligation to update any forward-looking statements, except as required under applicable law. Our earnings press release and this call include discussions of certain unaudited non-GAAP financial measures. Our press release contains a reconciliation of the unaudited non-GAAP measures to the unaudited most directly comparable GAAP measures and is available on our IR website at ir.baidu.com. As a reminder, this conference is being recorded. In addition, a webcast of this conference call will be available on Baidu's IR website.

I will now turn the call over to our CEO, Robin.

Hello, everyone. In Q2, Baidu General Business generated total revenue of RMB 25.2 billion, with Baidu Core AI-powered Business continuing to represent half of the total, reinforcing AI's position at the core of our business. AI Cloud Infra delivered another quarter of strong growth with overall revenue increasing 50% year-over-year, once again outpacing the broader market. Within AI Cloud Infra, GPU Cloud revenue nearly quadrupled year-over-year, growing 283% and accelerating significantly from an already strong 184% growth rate last quarter. With AI-powered business now at the core of our revenue mix, we are focused on building a stronger foundation for its next phase of growth across our full AI stack from chips and cloud infrastructure to models and applications. We are continuing to strengthen the capabilities that will support sustained innovation, power our future growth and reinforce our long-term competitiveness.

Let me now turn to the key business highlights of this quarter, starting with our proprietary AI chips, Kunlunxin. In Q2, Kunlunxin continued to demonstrate strong business momentum with demand remaining robust and broadening across industries. A growing number of customers are adopting its chips for an expanding range of AI workloads, reflecting increasing market recognition of Kunlunxin's stability, efficiency and versatility at scale. Kunlunxin continued to strengthen its software ecosystem, broadening compatibility with leading models and frameworks and improving ease of deployment across enterprise environments. Building on its support for ERNIE and other leading foundation models in China, Kunlunxin further extended its coverage in Q2 to include newer versions of major Chinese foundation models such as Kimi K3, GLM5.2, Minimax M3 and Hunyuan 3. It also improved inference throughput and overall compute efficiency, strengthening its ability to support diverse and demanding AI workloads at scale.

Over more than a decade, Kunlunxin has successfully developed and commercialized three generations of AI chips. Building on this track record, it continued to advance a clearly defined product road map, including the latest M100 optimized for large-scale inference and the upcoming M300s. This road map reflects Kunlunxin's deep understanding of evolving AI technology, workloads and their compute requirements, positioning it to support the next wave of AI innovation. As we continue to advance our AI infrastructure capabilities, we believe Kunlunxin will play an increasingly important role within our full stack AI architecture and enhance our ability to deliver high-performance, reliable and cost-efficient AI computing at scale. As demand for AI computing in China continues to grow, we believe our proprietary AI chips and full stack capabilities will become increasingly valuable. Supporting the future growth of our AI businesses and reinforcing our long-term competitiveness in AI.

Building on our strength at the infrastructure layer, AI Cloud Infra delivered another quarter of strong growth. In Q2, AI Cloud Infra revenue increased by 50% year-over-year, continuing to outpace the broader industry. Several factors combined to drive this sustained growth momentum. First, AI Cloud Infra continued to benefit from strong demand for AI computing. Demand remained robust across both training and inference workloads, while computing supply remains constrained across the market. Second, our existing key clients, including leading companies in online gaming, e-commerce and lifestyle content continue to increase both their usage and spending with us. Meanwhile, our overall customer count grew rapidly with new clients spanning companies at varying sizes.

Third, demand remained broad-based across industry verticals, including Internet, embedded AI, autonomous driving, smartphones, financial services and more. Within this mix, Internet and autonomous driving sustained strong growth, while embedded AI revenue grew approximately sixfold year-over-year in Q2. Based on these trends, we believe AI Cloud Infra revenue growth will remain strong in the second half with the potential for further acceleration. Importantly, the growth in AI Cloud Infra was accompanied by rapid profit growth and expanding margins on a year-over-year basis, reflecting continued improvement in the overall health and quality of the business. Within AI Cloud Infra, GPU Cloud revenue growth accelerated sharply to 283% year-over-year, building on an already high base of 184% growth last quarter. This momentum reflects strong underlying demand for scalable AI compute in the public cloud. The mix of our business continued to shift towards higher-quality revenue streams with GPU Cloud accounting for a growing share of AI Cloud Infra revenue. Given its more attractive margin profile, this shift is contributing to a healthier revenue mix and strengthening the long-term profitability of our cloud business.

On MaaS, our Qianfan MaaS platform offers one of the most comprehensive model libraries covering Baidu's ERNIE family as well as virtually all of China's leading models. A key priority for Qianfan is to make model inference at scale more reliable and cost efficient for customers. Leveraging our deep expertise in AI infrastructure and engineering, we further enhanced model serving through continued inference optimization, delivering higher throughput and greater service stability while reducing latency and inference costs. In Q2, revenue from external customers' token usage on Qianfan grew more than ninefold year-over-year, primarily driven by rapid growth in daily average token consumption among these customers.

Turning to foundation models. Advancing ERNIE and our overall model capabilities remains important to our next phase of AI-driven growth. Our commitment to foundation model innovation remains unwavering. As discussed in prior quarters, we reorganized our model teams into two groups with clearer mandates and greater focus across foundation models and applications. More recently, we welcomed a new generation of top AI talent to work on foundation models, further demonstrating our determination to compete and innovate at the forefront of AI. We believe these efforts will support the continued evolution of ERNIE and strengthen the foundation for future innovation across both models and AI applications.

Moving next to AI applications, where we continue to enhance product capabilities and expand real-world use cases. Let me begin with digital humans. As our digital human technology continues to advance, it is delivering stronger performance at lower cost and enabling an expanding range of use cases from e-commerce live streaming and digital human videos to real-time interactive digital humans and our newly introduced video podcast. These advances are opening up far broader possibilities for how digital humans can be used across industries. Our digital human capabilities are gaining increasing recognition from clients. In Q2, we continue to win new clients, including leading companies across industries, while existing clients also meaningfully scaled their usage. Some of our clients started with a pilot and after seeing what our digital human technology could deliver, expanded their usage.

A well-known Chinese Internet company, for example, expanded its digital human live streaming deployment to approximately 2.5x the previous level after just one quarter of use. Meanwhile, we continue to advance the global expansion of our digital human capabilities. Since launching our overseas digital human platform last quarter, we've seen encouraging momentum with its differentiated capabilities, delivering compelling results for merchants and creators overseas. As demand continues to unfold across more industries and regions, we believe the long-term growth potential for digital humans remains substantial.

Turning next to Miaoda, our vibe coding platform. With the launch of Miaoda 3.0 last quarter, users can now generate stand-alone mobile apps for both Android and iOS using natural language. Applications that once required a professional development team, a lengthy development cycle and significant investment can now be completed far more easily through Miaoda, even directly from a phone. We are seeing users engage with Miaoda more deeply. An increasing number of users are moving beyond one-off experimentation and returning to Miaoda to continue developing, iterating on, and refining their applications over time, reflecting stronger user stickiness. In June, Miaoda's monthly active users increased by 67% compared with March. Adoption is also expanding across industries, ranging from technology and education to health care, manufacturing, financial services and logistics, demonstrating Miaoda's applicability across diverse business scenarios and its broader commercialization potential.

We are also applying AI to help enterprises solve complex operational problems. A good example is Famou Agent, which can autonomously explore possible solutions to identify the best ones. Following the launch of Famou Agent 2.0 last quarter, we have continued to improve its usability and expand the scenarios it can address. Famou Agent has attracted growing interest from leading enterprises and begun to gain early commercial traction this quarter. We are pleased to see Famou Agent moving beyond efficiency gains to help enterprises optimize their operations and deliver real tangible business value. As its capabilities continue to advance, we believe its potential will continue to grow.

Another key direction for our AI applications is general-purpose agents. Earlier this year, we launched DuMate, our general-purpose agent for everyday productivity with seamless access across PC and mobile. In Q2, we introduced an enterprise version and continued to expand DuMate's proprietary Baidu skills and specialized toolkits, broadening the range and sophistication of tasks it can support. Meanwhile, our flagship consumer-facing AI applications, Baidu Wenku and Baidu Drive continue to embrace AI across the board, introducing new AI capabilities, sharpening existing ones. And this quarter, rolling out an upgrade to GenFlow that brings AI more deeply into users' everyday workflows. In June, AI DAU penetration across Baidu Wenku and Baidu Drive increased by 27.4% year-over-year, reflecting broader adoption of their AI-powered features.

Turning to AI search. We continue to improve both the quality of AI-generated answers and the overall user experience. Users are increasingly receiving answers that are more reliable, better structured and more effectively presented. At the same time, hallucination rates remained low, while our models became more effective at assessing content quality, helping reduce the incidence of low-quality answers. Together, these improvements drove better user experience and higher user satisfaction. We also further integrated AI search with ERNIE Assistant, extending the search experience beyond onetime answers into more seamless and interactive conversations that can better address users' follow-up questions and broader needs. In June, ERNIE Assistant's daily active users grew 83% year-over-year, while daily average conversation rounds more than tripled, reflecting growing user adoption and deeper engagement with this evolving search experience.

Turning now to AI in the physical world. Let me discuss Apollo Go, our autonomous ride-hailing service. This quarter, we continued to advance global expansion while further enhancing safety, operational performance and the rider experience. Hong Kong marked an important milestone for Apollo Go this quarter. In June, we received Hong Kong's first permits for fully driverless testing and began testing on Airport Island in July. This made Apollo Go the first autonomous ride-hailing service provider globally to conduct fully driverless testing in a right-hand drive, left-hand traffic robotaxi market. Hong Kong is one of the world's most sophisticated urban mobility markets with a complex operating environment and rigorous standards for both technology and operations. Reaching this milestone in Hong Kong provides strong validation of the maturity and adaptability of our technology and operational capabilities.

The experience we have gained in Hong Kong is already helping us advance more efficiently in London. In July, Apollo Go began open road testing there in partnership with Uber and Lyft. Together, our progress in these two markets demonstrates our technology's ability to generalize across different operating environments, giving us greater confidence in expanding into more and more high-value right-hand drive, left-hand traffic robotaxi markets over time. We also made progress across several other international markets. In Dubai, we launched a fully driverless commercial operations in July and now operate at the largest scale among fully driverless autonomous ride-hailing services in the city with rides available through both the Apollo Go and Uber apps. In Switzerland, we began open-road testing in partnership with PostBus. We also signed a memorandum of understanding with Kazakhstan's Turlov Private Holding Limited to jointly explore autonomous ride-having services in the country.

Overall, Apollo Go delivered around 1 million fully driverless operational rides in Q2. As of June 2026, cumulative rides provided to the public by Apollo Go exceeded 23 million. Ride volume during the quarter was temporarily affected by operational adjustments in certain domestic cities due to regulatory considerations. Over this period, we conducted a systematic review to further strengthen the robustness of our autonomous driving systems and the rigor of our operational processes. As of August, operations in the affected cities have begun to resume on a stronger footing. Meanwhile, we continue to expand our operations across other domestic markets. We are confident that ride volume will regain momentum over the coming quarters as we steadily ramp up operations and pursue further expansion.

In Q2, we continue to raise the bar on safety and the rider experience. As of the end of June, our fully driverless vehicles recorded an average of approximately one airbag deployment, every 14.4 million kilometers, underscoring our industry-leading safety performance. We also enhanced pickup and drop-off point recommendations to reduce walking distances and avoid unsuitable stopping locations while further improving perception and motion planning capabilities to deliver smoother and more consistent rides. These improvements represent an even higher operating standard, one we intend to build on as we continue to integrate Apollo Go more seamlessly into urban transportation systems, making it a more convenient and trusted part of everyday mobility.

Looking ahead to the second half, our priorities for Apollo Go are clear: further enhance our safety standards and operational capabilities, advance our global expansion, scale our fleet and ride volumes, and bring more cities to unit economics breakeven. We believe progress across these priorities will further strengthen Apollo Go's leadership in autonomous ride hailing and lay a stronger foundation for scaling its operations safely and sustainably over the long term.

To summarize, the progress we made across our full AI stack this quarter reaffirms Baidu's transition into an AI-first company and further strengthened the foundation for our next phase of growth. We are also actively expanding our AI businesses into global markets and are encouraged by the progress we are already seeing, including in AI applications and robotaxi. With this stronger foundation, we believe we are well positioned to capture a broader range of opportunities across markets over time.

With that, let me turn the call over to Henry to go through the financial results.

Thank you, Robin, and hello, everyone. We were pleased with the continued momentum of Baidu Core AI-powered Business this quarter. Revenue from Baidu Core AI-powered Business reached RMB 12.5 billion and continue to account for half of Baidu General Business revenue. AI Cloud Infra revenue grew approximately 50% year-over-year, with GPU Cloud revenue growing 283% year-over-year in Q2. This performance reinforces AI-powered business as a key driver of Baidu's long-term growth. In addition, we advanced an important capital market initiative to broaden access to capital markets and create long-term shareholder value.

In July, our Board approved a motion to pursue the voluntary conversion of our Hong Kong listing to dual primary status. Since then, we have submitted our application and received Hong Kong Stock Exchange's acknowledgment. We also plan to convene an Extraordinary General Meeting on August 26 to seek shareholder approval for certain related matters. We currently expect the conversion to become effective within this year, subject to the approval of the shareholders and the Hong Kong Stock Exchange. Once effective, we believe the dual primary listing will broaden our investor base, enhance the liquidity of our shares and provide greater flexibility in accessing capital across both markets. As we move forward, AI remains central to Baidu's long-term growth and competitiveness. We will continue to invest with conviction and discipline in the capabilities and businesses where we see the strongest long-term opportunities.

Now let me walk through the details of our second quarter 2026 financial results. Total revenue of Baidu was RMB 31.3 billion, decreasing 2% quarter-over-quarter and 4% year-over-year. Revenue from Baidu General Business was RMB 25.2 billion, decreasing 3% quarter-over-quarter and 4% year-over-year. Revenue from iQIYI was RMB 6.3 billion, increasing 1% quarter-over-quarter and decreasing 5% year-over-year. Cost of revenues was RMB 19.1 billion, decreasing 3% quarter-over-quarter, primarily due to a decrease in costs related to AI Cloud business, partially offset by an increase in traffic acquisition costs and increasing 4% year-over-year, primarily due to increases in costs related to AI Cloud business.

Operating expenses were RMB 9.2 billion, decreasing 1% quarter-over-quarter and decreasing 17% year-over-year. The year-over-year decrease was primarily due to decreases in 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 compared to RMB 626 million last quarter and RMB 4.9 billion for the same period last year. The year-over-year decrease was primarily due to a decrease in fair value gain from long-term investments and an increase in net foreign exchange loss arising from exchange rate fluctuation between Renminbi and U.S. dollar.

Income tax expense was RMB 1.0 billion compared to RMB 528 million last quarter and RMB 881 million for 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. Non-GAAP net income attributable to Baidu was RMB 2.6 billion. Non-GAAP net margin for Baidu was 8% and non-GAAP diluted earnings per ADS was RMB 7.22. We define total cash and investments as cash, cash equivalents, restricted cash, short-term investments, net long-term time deposits and held-to-maturity investments and adjusted long-term investments. As of June 30, 2026, total cash and investments were RMB 283.1 billion. Operating cash flow was RMB 3.4 billion. Baidu General Business had approximately 27,000 employees as of June 30, 2026.

With that, operator, let's now open the call to questions.

[Operator Instructions] The first question today comes from Alex Yao with JPMorgan.

So with multitrillion parameter models emerging rapidly and pushing the frontier on benchmark performance, how does Baidu think about ERNIE's competitive positioning from here? Following the recent addition of a senior foundation model talent, what are the key technical and product priorities for ERNIE? And what should investors expect from its next stage of development?

This is Robin. First, from an industry perspective, foundation models are still evolving rapidly, roughly every few months, different model takes the lead in some capability. This shows the field remains highly dynamic and the competitive landscape is far from settled. In a market like this, we believe long-term competitiveness ultimately comes down to sustained technology investment, application-driven approach and patience. Baidu has always been a company that believes in technology and is willing to commit to it for the long term. Our experience has repeatedly shown that meaningful technological innovation takes patience and persistence. Today, many of Baidu's important AI assets, including Kunlunxin and Apollo Go, are the result of more than a decade of sustained investment. They become a key source of our differentiated competitiveness. And their performance and commercial value are gaining increasing broad recognition. So we are very proud of that.

ERNIE has likewise always been an important part of Baidu's AI strategy and full stack AI capabilities. We were among the first companies in China to invest in foundation models. There were trials and errors along the way, but our commitment to make ERNIE competitive remains unwavering. Going forward, we will continue to invest resources needed to drive ERNIE's ongoing development. As part of this effort, we have further optimized our organization and recently brought in top AI talent. We are confident in accelerating AI's iteration and bring ERNIE back into the top tier of foundation models.

Looking ahead, we will continue to take an application-driven approach. Foundation models span a very broad range of capabilities and no single model can lead in every dimension at all times. We will, therefore, focus on capabilities that matter most to Baidu's applications and make ERNIE strongest in these areas. Spanning AI search, digital human, Miaoda, Famou and general-purpose agents like DuMate. These applications are vital parts of ERNIE's continuous improvement. I take AI search as an example, when we improve ERNIE's ability to understand user intent and assess content quality, we apply those improvements directly to search and feed. This lets us quickly see the results, identify what still needs work and feedback the relevant data into model training, which makes our model better at user intent understanding and content quality assessment. And we see this loop as an important path for ERNIE's development, one that translates technological progress more directly into better product experiences and real user and commercial value, and then ultimately benefiting a broader range of users and businesses.

The next question comes from Alicia Yap with Citigroup.

My question is on cloud. So Baidu AI Cloud Infra revenue has maintained strong growth. Could management discuss the key growth drivers and also your outlook for the revenue growth over the next few quarters? And also, how should we think about the long-term margin potential as the business scale?

Alicia, this is Dou. AI Cloud Infra revenue grew 50% year-over-year in Q2. This remains a robust growth rate and above the industry average, I believe. So over a longer horizon, our AI Cloud Infra has sustained rapid growth for several consecutive quarters, consistently outpacing the industry. The standout of this quarter was the GPU Cloud, whose revenue grew 283% year-over-year, marking its fourth straight quarter of triple-digit growth and accelerating further from 184% in Q1. Looking ahead, we see several drivers supporting continued growth.

Currently, demand for AI computing in China remains very strong. And as AI becomes more deeply embedded in real-world applications and business workflows, particularly as inference continues to scale rapidly, we expect demand to grow further. Meanwhile, our customer base is also expanding rapidly with new customers of different sizes adopting our AI Cloud Infra, while existing key customers keep increasing both usage and spending. Demand is also broadening across industries and use cases, including internet, gaming, embodied AI, autonomous driving, smartphones, financial services, and others. Actually, most importantly, we have built and continued to strengthen differentiated full stack AI architecture, spanning chips, cloud infrastructure, models, and applications with competitive offerings at every layer.

At the application layer, in particular, we moved early to build a portfolio of agents and AI applications with products such as Famou Agent, DuMate, Miaoda, and Baidu Yijing gaining traction and strengthening our ability to capture an increasingly diverse range of AI opportunities. So based on current demand trends, our customer pipelines and these differentiated advantages, so we feel confident that AI Cloud Infra can maintain strong growth in the second half with the potential for further acceleration.

On the profitability side, you just mentioned, we are pleased with the continued improvement alongside rapid revenue growth. In Q2, AI Cloud Infra profit and margins both increased year-over-year. Going forward, we think several factors should support further margin expansion. First, GPU Cloud is growing significantly faster than the AI Cloud Infra in general and continues to represent a large share of the mix. It also carries a more attractive margin profile than traditional CPU Cloud, with further room for margin improvement as it scales, supported by continued optimization of its product and customer mix, better resource utilization and greater operating efficiency. So as GPU Clouds contribution increases, the mix shift should continue to lift overall margins.

Second, on MaaS. Revenue from external customers token calls on Qianfan is growing very fast. While MaaS still represents a relatively small share of our AI Cloud Infra revenue today, so the early momentum we are seeing is very encouraging. As usage skills and unit inference costs keep coming down, we believe over the longer term, MaaS-related businesses will be able to unlock more profit potential and become an increasingly meaningful contributor to margins. Finally, our full stack AI capabilities and self-developed chips also provide end-to-end cost advantages that should support margin expansion. So taken together, we think there's still a lot of room for AI Cloud Infra margins to improve over the long term. Thank you.

The next question comes from Miranda Zhuang with Bank of America Securities.

My question is about margin. So with AI-powered business now accounting for half of the revenues and also CapEx continuing to ramp, so how do you think about Baidu's operating margin trajectory? And how will management balance the continued AI investments with profitability?

Thank you, Miranda. This is Henry. This quarter, AI-powered business continued to account for half of the Baidu General Business revenue, further underscoring AI's positioning at the center of our business. Within AI-powered business, AI Cloud Infrastructure sustained rapid revenue growth with profit also growing quickly and margins improving year-over-year. Within AI Cloud Infra, our GPU Cloud business, which typically carries a better margin profile, continue to increase as a percentage of revenue. As this favorable mix shift continues, together with the strong market demand and the cost advantage we get from our self-developed chips and full stack AI capabilities, we believe there is still meaningful room for AI Cloud Infrastructure margins to expand over the long term.

As the business scales, we also expect better resource utilization and greater operational leverage to provide further support for the margin expansion. We also see attractive long-term profitability potential in our AI applications. Many of these applications are sticky and subscription-based by nature with the potential to deliver increasingly attractive margins over time as they scale. As adoption growth and monetization progresses, we expect them to become a more meaningful contributor to overall profitability.

Meanwhile, I think we are still in an AI investment cycle and our commitment to that investment is unwavering. We invest with conviction, but just as importantly, we spend wisely and stay closely focused on the ROI. Our investments are driven by a clear demand from both customers and our internal business, allowing much of where we will invest in to be able to work quickly and begin contributing to revenue relatively soon. Meanwhile, we are continuously strengthening our supply chain management capabilities, which we believe will increasingly help us improve capital efficiency as we scale. Together, these strengths give us a good visibility into returns and are confident in our ability to improve investment efficiency over time.

That said, different AI investments play out on different time lines, and we will take longer to fully deliver their value. Some of them will take longer. We are now in a critical phase of investment, and we intend to keep investing decisively in the areas that matters most to our long-term competitive position while maintaining the same discipline around ROIC, operating efficiency and cash flow. As our AI business scale further and monetization matures, we believe these investments will increasingly translate into some and more sustainable profit growth. Thank you...

The next question comes from Lincoln Kong with GS.

Could you update us on the progress of Kunlunxin's proposed listing and the key milestones ahead? I'm wondering what will drive its future growth? And how does management view its long-term commercial potential and strategic role within Baidu AI ecosystem?

Okay. I'll take it. This is Dou. The listing process for Kunlunxin is still ongoing, and we will update the market as soon as we have more to share. From a business perspective, we remain very confident in Kunlunxin's long-term growth and commercial potential for a few reasons. First, across the industry, demand for AI compute continues to grow across both training and inference. As model capabilities keep improving and more applications moving into real-world use, especially as agents advance and expand into a wider range of use cases. So we are seeing inference pick up pace in particular. So we believe this trend will continue, creating a long-term structural growth opportunity for the AI chip industry.

Secondly, the domestic market carries significant growth potential, while with supply likely to remain constrained for some time. So against this backdrop, customers are increasingly seeking high-performance, reliable and cost-efficient domestic AI chips. So we believe this creates substantial opportunities for chip providers with strong technical capabilities and the ability to deliver at scale. Following more than a decade of investment, Kunlunxin has built solid capabilities in the chip performance, hardware software integration, compatibility with the mainstream models and frameworks and large-scale deployment, earning growing recognition from customers. Those are the things that put Kunlunxin in a good position in this market and capture the commercial opportunities arising from China's growing AI computing needs.

Thirdly, within Baidu's AI ecosystem, Kunlunxin is an important part of the infrastructure layer in our full stack AI architecture, spanning chips cloud infrastructure models and applications. The close coordination across these layers enables end-to-end optimization, allowing us to deliver greater performance, reliability and cost efficiency. So this supports the long-term deployment of AI Cloud Infra and our other AI businesses while further strengthening the competitiveness of Baidu's full stack AI capabilities. Looking ahead, we expect Kunlunxin to keep playing a meaningful role in our AI infrastructure, capturing a broader range of commercial opportunities and serving a wider range of market needs. Thank you.

The next question comes from Wei Xiong with UBS.

Could you walk us through the expected time line for the Hong Kong dual primary listing conversion and potential Stock Connect inclusion? Also, what's the strategic rationale? And how could it affect Baidu's investor base, share liquidity, and valuation over time?

Thank you. This is Henry. Let me start with the time line. Our Board has approved the conversion to a dual primary listing back in July. We also filed our application with the Hong Kong Stock Exchange and received its acknowledgment. The next step is our extraordinary general meeting scheduled on August 26. During that meeting, we will seek shareholder approval for certain matters required in preparation for conversion. And from there, we expect the conversion to take effect within this year, subject to approval of the Hong Kong Stock Exchange and other applicable conditions.

On Southbound Stock Connect, we are also actively preparing for potential inclusion following the conversion and hope our shares can be included at the earliest opportunity. And of course, this will remain subject to the applicable eligibility requirements and review of procedures and the decisions for the relevant exchange. As for the rationale, due primary listing is really about broadening our investor base, enhancing the liquidity of our shares and giving us greater flexibility in accessing both the Hong Kong and the U.S. capital markets. It also allows more investors, particularly in Asia, to better understand and participate in Baidu's value as an AI-first company.

Looking further out, if we achieve Stock Connect inclusion down the road, we would expect that to meaningfully expand participation from Mainland China investors specifically, which should support an even more diversified shareholder base over time. We will be happy to keep you updated as we make further progress.

The next question comes from Thomas Chong with Jefferies.

Could management update us about AI search progress across product capabilities, user experience, and monetization. We are seeing online marketing revenue remained under pressure in Q2. What were the main factors? And how does management expect the business to trend in the second half?

Thomas, this is Julius. Let me take your question. I think over the past few quarters, our focus on the AI search transformations have been improving the quality of our AI answers through enhancing the user experiences a lot. Accuracy and authority has always been our core strength, and we have kept reinforcing that as the AI transformation moves forward. And now our AI search can better understand what users are looking for. The answers are more reliable, better structured and presenting in richer formats. And meanwhile, hallucination rates remains low, and our models now are getting better at telling good content from bad. So we are facing more high-quality answers and fewer weak ones. Users have responded quite well on these changes, and we are seeing steady improvements in user satisfaction, the willingness to search and retention.

In this quarter, we further integrated AI search with the ERNIE assistant, turning one of the search answers into more coherent interactive multi-round conversations that better address the follow-up questions and broader user needs. And we are also continuing to strengthen the to-use multistep planning and complex task executions to help users to get more done. Recently, the ERNIE Assistant's Task Agent topped two influential third-party agent benchmarks, the Pinch Bench V2, which is a global benchmark focused on the real-world complex tax complexion and the SuperCLUE XCloud evaluation of the leading domestic agent products. I think these results can help to reinforce the ERNIE Assistant's leading capabilities in the to-use multistep planning and task executions.

That said, the competition in this industry remains very intense and as new product forms like AI chatbots continue to get traction, the ways users discover and consume information is keep evolving and competition for users' time and attention has intensified further. Meanwhile, we have continued to push forward with the AI search transformations while deliberately holding back on monetizing the AI search, both of which we have weighted on our advertising businesses in the near term. Given these dynamics are likely to persist, we expect our advertising business to remain under pressure in the second half. On monetization, our priority right now is still to getting the products and the user experience right. As model capabilities, the user experiences and the task complexions continue to improve, we believe that more monetization opportunities, which fit naturally into the AI experiences will emerge in the future. Thank you, Thomas.

The next question comes from Ellie Jiang with Macquarie.

I have a question on robotaxi, please. So with China's recent introduction of the new robotaxi policies, how does management view the evolving regulatory environment? How should we think about Apollo Go's relative focus and also the pace of expansion across domestic overseas market? And it would be great if management can talk about the progress that Apollo Go has made in the overseas commercialization side.

This is Robin. Let me answer this question. The global robotaxi industry is evolving very quickly. In the past, the industry's focus was on whether robotaxis could deliver a safe, comfortable riding experience. Today, that focus has expanded to whether robotaxis can operate reliably at scale and fit into the broader transportation system. In line with this trend, major markets around the world are also iterating on and refining their regulatory frameworks for robotaxis.

In China, for example, the country's first mandatory national standard on safety requirements for Level 3 and Level 4 automated driving systems was recently issued. And Apollo Go contributed its extensive technical and operating experience to the L4 requirements under this standard. Safety has always been our top priority, and we maintain an industry-leading safety record globally. We will continue to uphold high standards on safety and operations. More broadly, a clearer, more systematic regulatory frameworks will help raise operational standards across the industry, build public trust and lay a stronger foundation for the long-term orderly growth of robotaxis. Against this backdrop, we remain positive on Apollo Go's global expansion.

We do not view domestic and international markets as an either/or choice. We are highly open and adaptive. We assess each city based on its regulatory framework, mobility demand, right pricing, road conditions, and commercial viability and setting our pace of entry and expansion accordingly. Backed by proven technology and operating experience, we are ready to move quickly and scale efficiently in any city where regulations and market conditions allow. Our goal is to go deep and build a solid presence in every city we enter regardless of country boundaries.

This is reflected in our progress across different cities. In Dubai, Apollo Go has entered fully driverless commercial operations and is scaling up, and we now operate at the largest scale among robotaxi services providers over there. In London, we are advancing testing and development with partners, including Uber and Lyft. In Hong Kong, we became the first robotaxi service globally to conduct fully driverless testing in our right-hand drive, left-hand traffic market. In Shenzhen, the number of rides are picking up very quickly, making it one of our largest markets.

As our fleet expands and our operating model matures, we expect vehicle and operating costs to keep coming down, while scale brings additional efficiency gains. In the past, Apollo Go achieved unit economics breakeven in a market with relatively low taxi fares. In the future, in overseas markets with higher ride prices, our low-cost vehicles and proven operating model have the potential to deliver even stronger unit economics. The international market outside of U.S. and China is also larger than the domestic China market. So the addressable opportunity is quite substantial. Looking ahead, supported by our advantages in technology, cost and operations, we are confident in bringing more cities to unit economic breakeven. Thank you.

Ladies and gentlemen, that does conclude our conference for today. Thank you for participating, and you may all disconnect.

Hello, and thank you for standing by for Baidu's First Quarter 2026 Earnings Conference Call. [Operator Instructions] Today's conference is being recorded. If you have any objections, you may disconnect at this time.

I would now like to turn the meeting over to your host for today's conference, Juan Lin, Baidu's Director of Investor Relations.

Hello, everyone, and welcome to Baidu's First Quarter 2026 Earnings Conference Call. Baidu's earnings release was distributed earlier today, and you can find a copy on our website as well as on newswire services. On the call today, we have 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. After our prepared remarks, we will hold a Q&A session.

Please note that the discussion today will contain forward-looking statements made under the safe harbor provisions of the U.S. Credit Securities Litigation Reform Act of 1995. Forward-looking statements are subject to risks and uncertainties that could cause actual results to differ materially from our current expectations. For detailed discussions of these risks and uncertainties, please refer to our latest annual report and other filings with the SEC and Hong Kong Stock Exchange. Baidu does not undertake any 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. Our 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.

As a reminder, this conference is being recorded. In addition, a webcast of this conference call will be available on Baidu's IR website.

I will now turn the call over to our CEO, Robin.

[indiscernible] RMB 26.0 billion in total revenue in Q1, up 2% year-over-year, marking a return to positive growth. Revenue from our core AI power business reached RMB 13.6 billion, up 49% year-over-year. For the first time, it accounted for more than half of Baidu general business revenue reaching 52%. This is an important milestone as AI-powered business has now become the majority of our revenue mix.

AI cloud infra delivered exceptional momentum in Q1 with overall revenue growing 79% year-over-year. Within AI Cloud infra, GPU cloud revenue continued its strong trajectory from last quarter's 143% growth, accelerating further to 184% year-over-year.

Apollo Go also had a strong quarter. We delivered 3.2 million fully driverless rise in Q1, sustaining triple-digit growth in total rise year-over-year, reflecting the continued scaling of our operations. Together, these results confirm that AI has clearly become the primary growth driver of Baidu, reinforcing our position as an AI-first company.

As AI adoption continues to accelerate, real-world applications are expanding, opening up new and increasingly diverse demand for AI capabilities. We are confident in our ability to capture these opportunities as they unfold and believe AI will continue to drive the next phase of Baidu's growth.

Now let me walk you through the key highlights of this quarter, starting with AI Cloud Infra. As AI adoption accelerates across industries, we continue to see demand surge across both training and inference workloads with influence ramping, especially fast and accounting for a growing share of overall demand.

Q1 was a quarter of significantly accelerated growth for our AI Cloud Infra with revenue growth well above the broader market. The mix of our business continues to shift toward higher-quality revenue streams, GPU Cloud, which typically carries stronger margins, has become a meaningful contributor to our total AI Cloud Infra revenue, underscoring the ongoing improvement in overall business health.

A key driver behind this momentum is a differentiated advantage of Baidu's full stack AI capabilities, one that very few companies globally can truly claim. With proprietary components at every layer from underlying infrastructure to applications, we were able to ensure stable and reliable compute supply while also optimizing end-to-end across the entire stack, continuously improving performance, reducing costs and delivering compelling cost effectiveness for our customers. As AI applications continue to proliferate, this full stack advantage becomes increasingly pronounced, enabling us to capture a broader and more diverse range of opportunities.

At the infrastructure layer, we hold a distinct advantage through Kunlunxin, our self-developed AI chips. We have seen strong and expanding demand for Kunlunxin with a growing number of customers across diverse industries adopting it for a broadening range of AI workloads. This reflects growing market recognition of Kunlunxin's stability, efficiency, compatibility and versatility. It is also among the first domestic AI chips to achieve large-scale commercial deployment in a single AI computing cluster of over 30,000 accelerators with industry-leading cluster performance and stability.

Built on a comprehensive software stack, Kunlunxin delivers [indiscernible] compatibility with different models and frameworks as well as strong usability across enterprise environment. To date, it has been optimized and validated for workloads across various models covering the latest versions of ERNIE and other mainstream foundation models with inference support recently extended to DeepSeek V4 GLM 5.1 and MiniMax M2.7. As an important component of our AI infrastructure, Kunlunxin further strengthens the foundation of our infrastructure layer, enabling Baidu AI Cloud to support customers' AI deployment with greater efficiency, reliability and cost effectiveness and enhancing the overall competitiveness of our cloud offerings.

These advantages are translating into strong client momentum on the infrastructure side. Baidu AI Cloud has become a trusted infrastructure partner for a growing number of major companies across a broad range of industries, including Internet, gaming, embodied AI, autonomous driving, smartphones, financial services and more. This quarter, we added several prominent new clients, including leading model companies.

Our client base also includes leading names such as Unity, Honor, Opel and Vivo. At the same time, existing top-tier clients continue to deepen their collaboration with us and scale their usage, driving healthy expansion across our client base.

On the MaaS front, as open cloud gained traction across the industry, we moved quickly to expand the model library, our [indiscernible] Mass platform. In addition to ERNIE, [indiscernible] now supports an expanding set up in demand models, including popular ones from [indiscernible] AI, MiniMax, [indiscernible] and DeepSeek keeping our model library comprehensive and up to date.

In March, daily average token consumption from external customers grew to nearly 7x the level of a year ago while our MaaS revenue also scaled rapidly. We believe the MaaS platform still has significant untapped potential as the ecosystem around agents and AI applications continues to evolve.

On foundation models, we recently launched ERNIE 5.1, which delivers stronger tax capabilities, a more compact model of size and enhance the reasoning compared to its predecessor. We also made advances in key areas, including cold generation, agent capabilities and deep search. Recently, on the LM Arena, ERNIE 5.1 ranks first among Chinese models on the text leaderboard. ERNIE 5.1 also topped the LM Arena search leaderboard, among Chinese models, ranking fourth globally making it the only Chinese model to appear on that leaderboard as well. Looking ahead, we remain firmly committed to advancing ERNIE through an application-driven approach, continuously iterating based on real-world needs to keep earning at the forefront of AI capabilities.

Now let me turn to AI applications. We have long believed that the true value of AI is ultimately realized through applications and we have been early and persistent in building a comprehensive portfolio serving both enterprises and individual users. This quarter, we continue to see encouraging progress across several high potential directions.

Let me highlight a few examples. The first is to make our AI agent for everyday productivity, which we recently showcased at Baidu Create, [indiscernible] is designed to execute complex multistep workflows across applications and files autonomously, handling long-running tasks from start to finish. Available across both PC and mobile, it enables users to initiate tasks anytime and from anywhere while operating continuously in the background as a 24/7 AI assistant. Users simply describe what they need and come back to results.

What truly differentiates [indiscernible] is its seamless integration with Baidu's proprietary skills, including AI search [indiscernible] and more as we continue to expand [indiscernible] scale ecosystem, we believe it will be able to better tackle an ever wider range of office workflows and complex, real-world tests, helping users complete the end to end more effectively.

Turning to digital humans. Our hyperrealistic digital human technology continued to advance with improved performance and increasing readiness for large-scale deployment. On the cost front, we achieved around 80% cost reduction over the past 2 quarters, lowering the adoption barrier and making our digital humans more affordable and accessible for a broader range of clients. Meanwhile, we are also taking our digital human capabilities global.

At the recent Baidu Create, we launched an overseas digital human platform that enables merchants and creators to easily generate digital human content from e-commerce live streams to digital human videos and beyond. To make our digital humans truly work for global markets, we have built in deep localization from the ground up, supporting 24 languages, including Spanish, French and Thai with scripts and presentation styles culturally adapted to resonate with local audiences. This helps merchants run around the clock, digital human live streams that feel authentically native unlocking new levels of efficiency and conversion potential across global markets. Our growing partner base in China and overseas includes [indiscernible], TikTok and [indiscernible] with several partners deepening their collaboration with us.

Next is Miaoda, our vibe coding platform. Miaoda empowers anyone to bring their ideas to life without writing a single line of code, and we are seeing this value increasingly recognized. In March, monthly active users of Miaoda grew around 70% quarter-over-quarter, while our domestic paying user rate reached approximately 3x the level at the end of last year. At Baidu Create, we launched Miaoda 3.0, introducing an enterprise version and a mobile app, enabling broader adoption across both individuals and enterprises as well as more flexible usage across time and use scenarios. Notably, Miaoda now supports the generation of stand-alone mobile applications further expanding what users can create with Miaoda.

Another example is Famou agent. Our self-evolving agent designed to address complex operational challenges across industries and help enterprises unlock meaningful productivity gains. With the launch of Famou agent 2.0 at Baidu Create, we further expanded its accessibility. While earlier versions were primarily used by developers and technical teams, Famou agent 2.0 lowers the barrier to entry by enabling domain experts to interact with the agent directly through natural language, no coding expertise required. For example, at Qingdao Port, one of the world's leading ports with highly sophisticated scheduling system and deeply complex operational logic, Famou agent is helping push the efficiency of an already advanced system even further.

In an environment where thousands of interdependent variables must be coordinated in real time, Famou agent autonomously explores the solution space to identify optimal decisions across birth scheduling, equipment allocation and cargo prioritization even on top of an already highly optimized baseline, Famou agent continues to unlock incremental efficiency gains, further enhancing the overall operational performance.

Next is AI Search. In Q1, we continued to advance our AI Search transformation with a particular focus on improving user satisfaction and the overall search experience. Through ongoing enhancements in model capabilities, we further improved our search results are planned, structured and generated, enabling better assessment of content quality, broader distribution of high-quality information and a significant reduction in low-quality content. Meanwhile, ERNIE assistant continued to see strong user engagement driven by ongoing improvements to its interaction experience. In March, daily active users of early assistant nearly doubled year-over-year while daily average conversation rounds more than tripled over the same period.

Next day, retention also improved meaningfully, reflecting stronger user stickiness. Looking ahead, we will continue to deepen the integration between AI Search and ERNIE assistant, further enhancing the experience across information discovery, content understanding and task completion.

Beyond the digital world, AI is reshaping the physical world in profound waste and robotaxi stands as the most powerful environment of this transformation. In Q1, Apollo Go, our robotaxi business, maintained strong momentum with fully driverless rigs continuing to grow and our safety record remaining industry-leading. We also continued to advance our international expansion, making steady progress across key overseas markets. In Europe, we are on track to commence open road testing in Switzerland and our first vehicles have arrived in London in preparation for testing with Uber and Lyft expected to begin soon.

In the Middle East, Apollo Go's fully driverless operations are now running across multiple zones in Dubai. Also in late March, we officially launched the Apollo Go app, making us the first and only autonomous ride hailing service with its own stand-alone app there.

Beyond traditional ride hailing, we are exploring new use cases that broaden Apollo Go's commercial reach. In Hainan, one of China's most popular tourist destinations, we partnered with [indiscernible] on a rental model with our fully driverless vehicles, station directly at the arrival level of [indiscernible] airport. So Apollo Go is right there waiting as visitors step out of the terminal. We believe robotaxi can deliver value well beyond daily commuting opening up new monetization opportunities in the process.

As Apollo Go continued to scale fully driverless operations, we have encountered a broader and increasingly complex range of real-world scenarios, including system and operational complexities that only emerge at larger scale. When such situations arise, we handle them with rigor and use them to continuously strengthen our operations. More broadly, Apollo Go has moved well beyond technology demonstration and small-scale pilots. At this scale, we are addressing a new frontier centered on how robotaxi services fit more naturally into public transportation, city operations and everyday life. These experiences are helping us build the expertise and knowledge needed for Apollo Go to coexist more seamlessly with the broader transportation ecosystem over time and ultimately, to become a more convenient and trusted service for the people we serve.

In closing, our AI-powered businesses delivered strong momentum across the board in Q1. AI has now become the core driver and the majority of our business, and we see this role only growing from here as we scale AI across an increasingly diversified portfolio and extend our reach into global markets from AI applications to autonomous ride hailing, we see significant opportunities opening up on multiple fronts. We are confident in our ability to capture them.

With that, let me turn the call over to Henry to go through the financial results.

Thank you, Robin, and hello, everyone. In Q1, we continue to make progress on our key priorities, enhancing this closure, growing our core AI-powered business and improving operational efficiency. I'd like to highlight a few results from the quarter.

Total revenue of Baidu General business grew 2% year-over-year, returning to positive growth after several quarters of decline. Non-GAAP operating income of Baidu General business increased 39% quarter-over-quarter to RMB 4.0 billion. Operating cash flow for Baidu remained positive for the third consecutive quarter at RMB 2.7 billion, reflecting the continued improvement in our operating efficiency and overall business health. We also reached an important milestone our core AI-powered business accounted for more than half of Baidu General business revenue for the first time.

Revenue from Baidu Core AI-powered business exceeded RMB 13 billion, up 49% year-over-year. Within this, AI Cloud Infra growth significantly outpaced the broader market, while our AI applications portfolio continue to flourish across multiple fronts. Combined, AI Cloud Infra and AI applications drove our total AI cloud revenue to RMB 11.3 billion in the first quarter of 2026. Beyond cloud, Apollo Go further reinforce its position as a global leader in autonomous ride-hailing and continue to expand its operations. Collectively, these results point to a business that is becoming both more AI-driven and more financially healthy.

Now let me walk through the details of our first quarter 2026 financial results. Total revenue of Baidu was RMB 32.1 billion, decreasing 2% quarter-over-quarter. Revenue from Baidu General business was RMB 26.0 billion, increasing 2% year-over-year and remaining flat quarter-over-quarter, among which the increase in others was primarily driven by the growth of AI cloud business.

Revenue from IT was RMB 6.2 billion, decreasing 8% quarter-over-quarter. Cost of revenues was RMB 19.6 billion, increasing 7% quarter-over-quarter primarily due to an increase in costs related to AI cloud business, partially offset by decreases in content costs and traffic acquisition costs.

Operating expenses were RMB 9.3 billion, decreasing 28% quarter-over-quarter, primarily due to decreases in expected credit losses and personnel-related expenses. Operating income was RMB 3.2 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 626 million compared to RMB 1.2 billion last quarter. Income tax expense was RMB 528 million compared to RMB 1.0 billion last quarter.

Net income attributable to Baidu was RMB 3.4 billion, net margin for Baidu was 11% and diluted earnings per ADS was RMB 8.76. Non-GAAP net income attributable to Baidu was RMB 4.3 billion. Non-GAAP net margin for Baidu 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 were RMB 279.3 billion. Operating cash flow was RMB 2.7 billion. Baidu General business had approximately 28,000 employees as of March 31, 2026.

With that, operator, let's now open the call to questions.

[Operator Instructions] Your first question today comes from Alex Yao with JPMorgan.

And congrats on the impressive acceleration in AI Cloud Infra revenue this quarter, could you guys share more color on the key drivers behind this revenue momentum? And do you have sufficient compute capacity to support future growth and then how should we think about the margin profile of AI cloud compared with a more tradition or, for example, CPU cloud and the long-term margin trajectory?

Thank you, Alex. This is Dou. We have seen a remarkable strong enterprise demand for AI infrastructure [indiscernible] showing particularly strong momentum, which is a pretty healthy signals. It tells us that customers have moved beyond training models and we're now running AI across more parts of their business and accelerating pace. Closely related to this, our MaaS platform is seeing a strong traction. [indiscernible] is one of the [indiscernible] few MaaS platforms in China.

As Robin just mentioned, besides ERNIE we have quickly expanded [indiscernible] model library to include the most in-demand models like DeepSeek, Gemini, MiniMax and others. And we're seeing continued growth in token consumption from external customers. Importantly, supporting new models quickly is not a simple plug and play process. It requires higher throughput inference and efficient model serving capabilities, so we can run these models reliably at scale and serve more token demand with the [indiscernible] of 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 like 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 the supply is relatively tight. We are actively expanding capacity and improving resource efficiency to better support the growing customer needs.

Our confidence in capturing this demand comes from our differentiated full-stack AI capabilities, which provides 2 tangible advantages. First, efficiency. Owning and optimizing across the full stack enables us to deliver high -- highly competitive price performance for our customers. And secondly, 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. Firstly, GPU cloud is technically more complex with much higher barriers to our entry. Baidu was actually one of the earliest cloud providers in China to build GPU cloud at scale, and we remain at the forefront. Secondly, demand remains very strong, while high-quality supply is relatively tight, customers prioritize stability and availability, not just cost. Certainly, our [indiscernible] AI chips and full-stack capabilities gives us more room to optimize costs and continued improvement in our customer mix further supports margin expansion.

So GPU cloud takes a large and larger share of our total cloud infrastructure revenue, we believe the blended margins for cloud improved structurally. And that's a durable ongoing trend. So we're confident in the long-term profitability trajectory of our cloud business.

Your next question comes from Alicia Yap with Citigroup.

Also congrats on the solid quarter results. I have a question related to your foundation model. So how does Baidu view the positioning of ERNIE models in this increasingly competitive landscape? And looking ahead, what are your investment plans and the key direction for future model iterations?

Alicia, this is Robin. The model landscape is moving very quickly with active releases from players, both in China and globally. We believe model capabilities will continue to advance rapidly and strong in-house foundation model capabilities remain essential. So we will continue to invest in ERNIE with conviction. Meanwhile, we have always believed that the models ultimately create value through applications. That's why we have consistently taken an application-driven approach.

Each iteration of ERNIE is guided by real product needs and business scenarios. Most recently, we released ERNIE 5.1, which achieved leading results on LM Arenas, Text and Search leaderboard, demonstrating earnings continued progress in text capability, reasoning and search.

Going forward, we will continue to iterate ERNIE in line with the needs of our key applications that's AI search, digital humans, [indiscernible] Famou agent. These are among the application areas we believe how the greatest value. And our goal is to build the strongest capabilities where they matter most. For example, we will keep improving ERNIE's capability to understand user intent and assess content quality. So AI Search can deliver more accurate, higher quality and more intelligent results.

We also strengthened ERNIE's text and multi-model capabilities to make our digital humans more vivid and hyper realistic and more effective at engaging users and driving sales in e-commerce live streaming. We will enhance the coding capabilities to better support vibe coding, enabling users to build applications through natural language. And as coding becomes more -- becomes and increasingly conditional capability in the era, this will be a growing area of focus.

And we'll continue to strengthen ERNIE's capability to identify better and better solutions across complex real-world scenarios, helping enterprises in a wide range of industries achieve greater efficiency gains. To better understand -- to better support this direction, we have also made organizational adjustments to our model teams and will continue to evolve our structure as needed. We are confident that ERNIE will keep getting stronger across all of the areas.

Besides ERNIE, we also have a range of smaller, faster and more efficient models as well as model combinations optimized for specific scenarios. Different use cases have different requirements for capability cost, latency and deployment efficiency. Our goal is always to deliver the best outcome for each application. Over the longer term, we believe the fourth potential AI execution is still far from real life. As more AI use cases unfold, the value of our application-driven approach will become even clearer and ERNIE will become more capable and more valuable along the way.

The next question comes from Wei Xiong with UBS.

Congrats on very strong cloud momentum. I want to get your thoughts on the margin side. As AI Cloud Infrastructure revenue continues to grow rapidly and now with AI powered businesses accounting for over 50% of total revenue, how should we think about Baidu's long-term operating margin and the key drivers for margin expansion going forward?

Thank you, Wei. This is Henry. In Q1, as you noticed, our Baidu Cloud AI-powered business, which mainly includes business beyond traditional online marketing, already exceeded 50% of our total revenue for the first time. This is an important milestone reflecting both AI growing contribution and a more diversified revenue base. Many of these fast-growing businesses are still scaling and increasing. As it became a larger part of our revenue mix, we expect them to contribute not only to revenue growth but also to margin expansion, giving us multiple drivers for sustainable profit improvement over time down the road.

At this stage, we are investing in the most strategic opportunities with conviction. We care a lot about ROI of these investments. And I believe what we are building today will shape our margin structure for the years to come and create durable competitive advantages.

Let me walk through the key businesses where we see this playing out. First of all, as you mentioned, AI Cloud Infrastructure. The GPU Cloud is structurally higher margin than traditional CPU Cloud, driven by stronger demand, tighter supply chain, higher technical barriers and pricing power. So as it became a larger part of our mix, we expect it to be an important driver of margin improvement and expansion.

Second, AI applications. This is a naturally high margin business, driven by sticky and subscription-based models and operating leverage over time.

Third, robotaxi. Our unit economics has improved consistently since we have achieved breakeven in Wuhan City. We are still in the investment phase but the path forward to profitability is becoming clear as we scale up.

In addition, at the corporate level, a few additional levers worth highlighting today. First of all, we continue to drive cost optimization and operational efficiency across the entire organization. And second, we are deploying AI extensively to improve internal productivity as well. So not the least, third, on the infrastructure side, we are continuously improving server utilization rates which flows directly to the margin over time.

So as I mentioned in summary, our revenue mix is rotating towards higher-margin, faster-growing businesses, our full stack AI capability drives cost efficiencies and the company-wide productivity gains compounding over time. We believe the medium- to long-term margin trajectory is compelling for us, and we think it's sustainable. Thank you.

Your next question comes from Gary Yu with Morgan Stanley.

And congrats again on the strong AI Cloud Infra results. So my question is regarding robotaxi, management provide an update on your overseas robotaxi operations? And how should we think about the operating scale as well as the revenue mix between domestic and overseas for Robotaxi? How do we think about margin profile comparisons? And in the longer term, how does Baidu see its role in the robotaxi ecosystem as an operator, a technology provider or a platform?

This is Robin. First, on scale, apollo Go remains a global leader. We've completed over 22 million cumulative rises as of April. Chine is one of the most open market in the world. So it's natural that our domestic scale today is significantly ahead of overseas markets. But we are also seeing more markets globally opening up for robotaxi with regulatory environment turning more positive. We are really happy that our domestic operational experiences prepare us well for international expansion.

We have made a significant progress in a very short period. We only began accelerating our international expansion a few quarters ago, and our footprint has expanded across key markets in Europe, Middle East and Asia. That pace reflects the scalability of both our technology and our operations across different market environments. Our confidence in overseas expansion expected by the large-scale operational capabilities we've proven out in China. Through years of real-world fully drive operations, we have accumulated deep experience in complex road conditions, operational challenges and [indiscernible] that only emerged at certain scale.

And this is not something that can be built overnight. These experiences have continuously sharpen our algorithms and operational standards, making our robotaxi operation progressively more robust heavy mile driven. So when we expand globally, that accumulated experience travels with us and help us move faster. A good example is our progress from Hong Kong to London. Hong Kong has been an important right and dry global taxi market for us. Over the past year as we have accumulated valuable experience there and that know-how has helped to support our recent entry into London, another major right-hand drive market.

Regarding to profitability. Apollo Go has already achieved UE breakeven in its largest operational city in China despite very low fare levels. As we expand globally, the pricing environment becomes much more attractive, we believe our overseas operations have the potential to deliver much stronger profitability as they continue to ramp up. And the overall international market outside of U.S. and China, it's also bigger than China domestic market.

I think, finally, on our long-term role in the ecosystem, I think it's still early to call. The robotaxi industry is still evolving, and both the value chain and business models are still taking shape. What we're focused on right now is continuing to scale, deepening our technology and operational advantages and maintaining our global leadership. With that foundation in place, we'll have the strategic flexibility to define our role as the ecosystem mature and capture long-term value accordingly.

The next question comes from Thomas Chong with Jefferies.

And congratulations on a solid set of results. As AI investment continues to ramp up across the industry, how should we think about Baidu CapEx level in 2026? And how does management prioritize capital allocation between AI investment and shareholders' return? And finally, could management provide updates about the company's potential for dual primary listing in Hong Kong?

Thank you, Thomas. This is Henry. On the topic you mentioned on CapEx, our overall approach is to maintain strategic investment intensity while preserving financial discipline. AI remains Baidu's most important long-term opportunity, and we will continue to invest in foundation models to stay competitive at the frontier but also across our full stack AI capability more broadly. From a financial standpoint, we have the capacity to support this level of investment.

Thomas, as you noticed, our operating profit and operating cash flow remain healthy. With our total cash position also at a healthy level. Our operating cash flow for Baidu continue to be positive at RMB 2.7 billion in Q1 this quarter. The third consecutive quarter since turning positive in Q3 last year. And also meanwhile, we are drawing on a mix of financing channels and different instruments, including operating leases, financial leases and other low-cost bank borrowings as well to fund our AI investments while remaining and maintaining a sound cash position, a healthy balance sheet and a long-term financing structure.

And also, as you mentioned, our shareholder returns, it is also our priority and we still keep as it is. Last quarter, we announced a new buyback program and also introduced the first dividend policy. We will continue to balance long-term AI investment with shareholder returns, and we remain committed to creating sustainable value for our shareholders.

The last point regarding the Hong Kong dual primary listing, as you mentioned, we continuously evaluate initiatives that would help unlock long-term value for the company, including capital market initiatives. We believe Baidu's underlying value is substantial and we will be flexible and proactive in exploring ways to surfacing it. We will consider market conditions, regulatory requirements and shareholder interest, and we will communicate with the market when there are meaningful development.

Your next question comes from Miranda Zhuang with Bank of America Securities.

Congrats on the very strong results. So my question is about the AI agent. So as we are quickly moving into the agentic AI era, how to think about Baidu's strategy for AI applications and agents? What's management's view on the monetization methods, for example, like the token-based pricing or subscription-based model or project-based model?

Miranda, this is Robin. As I talked about at the Baidu Create conference last week, over the past 3 years, the biggest AI moment were driven by model breakthroughs. But this year, it's different. For the first time, it is an agent, an AI application that has captured the world's attention. That shift validates something we've long believed. In the AI era, value ultimately gets realized at the application layer. AI applications have always been a strategic priority for us. And we have built a portfolio around build user needs and business scenarios, spanning AI native products and AI transformations of our existing products, serving individuals, enterprises and industry verticals alike.

Within this portfolio, there are several directions we see particularly valuable. The first is search, our largest consumer-facing product and one we have been consistently transforming with AI. We are pioneering the AI Search experience globally, bringing the latest AI capabilities to our hundreds of millions of users and enabling search to deliver more intelligent, structured and generally helpful answers at scale.

And the second is digital humans. We continue to push towards higher [indiscernible] lower cost and greater scalability. As costs come down, digital humans are becoming much more accessible to merchants and can be deployed at larger scale. In e-commerce live streaming, they are proving increasingly effective at driving engagement and conversion with performance in many cases, comparable to or even better than human host.

And third one, Miaoda, our vibe coding product, as its capabilities evolve, Miaoda supports a broader range of applications and workflows, lowering the barrier for users and enterprises to build AI applications through natural language. And fourth is Famou agents focused on enterprise scenarios. It helps enterprises navigate complex dynamic real-world environment, continuously evolving to identify better solutions and drive meaningful efficiency gains. And as it handles more and more complex scenarios across more and more industries, it keeps getting better.

So for this direction, I think that the first direction is clearly more advertising-based and the other 3 directions is more subscription or usage-based or token-based pricing. And I think for search, it's more of a consensus that it's a meaningful market for AI native applications. And also for Miaoda, the vibe coding product, they are also comparable ones outside of China. But for digital humans and Famou agent, these are not consensus [indiscernible] our own belief, we think they both represent huge market potential, but not many companies putting into enough resources and efforts to do this kind of things.

And beyond this, we are also exploring new AI native product forms. And DuMate is one recent example. It is a general purpose agent. It brings AI into the workloads and tasks people deal with every day. It is [indiscernible]. It remembers a lot of things about you. So it's going to be quite sticky application, we think. And on monetization, the industry is at a very early stage globally and the monetization models are still evolving.

Today, token-based pricing is more common. People are essentially paying for foundation models. But over time, AI applications and agents will become more capable of completing real cash like a human being. We believe monetization will become broader and more results-oriented. That means users are paying for productivity gains, time savings, better experiences and tangible results. So in the future, people will pay for agents or applications and the market for this should be much larger than token.

Your next question comes from Lincoln Kong with GS.

Congrats on solid results. I wonder could management share your view on the growth outlook and competitive landscape for the domestic AI chips in China? So how is Kunlunxin positioned within this market? And what recent demand trends are we seeing actually?

Thank you for your question. This is Dou. I think China's domestic homegrown AI chip market is still early. But moving fast. We're seeing a structural shift in AI compute demand from [indiscernible] to a growing mix of inference. As agentic applications, industry-specific use cases and new forms of applications continue to emerge, inference is becoming a growing part of the picture. The open client is a good example, driving a wave of inference demand that is higher frequency, more real time and more endeavors. On supply, domestic AI chips still face near-term challenges around the capacity and the supply chain maturity. [indiscernible] demand is growing faster than supply.

But over the long term, China's semiconductor industry is developing quickly, but supported by a strong manufacturing and supply chain foundation. We believe domestic supply capabilities will keep improving over time. If that happens, competition, we will increasingly depend on not just delivering chips, but whether those chips can perform reliably and efficiently across endeavors real-world workloads. Domestic chips [indiscernible] still catching up with the most advanced global products in certain frontier training scenarios.

The inference is an area where domestic chips can be highly relevant and competitive. As inference continues to grow, that advantage becomes increasingly relevant. For enterprises, it's not only about the peak chip performance. What matters more is the stability at scale, compatibility with mainstream models and frameworks migration cost and friction support for our large-scale cluster deployment and ultimately, cost efficiency. So we think the market will increasingly consolidate around players who can deliver on all of these dimensions.

Kunlunxin is well positioned on each of these brands. Furthermore, Kunlunxin is not just a stand-alone chip product. It is a critical part of Baidu's full-stack AI capabilities. Spanning from infrastructure to applications, we can continuously optimize across the intersect improving model efficiency, reducing inference costs and delivering AI infrastructure that is more cost effective, stable and easier to deploy. We are seeing a strong and growing customer demand for Kunlunxin with adoption expanding across industries. So we believe Kunlunxin is well positioned to capture the opportunities ahead.

There are no further questions. at this time, and that does conclude our conference for today. Thank you for participating. You may now disconnect.

Hello, and thank you for standing by for Baidu's Fourth Quarter and Fiscal Year 2025 Earnings Conference Call.

[Operator Instructions]

Today's conference is being recorded. If you have any objections, you may disconnect at this time. I would now like to turn the meeting over to your host for today's conference, Juan Lin, Baidu's Director of Investor Relations.

Hello, everyone, and welcome to Baidu's Fourth Quarter and Fiscal Year 2025 Earnings Conference Call. Baidu's earnings release was distributed earlier today, and you can find a copy on our website as well as on Newswire services. On the call today, we have Robin Li, our Co-Founder and CEO; Julius Rong Luo, our EVP, incharge of Baidu Mobile Ecosystem Group, MEG, Dou Shen, our EVP in charge of Baidu AI Cloud Group ACG; and Henry Haijian He, our CFO.

After our prepared remarks, we will hold a Q&A session. Please note that the discussion today will contain forward-looking statements made under the safe harbor provisions of the U.S. Credit Securities Litigation Reform Act of 1995. Forward-looking statements are subject to risks and uncertainties that may cause actual results to differ materially from our current expectations. For detailed discussions of these risks and uncertainties, please refer to our latest annual report and other filings with the SEC and Hong Kong Stock Exchange.

Baidu does not undertake any obligation to update any forward-looking statements, except as required under applicable law. Our earnings press release and this call includes discussions of certain unaudited non-GAAP financial measures. Our press release contains a reconciliation of the unaudited non-GAAP measures to the unaudited most directly comparable GAAP measures and is available on our IR website at ir.baidu.com.

As a reminder, this conference is being recorded. In addition, a webcast of this conference call will be available on Baidu's IR website. I will now turn the call over to our CEO, Robin.

Hello, everyone. In Q4, Baidu General business total revenue was RMB 26.1 billion. Revenue from our core AI power business exceeded RMB 11 billion, accounting for 43% of Baidu General business revenue in AI Cloud Infra, subscription-based revenue from AI accelerator infrastructure grew 143% year-over-year accelerating further from 128% in Q3. Meanwhile, Apollo Go maintained its robust momentum delivering 3.4 million fully driverless operational rise in the quarter.

Total rise increased by over 200% year-over-year. 2025 marked the third year of our journey in GenAi and the pivotal year, where AI became the new core of our portfolio. In 2025, we made substantial progress in scaling AI across our businesses, accelerating AI cloud growth, expanding robotaxi operations with improved unit economics and deepening AI integration into our mobile ecosystem. Looking at our portfolio through an AI-native lens momentum across our core AI-powered businesses continue to build in 2025. AI cloud infra gained strong traction through its highly efficient and cost-effective training and inference capabilities.

Revenue from AI Cloud Infra reached approximately RMB 20 billion in 2025, up 34% year-over-year, outpacing industry growth. Our AI application portfolio is among the most comprehensive in the industry, combining AI-empowered flagship products with AI native offerings that unlock entirely new use cases. For the full year 2025, revenue from AI applications exceeded RMB 10 billion. Apollo Go achieved a significant landmark -- we delivered over 10 million fully driverless operational rides in 2025 alone.

To date, we have provided a total of over 20 million rise to the public cumulatively with our accelerated global expansion. Apollo Go's footprint has now reached 26 cities worldwide, reinforcing our leadership in autonomous ride-hailing services. Lastly, our AI native marketing services, including digital humans and agents, sustained strong growth with revenue up 110% year-over-year. Collectively, these results demonstrate AI's growing contribution to Baidu's value creation and our ability to translate AI capabilities into scalable commercial impact.

Now let me share the key highlights of the quarter, starting with our proprietary AI chips. This quarter, we announced the proposed spin-off and separate listing of [indiscernible] after more than a decade of steadfast investment in self-developed AI chips, we are proud to see the market increasingly recognize their value and proven performance. This milestone validates our long-term strategic vision and unlocks new opportunities for value creation. Our AI chips are built on a proprietary architecture developed in-house from day one. They deliver stable, high-performance AI computing at scale with broad compatibility across different models and frameworks.

This enables customers to deploy faster with lower integration costs. What distinguishes our AI chips is a proven track record of large-scale real-world deployments with leading enterprises across diverse industries, spanning financial services, telecommunications, energy and Internet sectors. Customers choose our chips for reliable performance, stable supply at scale, exceptional software compatibility and strong efficiency especially in inference workloads.

Looking ahead, we see significant opportunities for both Baidu and Kunlunxin infrastructure demand continues to accelerate. Next, I will turn to our AI Cloud Infrastructure. Our infrastructure is among the most advanced in China powered by a diverse mix of domestic and international high-performance computing resources. In Q4, subscription-based revenue from AI Accelerator Infrastructure grew 143% year-over-year, achieving triple-digit growth for the full year 2025.

Importantly, we saw a continued shift toward a more recurring, structurally healthier revenue model. The robust growth was fueled by rapidly expanding enterprise AI adoption. As customers integrate AI into core operations, the unique value of our full stack end-to-end AI architecture becomes increasingly evident by owning and optimizing across all 4 layers we achieved sustained advantages in stability and cost effectiveness, better addressing enterprises needs for AI deployment.

These advantages are translating into tangible market momentum fueling accelerated adoption of our AI cloud Infra. In Q4, we further broadened our client reach. Leading enterprise clients deepen their partnerships with us, driving increases in both usage and spending. We also saw healthy growth contribution from our mid-tier clients. We continue to strengthen our presence in diverse industries like Internet services, gaming, autonomous driving and embodied AI, underscoring the versatility of our infrastructure.

Embodied AI, in particular, showed notable momentum. Revenue from this vertical doubled quarter-over-quarter in Q4, we onboarded a new wave of leading humanoid robotics companies cementing our position as the go-to cloud service provider for China's fast-growing embodied AI industry. Next, I'll cover our foundation model progress, which is a critical part of our AI capabilities. We remain fully committed to advancing our proprietary foundation model, ERNIE. Following the unveil of ERNIE 5.0 last quarter, we launched an updated version in January. As we advance ERNIE, we remain guided by a clear application-driven approach, making ERNIE strongest where it matters most for our portfolio.

To execute this approach more effectively, we recently restructured our model development organization into 2 dedicated teams. One team advances early state-of-the-art foundation model capabilities, maintaining our technological edge in this fast evolving space. The other team tailors models for specific business needs reducing costs, improving response latency and optimizing model size and efficiency to ensure our technologies are not just cutting edge, but readily scalable across our businesses. Close collaboration between both teams ensures our technologies stay grounded in real-world needs while our applications benefit from continuous technological advancement.

Now turning to AI applications. This is where we believe AI's greatest value will ultimately reside. We are pioneering AI applications to solve complex real-world problems for both individuals and enterprises. Let me share our progress across multiple key areas, starting with AI-powered search. In Q4, we continued our AI search transformation, pursuing one of the most comprehensive and ambitious transformation globally. Our focus remains on continuously improving the quality of AI search results while expanding what users can accomplish directly within search.

This quarter, for example, we introduced AI generated infographics into our search results, realizing tax-based information where appropriate to make key insights immediately clear and digestible. We've also integrated more MCP capabilities across key scenarios, including e-commerce health care and local services. This enables actions such as shopping, booking and health care consultation to be completed seamlessly within the search experience.

During the Chinese New Year, we moved quickly to embrace the latest AI agent innovation by integrating Open Claw, a recently popular open source agent framework directly into Baidu app with 1 click access, enabling our users to immediately benefit from cutting-edge agentic AI capabilities with an MAU of around $700 million. We provide easy access to Open Claw for almost half of the Chinese population.

For ERNIE assistant, which is the AI chatbot integrated across our platform, we enhanced the user experience by introducing broader multimodal capabilities. This improvement have been well received by users driving ERNIE assistance MAU to exceed 200 million in December. We are also scaling our AI search API, adoption has accelerated in Q4 and with core volume up over 110% quarter-over-quarter with industry-leading authority, comprehensiveness and newly added multilingual capabilities our AI search API is now opening up broader possibilities for the international market. Next is digital humans, which represent a compelling form of AI application. They combine visual presence voice and real-time interaction to create more engaging and effective experiences.

In December 2025, the number of digital humans live streaming on our platform increased nearly 200% year-over-year. Beyond Baidu's own platforms, our digital human technology is expanding to empower the broader industry. Leading companies have partnered with us including Jingdong, Toba and TikTok, validating the performance and efficiency of our digital humans.

On the technology front, we believe our hyper realistic digital human represents the next generation of capabilities. This quarter, production costs declined to roughly 1/3 of previous quarter levels, bring industry leading cost performance and positioning this technology for broader adoption. Another area of progress is Miaoda, our vibe coding platform, which enables users without coding experience to build applications through natural language, including WeChat mini programs, websites, mini games and more.

Following the Q4 launch of Midas international version, Miaoda, users globally have created over 1 million AI applications as of early February, all without writing a single line of code. Looking ahead, we see meaningful opportunities to unlock even greater possibilities in AI application development. Lastly, we are using AI to solve operational problems and drive efficiency gains across the industries.

One example is EDN, our advanced visual intelligence platform. EDN enables enterprises to automate operational compliance and safety checks through intelligent visual analysis, while known brands across coffee chains, quick service restaurants and fine dining are now using EDN to ensure high standard operations across their thousands of locations. Another example is FM agent, our self-evolving agent designed to solve complex operational challenges by autonomously reasoning across data, rules and real-world constraints, it simulates countless scenarios to identifying best solutions.

We've seen strong validation both internally through our own cloud resource optimization and externally across industries like manufacturing, energy, finance and logistics, where efficiency improvement is a universal priority. On the organizational front, we recently established the Personal Super Intelligence business group or PSIG. PSIG unified Baidu Wenku and Baidu Drive our 2 flagship consumer-facing AI applications. Even before this organizational integration, the 2 teams had already collaborated at the product level to deliver innovations like free chemists and gemflow. This new group enables even deeper collaboration going forward. as we accelerate the rollout of new applications to foster a robust growth curve, driven by application layer innovation.

Shifting to physical AI, Apollo Go represents our largest AI application in the physical world. 2025 was a year of accelerated scaling for Apollo Go where we reinforced our leadership in operational scale and achieved significant progress in global expansion. We continue to expand fully driverless operations at pace delivering 3.4 million fully driverless operational rise in Q4 with weekly rise peaking at over 300,000. Total rights grew by over 200% year-over-year. Cumulative rights provided to the public have surpassed 20 million as of February 2026, firmly cementing our position as the world's leading autonomous ride-hailing service provider.

We entered 2026 with momentum across key international markets. In the U.K., we advanced our partnerships with Uber and Lyft moving forward with plans to pilot autonomous vehicles in London with testing expected to begin in the first half of 2026. This represents an important step in Apollo Go international expansion, extending our right-hand drive robotaxi capabilities from Hong Kong to another strategically important market. In Switzerland, we initiated testing in St. Gallen following our market entry last quarter.

In the Middle East, we achieved progress in both Abu Dhabi and Dubai. In Abu Dhabi, we launched a fully autonomous ride hailing services on Yas Island in January with Auto Go. In Dubai, we secured the city's first fully driverless testing permit from the road and transport authority. We also announced the next phase of our global partnership with Uber to bring our fully autonomous ride hailing services to Dubai via the Uber platform. These are critical milestones that accelerate our progress across the Emirates. In Asia, we entered a new market, South Korea, starting with the sole metropolitan area, further expanding our presence across the Asian region.

Meanwhile, in Hong Kong, we expanded our open road testing into Sunwin and initiated cross-district testing between Airport Island and Tongcheng, bringing us closer to commercial readiness there. As of February 2026, Apollo Go's global footprint reached 26 cities, demonstrating the scalability of our autonomous driving technology across diverse regulatory and operational environments.

Looking ahead, we are focused on accelerating expansion to more cities globally while continuously improving operational excellence and unit economics. Our growing experience across diverse markets gives us confidence in our ability to scale further and we expect more cities to achieve positive unit economics over time. Underpinning this expansion, safety remains our top priority and the foundation of everything we do. Our autonomous ride hailing service. It's the safest globally with our fully driverless vehicles, experience and airbag deployment accident, only once every over 12 million kilometers. As we scale, we will continue strengthening safety standards and ensure sustained reliability.

Ultimately, our mission is to harness AI to transform mobility, making it fundamentally safer, more affordable and more comfortable and improving how billions of people move, work and live. In summary, with AI now firmly integrated across our portfolio, we believe we are well positioned to deliver sustainable value and shape the next phase of the AI era.

With that, let me turn the call over to Haijian He to go through the financial results.

Thank you, Robin, and hello, everyone. We are making progress on our key focus areas. Over the recent quarters, we've enhanced disclosure for greater transparency and driven operational efficiency improvements. This quarter, we took a significant step to unlock value from our strategic AI chip investments through the proposed Kunlunxin spin-off and separate listing a milestone we are particularly pleased with. We've also announced a new USD 5 billion share repurchase program and adopted a dividend policy for the first time.

Additionally, we've sharpened our strategic focus on high potential AI applications by forming a PFIG business group, integrating Baidu Wenku and Baidu Drive. These actions reflect our consistent execution and ongoing focus on creating shareholder value. Looking at Q4 results, we saw positive momentum. Baidu General business total revenue increased 6% quarter-over-quarter with non-GAAP operating profit, expanding 28% sequentially to RMB 2.8 billion, Operating cash flow for Baidu turned positive in Q3 and remained positive in Q4, generating a combined RMB 3.9 billion across both quarters.

In terms of our core AI-powered business, in Q4, revenue exceeded RMB 11 billion, accounting for 43% of Baidu general business revenue. We are seeing strong momentum across several areas. AI Cloud Infra continues to gain market traction and outpace industry average. Our AI application portfolio is expanding rapidly with strong enterprise adoption. Combining AI Cloud Infra and AI applications, our coud revenue reached RMB 30 billion for the full year 2025. Meanwhile, Apollo Go reinforces its position as a global leader in autonomous ride-hailing with one of the industry's largest footprints and the strongest growth momentum.

And AI native marketing services is growing fast. These results demonstrate our progress, and we believe this is just the beginning. We have a robust pipeline of initiatives ahead, and we are confident in our ability to create lasting shareholder value. Now let me walk through the details of our fourth quarter and full year 2025 financial results. Total revenues in Q4 were RMB 32.7 billion, increasing 5% quarter-over-quarter, primarily due to an increase in Baidu core AI-powered business. Total revenues for the full year 2025 were RMB 129.1 billion, decreasing 3% year-over-year. primarily due to a decrease in legacy business, partially offset by an increase in Baidu core AI-powered business.

Cost of revenues was RMB 18.3 billion in Q4, which remained flat quarter-over-quarter. Cost of revenues was RMB 72.4 billion in 2025, increasing 10% year-over-year, primarily due to an increase in costs related to Baidu core AI-powered business. Operating expenses were RMB 13.0 billion in Q4, increasing 10% quarter-over-quarter, primarily due to an increase in expected credit losses and a onetime employee severance costs to improve efficiency.

Operating expenses were RMB 46.3 billion in 2025, increasing 1% year-over-year. Impairment of long-lived assets was RMB 16.2 billion in 2025, attributable to an impairment loss of core asset group. Operating income was RMB 1.5 billion in Q4, and operating margin was 5%. Operating loss was RMB 5.8 billion in 2025 and operating loss margin was 5% excluding impairment of long-lived assets, operating income was RMB 10.4 billion in 2025. Non-GAAP operating income was RMB 3.0 billion in Q4. and non-GAAP operating margin was 9%. Non-GAAP operating income was RMB 15.0 billion in 2025, and non-GAAP operating margin was 12%.

In Q4, total other income net was RMB 1.2 billion compared to RMB 1.9 billion last quarter. Income tax expense was RMB 1.0 billion compared to income tax benefit of RMB 1.8 billion of the quarter. In 2025, total other income net was RMB 12.5 billion compared to RMB 7.4 billion in the same period last year. Income tax expense was RMB 1.3 billion compared to RMB 4.4 billion in the same period last year. In Q4, net income attributable to Baidu was RMB 1.8 billion, net margin for Baidu was 5% and diluted earnings per ADS was RMB 3.71.

Non-GAAP net income attributable to Baidu was RMB 3.9 billion, non-GAAP net margin for Baidu was 12%, and non-GAAP diluted earnings per ADS was RMB 10.62. In 2025, net income attributable to Baidu was RMB 5.6 billion, net margin for Baidu was 4% and diluted earnings per ADS was RMB 11.7%, excluding the impact of impairment of long-lived assets. Net income attributable to Baidu was RMB 19.4 billion. Non-GAAP net income attributable to Baidu was RMB 18.9 billion. Non-GAAP net margin for Baidu was 15% and non-GAAP diluted earnings per ADS was RMB 53.41%. 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 December 31, 2025, Total cash and investments were RMB 294.1 billion. In Q4, operating cash flow was RMB 2.6 billion. In 2025, operating cash flow was negative RMB 3.0 billion, which remained positive for the past 2 consecutive quarters. Baidu General business had approximately 29,000 employees as of December 31, 2025.

With that, operator, let's now open the call to questions.

[Operator Instructions]

Your first question comes from Alicia Yap with Citigroup.

Robin, Henry, Jan.

Have questions related to the model. So we have noticed very active model iteration recently. How does management view the current competitive landscape? And then Baidu recently also released updated ERNIE 5.0 and also make some organizational adjustments. So could management discuss the strategic rationale behind these moves and also how the company thinks about the relationship between the model evolution and also the application in your overall AI strategy.

Alicia, this is Robin. We did see very active model releases recently. The market is highly competitive and moving fast. But amid all the competition, we've always believed that applications matter more than models because models ultimately create value through applications. That is why we always take an application-driven approach with ERNIE. Model improvements are guided by the most valuable and promising use cases. And this has been consistent across every iteration of ERNIE. As I just mentioned, recently, we released the updated version of ERNIE ERNIE 5.0. At the same time, we've been proactively making organizational changes to stay agile in the fast-moving market.

We restructured our model team into different focus areas. One team continues pushing frontier capabilities at the foundation model level to maintain technical leadership. ERNIE has clear strength in several key areas such as creative writing, omniodel understanding and instruction following. We are confident we will keep improving ERNIE's performance across key application scenarios. Meanwhile, this high-value application scenarios continuously provide ERNIE with real data and feedback, driving model iteration and making ERNIE better and better.

The other team works much closer to specific business needs and application scenarios focused on reducing costs, improving speed and increasing efficiency or leveraging the best available models for specific use cases, all aimed at helping businesses better leverage AI based on their actual needs. We recognize that model capabilities are broad and application scenarios can be highly diverse. And no single model can lead everywhere. So we fully leverage ERNIE where it has clear strengths, and we are open to using other models where they are better suited. The goal is always to achieve the best application outcomes. So to sum up, we will continue with our application-driven approach using real application needs to continuously iterate and optimize our models while also keep refining applications themselves to deliver better and better results, ultimately, creating tangible value for users and businesses.

Your next question comes from Alex Yao with JPMorgan.

I have one question about the Baidu AI Cloud. We noticed that Baidu AI Cloud revenue delivered strong growth for the full year 2025. Can you elaborate and help us understand the key growth driver behind the robust revenue growth number? And how should we think about the AI cloud revenue growth outlook in 2026?

Thank you, Alex. This is Dou. For 2025, our AI cloud revenue, which includes revenue from AI cloud infra and AI applications, reached RMB 30 billion. Revenue from AI cloud infra grew 34% year-over-year, outpacing the broader market. Within AI cloud infra, subscription-based revenue from AI accelerator infrastructure grew 143% year-over-year in Q4 and has become the primary growth driver, demonstrating strong momentum. We remain highly confident in sustaining strong growth momentum in 2026. Underpinning our growth is the accelerating enterprise AI adoption. We are seeing demand growth in both training and inference workloads, and we expect the demand for AI computing to keep expanding, creating significant opportunities ahead.

Baidu's full stack end-to-end AI architecture is a key differentiator in capturing such opportunity. At the foundation of this architecture is our industry-leading AI infrastructure, which achieves an excellent balance across performance, efficiency and cost. Our AI infra is powered by a diverse mix of chips. We have built deep expertise in heterogeneous computing and unified scheduling. -- which enables us to efficiently manage computing resources from different chip vendors and achieve industry-leading performance and efficiency. In the meanwhile, our proprietary chip capabilities provide a significant competitive advantage. As Robin just mentioned, our self-developed Kununqin AI chips deliver strong performance, compatibility and cost efficiency.

They have been deployed at scale with leading enterprise customers across financial services, telecom, energy and Internet sectors and the market feedback has been very positive. serves as a key component of our own cloud platforms computing power, playing an important role in our overall AI infra. As AI demand grows, the advantages of our AI infra will become increasingly evident. Beyond AI infra we just discussed, we are continuously evolving our best-in-class agent infra to help enterprises rapidly build and deploy AI agents at scale.

We keep bringing in the latest, most cutting-edge capabilities. For example, we recently launched simplified open cloud deployment on Baidu AI Cloud, which streamlines the process so that even users with no coding experience can quickly deploy their own open cloud agents. Then looking into 2026, as enterprise AI deployments deepen further, we are confident that our cloud business will continue to grow faster than the industry. We expect AI cloud infra to maintain strong momentum with AI accelerator infrastructure continuing to serve as a core driver, propelling our overall cloud business toward a more sustainable and high-quality growth mode.

Your next question comes from Lincoln Kong with GS.

So actually, this quarter, we see this AI-powered business continue to deliver a pretty solid growth. So how does management view the current stage development for those AI-powered business? So when should we expect this share to exceed, say, 50% of the Baidu General business? And what will be the key driver going forward for the AI-powered business?

Okay. Let me start by sharing how we think about our core AI-powered business. This includes AI cloud infrastructure, AI applications like Baidu Wenku and Baidu Drive and our robotaxi business, Apollo Go and our AI-native marketing services, including agents and digital humans. AI-powered business organizes our business according to the nature of our products and services, where AI is empowering each to create meaningful customer value and business impact. In Q4, AI-powered business revenue exceeded RMB 11 billion. That's like 43% of Baidu Chanel business revenue. This percentage has been rapidly increasing over the recent quarters, and AI-powered business is becoming the core driver of our overall revenue growth.

Each of our AI-powered businesses has clear strategic positioning and competitive advantage. First, it's AI -- AI cloud infra. We see enterprises scale AI from pilots to production and our full stack end-to-end AI capabilities enable strong performance at competitive cost. AI cloud infra revenue grow faster than the industry average in 2025 with subscription-based AI accelerator infrastructure revenue accelerating sharply in Q4. And second, it's AI applications. We've always believed AI's ultimate value will settle at the application layer, and we built one of China's most comprehensive AI application portfolios. As AI capabilities continue to evolve and new use cases emerge, we see significant expansion potential in this business. And then third is the robotaxi business, Apollo Go.

Apollo Go is scaling rapidly while expanding internationally. We lead globally in operating scale, safety record, efficiency and cost structure. And the fourth is AI native marketing services like agents and digital humans. They improved engagement and conversion, and we're seeing strong market adoption with great potential ahead. So looking to the mid- to long term, as enterprise AI deployment deepens, monetization capabilities of AI applications improve and physical AI applications such as autonomous driving continue to expand, and we're confident in the growth trajectory of our AI-powered business. This AI-powered business aren't isolated. They continuously reinforce each other through our full stack capabilities. And based on current visibility, we believe our core AI-powered business will become the majority of Baidu general business in the foreseeable future.

Your next question comes from Wei.

Could management elaborate on the framework that you use to allocate capital, including shareholder returns, organic investment and potential strategic opportunities? And also, could management comment on the long-term strategic positioning of Kun in within the Baidu Group?

Thank you,. This is Henry. I believe many of you may have noticed our recent series of initiatives. These include enhancing our disclosures, improving operational efficiency, optimizing our organizational structure, advising the proposed -- advancing the proposed Kunxin spin-off and separate listing and also announcing our new share repurchase program and the first dividend policy. And recently, we're also reforming the PSIG, the Personal Super Intelligent Group Business Group, integrating Baidu One Ku and Baidu Drive.

Altogether, these moves reflect a coherent execution framework, demonstrating our improved management execution and ongoing commitment to creating shareholder value. I think take our new share repurchase program as one example. We are very focused on providing clear and sustainable returns to shareholders. So in the recent -- in February, the Board has approved a new USD 5 billion share repurchase program, which we plan to execute on a regular basis in a very disciplined and transparent manner. We are also introducing Baidu First dividend policy.

We believe the introduction of the policy alongside with a sizable buyback program will further strengthen our shareholder return profile and attract a broader range of investors, thereby further diversifying our investor base. As you mentioned, the proposed spin-off and a separate list of Kunlunxin is another good example. We are making very good progress of the listing process. Kunlunxin is a result of over a decade of investment and represents a critical infrastructure component of our full-stack AI capabilities. We believe this spin-off and a separate listing will receive strong market recognition and unlock significant value for Baidu as a Group.

So looking ahead, we firmly believe the company has tremendous value, and we will continue unlocking it through various initiatives. We remain committed to deliver sustainable and consistent returns to our shareholders. So more initiatives will follow in due course. So stay tuned with us. Thank you.

Your next question comes from Gary Yu with Morgan Stanley.

My question is on robotaxi. And first of all, congratulations on the robotaxi expansion into more countries, especially to my hometown Hong Kong. Can you share your overseas strategy in 2026? And what are your key competitive advantages there? And also with Wemo recently valued at $126 billion, how is management thinking about unlocking Apollo Go's value? Would you consider a spin-off?

Gary, as I mentioned last quarter, I believe robotaxi has reached a tipping point globally. Through continuous delivery of safe autonomous drives and positive word of mouth, we're seeing more countries and regions creating supportive environment for robotaxi operations. We believe the industry will accelerate in 2026. Apollo Go is a clear global leader in this space. We've completed over 20 million cumulative rides. At peak period, our weekly fully driverless rides exceeded 300,000.

To date, Apollo Go fleets have accumulated more than 300 million autonomous kilometers, including over 190 million fully driverless autonomous kilometers with an outstanding safety record. And we continue advancing our industry-leading technology to make rides safer and more comfortable. We're also accelerating international expansion to capture global opportunities. Today, our global footprint spans 26 cities across different continents, covering both left-hand and right-hand drive robotaxi market.

Our autonomous driving system works reliably anywhere across different traffic patterns and different urban environments. Notably, very few players have entered right-hand drive robotaxi market, while we've already established a presence and are making rapid progress. Moreover, we have a fundamental cost advantage. RT6 is the world's first purpose-built production vehicle designed from the ground up for Level 4 autonomous driving. at under USD 30,000 per vehicle. RT6 offers the industry's best cost structure and combined with our leading operational efficiency, this enables us to achieve the lowest cost per mile globally while maintaining superior safety.

We were the first to achieve UE breakeven in Wuhan in late 2024. And as you know, most major cities have higher ride-hailing prices than Wuhan. To accelerate global expansion, we are leveraging diverse strategic partnerships -- for example, we are collaborating with Uber and Lyft in London to launch this year and in Dubai with Uber also. These partnerships drive faster, more efficient market expansion. We see Apollo Go as a strategic growth engine with significant long-term potential. Many major cities are short of human drivers.

More supply via robotaxi service not only offer safer rides, but also stimulate ride-hailing demand, therefore, add tax revenue to the government. It also releases precious land from parking spaces and provide additional monetization opportunities for these real estate assets. Our focus 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. As for strategic options, we will remain flexible and evaluate the best path that maximizes long-term shareholder returns. And of course, our focus is always on execution and sustainable growth. We believe the autonomous ride-hailing sector as a whole remains undervalued. Over time, we expect valuations across the sector to better reflect the transformative potential of this technology, which creates meaningful upside opportunity for Apollo Go.

Next question comes from Miranda Lang with Bank of America Securities.

Thank you, Management. Wish you a happy year of half -- so my question is about competition. We have seen that the consumer-facing AI competition is intensifying recently, especially during the Chinese New Year. How do you assess the current competitive dynamics? Where do you see by dose AI2C products such as the early assistance, differentiation and also positioning in this market? And lastly, how to think about the path to monetization?

This is Julie. The AI 2C product market is highly competitive. We have seen some competitors [indiscernible] very aggressive market strategies to rapidly scale their user base in the past 2 new year. However, as technology and products evolve rarely, we still believe our core strategy should remain grounding in actual user needs. We are highly committed to continuously enhancing our existing products and services capabilities through AI innovations to better serve our users.

In our flagship consumer-facing products Baidu app today, we have built the earning assistant to strengthen our service capabilities across the entire user journey. From information thinking to providing solutions and completing tasks. On information thinking we have significantly enhanced our users assess information through earnings assistance. For example, we have improved the answer accuracy and relevance and earning assistant maintains low ad rates with minimal locations, delivering the highly trusted is content to users. We have also integrated the multilingual AI such API capabilities that can enable the users to assess the broad information sources during conversations, improving the information richness and usability.

And especially for scenarios like travel planning, which is quite helpful. And in December, earnings assistance, MAU surpassed 200 million and with conversation loans and engagements are growing quite fast. For past complexion, we are integrating the MCP agents to connect users with tools and real word services. This quarter alone, we're adding nearly 100 service capabilities, especially in health care, travel, education and e-commerce. For example, through the Baidu Health MCP integrated into earning assistant, users can assess a range of health care capabilities, spending online to our plan services.

In e-commerce, our MCP module saw a very strong GMV growth quarter-over-quarter. Meanwhile, we are taking a different approach with the stand-alone ERNIE app. Our positioning as a platform for innovation and experimentation. Our earlier multi-model AI features have gained good traction with the young audiences. And more recently, we have added AI capabilities focused on the workplace productivity, tapping into earning book ability to handle the complex tasks in professional [indiscernible]. We are seeing the promising early signals in these productivity scenarios.

We take a measured approach to monetize the AI products, prioritizing the product excellence and the user experiences. Monetization will follow naturally as the products mature. Thank you for your question.

Your next question comes from Ellie Jiang with Macquarie.

My question is mostly focusing on the AI investment. How do you think about the AI-related CapEx over the next 12 to 24 months? How should we think about the return profile of these AI investments and the expected impact on the ROIC over time? Broadly speaking, where do you see further efficiency opportunities to support margin and cash flow improvements in the future? Thank you.

This is Henry. First of all, on CapEx and AI investment, since we have launched early in March of 2023, we have invested over RMB 100 billion in AI. Going forward, we will continue to maintain this level of the investment density. Second, we are very conscious about returns and understand investors' focus on the return on capital invested. That's why we have work to improve our financial performance, and we have delivered good results on key metrics over the past few quarters. For example, in Q4, gross profit for Baidu grew double digits sequentially. -- and non-GAAP operating income for Baidu increase about 35% quarter-over-quarter. We also performed better on the margin profile, both on market and leading margin increase.

Importantly, operating cash flow for Baidu turned positive in Q3 and remained positive in Q4. With the second half, operating cash flow reached nearly RMB 4 billion. Free cash flow for Baidu also turned positive in Q4. Thirdly, we have also found and explored alternate ways of supporting our financial needs including, for example, operational and financing leasing as well as we have access to the low-cost interest banking borrowing. For example, some of these bank borrowings and the leasing facilities carry the interest rates as low as below 2%.

Though these approaches help us maintain a healthy long-term financing structure while sustaining our AI investments and support our business growth. So in summary, we will continue to maintain our AI investment density. -- while balancing investor focus on profitability and return time lines, we believe that even with significant AI investment, our operating cash flow remain positive going forward as well. Thank you.

Thank you. Ladies and gentlemen, that does conclude our conference for today. Thank you for participating. You may all disconnect.

Hello, and thank you for standing by for Baidu's Third Quarter 2025 Earnings Conference Call. [Operator Instructions]

Today's conference is being recorded. If you have any objections, you may disconnect at this time. I would now like to turn the meeting over to your host for today's conference, Juan Lin, Baidu's Director of Investor Relations.

Hello, everyone, and welcome to Baidu's Third Quarter 2025 Earnings Conference Call. Baidu's earning release was distributed earlier today, and you can find a copy on our website as well as on Newswire Services. On the call today, we have 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 He, our CFO.

After our prepared remarks, we will hold a Q&A session. Please note that the discussion today will contain forward-looking statements made under the safe harbor provisions of the U.S. Credit Securities Litigation Reform Act of 1995. Forward-looking statements are subject to risks and uncertainties that may cause actual results to differ materially from our current expectations. For detailed discussions of these risks and uncertainties, please refer to our latest annual report and our filings with SEC and Hong Kong Stock Exchange.

Baidu does not undertake any obligation to update any forward-looking statements, except as required under applicable law. Our earnings press release and this call include discussions of certain unaudited non-GAAP financial measures. Our press release contains a reconciliation of the unaudited non-GAAP measures to the unaudited most direclty GAAP measures and is available on our IR website at ir.baidu.com. As a reminder, this conference is being recorded. In addition, a webcast of this conference call will be available on Baidu's IR website.

I will now turn the call over to our CEO, Robin.

Hello, everyone. In Q3, Baidu Core reported total revenue of RMB 24.7 billion, AI Cloud revenue reached RMB 6.2 billion, increasing 21% year-over-year sustaining value growth momentum. Apollo Go growth accelerated comply. We delivered over 3 million fully driverless operational rise in Q3, representing 212% year-over-year growth, up from 148% last quarter. .

This quarter demonstrated how AI is driving transformative value across our business. From enterprise services to consumer-facing products to smart mobility, our AI capabilities are delivering proven tangible impact at scale. Starting with the enterprise side, where our AI Cloud business continues to scale with healthy momentum and deliver measurable business impact.

In Q3, AI Cloud continued its strong growth trajectory. Within AI Cloud, the area's most central to AI achieved the fastest growth. In particular, subscription-based revenue from AI accelerator infrastructure surged to 128% year-over-year, becoming the primary driver of AI Cloud expansion.

This reflects both a healthy shift towards a more recurring, structurally healthier revenue model and the strong demand for our AI products and solutions. Our ability to serve this growing demand stems from our early and strategic deployment across Baidu's full stack AI architecture, spanning infrastructure, framework, models and applications, which allows us to support enterprises at this stage of their AI journey.

At the infrastructure layer, our AI infrastructure is among the most advanced in China powered by a diverse mix of domestic and international high-performance computing resources, including our own self-developed AI computing architecture. Through continuous technical innovation, we drive performance and efficiency improvements while consistently reducing inference costs. Additionally, our industry-leading resource management capabilities significantly boost utilization and scalability.

These advantages make our AI infrastructure reliable, scalable and highly cost effective for enterprise clients. And the model layer, we feature our self-developed early foundation model, which continues to iterate rapidly. At the recent Baidu World 2025, we unveiled ERNIE 5.0 our first native omni model foundation model with exceptional performance in omni model understanding, creative writing and instruction following.

ERNIE not only represents the cutting edge of our AI technology, but also serves as a backbone behind much of the AI-driven innovations across our businesses. At the application layer, we have a range of AI applications tailored to enterprise business needs.

Let me share some examples. The first is formal or FM agent, a self-evolving agent, we recently launched that significantly improved enterprise efficiency built on ERNIE, it autonomously explores countless possibilities and continuously evolve its strategies to identify best solutions for highly complex, constantly changing real-world problems.

FM agent is now deployed across industries including transportation, energy, logistics and ports, optimizing complex operations that traditional approaches struggle to handle. Its capability is particularly valuable in China, where we have diverse industrial sectors with numerous scenarios demanding efficiency improvements, when you can meaningfully boost efficiency across such varied use cases, the social impact is profound.

Another example is the Daniel Wu English coach and ERNIE powered digital employee we created for Yash Education, featuring the lightness of the well-known actor. It enables users to engage in one-on-one real-time English conversation practice anytime, anywhere. This directly addresses a key challenge. Yash's users require frequent on-demand speaking practice, which is difficult to scale with human instructors.

The digital employee provides unlimited availability and an immersive engaging learning experience. Besides enterprises, our AI applications are creating value for individuals by enhancing productivity. Baidu Wenku and Baidu Drive, our largest AI applications for individuals have been revitalized with AI. Their combined MAU has approached 300 million. In August, Wenku and Drive jointly launched a general-purpose AI agent platform that orchestrates hundreds of specialized agents to complete complex end-to-end tasks through Singapore natural language interactions.

The platform has gained strong tracking since launch, demonstrating how AI can meaningfully enhance personal productivity at scale. In the physical world, autonomous driving exemplifies the transformative value of AI, unlocking new possibilities for mobility, safety and efficiency.

In Q3, Apollo Go growth significantly accelerated to new heights. During the quarter, we provided over 3 million fully driverless operational rights to the public, representing a remarkable 212% year-over-year surge compared to 148% growth last quarter.

In October, weekly average fully driverless operational rides exceeded 250,000, marking one of the highest levels achieved in real-world Robotaxi operations globally. To date, our fleets have accumulated over 240 million autonomous kilometers with more than 140 million of those being fully driverless, maintaining an outstanding safety record throughout Achieving this rapid expansion while delivering exceptional safety performance is a powerful validation of our technologies maturity and operational capabilities.

We are proud to see our decade-long commitment to autonomous driving, now bearing fruit in large-scale operations. Reaching this scale requires maturity across multiple fronts advanced technology, a rigorous and widely recognized safety record, demonstrated business viability, deep operational expertise and the ability to expand rapidly.

These years of unwavering investment have not only given us a first-mover advantage, but more importantly, have better comprehensive capabilities that position Apollo Go as the undisputed global leader in this field.

With this strength, Apollo Go has now entered a phase of rapid global expansion. As of October, Apollo Go's global footprint expanded to 22 cities, an increase from 16 last quarter. In October, Apollo Go entered Switzerland through a strategic partnership with the country's leading public transport operator.

Together, we plan to launch autonomous ride-hailing services in Eastern Switzerland, representing a key step in our European market expansion and another milestone in our global journey. In the Middle East, we secured one of the first fully driverless commercial operation permits in Abu Dhabi recently and deepen our collaboration with local partners. In Dubai, Apollo Go was granted exclusive authorization to conduct self-driving trials on open roads at the fourth Dubai World Congress for self-driving transport in September.

RT6 provided trial rides to global attendees, including government officials, business leaders, media and investors showcasing our technology's maturity on an international stage and demonstrating our global leadership. In Hong Kong, where Apollo Go has established by far the strongest presence in right-hand drive Robotaxi markets. We expanded our open road testing zones to include Kolon and Kung Tong District recently, further strengthening our position in the strategically important market.

These milestones spanning Europe, the Middle East and Asia validate both our technologies adaptability and our ability to partner effectively with leading local operators in different regulatory environments.

Looking ahead, we will expand to more markets with strong commercial potential and partnership opportunities, maintaining our unwavering focus on safety and operational excellence as we work toward making smart mobility widely accessible. In our mobile ecosystem, agent and digital humans represent AI-native monetization innovations that are transforming our online marketing business, creating substantial value for advertisers through higher engagement, better lead conversion and stronger ROI.

Our agent health advertisers effectively clarify user intent through intelligent multi-round conversations and quickly find out the most relevant high-quality sales leads. This ensures advertisers receive more precise and qualified leads compared with traditional approaches.

Building on this capability, agents have evolved into multiple forms, techs, voice-enabled and visually embodied digital humans, each designed to address different scenarios and interaction needs. Such versatility enables advertisers to choose the most effective format for their specific use cases, achieving broader scenario coverage and higher conversion efficiency. As a result, agents have gained strong traction across diverse industries, including health care, business services and lifestyle services.

In September, around 33,000 advertisers generated ad spending through our agents on a daily basis. Digital humans also saw strong momentum powered by ERNIE, our digital humans provide 24-7 AI-powered live streaming for advertisers at low cost, making professional live streaming accessible across more scenarios and industries.

The technology continues evolving, delivering greater realism, more natural interaction and real-time engagement with viewers. This enables performance that surpasses human health in many cases, making our digital humans increasingly attractive to advertisers.

Adoption has broadened beyond merchants to sectors such as health care, automotive and legal services are seeing both existing clients increase their budgets and new clients rapidly coming on board. In September, the number of digital humans live streaming on our platform almost tripled year-over-year, underscoring quick adoption across industries and growing monetization potential.

These innovations are already generating significant revenue with fast growth rates. In Q3, combined revenue from agents and digital humans reached RMB 2.8 billion, up 262% year-over-year, validating the strong market appetite for our AI native monetization approaches.

Looking ahead, we see substantial opportunities to scale these innovations further, broadening adoption across more verticals and deepening penetration with existing advertisers. Now let me review the key highlights for each business.

In our AI Cloud business, our client portfolio continued to improve in Q3, demonstrating deeper collaboration across the board, leading enterprise clients increased spending and expanded usage beyond AI infrastructure. Mid-tier enterprise clients delivered healthy growth with both subscription-based revenue and client count rising.

Several key verticals saw strong momentum. In embodied AI, our client base expanded to 35 from 20 last quarter, covering nearly all major industry players in China. The automotive vertical also delivered strong growth with revenue nearly doubling year-over-year.

In addition, this quarter, we entered into new collaborations with leading players, including [ Neolix ] a major provider of autonomous delivery vehicles in China, Collectively, these results affirm the broadening adoption and strong recognition of Baidu AI Cloud.

To address fast-growing demand, we strategically upgraded our MaaS platform to be agent-centric. is now the vision to provide not only leading model services with a constantly enriched model library, but also cutting-edge agent development capabilities and best-in-class agent infrastructure. By integrating high-quality proprietary and third-party capabilities and tools, enables seamless agent creation and empowers enterprises to accelerate AI native application development.

At the application level, we are driving productivity gains, both internally and externally. Internally, our developers widely leverage our AI coding assistant for developers. In September, AI contributed to over 50% of new code generation under developer oversight substantially improving our overall engineering and R&D productivity. exemplifies our belief that AI should liberate humans from repetitive tasks and deliver immediate efficiency gains.

Externally, we are democratizing AI from, our no-code platform. After continuous capability enhancements, we launched the International version, named Midu in November, bringing powerful no-code capabilities to global users. By removing areas like specialized training, we aim to empower more people worldwide to innovate and create with AI.

On intelligent driving, Apollo Go provided 3.1 million fully driverless operational rise in Q3, up 212% year-over-year. As of November 2025, cumulative rides provided to the public have surpassed 17 million. In terms of geographic expansion, Apollo Go added 6 new cities, bringing its global footprint to 22 cities as of October 2025.

In Chinese Mainland, Apollo Go has already achieved 100% fully driverless operations in multiple cities, including Beijing, Shanghai, Shenzhen, Chengdu, Chongqing, Wuhan,, and more. These are not pilot zones, but represent real services already open to the public, which speaks to the maturity of our technology and creation.

On our asset-light model and domestic partnerships, we also made good progress this quarter. The asset-light approach allows us to expand our autonomous driving services through partnerships and facilitate faster and more capital-efficient expansion. Following the launch of fully autonomous vehicle rental services with Car Inc, Apollo Go now enables cross-city travel in Hainan province with fully driverless rental vehicles, offering users a differentiated experience, not typically available through traditional car rental services, particularly for tourism and leisure travel.

In addition to our partnership with Hello Ride, we achieved scaled fully driverless operations in 2 cities in China, further validating the feasibility of the asset-light model. Looking ahead, we will continue to expand rapidly while prioritizing safety, accelerating the adoption of autonomous ride-hailing services across broader markets.

In our mobile ecosystem, the AI transformation of Baidu Search continued to progress in Q3. At the end of October, roughly 70% of mobile search result pages contain AI-generated content. We believe this represents an optimal and sustainable level. This quarter, we prioritized enhancing the quality of multimodal content within AI search results while expanding our overall content ecosystem. AI generated multimodal content saw rapid growth in both volume and quality.

In particular, with daily AIGC video generation consistently at the scale of million, our total AIGC video content continues to expand quickly while daily distribution within Baidu App is also seeing strong growth. As content supply improves, users experience richer, more relevant and engaging search results.

User metrics continue to improve. In September, Baidu App MAU reached 708 million, up 1% year-over-year. The daily average time spent per user in Q3 increased 2.3% year-over-year. We are also extending our AI search capabilities to external partners through the Baidu AI search API, which enables integration of our industry-leading search technology that delivers superior accuracy, authority and comprehensiveness.

Leading companies such as Samsung, Xiaomi and Honor have already adopted the API. This strategic initiative expands our technology reach beyond our own ecosystem, unlocking new business models and creating broader value across the industry.

Underpinned by our full stack AI capabilities, each business group within Baidu has seen rapid progress with AI driving both product innovation and business growth. From an AI-native perspective, our portfolio costs across business groups with a comprehensive range of AI-powered businesses from AI Cloud Infra to AI Applications, such as Baidu Wenku and Baidu Drive and to AI native marketing services, including agents and digital humans, all of which are showing strong growth momentum.

In the physical world, Apollo Go, our largest AI application continues to scale rapidly, and these are just a few examples, underscoring the broad-based growth of our AI-powered businesses and the meaningful business impact our AI capabilities are already delivering at scale.

Looking ahead, we will continue to expand our AI-powered revenue streams and strengthen our position to capture the long-term opportunities ahead. We are confident that our AI capabilities will bring even greater transformative value across our portfolio in the years to come.

With that, let me turn the call over to Henry to go through the financial results.

Thank you, Robin, and hello, everyone. Robin just mentioned our AI-powered businesses, and I'd like to elaborate based on ongoing feedback from investors and to better reflect valuation drivers based on our current portfolio. We are introducing a new AI native view this quarter across business groups to track AI-empowered assets company-wide.

This new view organized our business according to the nature of our products and services, helping investors better understand the fundamental valuation drivers across our diverse product and service offerings.

Going forward, we will provide business updates through this AI native view on an ongoing basis, while continuing to disclose results under the existing reporting methods, giving investors complementary lenses to assess the value of our portfolio. From this AI native view, we have a rich array of AI in power assets.

We are highlighting 3 categories this quarter. AI Cloud Infra, AI applications and AI native marketing services. First, AI Cloud Infra, which refers to the AI infrastructure and platform services we provide to enterprises and public sector. In Q3, revenue from AI Cloud Infra reached RMB 4.2 billion, up 33% year-over-year. We operate one of China's most advanced AI accelerator infrastructure, enabling highly efficient and cost-effective training and inference across diverse enterprise workloads. Within AI Cloud Infra subscription-based AI accelerator infrastructure revenue grew 128% year-over-year.

Second, AI Applications. These are AI-native or AI-powered product offerings addressing specific use cases for individuals and enterprises, including our flagship software products such as Baidu Wenku, Baidu Drive and digital employee. AI is transforming how applications create value, enabling far more powerful capabilities that address real-world scenarios more effectively. We built a leading and comprehensive portfolio across both individuals and enterprises.

Most of our AI applications are based on sticky subscription models, delivering high-quality revenue. In Q3, AI Applications generated revenue of RMB 2.6 billion. Third, our AI native marketing services, such as agents and digital humans continue to scale rapidly. This represents our second growth curve beyond our legacy business.

These innovative products are gaining strong traction with customers seeking performance-driven AI-native solutions. Customers are increasingly willing to pay a premium for cutting-edge AI technology that delivers miserable improvements in productivity, marketing efficiency and ROI.

In Q3, revenue from AI-native marketing services reached RMB 2.8 billion, representing a robust 262% year-over-year increase, accounting for 18% of Baidu Core's online marketing revenue. Now let me walk through the details of our third quarter financial results.

Total revenues were RMB 31.2 billion, decreasing 7% year-over-year. Revenue from Baidu Core was RMB 24.7 billion, decreasing 7% year-over-year. Baidu Core's online marketing revenue was RMB 15.3 billion, decreasing 18% year-over-year. Baidu Core's online marketing revenue was RMB 9.3 billion, up 21% year-over-year, driven by the boost of AI Cloud business within Baidu Core's online marketing revenue. AI Cloud revenue was RMB 6.2 billion, increased by 21% year-over-year.

Revenue from IT was RMB 6.7 billion, decreasing 8% year-over-year. Cost of revenues was RMB 18.3 billion, increasing 12% year-over-year, primarily due to an increase in costs related to AI Cloud business and content costs. Excluding impairment of long-lived assets, operating expenses were RMB 11.8 billion, increasing 5% year-over-year.

And Baidu Core's operating expenses were RMB 10.4 billion, increasing 5% year-over-year. Baidu Core SG&A expenses were RMB 5.7 billion, increasing 14% year-over-year, primarily due to an increase in expected credit losses and channel spending expenses. SG&A accounted for 23% of Baidu Core's revenue in the quarter compared to 19% in the same period last year.

Baidu Core R&D expenses were RMB 4.8 billion, decreasing 3% year-over-year. R&D accounted for 19% of Baidu Core's revenue in the quarter, which was basically flat from last year. Impairment of long-lived assets was RMB 16.2 billion, attributable to an impairment loss of core asset group with our rapid progress in high-performance computing capabilities. We proactively conducted a comprehensive review of our asset base and impaired including, but not limited to, existing infrastructure that no longer aligns with current computing efficiency requirements.

This results in a healthier and more optimized asset portfolio that better supports the future growth of our AI native business. Operating loss was RMB 15.1 billion. Baidu Core's operating loss was RMB 15.0 billion and Baidu Core's operating loss margin was 61%.

Excluding impairment of long-lived assets, operating income was RMB 1.1 billion and Baidu Core operating income was RMB 1.2 billion. Non-GAAP operating income was RMB 2.2 billion. Non-GAAP of Baidu Core operating income was RMB 2.2 billion, and non-GAAP Baidu Core operating margin was 9%. Total other income net was RMB 1.9 billion compared to RMB 2.7 billion in the same period last year.

Income tax benefit was RMB 1.8 billion, compared to income tax expense of RMB 814 million in the same period last year. Net loss attributable to Baidu was RMB 11.2 billion and diluted loss per ADS was RMB 33.88. Net loss attributable to Baidu Core was RMB 11.1 billion, and net loss margin for Baidu Core was 45%.

Excluding the impact of impairment of long-lived assets, net income attributable to Baidu was RMB 2.6 billion, and net income attributable to Baidu Core was RMB 2.7 billion. Non-GAAP net income attributable to Baidu was RMB 3.8 billion. Non-GAAP diluted earnings per ADS was RMB 11.12. Non-GAAP net income attributable to Baidu Core was RMB 3.8 billion, and non-GAAP net margin for Baidu Core was 16%.

We define total cash and investments as cash, cash equivalents, restricted cash, short-term investments, net long-term time deposits and held to maturity investments and adjusted long-term investments as of September 30, 2025, total cash investments were RMB 296.4 billion, and total cash and investments, excluding iQIYI were RMB 298.4 billion. Operating cash flow was RMB 1.3 billion, and operating cash flow, excluding iQIYI was RMB 1.5 billion. Baidu Core had approximately 31,000 employees as of September 30, 2025.

With that, operator, let's now open the call to questions.

[Operator Instructions]

Our first question today comes from Alicia Yap with Citigroup.

My question is on ERNIE 5.0 that was unveiled at Baidu World recently? And then so how will the new model drive the next stage of application such as the digital humans? And what are the key focus area for earnings, future iterations and also the differentiation?

Alicia, this is Robin. Over the past couple of years, I've been repeatedly saying that we're taking an application-driven approach when it comes to earnings iteration at the Baidu World just a few days ago, we unveiled ERNIE 5.0, our first native omni model foundation model. It has reached world-class levels in omni model understanding, creative writing and instruction following, which have very important capabilities to our current and future product portfolio.

From ERNIE 4.5 and ERNIE X1 in March to ERNIE 5.0 in November, ERNIE keeps getting better. Digital humans are a good example. Powered by ERNIE, they deliver float, contractually accurate and highly expressive dialogue. These are capabilities rooted in ERNIE's language strength. Beyond language, our model also drives visual real appearance, movement and even subtle micro-expressions, all synchronized with the conversation.

When these elements come together, the performance of our digital humans is truly exceptional and genuinely persuasive capable of driving user engagement and purchasing decisions. ERNIE also powers FM agent, our self-evolving agent that significantly improved enterprise efficiency.

It has proven to be very effective in industries like manufacturing, energy, finance, transportation and logistics. Similarly, our AI search and the Cloud business benefit from earnings capabilities too. Although ERNIE has delivered remarkable results for these applications, we see a lot of room for improvement.

We like to see digital humans sell better than real humans in all kinds of live streaming e-commerce across many product categories. We like to see FM agents find better and better solutions in more complicated scenarios in all industries.

We like to see AI-generated content to match users' interest better than KOL-generated contents. We like to see ERNIE-based agents to be able to tell which piece of content has better quality regarding certain topics and so on and so forth. These are the areas where none of the existing models do a good job, not even close.

So we aim to solve this problem. The application-driven approach actually reflects our deep conviction in where AI value will ultimately reside. While economic value today sits largely as the infrastructure layer, in a healthier AI ecosystem, the greatest value should come from applications where products deliver real impact to users, advertisers and enterprises.

Going forward, I think no foundation model can be better than anyone at any aspect. We will continue to focus on making ERNIE strongest where it matters most for our portfolio. Baidu has always been a company with strong belief in technology, and we will continue investing decisively in areas where technology can create will measurable value. So staying close to applications ensures a sustainable path forward for AI.

And our second question today comes from Lincoln Kong at GS.

So my question is about the Cloud business. So in the third quarter, we have seen a Cloud growth has slightly moderated. So are we seeing any changes in terms of the Cloud demand? And should we expect a re-acceleration in the coming quarters? So what's your outlook for the next year? And what are the key drivers that should support the sustainable growth of our cloud business. .

This is Dou. Thank you, Lincoln. If you look at our year-to-date performance, our Cloud business is growing well above the industry average. Well, for quarter-to-quarter, there can be some variability, but the overall trend is strong, and we remain very confident about this growth trajectory going forward. On the demand side, enterprises or applying AI across every aspect of the operations, driving strong broad-based demand for AI-centric Cloud services.

Within AI Cloud, the area most closely tied to AI workloads is gain. Our clients were using our Cloud, not only for model training, but increasingly for inference talks. In Q3, AI Cloud Infra revenue reached RMB 4.2 billion, up 33% year-over-year outpacing overall cloud growth and the subscription-based AI exploratory infrastructure revenue grew 128% year-over-year, accelerating from around 50% last quarter.

These results, both strong reflects both strong AI engagement demand and a healthier revenue mix. This momentum is supported by our full stack AI capabilities. At the infrastructure layer, our high-performance AI infrastructure, especially self-developed AI computing architecture continues to see strong adoption driven by superior performance, efficiency and cost effectiveness.

Many can start with AI Infrastructure and then expand to additional offerings over time. Also our Qianfan MaaSmass platform has been upgraded to be agent-centric with expanded model libraries and greatest tools and strengthen support for complex aging workflows. Qianfan provides best-in-class agent infrastructure, enabling enterprises to easily build and deploy AI agents at good scale.

At the application layer, we provide applications that can be readily applied to real business scenarios. Our Cloud growth is not just about investment in AI infrastructure. We attract huge importance to applications. We have a comprehensive portfolio of AI products and solutions that is growing very fast, including digital employee. In, FM agent and so on.

So we firmly believe AI applications will create substantial value in our Cloud businesses in the long term. So to sum up, if we look at our cloud business on an annualized basis, we believe that our full-stack AI capabilities and the strong demand for AI-centric cloud services will enable healthy scalable and sustainable growth in the future. Thank you.

And our next question comes from Alex Yao with JPMorgan.

The Baidu Application evolves into an AI application and web search becomes a bidding feature for AI chatbots. The line between search and chatbot is getting blurry. How are -- based on your observation, how user behavior is changing and what is your competitive strategy going forward?

Alex, let me answer your questions. And is very and work very quickly. So it's necessary to stay flat of products for different scenarios. And within Baidu, we leverage the travel capabilities from 2 continental offerings. The first one is the earning distance, which is the building travel inside by Baidu App. This supports conversations function calling and thanks all for its integration with search. Since many users assessed ERNIE assistance effectively from search. So you can draw on inquiry contacts and interruption history to deliver a more relevant and personalized answers. .

And it also connects to a set and allowing the users to move seamlessly from information discovery to test compaction. And distance is growing quite fast in our app. You can see that the conversation loans have increased around fivefolds year-over-year and Baidu has surpassed 2 million with a very strong month-over-month momentum. And we expect this churn to be further continue in coming few quarters. In parallel, we also offer a sustained alone travel application app has increased fivefold new shares and core facilities behind the ERNIE assistance. The ERNIE takes a more experimental innovative approach with near-term focus on improving retention and the long-term ambition to compete at the forefront of the travel category.

For example, we provide some cutting-edge multi-model features such as the AI images or style generations and which have been especially popular among the younger users. And looking ahead, we believe the chatbots are not only all the ultimate form of AI applications. The future of AI interactions will be multi-model, real-time generative and interactive. And for example, the most recent by Baidu World, we have showcased the creative digital human, which is capable of the instant interactions through the real-time voice and video like live conversations.

As many of you may recall, that we even have very small technical hiccup during the live demo, which actually proved that it was truly real time and not prerecorded. Once resolved the responding very vividly and delivered a dynamic force conversations that feel generally human. So we will continue to bring these advanced capabilities into search, making more intelligent, personalized and a couple of completing tasks -- this continuous innovation is how we intend to capture the opportunities in AI and strengthen our competitive advantages. Thank you, Alex.

Our next question today comes from Gary Yu of Morgan Stanley.

And also appreciate the additional disclosure on AI-powered businesses. Could management share more on the growth outlook and also the profitability of Baidu new AI-powered businesses? And how will these categories help accelerate our overall revenue growth going forward? .

Thank you, Gary. This is Henry. Let me provide some background on this new AI-native views. Based on the investor feedback, we are seeing a need for greater transparency into our high-growth AI businesses. These fields organize our portfolios by product nature, giving investors clearer visibilities into the underlying value drivers.

We will maintain both this AI-native view and our existing reporting methods in parallel, offering complementary perspectives on our business performance. From this new perspective, I think we have a rich portfolio of AI-empowered assets.

Let me give some highlights here. First of all, for the AI Cloud Infra, this part includes our industry-leading AI infrastructure containing self-developed AI computing architecture, cloud infrastructure and a best-in-class mass platform.

As AI adoption accelerates, demand for robust infrastructure is growing, and our differentiated capabilities position as well. We are capturing long-term sustainable revenue and expect margin to improve as utilization increases. Secondly, for the AI applications. This part include flagship products, for example, of Baidu Wenku and Baidu Drive.

AI has significantly enhanced functionalities across these products. We have one of the China's broadest AI application portfolios, and most of these applications on a subscription basis and contributing to a higher quality revenue and margin. Third, for the AI-native marketing services, including agents and the digital humans. This reflects how a lot greater efficiency and drive the second growth curve in our advertising business throughout enhanced engagement, conversation and ROI. This quarter, AI-native marketing services reached 18% of our Baidu Core online marketing revenue, up from 4% a year ago.

And we expect penetration to continue rising as adoption broadens. Customers are increasing these results-driven AI solutions and are willing to pay for tangible gains in productivities and marketing efficiency. Importantly, this AI-empowered business reinforce one another across Baidu's ecosystem.

And AI invest deeper across products. So we expect accelerating growth of these businesses. So when we're looking ahead, we remain confident in the revenue and profitability potential, which we believe we will support for a stronger growth trajectory for Baidu over time. Thank you, Gary.

And our next question today comes from Miranda Zhuang with BofA Securities.

The question is about the Robotaxi business. So Apollo Go has been accelerating growth this year. So looking ahead, can management update us on Apollo Go for next year and beyond, including your growth expansion plans. And how do unit economics look across different markets? And how does management view the long-term profitability potential of the Robotaxi business?

This is Robin. If you remember, our Robotaxi's journey started in 2013. So this is our 13th year. Today, Apollo Go is one of the world's largest robotaxi service providers. And as of November, we have provided over 17 million rides cumulatively, a level of very few players globally have achieved.

In China, we are the undisputed market leader through the first 3 quarters this year, our ride volumes were over 15x higher than our nearest domestic peers according to publicly disclosed data. All described are fully driverless, demonstrating unmatched operational scale and technological excellence.

Scale matters a lot. The reason we are able to achieve a leading position in autonomous driving technology on a global basis is that we have the scale. We have encountered many issues,-- others have not seen, we were able to train our models to handle those cases and smarter and smarter.

I think robotaxi has reached a tipping point, both here in China and in the U.S. There are enough people who have a chance to experience driverless drives and the world of miles has created positive social media feedback, which I think will propel the opening or loosening of related regulations.

For 2026 and beyond, we will continue to scale up our operations, both domestically and internationally. We will add more cars in our existing cities. We will expand to more cities. We will accumulate more fully driverless mileage and further improve our technology based on the operational data we gathered on the road. And yes, we need more data to train our models, better models make the cars safer and faster.

We will continue to drive down the cost per mile through technological innovation and operational efficiency. Now a few cities have achieved positive unit economics. As we scale, we hope to see more cities turn positive in 2026.

Also with scaling flexible business models, including asset-light models, we are ready to enter any city quickly once regulatory and market conditions allow. As of October, Apollo Go's global footprint reaches 22 cities with significant progress in Europe, Middle East and Hong Kong. We're confident that will continue to improve as we scale. So in summary, for 2026 and beyond, we expect strong growth across 3 areas: Rapid growth in ride volumes and fee size, geographic expansion in new markets into new markets and accelerated adoption of new business models. We believe Apollo Go is well positioned for continued global expansion and long-term profitability. Thank you.

And our next question today comes from Thomas Chong of Japanese.

My question is about how is AI search monetization progressing? And what feedback are you seeing from advertisers and users? Can AI native marketing services offset traditional ad business? And how should we think about core advertising profitability going forward?

Thomas, this is Julius. In October, nearly 70%, 7-0 of the mobile search result pages have content AI generated and the multi-model first content. This format is quite unique to us, and we are the first only one doing this. We expect this level to remain relatively stable as we have largely covered a query types where the AI meaningfully improve the user experiences. .

Our books now has shifted to improving quality, particularly the rich media content like images, videos, and we are seeing very clear improvement in content quality this quarter. which translate directly into the back to use experiences. And we can see that the usage retention is higher and users exposed to AI search results are initiating 6% more queries and spending more time with us. This tells us the users are finding real value in AI search and engaging more deeply. Our monetization, we are actively testing and seeing some encouraging early results.

First, that we are testing MCP in the commercial modules in AI search. For example, our e-commerce MCP module peaked at nearly RMB 6 million in GMV during the recent Double 11 shopping festival. This is a very early stage, but the results are quite encouraging.

Second, agent advertisers are generating over RMB 25 million in daily revenue, and we expect this to grow as we bring more agents into the ERNIE assistance as well. And third, we have started testing the digital human live streaming with the realtime interaction capabilities, try to explore new ways to create engaging commercial experiences. And looking ahead, we've seen a significant monetization potential for AI search, and we will continue testing actively. However, our near-term priority still remains the use experiences over immediate monetization. This AI transformation is necessary for long-term competitiveness and will inevitably create near-term pressure on both revenue and margins. So we believe this is the right trade off to capture the large opportunities ahead. Thank you.

And our final question today comes from Wei Xiong with UBS.

Actually, I have a few questions here. First, just a quick one. Could you please explain this quarter's asset impairment and its rational? And second, what are your CapEx plans for next year? And how should we think about the margin trajectory as AI revenues grow? And lastly, could we please have an update on shareholder returns or once the current buyback program expires?

Thanks, Xiong. First of all, on your first question on asset impairment. The background is we are accelerating investments in the latest AI completion technologies without any hesitation. So as part of this effort, we have conducted a comprehensive review of our infrastructure portfolio. Some of the existing assets no longer meet today's computing efficiency requirements. So we actually proactively give some impairment. After this onetime of impairment, our asset base and portfolio profile is in a much healthy position and better align with advanced AI computing demand and higher value avocation scenarios going forward.

Second, on the capital expenditures. We are maintaining a high level of investment just give one example. Since Baidu launch earnings in the March of 2023, we have invested well above RMB 100 billion in the AI investment.

Going forward, we will continue increasing our investment intensity in the AI area. We do expect to see greater operational leverage as our AI business deals. We're acting on 3 fronts. First of all, the asset review and impairment have left us with a leaner and more efficient asset base. Second, we are investing with the discipline to ensure capital efficiency.

And of course, thirdly, we are enhancing utilization of our AN infrastructure, for example, through dynamic allocation of capacity across internal products and external cloud services. So as a result of these initiatives, we believe Q3 represents a lower point for margins.

Looking to next year, we will drive to improve our non-GAAP operational income and the margins as benefits start to flow through. So on your last point regarding shareholder returns. Under the plan and program authorized in 2023, we have already bought back worth of USD 2.3 billion in shares. We are currently reviewing the future buyback mechanism.

We understand that we also think it is important to provide a greater certainty and clarity to reduce volatility of buyback programs going forward. We're also actively exploring diversified return mechanisms. For example, setting a dividend policy potentially. Together, these efforts aim to deliver more consistent value to our shareholders.

Ladies and gentlemen, that does conclude our conference for today. Thank you for participating. You may now disconnect.

Hello, and thank you for standing by for Baidu's Second Quarter 2025 Earnings Conference Call. [Operator Instructions]. Today's conference is being recorded. [Operator Instructions]

I now like to turn the meeting over to your host for today's conference, Juan Lin, Baidu's Director of Investor Relations. .

Hello, everyone, and welcome to Baidu's Second Quarter 2025 Earnings Conference Call. Baidu's earnings release was distributed earlier today, and you can find a copy on our website as well as our Newswire services. On the call today, we have 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 He our CFO. After our prepared remarks, we will hold a Q&A session.

Please note that the discussion today will contain forward-looking statements made under the Safe Harbor provisions of the U.S. Credit Securities Litigation Reform Act of 1995. Forward-looking statements are subject to risks and uncertainties that may cause actual results to differ materially from our current expectations. For detailed discussions of these risks and uncertainties, please refer to our latest annual report and other filings with SEC and Hong Kong Stock Exchange. Baidu does not undertake any obligation to update any forward-looking statements except as required under applicable law.

Our earnings press release and this call include discussions of certain unaudited non-GAAP financial measures. Our press release contains a reconciliation of the unaudited non-GAAP measures to the unaudited most directly comparable GAAP measures and is available on our IR website at ir.baidu.com. As a reminder, this conference is being recorded. In addition, a webcast of this conference call will be available on Baidu's IR website.

I will now turn the call over to our CEO, Robin. .

Hello, everyone. In Q2, Baidu Core's total revenue was RMB 26.3 billion. Our AI cloud continued to gain momentum, growing 27% year-over-year to RMB 6.5 billion. Notably, Baidu Core's non-online marketing revenue exceeded RMB 10 billion for the first time. That's up 34% year-over-year. This performance helped offset the near-term headwinds in our online marketing business. This year marked Baidu's 20th anniversary as a public company. Over the past 2 decades, we remain grounded in our belief in technology and innovation. Today, technological advancement is unfolding at an unprecedented pace. We've embraced the mega trend with open mind, experimenting boldly adapting quickly and engaging deeply with AI frontiers.

Amid rapid evolution, we've identified and doubled down on a few directions, we believe holds the greatest long-term value and are deepening our efforts with increasing clarity and confidence. Foundation model development remains a key focus area where we are actively exploring the frontier of foundation model research and pushing the boundaries of AI capabilities. With an application-driven approach, we see earnings intuition towards areas with real world application value. such as the fundamental AI transformation of Baidu Search and our industry-leading digital human technology.

Take digital human as a prime example, which represents 1 of the best applications of our earning models. This quarter, powered by ERNIE, our digital human technology reached new levels of rareness and capabilities, matching or even exceeding human performance in certain scenarios. Standout case was a live stream featuring the digital human of volume car, a top influencer in China. The 7-hour live stream generated tens of millions in GMV fully powered by ERNIE series models. ERNIE 4.5 Turbo generated a complete script, including dialogue, tone and action queues. closely mirroring the real person communication style.

Our multi-model capabilities delivered industry-leading visual realism with nuanced facial expressions, gestures and body movements that responded naturally to conversation flow in real time, achieving next level of performance that sets new standards in digital human technology. Beyond this flagship case, digital humans are empowering our broader merchant base with performance that already surpasses human live streaming hosts in many scenarios. Going forward, we will continue accelerating foundation model integration, strategically focusing our efforts on areas with application value where we can maintain our most competitive capabilities.

Beyond the model capabilities, our unique 4-layer end-to-end AI architecture has become a core competitive advantage and represents a key focus in our AI cloud business where our full stack AI capabilities are driving healthy growth. As the infrastructure layer, we achieved a critical system engineering breakthrough this quarter by completing the large-scale stable deployment of prefiled detailed separation architecture. This breakthrough significantly improves inference on currency and resource utilization, while substantially reducing inference costs.

The achievement was made possible at the end to end optimization enabled by our unique 4-layer AI architecture. spending infrastructure framework models and applications, which allows us to comminate improvements across all layers. At the same time, each layer remains open giving customers flexible choices between Baidu's proprietary and third-party options. As a result, we continuously improving the cost effectiveness of our AI cloud products and solutions reinforcing our position as China's top-tier cloud provider in the AI era. Meanwhile, our industry-leading MaaS platform, Qianfan continue to evolve to better support enterprise clients in building models and facilitating AI application development.

China features a comprehensive model library covering nearly all mainstream foundation models on the market. This quarter, we further expanded the library with a range of new models, including our newly often sourced early 45 series, additional third-party multi-model models and other leading options enabling greater flexibility across enterprise use cases. Leveraging our breakthrough in cloud infrastructure, Shenzhen delivers enhanced stability, higher currency and lower inference costs when running models, meeting our superior price performance.

On the tool chain front, Shipments tool chains are among the most comprehensive with industry-leading reinforcement learning and post training tools for our model development. In Q2, [indiscernible] was further enhanced to support a wider range of AI tools and functions that can be caused via MCP or API including Baidu's proprietary capabilities such as Baidu Search, Baidu Wenku, Baidu Maps as well as selected third-party capabilities like payment services. These enhancements helped simplify AI application development and continue to solidify [indiscernible] leadership as 1 of China's top MaaS platforms.

Autonomous driving remains 1 of the most promising areas with long invested, which represents a critical frontier in physical world AI. Following the successful validation of our urban model at the end of last year, Apollo Go is now scaling rapidly. In Q2, Apollo Go provided over 2.2 million fully driverless drives to the public. That's up 148% year-over-year, marking our strongest quarterly growth in 2 years. Also, Apollo Go global expansion has gained solid momentum, highlighted by 2 strategic partnerships with leading global life saving platform. In July, we announced a multiyear strategic partnership with Uber. Under this partnership, thousands of Apollo Gos fully autonomous vehicles will be deployed on the Uber platform across multiple international markets with initial rollouts planned for Asia and the Middle East. This milestone was followed by our partnership with Lyft in August, which will also bring thousands of our fully autonomous vehicles to key European markets through the Lyft platform. starting with Germany and United Kingdom and selling across Europe over time.

Our expansion into international markets is build on a strong foundation. In China, we have already achieved positive unit economics in markets where wide fares are much lower than those in major overseas markets. That's why this global partnerships are both logical and strategic positioning us to capture greater value in higher fare markets while scaling efficiently. Our partners, global market presence, leveraging our partners' local market presence, we can accelerate market entry across different continents and achieve faster deployment. while maintaining a cost-efficient asset-light business model.

In markets we've already entered, we continue to make encouraging progress in Hong Kong, 1 of the world's most complex right-hand drive city. We recently expanded our testing coverage to include Tongcheng and Southern District advancing our open node testing into more complex urban scenarios across both commercial and residential areas. Also, we further strengthened our presence in the Middle East in Dubai and Abu Dhabi, we started open road testing in designated areas in August. Notably, Apollo Go leads the world in right-hand drive robotaxi market. This is a space where hardly any companies of our kind have entered, and we've made by far the most progress. The rapid progress we are making in Hong Kong really shows our global leadership. It's proof of how adept our technology is and how mature our operations have become across all kinds of environment.

Our experience there provides us valuable insights for entering other right-hand drive markets, strengthening our confidence in scaling Apollo Go globally. With solid progress quarter by quarter, we are more confident than ever in Apollo Go's international potential. As China's largest autonomous ride [indiscernible] service provider and a global leader in this space, Apollo Go continues industry-leading technology, Apollo Go combines industry-leading technology, extensive operational experience and extraordinary safety records to bring safe, comfortable and affordable autonomous ride-hailing services to more markets worldwide than anyone else.

In our mobile ecosystem, transforming our products with AI remains a strategic priority, especially our legacy consumer-facing product Baidu Search. Baidu is at the forefront of applying AI to transform search globally rather than simply inserting AI summaries into search results we are fundamentally revolutionizing the third experience by completely replacing static actual hyperlink with intelligent, structured and multi-model first AI-generated responses. These responses start with relevant multi-model content right at the top. making complex information more accessible to a broader user base and therefore, creating a more intuitive experience.

In Q2, our AI transformation continued to accelerate, with AI-generated content reaching over 50% of mobile search result pages by the end of June up from 35% in April. By July, 64% of mobile search result pages contained AI-generated content presented in a structured and multi-model first format, marking the broader rollout of our innovative AI search experience. This AI transformation reached over 90% of Baidu App's monthly active users in July, with over 60% of such search result pages, starting with rich media elements such as images or videos. As we advance our AI transformation, the expanding content ecosystem across Baidu provides meaningful support leveraging ongoing progress in Gen AI and multi-model capabilities, Baidu's AI-generated content has grown significantly in both scale and quality providing more high-quality content for search result. AI-generated multi-model content, in particular, has been a seen rapid expansion. For example, daily IGC video generation reached millions of units starting from May and daily AI GC video distribution within Baidu App has grown rapidly. We're delighted to see sustained improvements in user metrics. In June, Baidu App MAU reached 735 million representing a 5% year-over-year growth. The daily average time spent per user in Q2 increased by 4% year-over-year.

Building on search ability to satisfy user intent, we are expanding its boundaries from providing smart answers to completing tasks and connecting real-world services. For instance, our agents engaged users in multiround conversations, connect them with relevant service providers when needed. and facilitate end-to-end task completion across multiple verticals. We believe this represents a meaningful expansion of what search can achieve enabling users to seamlessly move from information to action.

Now let me review the key highlights of each business sector this quarter. AI cloud revenue reached RMB 6.5 billion in Q2, up 27% year-over-year with non-GAAP operating profit achieving year-over-year growth. The growth was primarily driven by the growing demand for our highly cost-effective end-to-end AI products and solutions. Within the enterprise cloud, which contributes the vast majority of AI cloud revenue, subscription-based revenue grew at a solid pace, signaling a healthier and more sustainable revenue structure.

On the infrastructure layer, we continuously enhanced our resource management capabilities, achieving higher and higher infrastructure utilization. Through ongoing end-to-end optimization across our 4-layer AI architecture, combined with increasingly refined and efficient GPU resource management capabilities our large-scale key clusters have achieved over 90% utilization rates recently for key tasks. Our enhanced capabilities allow us to deliver better performance at lower cost and provide more competitive pricing for enterprise customers, establishing a virtuous circle, where our growing customer base and diversified workloads further improve resource utilization, reinforcing our sustainable revenue model. In Q2, our customer portfolio continued to improve. existing clients deepen the collaboration and increased spending, while mid-tier enterprise clients demonstrated strong growth momentum. Additionally, this quarter marked several strategic partnerships with prominent companies across key verticals, including a leading lifestyle platform and top-tier gaming company in China.

In the embodied AI industry, we have partnered with 20 companies cumulatively, including changing Institute of artificial intelligence and robotics for society. In autonomous driving, we established a partnership with Black System Technologies on AI cloud infrastructure. This partnership reflects the strong recognition of Baidu AI cloud and affirm our competitive positioning in China's AI cloud market. Building on our full stack AI capabilities, we are not only serving enterprise clients but also driving internal productivity and mass-market AI adoption at the application layer. Internally, we widely adopted [ Comet ] our AI coding assistant for developers. Comet capabilities continue to improve, enabling more agentic and efficient development workflow.

In July, AI contributed to generating over 45% of our new code with [indiscernible] providing oversight and approval. This has significantly boosted our engineering productivity and meaningfully enhanced our internal R&D efficiency. Externally, [ Minda ] extends this AI development capabilities to the broader community. Following Midas official launch last quarter, we are now delivering 2 low-code capabilities that enable users to create applications from mini games to utility tools and websites through simple natural language conversations with AI, no programming expertise required.

As of July, users have created around 200,000 applications on [ Milda ] are built completely without writing a single line of code. We are continuously enhancing low-code capabilities as we work toward our mission to democratize AI and empower anyone to innovate.

Moving to intelligent driving. In Q2, Apollo Go provided over 2.2 million fully driverless to the public, up 148% year-over-year. As of August, cumulative rights provided to the public have surpassed 14 million, underscoring the scale and maturity of our fully drivers operations. As of June, Apollo Go's global footprint spans 16 cities. To date, our fleet have accumulated over 200 million autonomous kilometers with an outstanding safety record, which is a testament to the ability and safety of our autonomous driving technology.

Beyond global partnerships like Uber and Lyft, we are accelerating the rollout of asset-light business models domestically. This quarter, we established new partnerships with [indiscernible] and [indiscernible] screen mobility, expanding our collaborative network with leading mobility service providers. Additionally, building on the partnership announced last quarter, Apollo Go's fully autonomous vehicle rental service, officially went live on the Car Inc. app. offering users a new access point to our Apollo Go fleet. These partnerships enable us to rapidly scale our services while leveraging partners operational expertise and existing customer bases, creating an efficient path to broader market penetration.

Going forward, we are confident to further accelerate our global expansion and capture significant value across multiple markets worldwide. For our mobile ecosystem, we continue accelerating AI transformation of search in Q2. In today's highly competitive mobile Internet market, where new products and technologies are emerging and evolving faster than ever, user needs and behaviors are constantly shifting, making it essential for us to keep waiting at a rapid pace. While our AI transformation is progressing rapidly, it is still in the early stages with considerable room for optimization before reaching its full potential. And we are not yet at the stage for large-scale monetization. Against this backdrop, we began prudent small-scale monetization testing in Q2 with user experience remaining our top priority.

Early results have been satisfying for example, some queries that we previously -- difficult to monetize are now showing potential. Agents maintained strong performance in driving better conversion efficiency, further validating our effectiveness. In Q2, revenue generated by our agents for advertisers grow 50% year over -- quarter-over-quarter. contributing 13% of Baidu Core's online marketing revenue. That's up from 9% in Q2 than in Q1. In parallel, digital human gained traction as an innovative monetization avenue for our advertising wins, particularly through AI-powered live streaming.

We've seen steady growth in digital human adoption over recent quarters. Beyond serving live streaming hosts for merchants, they were being adopted at growing scale in health care legal services, education and automotive sector. More advertisers recognize their value in boosting conversion performance through real-time user interaction and round-the-clock availability, leading to increased ad budget allocation towards digital humans. In Q2, revenue generated by digital human increased by 55% quarter-over-quarter, contributing 3% of Baidu Core's online marketing revenue.

To sum up, as we look ahead, Baidu will stay anchored in our long-term mission and move forward with greater focus and resolve as we continue to translate AI innovation into real-world value.

Before we move to Q&A, I'd like to take a moment to welcome Henry, Mr. Haijian He, who recently joined us as Chief Financial Officer.

With that, let me turn the call over to Henry to go through the financial results.

Thank you, Robin, and hello, everyone. I'm delighted to join the Baidu team and looking forward to working with all of you. Now let me walk through the details of our second quarter financial results. Total revenues were RMB 22.7 billion, decreasing 4% year-over-year. Revenue from Baidu Core was RMB 26.3 billion, decreasing 2% year-over-year. Baidu Core's online marketing revenue was RMB 16.2 billion, decreasing 15% year-over-year. Baidu Core's non-online marketing revenue was RMB 10 billion, up 34% year-over-year, primarily driven by the boost of AI cloud business. Within Baidu Core's non-online marketing revenue, AI cloud revenue was RMB 6.5 billion, increased by 27% year-over-year.

Revenue from ITE was RMB 6.6 billion, decreasing 11% year-over-year. Cost of revenue was RMB 18.4 billion, increasing 12% year-over-year, primarily due to an increase in costs related to AI cloud business and content costs. Operating expenses were RMB 11.1 billion, decreasing 4% year-over-year, primarily due to a decrease in personnel-related expenses, partially offset by the increase in channel spending expenses. Baidu Core's operating expenses was RMB 9.7 billion, decreasing 5% year-over-year. Baidu Core's, SG&A expenses were RMB 5 billion, increasing 6% year-over-year. SG&A accounted for 19% of Baidu Core's revenue in the quarter compared to 18% in the same period of last year.

Baidu Core R&D expenses were RMB 4.7 billion, decreasing 14% year-over-year. R&D accounted for 18% of Baidu Core's revenue in this quarter compared to 20% in the same period of last year. Operating income was RMB 3.3 billion. Baidu Core's operating income was RMB 3.3 billion, and Baidu Core's operating margin was 13%. non-GAAP operating income was RMB 4.4 billion. Non-GAAP Baidu Core operating income was RMB 4.4 billion, and the non-GAAP Baidu Core operating margin was 17%.

Total other income net was RMB 4.9 billion, increasing 531% year-over-year, primarily due to an increase in the fair value gain and a pickup of earnings from long-term investments. partially offset by increase in the net foreign exchange loss arising from exchange rate fluctuation between RMB and the U.S. dollar. Income tax expenses was RMB 881 million compared to RMB 1.1 billion in the same period of last year. Net income attributable to Baidu was RMB 7.3 billion, and the diluted earnings per ADS was RMB 20.35. Net income attributed to Baidu Core was RMB 7.4 billion, and the net margin for Baidu Core was 28%.

Non-GAAP net income attributed to Baidu was RMB 4.8 billion. Non-GAAP diluted earnings per ADS was RMB 13.58. Non-GAAP net income attributed to Baidu Core was RMB 4.8 billion, and non-GAAP net margin for Baidu Core was 18%. As of June 30, 2025, cash, cash equivalents restricted cash and short-term investments were RMB 124.2 billion. And the cash, cash equivalents, research cash and short-term investments, excluding IT, were RMB 119.9 billion. As of June 30, 2025, cash, cash equivalents, short-term investments and the long-term time deposits and had to maturity investments for Baidu Core were RMB 229.7 billion.

Free cash flow was negative RMB 4.7 billion, and free cash flow, excluding iQIYI was negative RMB 4.6 billion, primarily due to the increase of investment in AI business. We define net cash position, as total cash, cash equivalents, restricted cash, short-term investments, net long-term time deposits held to maturity investments and others. less total loans, convertible senior notes and notes payable.

As of June 30, 2025, net cash position for Baidu was RMB 155.1 billion. Baidu Core had approximately 31,000 employees as of June 30, 2025.

With that, operator, let's now open up the call for the questions. Thank you.

[Operator Instructions]

Your first question comes from Alicia Yap, Citigroup.

And also welcome Henry as the new CFO. I have a question on your AI model. with the rev model iteration, how do you view the current landscape? How do you position ERNIE strategically in the market and alignment with Baidu broader business strategy? And we also have heard that you are planning to launch ERNIE 5.0, could management share plans for ERNIE in the second half this year and also the key focus area for this next version.

Alicia, this is Robin. Let me first give you our take on the current landscape. The pace of model iteration is faster than ever. We see multiple new models launched almost every week and each new generation is is stronger than the last. In recent months, we've seen models grow more capable, reaching the stage where they're deeper logic and greater creativity now enable them to proposed entirely new solutions to never saying before. And I believe this is this kind of innovative ability is getting stronger. Meanwhile, the foundation model landscape is becoming more numbers and clearly not a one-size-fit-all situation, especially in China, similar to EVs you always have a lot of choices, different models excel at different tasks. Some are stronger in [indiscernible] some are in coding and some in multimodality.

So we will continue to see a market where multiple models coexist at very reasonable prices. And the value creation will happen at the application level more than at the model level. Against this backdrop, ERNIE positioning is clear. We take an application-driven approach to innovation. In fact, we've taken this approach since the launch of ERNIE -- first launch of ERNIE more than 2 years ago. Rather than spreading efforts across every possible direction, we stay focused on the strategically important areas that's valuable to us. We think we can deliver meaningful impact and sustain our leadership.

For example, as we advance AI search transformation, we direct our model capabilities towards generating and selecting multi-model search results. And our users love it. Our cloud customers also love it. They're paying for our search API for the purpose of RAG in their Gen AI applications. Another example is our hyper realistic digital human technology, which now matches and even exceeds real human performance in the live streaming e-commerce scenario. Our model is just better at convincing people to buy.

Cloud customers are paying for these capabilities, too. As we move into the second half and beyond, we will continue this acceleration. We're currently working on the next flagship version of ERNIE with significant improvements across key capabilities and expect to launch it as we are ready. In the meantime, we will continue to roll out iterations and updates on an ongoing basis for our existing model. We also keep monitoring industry developments to ensure our technology road map captures the most promising market opportunities. Thank you.

Your next question comes from Alex Yao from JPMorgan. .

And Henry all the best to your new role. So here is my question, how is the AI-powered search upgrade progressing in Q2 and Q3? Could the management share update updated metrics on how user behavior is shifting with the new experience, how should we think about the end game of AI search in terms of product format user reach and lastly, commercial potential.

Alex, thank you so much for your question. This is Luis. I think Q2, we continue to accelerate transformation. And as Robin has just mentioned, Baidu leads globally in using AI to transfer search and we are maybe were the most aggressive in revolutionized search. And we're probably the only company that has completely replaced the traditional links with intelligent AI answers that start with the multi-model content. This creates a more efficient, intuitive user experiences and unlike the current AI answers -- travel that remains mostly still as a tax base.

Our focus remains delivering the better user experiences beyond the MAUs and the time spend improvements and the user exposed to AI search now shows higher UV and retention indicating our next generation, such experience is driving the stronger user satisfaction. As for the end game of AI Search, I think still an open question, but our part is quite clear. We are fundamentally restructuring search. First, instead of just indexing or link into information, we are delivering the intelligent AI generating answers that began with the relevant multi-model content. Multi modern content now appears more at the very top of AI answers with increasing portion being AI generating as high-quality AI GC content expensed on our platform directly enrich search results and broaden what we can offer to the users.

And meanwhile, the AI is also empowering the people across our ecosystem from users, content creators, advertisers and service providers to produce the more and back to content. This includes enabling those who were not traditional content creators to participate in making our whole ecosystem more vibrant. And for example, in July, we have launched our mill steamer our appropriate trade video generation model to facilitate AI GC video creation at scale. And the latest version of new steamer with significant update will be launched tomorrow afternoon and [indiscernible], we were moving fast.

And second, we're also evolving from our fight information to completing tasks and connecting users with the real world services. For example, through the MCPs, we have connected search to external capabilities like our British [indiscernible] metropolitan MCPs that can provide ambitious explanations right in search. For more complicated needs, our agents help understand the user's intent and connect them with the services providers when the off-line services are required. What we have done today is only the beginning and search will continue advancing in capabilities and reach over time.

In these places, we are also working towards the shift from the general results to personalized -- personalized pages. While AI search understands the individual contacts, memories and preference could generate the tailor-made responses delivering more intelligent and relevant and personalized answers while better matching users with the tools and services they need. And looking ahead, we will continue to accelerate transformation which, in the short term, will wait on our revenue. But over time, we believe that AI search will unlock exciting commercial possibilities and upside is substantial. Thank you for your question, Alex.

Next question comes from Gary Yu, Morgan Stanley. .

I have a question regarding the AI cloud revenue. Can management provide a breakdown of the current revenue mix and margin profile. What's the split between subscription based and project-based revenue? And how do you see them evolving in the coming quarters? And also what's the margin profile looking like in the near term and over the long term?

Thank you, Gary. This is Dou. In Q2, AI cloud revenue grew 27% year-over-year to RMB 6.5 billion. For the first half of 2025 AI cloud revenue increased 34% year-over-year, accelerating from the low teens growth we saw in the first half of 2024. Enterprise Cloud has consistently outgrown our overall AI cloud business and remains the main growth driver. Then within the Enterprise Cloud, subscription-based revenue accounts for more than half of the total and continued growing steadily in Q2. The growth was driven by strong momentum in subscription-based AI infrastructure, which grew over 50% year-over-year.

We are seeing a good traction with both top-tier and mid-tier customers. Our mid-tier customers, in particular, delivered notable revenue growth as they continue expanding with this. reflecting our broadening customer base. The other part of the enterprise cloud is project-based revenue. Project-based revenue is typically linked to customer deployments and will inevitably fluctuate from quarter-to-quarter based on contract timing and project schedules. We are currently conducting a careful review of our project portfolio and aim to gradually reduce the propulsion of project-based revenue for greater revenue stability.

Turning to Personal Cloud, which is a smaller part of our overall AI cloud business. Over the recent quarters, we've integrated Baidu with Wenku and launched multiple new AI features. Recently, we opened up select AI features for free to encourage wider adoption. While this may involve some near-term trade-offs, we believe it helps deepen user engagement and positions us to benefit from a broader AI adoption.

On profitability, we achieved year-over-year growth in non-GAAP operating profit and maintained a healthy margin driven by a healthier revenue mix duty towards higher value offerings, while margins can move around from quarter-to-quarter due to dynamic market environments, we see clear potential for the future improvement over the long run as we scale and optimize our mix. Thank you Gary.

Our next question comes from Lincoln Kong from Goldman Sachs. .

So my question is about the search. So actually, can you management share more color on the reason AI search monetization testing because the following your earlier comments on AI search opening up new monetization opportunity. Could you elaborate on that? And how are all pricing format and business model evolving? What will be the margin look like?

Thank you Lincoln, the master question is curious. I mentioned the bottom monetization opportunities earlier, and let me elaborate over here. And from a product set, while our air transformation already covers a large portion of search results. We are still in the early stage with substantial room for improvement. We aim to further increase the penetration of multi-model content search results and continue building and reaching our AI native ecosystem through MCP agents and on the foundation enable a deeper or broader or higher-quality connections to the real world services. As the transformation once and the user experience improves monetization opportunities will naturally follow. And also, the AI search brings the native apps that fuels intuitive and integrate, enhancing rather than interacting the user experiences. So the vast majority of keywords that were previously very difficult to monetize now can be monetized under AI search, which should be significantly expand our advertisement inventory over time.

While initially, we will be very conservative with monetization to ensure we get the user experience right and by the long-term upside is much higher. And during our transformation, we are moving from safely generating sales leads to enabling the real world service delivery -- this is a shift made possible by new AI-native commercial products such as agents or digital humans. These innovative products allow us to better capture and serve the user needs through interactive conversations and while connecting the users service providers in verticals like health care, travel and education, where our agents and digital humans have already proven for [indiscernible] capabilities and over the long term, the ability to fulfill the user needs end to end position us well to drive a gradual transition from CPC to CPS, which offers a much higher ceiling for monetization.

And in Q2, we have already begun the early testing of the AI search monetization. While it's still in early days, we have seen the more encouraging signals, and we believe this stream will continue over time. that say that we always pull user experiences first. So we do a very deliberate approach to AI search monetization and large-scale monetization has not started yet. And at the same time, we have been aggressively accelerating the AI search transformation, including changing those queries with the highest monetization capabilities. So in the near term, we do expect the revenue margin will remain under significant pressure. But long term, we believe this positions us well for stronger growth. Thank you.

Our next question comes from the line of Miranda Zhuang with Bank of America Securities.

My question is about cloud and GPU. So how should we assess the sustainability of the AI patent cloud demand especially against the backdrop of a soft economy and intensifying market competition and also with the easing of the H20 chip restriction, has management seen any meaningful improvement in supply? How do you think about the chip constraint? Will it remain as a limiting factor for growth going forward?

Miranda, I would do questions. Actually, we were seeing strong and growing demand for AI driven cloud services as the China's cloud market continues its shift towards AI centric computing. The adoption of Gen AI and foundation models is accelerating, and AI has become a strategic focus for more and more companies. From what we've observed, demand is picking up across a wide range of sectors, not only from early adopters like [ Tican ] Internet companies, but also from a broader side of the industries like utilities, financial services and the public sector, where interested in AI-driven cloud solution is rising quickly.

Meanwhile, technological advances or strong new demand from emerging sectors, such as embodied AI. We are effectively capturing new opportunities and working with leading players in this field, including 20 of China's promising embodied AI startups and 4 which are China's top humanoid robot companies. The reason we were able to capture opportunities so quickly is our unique competitive positioning. What differentiates us is our ability to deliver highly cost-effective end-to-end AI cloud products and solutions, thanks to our full stack abilities.

Taking our AI infra as an example, we keep improving utilization and efficiency through our industry-leading resource management capabilities by dynamically allocating computing resources, we can better match workloads with the suitable resources and manage demand fluctuations, delivering better performance at lower costs. As a result, we can provide cost-efficient, reliable and scalable cloud services that makes it really easy for companies to adopt AI with minimal effort and scale it into real business impact. On your question about chips. Our focus remains on building a flexible AI architecture that maximize GPU utilization and supports a variety of chips, including domestic chips. This enables us to better serve customers as the supply environment evolves.

Looking ahead, we believe that a self-sufficient supply chain together with increasingly major homegrown software stacks will form a solid foundation for sustainable innovation in China's ecosystem. And clearly, Baidu is well positioned to lead the transition. Thank you.

Next question comes from Wei Xiong, UBS.

Given the near-term headwinds on ad revenue, and continued investment in AI. I wonder what are the plans for cost optimization and efficiency improvement that can help protect margins? How should we think about the margin trend in 2026 and beyond.

Thank you. This is Henry. First of all, on the AI investment, we remain committed to investing in AI and have made substantial investments throughout this year particularly in AI transformation of search. As Juan mentioned earlier, our core legacy product search is ongoing a radical transformation. Over the past several quarters, we have ramped up investments to accelerate this transformation, which we believe is critical to drive long-term value. However, since the AI search monetization is still in very early stages and has yet to scale, our revenue and margins are on the considerable pressure in the near term with Q3 expected to be especially challenging. To help pushing the near-term impact, we will actively drive internal efficiency gains. This includes strengthening resource coordination efforts across different business groups and improving overall resource utilization efficiency.

But also on the other hand, while we are remaining committed to long-term AI investment, we'll be very prudent in managing the pace to avoid future deterioration of fluctuation in margins. Looking further ahead, we see potential for margin improvement as our core advertising business recovers and stabilized and our non-advertising business both expand their revenue share and improving their own profitability. We believe our strategic direction and the disciplined execution should support a gradual recovery in profitability over time. Obviously, on the outlook front, I think before or around end of this year, we expect to have a greater visibility into next year. And at that time, we will have -- we will provide a clear outlook beyond the current quarters for the long term.

In parallel, we are carefully assessing different approaches to present and to unlock the hidden and unstated value of our assets. By doing so, we aim to strengthen our portfolio, create significant long-term value for shareholders, but also support sustainable growth over the long term of business. Thank you.

The next question comes from Thomas Chong, Jefferies.

My question is about the global autonomous driving landscape. We see it becomes increasingly competitive -- how does Apollo Go assess its long-term differentiation against peers? And how do the recent Uber and its partnership fits into your global expansion strategy? And what is the road map for achieving sustainable profitability .

Yes. I think autonomous driving is 1 of the most exciting frontiers where AI is transforming the physical world. and success in this field requires cutting-edge technology, massive sustained investment and disciplined execution over many years to achieve commercial operations at scale. One of the earliest entrants, we have built an unparalleled foundation across all these areas and become an undisputed global leader in this field. With our business model validated, our current focus is on running real-world operations at scale. And we have established global leadership in both left-hand and the right-hand drive robotaxi market. In the left-hand drive market, we were the first to achieve UE breakeven and all of our current operations in China, in Mainland China are fully driverless.

Globally, we are among the very few capable of scaled fully driverless and commercial operations in a single complex, large population area. And in the right-hand drive market, we lead the industry globally. In Hong Kong, we've rapidly expanded our operating of testing and advanced into increasing the complex urban scenarios following regulatory approval. Our technology stack and operational expertise are highly transferable across geographies, allowing us to adapt efficiently to new markets and regulatory environment.

We also built a major advantage with RT6, the world's first and only production vehicle designed especially for level 4 autonomous driving from day one. Unlike some retrofitted cars, RT6 is a perfect speed from the ground up is a focus on safety system integration and cost efficiency. It has the lowest unit cost globally for Level 4 and is already running on our commercial operations at scale, giving us a big edge for broader rollouts with this strength, we are confident about expanding to more cities worldwide, especially those with higher ride fares.

Here's the picture. We have the lowest cost level 4 vehicle and most efficient operation. We first achieved UE breakeven in Wuhan where taxi fares over 30% cheaper than the Tier 1 cities in China and far below many overseas markets. Yet we still managed to prove [indiscernible] model there. Such operational excellence and cost efficiency is unmatched globally. For us, expanding overseas means going from low fare markets to high fare markets. opening fare several times higher. Our huge cost advantage can deliver much stronger unit economics in most major cities worldwide.

To accelerate our global expansion, we are also taking a proactive approach to global partnerships. As we mentioned in our prepared remarks, we announced a partnership with Uber in July and followed by Lyft in August. Our partnerships with this world's leading mobility platforms will help us enter and scale more quickly into global markets like Middle East, like Asia and Europe. Looking ahead, we anticipate accelerating growth in ride volumes with our global operational fleet size multiply. With that momentum, we are confident that Apollo Go will continue to lead the market and stay at the forefront of autonomous driving worldwide. Thank you.

That does conclude our conference for today. Thank you for participating. You may now disconnect.