Baidu, Inc. (NASDAQ:BIDU) Q2 2024 Earnings Call Transcript

Baidu, Inc. (NASDAQ:BIDU) Q2 2024 Earnings Call Transcript August 22, 2024

Baidu, Inc. misses on earnings expectations. Reported EPS is $0.2693 EPS, expectations were $2.68.

Operator: Hello, and thank you for standing by for Baidu’s Second Quarter and Fiscal Year 2024 Earnings Conference Call. At this time, all participants are in listen-only mode. After management’s prepared remarks, there will be a question-and-answer session. 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.

Juan Lin: Hello, everyone, and welcome to Baidu’s second quarter 2024 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; Rong Luo, our CFO; and Dou Shen, our EVP in-charge of Baidu AI Cloud Group, ACG. 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 US 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 a detailed discussions of these risks and uncertainties, please refer to our latest annual report and other documents filed 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.

Robin Li: Hello, everyone. Baidu Core total revenue grew slightly to RMB26.7 billion in Q2, thanks to the continuous acceleration of our AI Cloud business despite the headwinds in our online marketing business. Even with the ongoing investment in AI, Baidu Core’s non-GAAP operating margin improved by close to 2 percentage points year-over-year to 26% and non-GAAP operating profit grew by 8% year-over-year, thanks to the operational efficiency gains we continue to implement. Despite the near-term pressure, we remain fully confident in our strategic direction and the transformative potential of GenAI and foundation models. As we move further into 2024, we are happy to observe a significant change as Baidu is scaling AI to address real-world problems and generate substantial value, both externally and internally.

Central to this effort is our commitment to making ERNIE increasingly affordable and accessible. I’d like to highlight some key points of our progress. The scaling of AI is accelerating at a breakneck pace, reflecting the real value that ERNIE creates for people and for businesses alike. We find that the most tangible benefit from AI comes from the adoption and use of applications built on top of LLMs. Just three months ago, we announced that ERNIE handled about 200 million API calls daily. Recently, it surpassed 600 million or over 1 trillion tokens are generated every day. To make ERNIE more affordable, we continue to expand our model portfolio and enhance our model capabilities to meet diverse customer needs. Last quarter, we launched three lightweight ERNIE models, and they have quickly gained traction among enterprises and developers.

Building on this momentum, we introduced ERNIE 4.0 Turbo in June. It offers superior capabilities compared to ERNIE 4.0 for typical use cases, yet it’s designed to be much cheaper and faster to run. Our current lineup now includes our flagship models, ERNIE 3.5 and ERNIE 4.0 and the enhanced model ERNIE 4.0 Turbo and several lightweight models. This diverse portfolio allows us to accommodate the varying needs of our customers, optimizing for performance, cost and latency. In May, we made a strategic move to make API cost free of charge for three lightweight ERNIE models, that’s ERNIE Speed, ERNIE Lite, and ERNIE Tiny. And in July, we significantly lowered the price of API cost for the two ERNIE flagship models, ERNIE 3.5 and ERNIE 4.0. This decision is rooted in our commitment to enabling wider access to ERNIE and making AI accessible to all.

At the same time, we continue to lower the cost of model inference. One optimization of note this quarter is the upgrade of PaddlePaddle, our open-source deep learning framework, to version 3.0. This upgrade significantly improves the framework’s compatibility with our AI infrastructure and ERNIE, which we expect will further help reduce model inference costs in the future. I briefly mentioned earlier that Baidu scales AI to address real-world problems and generate substantial value, both externally and internally. I’d like to discuss this in more detail. Externally, we have empowered our AI Cloud customers to achieve greater efficiency and scalability by using ERNIE. Our solutions have enabled clients to optimize their operations and realize significant benefits across various initiatives.

Let me offer you some use cases to illustrate the real business problems we have addressed and how we add value to our customers. In the healthcare industry, doctor burnout is one of the greatest challenges. By leveraging ERNIE Speed and our ModelBuilder, we helped a healthcare automation solution provider to train and finetune an industry-specific model that enables automatically-generated medical records for doctors. This solution significantly reduces the administrative burden on doctors, a major factor in burnout, and enhances clinical efficiency. After two months of model deployment, doctors using it were able to treat 50% more patients in — on average. In the recruitment industry, we have collaborated with a recruiting service company.

By leveraging ERNIE’s capabilities through API calls, our customer has significantly upgraded the matching process between job description and resumes. The smart matching upgrade has reduced the labor costs in this process by over 50%, measured by total working hours, while maintaining high-quality results. This automation allows the customer to expand its services to a broader client base, thereby increasing revenue-generating capability and demonstrating the tangible benefits of integrating ERNIE into its operations. The customer is very satisfied with the result and is now exploring new collaboration opportunities. In the public service sector, we utilized ERNIE Speed to deliver tailored, personalized and scalable services, addressing challenges like limited resources and efficiency needs.

This is particularly valuable in China’s grassroots public services, where shortages in manpower make large-scale personalized services challenging. We have collaborated with the customer to help over 6,000 villages in China improve public services, offering citizens more personalized and efficient support. Since the large-scale launch of this service in April, its daily usage has surged over 15-fold to more than 2 million times per day, helping locals handle tasks such as household registrations, social security inquiries and tax declarations. Internally, we have accelerated the renovation of Baidu Search with ERNIE, significantly enhancing our consumer-facing products at large scale. Generative AI is transforming user search experience, pushing beyond traditional boundaries.

Since the second half of last year, we have been testing ERNIE-powered features on a small scale. This quarter, our focus has shifted to substantially amplifying search capabilities with advanced AI features. Currently, 18% of search result pages contain generated content, up from 11% in mid-May. This AI-generated search results deliver more accurate and direct answers, enhancing content quality and providing previously unattainable information. This improvement has led to increased user satisfaction and engagement, as more users are now turning to Baidu Search for more complex queries, increasing Baidu Search versatility. AI-generated search results may reduce ad impressions and, therefore, have a negative short-term impact on monetization, but they provide significant value to our users.

By prioritizing long-term user experience over immediate revenue and profits, we see the strategic implementation of generative search as essential to driving future success. Our AI investments are fostering deeper user interaction. New interactive features enable users to refine their questions through multi-round conversations, enhancing the overall user experience. Additionally, there is a significant increase in users utilizing text and image creation tools within search. This development opened doors for long-term value creation. Last quarter, I mentioned that we planned to accelerate the distribution of ERNIE Agent. So far, we have seen more and more developers, advertisers and partners gather on our platform to develop innovative AI agents.

ERNIE Agent distribution within Baidu increased dramatically in July, exceeding 8 million daily, more than double the number in May. Currently, the most frequently used agents by users are for content creation, personal insights, such as personality testing, and tools, such as translation and schedule planning. At the same time, we also see a trend of ERNIE agents being more widely developed and applied, both internally and externally. Our first college application assistant agent is one example of how ERNIE agents are being distributed within Baidu to assist users with complex problem-solving and decision-making. Every year after China’s National College Entrance Examination, selecting universities after checking exam results is crucial for the 10 million-plus candidates.

In June, Baidu launched an innovative college application assistant agent, which is designed to meet personalized needs for selecting universities and majors. Following the National College Entrance Exam, the peak daily active users of the agent approached 2 million, highlighting its significant impact and utility. While all these new features for more sophisticated user needs have not yet been monetized, they are transformative for us for search and are crucial for our future success. This strategic focus positions us to capture substantial long-term growth opportunities, transforming the usage of search and solidifying our leadership in the AI-driven search landscape. Our one-stop shop for document creation is another example of how Baidu scales AI to address real-world problems and generate substantial value.

As our most pioneering internal product to embrace generative AI and LLMs, Baidu Wenku is now [revamping] (ph) the benefits of its product renovation efforts. In the second quarter, Wenku’s subscription revenue marked a year-over-year increase in high teens. AI is rejuvenating the platform with comprehensive content understanding and generation capabilities, while continuously improving in versatility. Since May, Baidu Wenku has launched new features resonating particularly well with young users, such as crafting lengthy documents with tens of thousands of characters, building and editing PowerPoint presentations, and creating children’s picture books with natural language guidance. The highlight — the heightened engagement underscores Wenku’s appeal to the next generation and signals promising future growth.

Moving from the digital realm to addressing changes and scaling AI in real — in physical world, our decade-long dedication to autonomous driving innovation and long-standing investment is bearing fruit. Apollo Go, our autonomous ride-hailing service, has recently achieved two significant breakthroughs. It’s establishing a robust foundation for making commuting more affordable and benefiting more people through technological innovation. First, since June 19th, building on our proven track record of safe operation, Apollo Go has successfully transitioned to offering 100% fully driverless ride-hailing services in practically the entire Wuhan municipality. This means all vehicles are now operating without the need for human safety officers on board, a major step forward in making autonomous ride-hailing [business] (ph) commercially viable.

Another milestone was the large-scale open-road testing of our sixth-generation autonomous vehicle, the RT6. Equipped with a battery-swapping solution, RT6 is competitively priced at below US$30,000 for mass production. After thorough testing, we plan to officially roll out RT6 into our fleet, establishing a strong foundation for further substantial cost reductions in Apollo Go operations. All in all, our efforts to scale AI to solve real-world problems fully align with our broader vision of creating a more efficient, equitable and sustainable future. Looking at how our efforts are yielding promising results, we believe we are well-positioned to capitalize on the growing demand for AI-driven applications. By leveraging AI to address complex challenges, we can continue our journey of innovation and growth and create long-lasting value for our stakeholders.

I want to thank our employees for their effort and contribution and our customers and shareholders for their trust in us. Now, let me review the key highlights for each business for the second quarter. AI Cloud revenue reached RMB5.1 billion, marking a consecutive acceleration to 14% year-over-year growth, while sustaining non-GAAP operating profitability in the second quarter. The strong growth is mostly attributable to the following two factors. First, GenAI-related revenue continued its robust momentum, accounting for nearly 9% of our total AI Cloud revenue in Q2, up from 6.9% in the previous quarter. As more enterprises integrate GenAI and foundation models into their daily operations, they increasingly come to us, thanks to our reputation as China’s most-advanced and cost-effective AI infrastructure provider and our excellent MaaS platform for model training and inference.

During the quarter, we further advanced our AI infrastructure management, enhancing our ability to combine GPUs from more vendors for optimal training and hosting of models. This ensures flexibility, reliability and efficiency, positioning us to capture a larger share of the GPU cloud and LLM market. We also continued to develop toolkits on our MaaS platform for our customers and partners, enabling the ERNIE family of models to deliver a superior price performance ratio. We’re proud that our ModelBuilder now supports the full model development lifecycle, including data management, finetuning, evaluation, optimization and prompt engineering. A major upgrade has been the introduction of diverse high-grade training data science, spanning from general to specialized industry-tailored data sets.

This enables efficient finetuning and ensures high performance for industry-specific applications, while retaining strong general LLM capabilities. Also, thanks to our continuous refinement to AppBuilder, the number of AI-native apps on our cloud now runs into the hundreds of thousands. These applications span a wide array of industries and scenarios, from online education and e-commerce to public service sectors, and are integrated into both online platforms and smart devices. The second major driver of AI Cloud revenue acceleration is cross-selling of our CPU Cloud services to our GPU Cloud customers. In Q2, we continue to observe an increase in CPU spending among our GPU Cloud customers. We see that our strong brand recognition in GPU Cloud is helping us win businesses in the CPU Cloud industry.

With revenue acceleration, AI Cloud business continue to deliver positive non-GAAP operating profit and improved margins. Legacy Cloud saw margin expansion, and GenAI-related businesses are expected to have higher normalized margin compared to the traditional cloud service. Overall, we remain confident in the strong growth outlook for our AI Cloud revenue, and we aim to continue generating non-GAAP operating profit going forward. For Mobile Ecosystem, Baidu Core’s online marketing revenue declined by 2% year-over-year in the second quarter due to broader macroeconomic challenges, competition and our aggressive AI-driven search renovation. Key offline verticals, such as real estate, franchising and automobile, remained subdued. Despite these challenges, we remain committed to optimizing our operations to maintain both a healthy margin and cash flow, while accelerating the AI-native transformation of our products.

In Q2, incremental ad revenue from GenAI and LLM enhancements to our advertising system continue to grow quarter-over-quarter, driven by the ongoing reconstruction of our monetization system and marketing platform. We envision ERNIE Agent as a compelling opportunity for our ad business. ERNIE Agent has the potential to revolutionize our traditional CPC model into a more efficient CPS model in the future, as it transitions from pre-sale consultations to catalyzing more direct sales on our platform. Serving as a virtual salesperson, ERNIE Agent empowers advertisers to provide more personalized content and product introduction, thereby ensuring better pre-sale consultation. Early adopters from the education, legal and B2B sectors are pioneering ERNIE Agent and have already seen a notable increase in effective sales leads.

We are proud to see that 16,000 advertisers already have their own ERNIE agents, which are now being distributed on the Baidu platform. Looking ahead, we believe ERNIE Agent is poised to deepen its penetration and broaden its reach across more sectors, unlocking promising potential for our advertising business. And moving into Intelligent Driving. I mentioned earlier that Apollo Go has made remarkable breakthroughs. In Wuhan, achieved 100% fully driverless operations in practically the entire municipality, reducing Q2 cost per vehicle by more than half compared to Q2 of last year. On top of that, Apollo Go has started scalable testing of the latest RT6 vehicles on open roads. These two factors are critical milestones in our strategy aimed at reaching UE breakeven in targeted cities.

In addition to cost reduction, Apollo Go’s operations continue to offer more expedient services. In Wuhan, its services is now available to 9 million people. The number of Apollo Go’s pickup points at the end of June has increased by over three-fold from the previous quarter, enhancing accessibility and convenience for passengers. The deepening of our operational footprint and increased station density have propelled utilization rate for our vehicles, significantly driving up the daily rides per vehicle and the distance per ride. Nationwide, Apollo Go provided about 899,000 rides to the public in the second quarter, marking a 26% year-over-year increase. In July, the cumulative rides provided to the public surpassed 7 million. Even with these milestones, our share in the entire ride-hailing service market is very small.

It will take many years for us to reach a meaningful market share in China or elsewhere. Looking ahead, we are committed to providing increasingly affordable, convenient and safe travel for more passengers and drive long-term sustainable growth. We will remain resolute in executing on our operational strategy, aimed at boosting efficiency and steering our Intelligent Driving business towards profitability. In summary, by extrapolating the ongoing progress from leveraging ERNIE to revolutionize product usage and transform business operations for our users, customers and developers, and driven by a deep belief in technological innovation, we aim to create value for society and contribute to the greater good. With that, let me turn the call over to Rong to go through the financial results.

Rong Luo: Thank you, Robin. Now, let me walk through the details of our second quarter financial results. Total revenue was RMB33.9 billion, which was basically flat from last year. Our revenue from Baidu Core was RMB26.7 billion, increasing 1% year-over-year. Baidu Core’s online marketing revenue was RMB19.2 billion, decreasing 2% year-over-year. Baidu Core’s non-online marketing revenue was RMB7.5 billion, up 10% year-over-year, mainly driven by the AI Cloud business. Our revenue from iQIYI was RMB7.4 billion, decreasing 5% year-over-year. Cost of revenue was RMB16.4 billion, increasing 1% year-over-year, primarily due to an increase in traffic acquisition costs and the costs related to AI Cloud business. Operating expenses were RMB11.6 billion, decreasing 9% year-over-year, primarily due to a decrease in personnel-related expenses, expected credit losses and the channel spending and promotional marketing expenses.

Baidu Core’s operating expenses were RMB10.2 billion, decreasing 10% year-over-year. Baidu Core SG&A expenses were RMB4.8 billion, decreasing 11% year-over-year. SG&A accounting for 18%, one-eight, of Baidu Core’s revenue in this quarter compared to 20% in the same period last year. Baidu Core R&D expenses were RMB5.4 billion, decreasing 8% year-over-year. R&D accounting for 20% of Baidu Core’s revenue in the quarter compared to 23% in the same period last year. Operating income was RMB5.9 billion. Baidu Core’s operating income was RMB5.6 billion and Baidu Core’s operating margin was 21%. Non-GAAP operating income was RMB7.5 billion. Non-GAAP Baidu Core operating income was RMB7 billion, and non-GAAP Baidu Core operating margin was 26%. Total other income, net was RMB771 million, decreasing 44% year-over-year, primarily due to a decrease in net foreign exchange gain and disposal gain, partially offset by a decrease in fair value loss and impairment loss from long-term investments.

Income tax expenses was RMB1.1 billion compared to RMB1.3 billion in the same period last year. Net income attributable to Baidu was RMB5.5 billion and diluted earnings per ADS were RMB15.01. Net income attributable to Baidu Core was RMB5.5 billion and net margin for Baidu Core was 20%. Non-GAAP net income attributable to Baidu was RMB7.4 billion. Non-GAAP diluted earnings per ADS were RMB21.02. Non-GAAP net income attributable to Baidu Core was RMB7.3 billion and non-GAAP net margin for Baidu Core was 27%. As of June 30, 2024, cash, cash equivalents, restricted cash and short-term investments were RMB162 billion, and cash, cash equivalents, restricted cash and short-term investments, excluding iQIYI, were RMB155 billion. Free cash flow was RMB6.3 billion, and free cash flow, excluding iQIYI, was RMB5.9 billion.

Finally, Baidu Core had approximately 31,000 employees as of June 30, 2024. With that, operator, let’s now open the call to questions.

Operator: Ladies and gentlemen, we will now begin the question-and-answer session. [Operator Instructions] Your first question comes from Alex Yao with JPMorgan. Please ask your question.

Alex Yao: Thank you, management, for taking my question. I have a few on AI development. So, first of all, how do you guys foresee the competition in AI evolving over the next two to three years? Will Baidu’s current first-mover advantage sustain? Or will the technology gap among key players narrow over time? How do you guys view the current competitive landscape compared to, for example, Kimi, Doubao, BABA and Tencent? In addition to technology leadership, what other factors are crucial to build a successful ecosystem in the AI industry? Thank you.

Q&A Session

Robin Li: Hi, Alex, this is Robin. Competition will be fierce over the next two to three years. Ultimately, I think whoever can make money from it will be able to survive. Last year, when China’s AI industry was still at the so-called “war of hundred foundation models” period, I predicted that the industry would go through consolidation, and only a few foundation models would survive. We are seeing a strong demand for our GenAI products, and I’m proud that ERNIE has continued to strengthen its market lead. And I’m confident we will maintain our leadership going forward. Our decade-long investment in AI provided us with the first-mover advantage and technological leadership. We were the first public company globally to launch a GPT-like model in March of last year.

And since then, we kept leveraging our self-developed four-layer AI architecture for model upgrade. Last October, we launched ERNIE 4.0. That’s China’s first GPT-4-type model. And this June, we launched ERNIE 4.0 Turbo, which offers superior capabilities over 4.0. We will continue to rollout upgraded versions of our flagship foundation models. And our application-driven approach further expands our competitive advantage and sets us apart from the others. We believe that foundation models hold little practical value without real applications built upon them that are widely used by users and customers. So, I mentioned earlier that currently, ERNIE handles about 600 million API calls and generates about 1 trillion tokens every day, which we believe is the highest in China.

We have been heavily using ERNIE to renovate our own products, like Baidu Search or Baidu Wenku into AI-native applications on a large scale. And this approach has helped us gain valuable insights from users and continuously supported the upgrade of ERNIE. And this experience allows us to create better solutions for our external customers when they use ERNIE. At the same time, we have been working really hard to make ERNIE increasingly accessible and affordable, so it can be used widely to address real-world needs. We continue to lower the cost of model inference, expand our model portfolio and develop toolkits for model building and app building, so that our customers, partners, developers can leverage the power of ERNIE more efficiently and effectively.

And this application-driven approach has helped us create a dynamic and expanding ecosystem. By the end of June, PaddlePaddle and ERNIE developer community had grown to around 14.7 million. Meanwhile, hundreds of thousands of enterprises are now engaged with us, highlighting the expanding influence and the utility of our ecosystem. Looking ahead, we will be patient. We will continue to iterate on ERNIE’s capabilities to ensure we stay at the forefront of this new AI wave. By leveraging real-world applications and feedback from our ecosystem partners, we gain valuable resources and insights to continuously refine our model. The synergy among our vibrant ecosystem, advanced model capabilities and an application-driven approach will continue to strengthen our leadership in this rapidly-evolving AI landscape.

Thank you.

Operator: Your next question comes from Gary Yu with Morgan Stanley. Please ask your questions.

Gary Yu: Hi. Thank you for the opportunity to ask questions. My questions are related to AI and search. So, what kind of progress has been made in renovating traditional search using GenAI and the large language models? What kind of KPI is the company monitoring for AI in search? And have you observed any changes in user behavior since implementing AI in the search business? Related to that, what is the rollout plan for AI search moving forward? Any specific KPI that the company aim to achieve by the end of the year? For example, certain percentage of search results generated by AI either by the end of this year or by in coming years? Thank you.

Robin Li: Yeah. We are very committed to using cutting-edge AI technologies to transform our search capabilities and blend search with speed in one app. So, KPIs related to Baidu App will override search-only KPIs. In the second quarter, we accelerated the pace of renovating our search by leveraging ERNIE. Our primary goal is to improve user experience, offer more capabilities and boost user engagement on Baidu. As highlighted in my opening remarks, we’ve made several important progresses. First, we are providing smarter search results at an expanding pace. Currently, 18% of search results are provided by GenAI. These results are more comprehensive and relevant than traditional search results, offering information that was previously unavailable or scattered around different places.

Furthermore, GenAI enables infinite supply of content on demand and can effectively hedge the decline of the open web system. Second, when GenAI results can already meet user needs from the current query, we will recommend other content or services, just like a user would find in a typical newsfeed context. This will enhance users’ time spent and retention. Third, the new interactive function in Baidu Search allows users to refine their questions and follow-up through multi-round conversations, reshaping how users express their need and making the search experience more natural and efficient. Additionally, I would like to point out that we are rapidly rolling out ERNIE Agent within search results to provide intelligent assistance to our users.

By opening up our platform, developers can create a wide range of agents tailored to meet the unique needs of each and every user. Our vision is to create an ecosystem where many developers contribute to building a robust network of agents, collectively enriching the user experience and outperform the most powerful frontier models, addressing a broad spectrum of challenges and opportunities. With all the efforts mentioned, the types of search queries on Baidu have become more diversified. Users are increasingly posing complex questions, seeking comprehensive and in-depth answers from Baidu Search. While these demands were once challenging to meet, they can now be efficiently addressed. On top of that, users are also using Baidu to create content, such as writing and editing text and images.

GenAI is transforming Baidu Search into a multi-functional creative platform, offering a wider range of services to users. We’re still in the early stages of an exciting yet uncharted opportunity. As we [re-invent] (ph) the product, we are listening to feedback from users, creators and partners. We’re focused on long-term sustainability over short-term monetization. We will be patient and make sure that we set the right foundation for new ecosystem and closely track user metrics. We believe this approach will deliver greater value to all our shareholders. Thank you.

Operator: Your next question comes from Lincoln Kong with Goldman Sachs. Please ask your question.

Lincoln Kong: Thank you, management, for taking my question. So, my question is about cloud business. So, what are the drivers behind this accelerating growth of cloud revenue this year? How much of this incremental revenue in cloud this year is AI related? And how does this contribute to the long-term AI Cloud revenue growth? Also, what’s the margin profile for the AI Cloud compared to the traditional cloud business? So, how should we think about the intense pricing competition among different players in the market? And ultimately, what should be optimal longer-term profit structure for our cloud business? Thank you.

Dou Shen: Hi, Lincoln. Thanks for your question. This is Dou. So, as you may have noticed, our AI Cloud revenue shifted from a year-over-year decline in Q3 last year to a positive growth in Q4 and then kept accelerating. So this quarter, AI Cloud revenue grew by 14% year-over-year. The rapid growth is driven by strong demand for our GenAI and LLMs across various industries. Thanks to our reputation for having the most powerful and cost-effective AI infrastructures in the MaaS platform, so we’re gaining recognition across a wide range of sectors, from early adopters like Internet companies and online education providers to the industries like automotive, financial services and utilities. Leading companies like State Grid, SF Express and Guomin Pension & Insurance and also the rising AI startups, such as Shengshu Technology, so they all intend to form in-depth partnership with us, leveraging our industry-leading AI capabilities.

So, as Robin just mentioned, right, a key revenue driver for AI Cloud is the fast-growing GenAI and foundation model-related revenue. It has increased from about 5% of the total AI Cloud revenue in Q4 last year to almost 9% this quarter. To meet the diverse needs of our customers, we offer a wide range of products and solutions. So, for those seeking model training and hosting services, we provide the most cost-effective and powerful AI infrastructure. This quarter, we have observed a notable increase in spending on our public cloud from GPU customers, reflecting their high satisfaction and trustiness. So, this strong customer endorsement reinforces our confidence in continuing to gain market share. For the customers who are looking to harness the power of LLMs through API calls, we offer the ERNIE family, which includes some of the most powerful LLMs available in the China market.

MaaS platform also provides comprehensive toolkits for easy finetuning of custom models and the development of applications tailored to specific needs. We see the LLM and GenAI market is still in its very early stages. We’re educating the market and making AI accessible or key. On top of the model capabilities and ease of use, wide adoption also depends on affordability. So, to encourage usage, we have made API calls free for three lead ERNIE models and cut price for our flagship models. So, this has led to a rapid increase in ERNIE API calls. As Robin just said, the total number of daily API cost has surged from around 200 million in mid-May to over 600 million, which means over 1 trillion [days] (ph) tokens. We believe that demand for ERNIE through API calls will continue to grow rapidly, given ERNIE’s superior and continuously improving efficiency and performance.

For our traditional cloud services, by cross-selling CPU cloud services to our existing GPU cloud customers, we have noticed a significant increase in CPU cloud revenue from these customers. So overall, we are pretty confident that our AI Cloud revenue growth will maintain strong momentum in the coming quarters and capitalize on significant long-term growth opportunities. So, on the profit side, we are committed to a sustainable and healthy growth. So, our AI Cloud business is healthier than ever and consistently generating non-GAAP operating profit, and we have been improving the margins. So, regarding the price competition issue in your question, we already know that GenAI and LLMs market is still in its very early stage. And accessibility with superior price and performance is essential for growing the market.

While we were working very hard to improve the capability of early models, we are also continuously lowering the inference cost. Considering the SKU effect as leading player in the market, I believe we will keep growing our market share. And also with the SKU, I believe the normalized margin for our GenAI-related cloud business should exceed that of traditional cloud business. Thank you.

Operator: Your next question comes from Thomas Chong with Jefferies. Please ask your question.

Thomas Chong: Hi, good evening. Thanks management for taking my question. Can management quantify the impact of the soft advertising revenue, distinguishing between weak macro, the cannibalization impact from generative AI implementation, as well as the competition, particularly in search separately? What is the core advertising growth for the second half this year? If we expect to see a trend around by year-end, is it more driven by better macro or better monetization in AI? Looking at 2025 and beyond, what’s the sustainable growth rate? Thank you.

Robin Li: Yeah. As I mentioned in my opening remarks, our advertising business is currently facing pressure caused by a combination of external factors and our proactive efforts to accelerate the AI-driven renovation of search. Externally, macro weakness and competitive pressure are the primary challenges. So, on the macro front, the recovery of consumer spending has been sluggish, causing many advertisers, especially the small- and medium-sized advertisers who heavily rely on offline activities, to adopt a very cautious approach to their ad spending. Certain verticals, such as real estate and its related sectors, franchising and automobile, have been particularly weak. In addition to this, Chinese Internet users’ behavior has changed a lot.

Users are increasingly concentrating their time on a selected few leading short video and social media platforms. The user time spend share change can be a proxy into the loss of our online advertising business. And as I mentioned earlier, 18% of our search results contain GenAI content, which is not monetized at this time. This number will increase over time. Of course, our aggressive rollout of GenAI features in the monetization system, partly offset the revenue sacrificed here. While these are all the big challenges for the time being, we strongly believe technological innovation will be the key for our future success. Specifically, we need to focus on what we can actively shape influence, leveraging our strengths and the resources to seize new opportunities within our reach and foster sustainable growth over time.

So, to this end, we are renovating our entire Mobile Ecosystem with GenAI and ERNIE. For the advertising system, the renovations are already showing some results. We talked about this during the previous quarter. The incremental ad revenue generated by GenAI and LLM continue to grow in Q2. This is mainly due to the improvement in the monetization system, which have improved ad conversion and efficiency. And on top of that, we’re actively exploring opportunities for ERNIE agents to further enhance our advertising capabilities. However, using AI to upgrade our existing advertising system, it’s just a near-term step. We believe that the two long-term value lies in the renovation of our consumer-facing products. With over 20 years of experience in general search and our industry-leading capabilities in GenAI and LLMs, Baidu is uniquely positioned to innovate and transform general search into AI search.

This transformation is not just about enhancing the traditional search experience. It’s about redefining what search can do for users. Since Q2, we have accelerated the pace to renovate Baidu Search using ERNIE. New features such as AI-generated search results, interactive features within search and ERNIE agents have been rolling out rapidly. This renovation has high potential, but it’s not something that will be completed in a short time. During this process, there may be a negative impact on revenue, but we are confident that the result will be a completely different new AI ecosystem, one that is much more versatile, productive and user-friendly. So, it may take some time before we see a noticeable revenue contribution from the AI enhancements in our search product, but we are confident that focusing on creating long-lasting user value rather than short-term revenue while ultimately create tremendous value and drive sustainable revenue growth.

Thank you.

Operator: Your next question comes from Miranda Zhuang with Bank of America Securities. Please ask your question.

Miranda Zhuang: Thank you, management, for taking my questions. So, my question is about the AI agent and the monetization. So, can the company explain how the AI agent is actually used in your users experience and monetization? How do you drive the adoption of this AI agent, especially among the SME advertisers, which have less data and less tech capability? What kind of ramp-up pace do you expect by the end of this year for the AI agents? And also, can management give us more color on the recent development in AI monetization, especially the progress from the CPC to the CPS model? What are the advertisers feedback on the effectiveness and the future budget allocation on the Baidu platform? Thank you so much.

Robin Li: Yeah, there are more than 10,000 ERNIE agents right now on our platform. The beauty of ERNIE agents is that anyone can create them, whether you are a user, a creator, a service provider, an advertiser or a company. The barrier to entry is very low. It’s really important to understand that the value of ERNIE agents extends far beyond just being a tool for monetization. It will significantly improve the user experience in our Mobile Ecosystem. A recent example is a lawyer agent created and distributed on our platform. A user had a conflict with a law enforcement officer and started this consultation with this agent. We had, I don’t know, 288 rounds of conversation and lasted for five hours. It’s hard to imagine a real human lawyer would have the time and patients to talk to a potential client like that.

But the benefit for the user, for the advertising law firm and for our platform are self-evident in this case. Today, ERNIE Agent is still in its early stages and it’s evolving quickly, but we are glad to see that advertisers are using ERNIE agents to get better sales fee conversions than the traditional search ads. As I mentioned before, more and more advertisers, especially in fields like legal, education and B2B are starting to use ERNIE Agent. These areas often require long decision-making time and personalized pre-sales consultation. ERNIE Agent with its interactive feature is getting better and better at addressing these kind of needs. Plus, as ERNIE Agent becomes smarter, it has the potential to drive direct sales and even more value for our advertisers, therefore, moving us from a CPC model to a more effective CPS model.

We believe ERNIE Agent has the potential to become the new website system in the AI era. Early-stage ERNIE agents might not always improve the user experience right way, but as they start providing greater interactions, boosting engagement and offering personalized services, monetization opportunities will naturally follow. Thank you.

Operator: Your next question comes from Alicia Yap with Citigroup. Please ask your question.

Alicia Yap: Hi, thank you. Good evening, management. Thanks for taking my questions. I have questions on the margin. So, Baidu Core continue to achieve the expansions in the operating margins in the second quarter. Just wondering, is this expansion sustainable? What is your outlook on the profit margin trend for Baidu Core in the entire year 2024? Thank you.

Rong Luo: Thank you for your question. Let me take your question. As you know, we don’t give the guidance for quarters ahead, but I would try to give you some insights into our business priorities and what could that mean for our cost and expenses. For AI Cloud, we continue to phase out the low-margin businesses, so we can continue generating operating profit and margins on a non-GAAP basis. Looking to the long term, the normalized margins for the GenAI-related cloud businesses should be higher than the legacy cloud business. And this kind of revenue today is already [9%] (ph) of our total cloud revenue. For Intelligent Driving, the other growth initiatives, we tend to reduce our losses and particularly our Intelligent Driving businesses with the operational efficiency gains and try to reach improvements in robotaxi business.

Overall, we believe that high-quality growth and the investments should be well balanced. For many years now, we have delivered a solid track record of the top-line revenue growth and with the resolute cost discipline, and intend to continue building our future on this basis. Sometimes, of course, though, the investment runs ahead of the high-quality growth. We continue to operate our legacy search businesses in a challenging environment. That pressures our ad revenue. To tackle these kind of challenges, it’s very important to focus on offering the users improved experiences, and in particular, a whole new AI-based search experience for Baidu Search. As Robin has mentioned just now, this quarter, we have accelerated the renovation of Baidu Search with ERNIE.

In the short term, we have seen a significant increase in AI-generated search results. 11% of all search result pages contain the generative content in mid-May, and this percentage has risen to 18%, one-eight, as of now. The scaling of the AI-generated search results and other new features require a lot more computing capacity and users interact more in [indiscernible] conversations. They ask more complicated questions and use the new content creation capabilities within search. All of this raised the total cost of the more computing power is needed, but the new AI search features has not yet been monetized now. So, the accelerated AI-native transformation of search will inevitably put some pressures on our margins in the short term. So, we reduced our powerful and efficient AI infrastructures and our technical expertise to make efficient gains over time.

Thank you so much for the question.

Operator: Your next question comes from Wei Xiong with UBS. Please ask your question.

Wei Xiong: Sure. Thank you, management. Good evening. Thank you for taking my question. My question is regarding the robotaxi business. So, with government support on accelerating the robotaxi license approval in more cities, how would that translate to Baidu’s robotaxi opportunities, and also cost associated to that? And what are the targets for Apollo Go this year and for the next few years? How do we plan to replicate Wuhan success nationwide? For example, do you consider partnership or collaboration for future development? And lastly, could management share your view on the competitive landscape change in the robotaxi market? Thank you.

Robin Li: Yeah. We’ve been investing in autonomous driving technology for over a decade, guided by steady focus and unwavering patients. Robotaxi services have long been considered as a breakthrough, so advanced that no one can predict exactly how long it will take to bring it to full fruition. For a long time, many doubted whether we or anyone else in this world would make it, but we kept moving forward. Now, as we see regional UE breakeven approaching, our vision is becoming a scalable reality. Our persistence, even in the face of doubt, is not only driving progress, it’s shaping the future. Our dedication has made Apollo Go the world’s largest autonomous ride-hailing service provider. This is measured by a number of rides.

Our track record of safe operations has earned public and regulatory trust, allowing us to expand services and reach more people. We’re happy to see that more and more passengers are choosing Apollo Go, not just for testing rides, but as their go-to-option for daily community. Wuhan stands as our largest operational area. It’s a fleet of around 400 vehicles offering 24/7 service to 9 million residents there. Our focus on cost reduction through innovation has also made our services increasingly affordable. Since launching in Wuhan in Q3 of 2022, we have steadily increased the number of fully driverless vehicles and rides, and I mentioned that 100% fully driverless ride-hailing operation has been achieved in practically the entire city. This has a significant impact on costs.

With our RT6 vehicles being tested at scale, we expect continued cost reduction once RT6 are officially put into operation. So please note, our share of ride-hailing market in Wuhan is still very small, at around 1%. And the scaling will be a gradual process and it could take many years. The potential of China’s ride-hailing market is enormous. As robotaxi inherently offers more affordable services, it will naturally drive demand and open up great new opportunities. This is where innovation meets the vast scale of the market and creating a powerful synergy that will reshape the landscape of transportation and beyond. As Apollo Go scales up, it will offer an increasingly affordable, convenient and much safer options to an expanding population.

Also need to note that entering the robotaxi market requires deep belief in technology, long-term investment and strong capital reserves. Apollo Go stands out in these critical areas, setting the benchmark for the industry. As for your question regarding to future business model, currently, all of our vehicles are owned by us, and we bear the hardware CapEx and depreciation costs, but we are open to various business models and partnerships. While our immediate focus is still achieving UE breakeven in key cities. In the long run, we aim to innovate this asset-light models, like technology licensing and fleet operation, to capture the huge market opportunities, and ultimately, benefiting society with better transportation solutions. Thank you.

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