This conference call transcript was computer generated and almost certianly contains errors. This transcript is provided for information purposes only.EarningsCall, LLC makes no representation about the accuracy of the aforementioned transcript, and you are cautioned not to place undue reliance on the information provided by the transcript.
spk04: Hello, and thank you for standing by for Baidu's second quarter 2023 earnings conference call. At this time, all participants are in a 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.
spk07: Hello, everyone, and welcome to Baidu's second quarter 2023 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, Doshan, our EVP in charge of Baidu AI Cloud Group, ACG, and Zhen Yu Li, our SVP in charge of Baidu Intelligent Driving. 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 Security Delegation 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 documents filed 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 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 it's 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.
spk03: Hello, everyone. I'm pleased with our financial performance in the second quarter. Revenue from online marketing increased by 15% year-over-year, reflecting improvement in advertiser sentiment. In Q2, non-GAAP operating profit for mobile ecosystem continued to strengthen, and AI Cloud once again delivered positive operating profit. At the same time, we are facing tremendous opportunities in foundation models, or more broadly, AGI. Today, I would like to share an update on our ongoing business transformation by utilizing Ernie and ErnieBot, as well as our achievements in this field. After that, I'll briefly go over the operational highlights for each of our bins. Over the past few months, foundation models have captured the imagination of people and businesses around the world. It is becoming increasingly clear that foundation models will fundamentally transform work across industries, boosting overall productivity and accelerate the wider democratization of AI innovation. Right now, we are in the midst of a thrilling period. This is due to the fact that the hard work we poured into generative AI over the years is now beginning to bear fruit. This pivotal and transformative opportunity is setting the stage for Baidu's future success, and we couldn't be more excited about it. On product and service front, we're using early and early bot to improve, rebuild, and create new offerings. On the product side, we are reinventing our products, building AI-native apps. For by research, AI has played an important role for many years in driving innovation and improving user experience. resulting in a consistent increase in the percentage of search queries satisfied by one-shot search results. It means the first search result displayed on the result page that can provide a satisfactory response to a search query. We're using EarlyBot to enhance search experience, making by research capable of answering complex questions that were previously unanswerable For example, please craft a marketing campaign proposal for a smartphone launch event. Or what kind of new energy vehicles would be suitable for a family of six with a budget of 300,000 RMB? ErnieBot also enables by-do search to assist users with more personalized and in-depth research on a topic or project. The feedback from users who have tested these new features has been positive. Another example of product innovation is Baidu One Cool, where many users search for articles, papers, books, or templates to create documents on a wide range of topics and in many formats. We are currently beta testing an AI assistance feature that can generate customized content for users based on their requests. User engagement and retention among Baidu Wenku's testing users for such features saw significant improvement. For example, users have doubled their time spent on Wenku after trying out the AI Assistant feature, and they are more likely to become paying users. Currently, WinQ has over 100 million monthly active users, and we believe such an intelligent feature will help WinQ attract even more users and convert non-paying users to paying users once it is rolled out on a large scale. Generative AI offers bright horizons for our online marketing business. and we are actively using the technology to build and rebuild our marketing products. Actually, our online marketing business has relied on AI for years to provide more value to advertisers. Last quarter, we shared our progress on using AIGC to automate app creation in text format. which leads to an increase in conversion rate. This quarter, we further enable advertisers to use AIGC to generate image and video ads with natural language. In Q2, we further improved our monetization system by using generative AI to broaden ad campaign keywords to user search queries, driving better monetization for advertisers. For example, and online professional education companies' ad conversion increased by about 16% in August after using these new features. We are consistently enhancing our system using generative AI and we expect to introduce more and more new features in the coming months. Additionally, we have been leveraging generative AI to enhance our auction system so that it can better match ads to search queries and search intent. Such improvement helped increase Baidu's search eCPM in Q2 and contributed to the year-over-year growth in online marketing revenue in the quarter. We're also in active conversations with advertisers to get their input on how to further make our marketing products work best for them. Meanwhile, we will continue to leverage Ernie and Erniebot to help advertisers create sophisticated ad campaigns and improve ROI on Baidu. One other initiative I'd like to mention here is that internally, we introduced an AI assistant within InfoFlow, Baidu's self-developed enterprise communication and collaboration platform. The AI Assistant automates various workflows such as summarizing meeting notes, chat histories, and workplace content, drafting documents, generating charts, facilitating knowledge Q&A, and completing tasks such as creating meeting invites, applying for vacation days, and conducting data analysis using natural language. This assistance helps our employees work more productively. We plan to open up the intelligent features of InfoFlow to our customers in the future. Alongside our own products, we are empowering cloud customers to build their transformative products and services using EarlyBot. In Q2, the number of corporates connecting to EarlyBot continued to grow. Foundation models Large language models and generative AI are expanding our total addressable market, attracting new customers while also increasing sales to our existing customers. We are also using Early and EarlyBot to help customers in various industries address real-world challenges with unprecedented effectiveness and competence. This further strengthens the capabilities of Ernie and ErnieBot, empowering them to take on more significant roles in solving industry-specific issues. For example, in the software development industry, we launched BaiduComet, an AI coding assistant to the public in June. BaiduComet aids people in coding and can be used in multiple programming languages. It has now been widely adopted by Baidu's own R&D team, who have reported a meaningful improvement in their productivity after using Comet. As of today, more than a hundred organizations have tested Baidu Comet, and some of them have already decided to purchase Baidu Comet to boost productivity in coding. In the healthcare industry, We are solidifying our presence by utilizing industry expertise and sales network developed through our collaboration with leading hospitals in China. Gu Sheng Tang Limited, a traditional Chinese medicine clinic chain, is using our early-based model developed for the healthcare industry to support doctors in their patient care. Our AI assistant, helps doctors write notes for patients, analyze medical images, and make informed clinical decisions among other tasks. It also helps patients find the most suitable doctors and departments and provide pre-consultation service online. Furthermore, we've developed tools for model training and fine-tuning. data processing, labeling, and more. These tools help to lower the threshold and cost for enterprises to use foundation models on Baidu and support them to train and operate their purpose-built models efficiently. Additionally, Baidu AI Cloud stands out as our AI optimized cloud infrastructure makes us a top performing platform for training and serving foundation models, including large language models. This is why more and more enterprises are upping their digital spend on our cloud infrastructure to capture the new AI opportunities. This is particularly obvious among internet and tech companies as they are early adopters of Gen-AI and foundation models. Shifting our focus to technology, at Baidu, we maintain an unwavering commitment to improve early. During Q2, we unveiled Early 3.5, our latest state-of-the-art foundation model powered by PedoPedo. Notably, PedoPedo's enhancements enabled our self-developed four-layer AI architecture to operate more seamlessly than ever before, significantly improve the framework's compatibility with Ernie and demonstrating engineering excellence. As a result, compared to Ernie 3.0 version in March, Ernie 3.5 has tripled its training throughput and its queries per second, or QPS, for inference have increased by more than 30 times. Additionally, I'm proud to note that IDC recently reported that Ernie 3.5 surpassed peers in many areas, such as algorithm, industry coverage, developer tools, and ecosystem. As compared to Ernie 3.5, 3.5 is capable of producing safer, more responsible, and more creative responses, making significant improvements in question answering, reasoning, and coding. Ernie 3.5 is now powering ErnieBot with plugins to expand its functionality to cover real-time and precise information, long text summary, data analysis and visualization, text-to-video conversion, and facilitate dialogue that include images. We are committed to enriching our ecosystem by adding more high-quality plugins, in particular from third parties, to contribute to the advancement of China's foundation model. On the ecosystem front, We are determined to make Ernie the most popular foundation model in China. Importantly, we believe foundation models should be able to well support problem-solving capabilities for various industries. As a result, we need to work collaboratively with our partners, such as enterprises and AI developers, to continually build industry models and solutions. Our 8 million developers on Pedal to Pedal are the anchor that we can leverage to build a community for early. To foster a vibrant ecosystem, we hosted Baidu Early Cup Innovation Challenge. It attracted almost 1,000 startups to submit their ideas and prototypes, covering various fields, encompassing productivity towards developers sales and marketing companies, entertainment companies, social platforms, middleware developers, as well as applications across different industries, including education, healthcare, finance, et cetera. We have also launched a venture fund of 1 billion RMB to support startups in developing all kinds of AI native applications, which will complement our organic growth. On regulation, Baidu was recently appointed as a leader of China's LLM Standardization Task Force at the World AI Conference. This position signifies national-level recognition and endorsement of our foundation models and AI capabilities. While we are still waiting for the green light for large-scale rollout of Ernie Bot, we have observed that Chinese government has been increasingly supportive of the development of Trend AI and LLM. As a market leader, we believe Baidu is well capable of benefiting from the opportunity and contributing to this mega trend in China. Now, let's have a quick look of the second quarter operating highlights for each business. Revenue from AI Cloud increased by 5% year-over-year to RMB $4.5 billion in the quarter, and AI Cloud maintained positive non-gap operating profit. Our business has become healthier than ever, laying a solid foundation for future growth. On intelligent driving, In Q2, the rights provided by ApolloGo increased by about 150% year-over-year to around 714,000, resulting in a cumulative right exceeding 3.3 million, which is greater than our closest competitors by orders of magnitude. This leadership in operation implies competitive advantages in data quantity and quality, thus better model and greater safety rates. In addition, ApolloGo has widened its footprint in fully driverless ride-hailing services in more cities. In mid-June, ApolloGo received a permit from Shenzhen Pingshan Government to offer fully driverless ride-hailing services to the public. And in early July, in Shanghai, we were allowed to conduct fully driverless testing on open roads as well. With the expansion of the area and fleet size for fully driverless operation, along with the improvement of operational efficiency, we witnessed growth not only in the total average daily orders, but also in the portion of fully driverless orders within the overall order portfolio. This highlights a promising pathway for improving the unit economics of autonomous driving business. In particular, in Wuhan, where we initiated fully driverless ride-hailing services one year ago, the UE continued to improve in the past few quarters. On the revenue side, the average daily order volume and revenue per order have surged, propelling total revenue to grow. Meanwhile, the cost per kilometer per car is decreasing as well, thanks to improved operational efficiency. Finally, on mobile ecosystem, our users continue to grow, with Baidu apps' MAUs increasing by 8% year-over-year to 600% 77 million in June. Also in June, videos distributed by Baidu app achieved double digit growth year over year. Baidu remains a significant platform for users who are seeking a wide range of content such as information, products, and services. The revenue growth rate for online marketing accelerated in Q2, adding by strength in various verticals with offline exposure, including healthcare, business services, local services, and travel. At the same time, revenue from e-commerce demonstrated strength and outperformed in the quarter. Non-GAAP operating profit for mobile ecosystem grew steadily year over year in the quarter. mobile ecosystem continue to generate robust cash flow to fund our investments in AI, particularly in foundation models and generative AI. We are confident that revenue, profits, and cash flow generated from our mobile ecosystem will remain strong in the future. With that, let me turn the call over to Long to go through our financial results.
spk11: Thank you, Robin. Now let me walk you through the details of our second quarter financial results. Total revenue was RMB 34.1 billion, increasing 15%, one-five year-over-year. Revenue from Baidu Core was RMB 26.4 billion, increasing 14%, one-four year-over-year. Baidu Core's online marketing revenue was RMB 19.6 billion, increasing 15%, one-five year-over-year. Baidu Core's non-online marketing revenue was RMB 6.8 billion, up 12% year-over-year. And in Q2, AI cloud revenue increased by 5% year-over-year to RMB 4.5 billion. Revenue from IT was RMB 7.8 billion, increasing 17.17% year-over-year. Cost of revenue was RMB 16.2 billion, increasing 7% year-over-year. Vital cost of revenue was RMB 10.6 billion, increasing 4% year-over-year. Operating expenses were RMB 12.7 billion, increasing 14% 1.4 year-over-year, primarily due to an increase in channel spending, promotional market expenses, server depreciation expenses, and cloud-related expenses, which supports the earning board research inputs, and partially offset by the decrease in personnel-related expenses. Vital cost operating expenses were RMB 11.3 billion, increasing 15% 1.5 year-over-year. By-to-call SG&A expenses were RMB 5.3 billion, increasing 34% year-over-year. SG&A accounting for 20% of by-to-call revenue in the quarter, compared to 17% in the same period last year. By-to-call IMD expenses was RMB 5.9 billion, increasing 2% year-over-year. IMD accounting for 23% of by-to-call revenue in the quarter, and decreased from 25% in the same period last year. Operating income was RMB 5.2 billion, Baidu Core's operating income was RMB 4.6 billion, and Baidu Core's operating margin was 17%. Non-GAAP operating income was RMB 7.3 billion, Non-GAAP Baidu Core's operating income was RMB 6.5 billion, and Non-GAAP Baidu Core operating margin was 25%. Total other income net was RMB 1.4 billion, compared to RMB 151 million in the same period last year, primarily due to the increase in net foreign exchange gain and net interest income, partially offset by the increase of fair value loss from long-term investments. Income tax expenses was RMB 1.3 billion, compared to RMB 25 million in the same period last year. The lower level of income tax expenses in the second quarter of 2022 is primarily due to the reversal of certain tax expenses based on the 2021 tax return. Apart from the reversals, the main reason for the increase of income tax expenses is the increase in profit before tax year over year. Net income attributable to Baidu was RMB 5.2 billion. and diluted earnings per ADS was RMB 14.17. Net income attributable to Baidu Core was RMB 5 billion, and net margin for Baidu Core was 19% . Non-GAAP net income attributable to Baidu was RMB 8 billion. Non-GAAP diluted earnings per ADS was RMB 22.55. Non-GAAP net income attributable to Baidu Core was RMB 7.7 billion. Non-GAAP net margin for Baidu Core was 29%. As of June 30, 2023, cash, cash equivalents, restricted cash, and shorting investments were RMB 201.5 billion. And cash, cash equivalents, restricted cash, and shorting investments, excluding ITE, were RMB 196.9 billion. Free cash flow was RMB 7.9 billion. And free cash flow, including ITE, was RMB 7.1 billion. Baidu Corp. had approximately 35,000 employees as of June 30, 2023. With that, operator, let's now open the call to questions.
spk04: Thank you. We will now begin the question and answer session. If you wish to ask a question, please press star 1 on your telephone and wait for your name to be announced. If you wish to cancel your request, please press star 2. If you are on a speakerphone, please pick up the handset to ask your question. The first question today comes from Alicia Yap with Citigroup.
spk08: Please go ahead. Hi, thank you. Good evening, management. Thanks for taking my questions. Congrats on the solid results. So can management provide update on the current state of the advertising industry, including how online and offline platforms are recovering? Can management also share about the pace of recovery for key industry and which ones have stronger seasonality? And then in addition, can management also discuss the main drivers for the growth in this industry? And finally, what are the medium-term growth targets for ads as we are now already behind the pandemic and with the current latest macro situation? Thank you.
spk03: Hi, Alicia. This is Robin. Let me answer your question. Our online marketing revenue increased by 15% year-over-year last quarter. I think a number of drivers contributed to this growth. On the vertical side, as we are coming out of the pandemic, many offline verticals like healthcare, bin services, local services, and travel continue to outperform. For example, healthcare, we saw solid year-over-year growth in both search queries and eCPMs. which leads to revenue strength. We're seeing this momentum continue. Offline sector contributes to a sizable amount of total online marketing revenue to Baidu. And in the meantime, e-commerce continue to grow in Q2 and remain top revenue contributor. For a market like e-commerce, it's highly competitive. So more and more merchants, they come to realize the value of Baidu, and they come to us to acquire new users and buyers, and also invite their existing buyers to come back and buy more. For the second half of the year, we should continue to see a pretty clear recovery trend for our online marketing business. Also remember, search remains the most effective form of performance-based ads. This is because users approach search with a clear intent, and search ads allow us to connect this intention directly with the most relevant product service offering. In fact, the eCPM for search outgrow other ad formats in the quarter, reflecting this effectiveness. Due to the relatively high comp base in Q3, the growth rate might not be as high as Q2, but it should well outperform China's GDP growth. Additionally, as we have more immersive video features, more and more users are watching the short videos on Baidu. Short videos distributed through the Baidu app. again grow by double digit in Q2. As a result, revenue from short videos continue to grow as well, contributing meaningfully to the overall online marketing revenue growth. Looking into the mid to long term, I believe both user traffic growth and monetization upgrades are important growth drivers for our online marketing business. And on the user side, I talked about our MAU. It grows by 8% year-over-year, Baidu APP's growth rate in June, in the month of June. If we look at search, we are renovating the Baidu search step-by-step using EarlyBot. We believe it will provide users with innovative and intuitive experience, which will help our mobile ecosystem gain traffic and increase user time spent. And on the monetization side, we've been leveraging AI to improve ad targeting capabilities in our bidding system for a number of years. And now with Ernie and Ernie Bot, we're further upgrading the game. I mentioned earlier that we have made some major improvements in advertising technology by using LLM and generative AI, along for more innovative and personalized ads. I believe there are lots of room for growth in this aspect, and I look forward to further breakthroughs. Such improvements in monetization technology have started to contribute to revenue and ECPM growth starting from this quarter and Q3. Thank you.
spk04: The next question comes from Gary Yu with Morgan Stanley. Please go ahead.
spk14: Hi. Thank you for the opportunity to ask questions and congratulate on the solid set of results. I have a couple of questions related to AI product and monetization. First is, can management share some insight on how users have responded to the integration of Ernie Bot in multiple products? Any new development or features that have come up with the Ernie Bot offerings? Help with market growth in China internet, and specifically in the mobile internet sector. And also, lastly, I would like to know more about the monetization that would play a role in your advertising business. And also, update in key industries and use cases related to B-side, can you provide additional information on the business model and also potential revenue, specifically, would like to gain a better understanding on financial impact on cloud performance. Thank you.
spk03: Hi, Gary. Let me answer the consumer-facing products part of your question, and I'll have Doug talk about the enterprise side, especially cloud business. I think it's clear we are in a paradigm shift in our business with our early commitment to AI we saw this time coming. We are reinventing our consumer-facing products with an AI-native mindset. As I mentioned in my opening remarks, we are rebuilding our search experience. Users who are testing our early bot-enabled Baidu search and Baidu app tend to ask more questions that were not frequently searched in the past. And they tend to have multi-round conversation and interaction. This is quite different from the user behavior for traditional search. Therefore, it's an incremental business for us. We are using generative AI to construct direct answers to user queries. This typically creates better and easier to understand answers than expert from existing web pages. We see increases in click-through rate and user retention rate for this kind of changes. We also launched the testing version of EarlyBot app, a standalone app that is built upon our latest generative AI and LLM. It aims to serve users as a personal assistant. We see EarlyBot as a potential new traffic gateway that can connect users to various applications, including Baidu and other third-party applications to address users' needs. We also launched EarlyBot plugins like a search, like a chat file, e-charge, et cetera. Developers will soon be able to submit their applications as early-bought plugins. For Baidu Wenku, with the AI Assistant feature, on average, user time spent more than doubled, and the seven-day retention rate has increased by almost 10%. With the AI Assistant, I will not be surprised that we can convert more users into paying accounts and charge a premium for the new AI function in the future. Overall, we are highly encouraged by user feedback. This is just the beginning for us to reconstruct all the products and businesses with LLM and generative AI. Very importantly, we have also seen promising opportunities to lift the monetization capability by rebuilding our advertising system. As I mentioned in the prepared remark, early and early bots allow us to continue improving our targeting and ad bidding system. All in all, I believe that generative AI will help us gain market share of many products and ultimately become a new growth driver for our online marketing business.
spk15: Hi, Gary. For your second part, for the impact on our cloud business, I would first mention that the beauty of the foundation model and generative AI is that it can help organizations in many industries to increase productivity and efficiency. We see more and more enterprises willing to adopt such latest AI technologies, and that's why we would say the new technology is increasing our time. Specifically, Ernie and Ernie Bot have been helping our customers from different industries to handle their business challenges more effectively, ranging from software internet, healthcare, education finance, and even to the public sector. So they use Earning and EarnBot to train models, to build applications, and to improve solutions for efficiency gain, such as improving their customer service, producing different kinds of content, writing code, providing knowledge searches within their organizations, as well as increasing their sales and marketing efforts. As to the business models you just asked, I think it's still at its very early stage and it's evolving. Though we have our preference, but we are pretty open and would like to take any effort to speed up the growth of generative AI. So in addition to providing AI computing infrastructure to our customers, we also consider charging customers for the usage of early and early bot or using our AI applications such as CoMate, Robin just mentioned. And also, we can charge our customers for retraining or operating their own models on our platform. Or we can deploy the foundation models to our customers' private cloud on the project basis. So we expect the revenue generated from this new opportunity to be gradual. But we believe the long-term potential is immense. So we'll continue to invest in Ernie and her bot to serve our customers well for the long term. That's it, Gary.
spk04: The next question comes from Lincoln Kong with Goldman Sachs. Please go ahead.
spk13: Thank you, management, for taking my question, and congrats on the solid quarter. So my question is about the cloud business. So could management help us understand sort of the main reason for the revenue slowdown for cloud this quarter, and when do we expect the revenue to pick up or accelerate again? When we're looking ahead into the second half of our entire year of 2023, how is the revenue growth trajectory expected to play out? and what factors will be driving that. And also on the profitability side, I'm also curious whether this profit break-even on a non-GAAP operating level is sustainable, and what's the current trend for the AI cloud profitability? Thank you.
spk15: Okay, thank you for that question. I'll take it. So cloud revenue grow 5% year-over-year in Q4. and we continue to generate profit on the non-GAAP operating level. It means that our business is becoming healthier, laying a solid foundation for sustainable growth in the future. The growth rate was mainly dragged by government-related projects, which usually take a longer time to recover after the pandemic. But many enterprise-related projects in China the key verticals like manufacturing, utilities, and internet service have demonstrated quite solid growth in Q2. As I just mentioned before, we see strong interest from existing and new customers in generative AI. So while some of them directly leverage Ernie or Erdibot to enhance their products, to improve their operational efficiency, so some of them actually choose our AI infrastructure as the foundation for their models and applications. Though it may take a little while to see significant revenue contribution, but this opens up a very new space for our cloud business, where we're already in a leading position. So on the profit side, we believe we can keep generating operating profit from AI Cloud, as we did in the previous quarters. So this indicates that actually our business is becoming increasingly healthy. But we are going to keep standardized our AI solutions that can effectively address AI critical pinpoints and then replicate them from one project to another, which we believe will further grow the margin and profit.
spk04: The next question comes from Kenneth Fong with Credit Suisse. Please go ahead.
spk12: Hi, good evening, management. Thank you for taking my questions and congrats on the solid set of results. I have a question regarding the generative AI. Regarding the recent introduced interim measures to regulate generative AI service by the government, could you provide insight into whether this suggests an imminent authorization of the large language models? Additionally, do you have a projected timeline for the rollout of the applications utilizing these models? Thank you.
spk03: Thank you for your question, Ken. We've seen the government continue to support for tech innovation, including AI and foundation models. I think a good sign of this is that the new guideline and specifications that came out in mid-July about how to use generative AI technology and related services in China. This is a new version, and this version went effectively, I think, on August 15th. Compared to the previous version, this version, you can probably tell, it's more pro-innovation than regulation. We are still waiting for the green light for large-scale rollout of early bot for its use in the consumer-facing apps. But as I said before, the government has increasingly recognized early and early bot. which we believe provides a good foundation for the eventual release of Ernie Bot on a large scale. Generative AI is pretty new, and it's understandable that people might have concerns about things like user privacy, IP protection, AI ethics. So there should be certain regulatory requirements in place. At Baidu, we have accumulated extensive experience in providing appropriate information to the vast amount of internet users. We are a leader in not only AI technology for commercial purposes, but also using AI for good. In fact, we are working closely with the regulators and other organizations to push the development appropriate usage of generative AI. We believe we are well positioned to benefit from this opportunity and contribute to this megatrend of generative AI in China. We don't have an exact date for everything. But the trend is very promising, and we are quite optimistic about the future of a better regulatory involvement.
spk04: The next question comes from Wei Zhong with UBS. Please go ahead.
spk06: Hey, good evening, management. Thank you for taking my questions, and congrats on the solid results. My question is also related to the foundation model. How can we assess Baidu's foundation model's performance comparing to other Chinese peers? And also related to that, when we think about industry-specific models versus foundation models, how the balance between these two will shape up the broader AI landscape in the future. And now we have more companies offering or focusing on models tailored to specific industries. So how does management perceive the landscape in terms of opportunities and competition? So what are our competitive edges? Thank you.
spk03: These are very good questions. When it comes to foundation model in China, early 3.5 clearly stands out. People have witnessed the fast iteration and improvement over the past five, six months. Our unique full-stack AI capabilities differentiate us from our competitors and give us an edge in the market. As mentioned in my prepared remarks, an IDC report recently published named Earning 3.5 as a leader across various aspects. Like many other technologies, foundation models' speed of innovation depend on the applications that use this model. Baidu is an AI company with a strong portfolio of internet products. These products are all being rebuilt and reconstructed with ErnieBot, therefore drives the innovation of Ernie into the right direction. In other words, we know what kind of problem to solve. and I think many other peers don't really know. They just use open domain testing sets to evaluate the effectiveness of their models. We also have tens of thousands of cloud customers testing our new bot. They provide valuable feedbacks that propel the improvement too. As for the Industry-specific models you mentioned, they should be built upon the most powerful foundation models. I think that with time, foundation models will rapidly evolve, given its strong capabilities to learn new subjects and gain industry insight. And the standalone industry-specific models will have a hard time to keep up with the pace of innovation. In fact, since Earning 3.5 became available, an increasing number of customers have told us that Earning 3.5 is visibly more advanced than Earning 3.0. And it has helped them solve a growing number of pain points. And as you can imagine, we are working toward Earning 4.0. should launch it by the end of the year. I believe that ultimately only a select few companies will reach this level of advancement for foundation models, and Baidu will be one of them. During this transitional period, we are focused on multiple initiatives to drive the adoption of foundation models. We're working hard to build early-powered applications and solutions for different industries and scenarios. It's such an example. With the upgrade to Ernie 3.5, we are swiftly rolling out to empower enterprises to create industry-specific models and applications that tackle their own challenges. Also, we offer a mass platform, which includes not only Ernie, but also a variety of models. This allows businesses to easily view, fine-tune, and operate their own models using their own data. We provide a set of valuable tools on our platform, enabling our customers to easily enhance their model training and application development. Of course, enterprises can directly access earnings capabilities through API to enhance their business. Recognizing that in the early stages, enterprises prioritize security. So we do provide the flexibility of using our foundation model services on both public and private projects. Our self-developed four-layer AI architecture is our core competitive advantage because we deliver better model performance when we serve our customers. as shown in the testing phase over the past few months. Our strengths in model training efficiency and cost effectiveness, in turn, attract more enterprises to partner with us. So we gain more and more industry know-how and insights, and building a reinforcing positive cycle to further improve earnings. Our ultimate goal is to cultivate an AI-native ecosystem centered around Earth. We believe that the true power of a foundation model lies in driving a wide array of high-quality AI-native applications, much like an operating system. Some of the applications are built by us, while others were built by I would say most will be built by our ecosystem partners to address industry-specific challenges. To enhance this ecosystem, we're happy to invest in other companies to complement our organic growth. Just as I mentioned in the prepared remarks, we hosted Baidu Early Cup Innovation Challenge and launched a venture fund to support startups in developing all kinds of AI-native applications. To sum up, we will continue to improve Ernie, make Ernie easy to use, and help enterprises in any industry to not only retain or fine-tune models on top of Ernie, but more importantly, leverage Ernie and Erniebot to develop their own applications. We aim to make earnings the preferred choice among AI developers so that more and more applications will be built upon earnings and making our ecosystem vibrant. Thank you.
spk04: The next question comes from Miranda Zhuang with the Bank of America. Please go ahead. Thank you.
spk05: Good evening, Majmin. Thanks for taking my questions. So lately, open source AI models have attracting significant market attention. So can management provide your insights into Baidu's perspective on this front and the strategy regarding the open source versus the closed AI models? Thanks.
spk15: Thank you, Miranda. This is a very interesting question, and I believe there are quite different views on this. I think both open source and closed source foundation models will coexist for a while. In fact, actually, we have already seen many models trained on top of the open source foundation models, powering tons of applications, even today on our cloud platform. But it's really tough, actually, for open source foundation models to keep evolving and getting better due to the lack of effective feedback loops, which play a super important role in improving the foundation models. And additionally, building and upgrading foundation models can be very expensive. So it is really important to build a durable business model to fund the sustainable improvement of the foundation models so that they can finally provide long-term value to the market. First, it's our priority that Ernie keeps evolving quickly and remains the market leader, which can then attract more and more customers to use our platform to create innovative apps and improve their business. with their feedback from the real world scenarios, we can further improve our foundation models. Looking ahead, I believe that there will be a limited supply of advanced foundation models, whether open source or closed source, or even industry specific, available in the market. So like we just talked, as the industry Progressives and enterprises will need advanced models to develop more sophisticated applications. And I bet Ernie will be one of the few foundation models that can satisfy those evolving demands in the long run.
spk04: The next question comes from Alex Yao with J.P. Morgan. Please go ahead.
spk10: Thank you, Benjamin, for taking the question. Congrats on a strong quarter. I have a couple of questions on the margin outlook. Considering the opportunity of AIGC and the large language models, how should we think about the investment trajectory for the second half this year? Additionally, once we start large-scale rollouts, what direct cost will be incurred? how should we compare the margin of early bot or large language model related to products, related products than the existing business, i.e. search, et cetera. More broadly, could mention to also provide directional color on the trend of ideas of core margins when we look into the second half this year. After the significant cost reduction, the efficiency improvements, incremental last year, what will be the key drivers of the project moving forward? Thank you.
spk11: Thank you, Alex. Let me ask you a question. This is Julius. I think previously, Robin has already talked about a lot of opportunities of the earning boss. So in the first place, I would like to literally that we are highly committed to the long-term investments in these promised areas. And to support the fast upgrade of earnings, we have invested in competition infrastructures. If you look into our cash flow statements, you probably can see that the buy-do cost capex in Q2 have increased over last year. The increase is mainly due to the hardware purchase to support the training and iterations of all the AI campaigns, such as Earnings and Earning Boss. But the impact on the income statement side, which is quite manageable, that's because all such power depreciations in general will spread over the next few years, and the impact on the near-term profitability is not substantial. The investment is expected to continue, and the magnitude of the future investments will be highly correlated to the pace of the actual business expansions, rather than only supporting the model training, which means that in the near future, the spending will be sponsored by the incremental revenue, which is generated from our earning boards and formulates our positive virtual circles. And regarding to the margin for the financial model and related products, I believe it's a little bit earlier to discuss at the current stage as the BIMS model is still evolving. But as Romy has mentioned earlier, empowered by earnings, the search advertising could become more customized, which will result in an increased ROI for advertisers. The addictions for cloud services, as Doug has just talked about there, The AGI and the financial models will empower various industries and enhance the efficiencies across different industries. So from a long-term perspective, we believe that our customers will be increasingly motivated to leverage our services to enhance their efficiencies and increasing our pricing power and resulting in higher margins as compared to the current cloud business. If we look into the second half of this year for mobile ecosystem, we will continue to serve as a cash call for the group. In the long run, our goal for mobile ecosystem is to consistently achieve high margins. For the AI cloud side, we aim to achieve the profitability on a non-GETOP level and improve margins over time. For the intelligent driving, while we firmly believe that it's a huge long-term opportunity, we will continue to make sure we invest at a measured pace. As a whole, our main focus is on developing sustainable growth strategy for each business, which prioritizes the long-term thinking. In addition, we plan to invest in earning and earning ball to take advantage of the huge opportunities presenting by the foundation models. Thank you, Alex.
spk04: The next question comes from Thomas Chong with Jefferies. Please go ahead.
spk09: Good evening. Thanks, management, for taking my questions, and congratulations on a strong set of results. I have two questions. The first is about our mobile taxi business. Management mentioned that the unit economics in some cities are improving. Could management share the logic and the efforts taken behind it? What's your plan in terms of the city expansion and the volume target for fully driverless operations in the upcoming quarters? In addition, when can we expect the LT6 vehicle to be launched? What's the projected size of the vehicle fit for this year and how could it potentially impact the cost. My second question is on auto solution. Which vehicle models will hit the road in the later half of this year and next year, and how should we size up their financial impact on the books? Thank you.
spk02: Thanks for the question. This is Jin Yu. I will start from the road taxi question. Our autonomous ride-hailing service is rapidly gaining traction in major cities. providing valuable data that fuels the development of our driverless technology. This data is instrumental in enhancing the safety, efficiency, and overall experience of our operations. As a result, we have gathered strong support from both customers and regulators, enabling us to expand into new locations and significantly increase our place. This positive feedback loop of large operations also drives the improvement of our model, and the improved model drives the expansion of operations, propelling us to the forefront of the global leading in autonomous driving technology. On the logic behind the UE improvement, here I would like to take Wuhan's operation as an example. We have continued to scale up our service in Wuhan in the past year. If you compare with the situation one year ago, you may notice that, firstly, our fully driverless fleet size has expanded significantly, going from five vehicles one year ago to almost 200 in August this year. Secondly, the operation area for fully driverless service has expanded by around 15 foot in Wuhan. Meanwhile, the operation hours have been further expanded, covering both pier hours in the morning and the evening, as well as the later night operations. Thirdly, ApolloGo is now serving more passengers than ever, becoming one of the largest airways providing fully driverless retailing service in the world. And fourthly, In terms of the order, the proportion of the driveless orders to all operations has increased from 10% in last August to 55% in July this year. All the progress mentioned above contributes to the improvement of OUE. Looking ahead, we will continue to focus on the operation in K-cities and operation regions. like Beijing and Wuhan, to further expand the fleet size and operation areas of driverless operations and thus gradually improve the unit economics. Next year, we aim to enter into more cities to conduct fully driverless ride-hailing services and expand the skills of fully driverless commercial operations. Regarding RT6, The timeline for mass production is proceeding as planned. It brings a significant cost advantage, with mass production costs lowered by 50%, compared with the previous generation. So it will be glad to become the majority in our operation recovery, providing the unit economic closer to profitability point. Last, on auto solution business. Currently, prevailing assistance-driven products do not fully meet the standards in terms of generalization and safety, failing to satisfy consumer demands for convenience and efficiency. Starting this year, the intense price competition in China's EV market has led to a general decline in the profitability within the automotive sector. And OEMs have been showing softened demand for intelligent driving and leaning more towards cost-cutting models to stimulate sales, which has affected and is also expected to offer auto solutions in the short term. However, we believe there are much opportunity ahead in the long run. Driven by the consumer demand for intelligent driving, we plan to launch our safety navigation pilot product, Apollo Safety Driving Map. at our own pace later this year. As we look ahead, we believe the change in the auto industry is heading towards intelligent driving with exciting opportunities in the future. And we will capitalize on this change with cutting-edge technologies and a sensible business model. Thank you.
spk04: This concludes our question and answer session and does conclude our conference for today. Thank you for participating. You may now disconnect.
Disclaimer