Title

Presentation

Baidu, Inc. misses on earnings expectations. Reported EPS is $0.2675 EPS, expectations were $2.3.

Operator: Hello, and thank you for standing by for Baidu’s first quarter 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 Li, Baidu’s Director of Investor Relations

Juan Li: Hello everyone and welcome to Baidu’s first quarter 2024 earnings conference call. Baidu’s earnings release was distributed earlier today and you can find a copy of our IR 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 U.S. Private Securities Litigation Reform Act of 1995. Forward-looking statements are subject to risks and uncertainties that may cause actual results to differ materially from our current expectations.

For a detailed discussion 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 statement 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. Our business continued to grow in the first quarter. Baidu Core total revenue increased by 4% year-over-year to RMB 23.8 billion, and non-GAAP operating margin reached 23.5%, an improvement from a year ago. In particular, revenue growth from Baidu AI Cloud accelerated to 12% year-over-year [indiscernible] while continuing to deliver operating profit on a non-GAAP basis. 2024 is the second year of our march on the gen-AI path. As we solidify our leadership position in foundation models, we are transforming the company from an internet-centric business to an AI-first business. Given that ERNIE is the most powerful LLM in China, we are aggressively pushing the envelope for both our 2C business and 2B business to adopt ERNIE to provide better user experience, to increase advertiser ROI, to enable developers to write agents and applications, to let customers enjoy more effective and more efficient models.

While we operate our legacy business in a challenging environment and experience lower revenue growth in the near term, we remain confident that AI will bring us sustained growth in revenue and profit in the long run. We expect our cloud business to accelerate and the loss of our robotaxi business to narrow for the rest of the year. We expect mobile business to be soft in the near term and start to recover when gen-AI becomes the new core of our existing products next year. Looking beyond the near term, gen-AI and foundation models will bring us tremendous opportunities, ushering in a new innovation cycle. Enterprise and individual developers have swiftly transitioned from the fear of missing out on this opportunity to leveraging foundation models like ERNIE to build AI applications.

Baidu is well prepared to benefit greatly from this technology transformation. We believe one of the most important long-term opportunities in model inferencing, which will be a key growth driver for our AI Cloud revenue in the future. In April, ERNIE handled about 200 million API calls daily, a significant jump from around 50 million in December last year. This considerable growth indicates that increasing adoption of ERNIE can point to strong future revenue potential from model inferencing. To accelerate the adoption of ERNIE, we are building a vibrant and healthy ecosystem around it. We believe ERNIE ecosystem will over time contain millions of applications, especially agents developed by a diverse community of enterprise and individual developers across various industries, meeting a wide range of needs in people’s everyday lives and work.

Our large user base in mobile and desktop will enable us to distribute these agents and apps to whoever needs it, whenever appropriate. The anchor of this ecosystem is the ERNIE family of models, including our flagship models, ERNIE 3.5 and ERNIE 4.0, as well as the lightweight models we introduced in Q1. Throughout the quarter, we continued improving ERNIE’s efficiency, leveraging our unique proprietary four layer AI architecture and our strong ability in end-to-end optimization. For example, ERNIE has increased its training efficiency to 5.1 times and its inference cost is only about 1% compared to the March 2023 version. To make ERNIE increasingly accessible and affordable, we now offer three sets of tools on our mass platform. Last quarter, we introduced AppBuilder and ModelBuilder for enterprise and individual developers to develop apps and build models.

In April, we took it a step further by introducing AgentBuilder, a platform encompassing tools for easily creating AI agents. This is because we envision AI agent will become one of the most important forms of applications powered by gen-AI and foundation models. With the ability to use natural language as the programming language, developers will be able to build AI agents without the need to write a single line of code. Currently, new early agents are created on our platform every day, and together they are distributed millions of times per day, serving a wide range of verticals including education, legal, B2B, travel, and more. All these initiatives derive from our extensive experience and insights in building and running ERNIE, as well as developing AI-native applications.

We believe that ERNIE’s true value will only be realized when numerous applications built on top of it are widely used by users and customers. I’m pleased to note that ERNIE is extending [indiscernible] across smart devices through API. Last quarter, we proudly announced a partnership between all the smart phone brands, such as Samsung China and Honor, assisting them in enhancing their native app experiences using ERNIE. This quarter, we are excited to extend our collaboration with more leading smartphone makers such as Oppo, Vivo and Xiaomi, who leveraged ERNIE APIs to elevate user experiences. Moreover, our reach now extends beyond smartphones to include PCs and electric vehicles. ERNIE APIs are now utilized by Lenovo, a top PC brand to empower its AI assistant in the default browser of its PC.

Nio, China’s leading smart EV manufacturer, began using the ERNIE API to enhance the in-cabin experience for its vehicles. This broadening of our partnerships into various smart devices opens up ample opportunities for large scale user adoption, paving the way for ERNIE-enabled applications to become the entry point into the world of generative-AI. In addition the brands I just mentioned, we have also acquired many notable new customers, such as Trip.com, [indiscernible] and Singapore Tourism Board. Another long term opportunity is in our consumer-facing business. We have been reconstructing all of our consumer-facing products . Our goal is to build our proprietary AI-native applications, potentially killer apps for the ERNIE ecosystem. By doing so, we should be able to generate new growth opportunities.

For example, after rebuilding with gen-AI and LLMs, Baidu Wenku, our one-stop shop for document creation, experienced double-digit year-over-year increase in paying users in the first quarter. Penetration of ERNIE for Baidu Search in feed took longer than expected because the user base is on the order of hundreds of millions and use cases are generally very sensitive to cost and response time. We needed a wide range of ERNIE models in different [indiscernible[ optimized for different scenarios for best price-performance ratio. After trial and error for a few quarters, we are firming up our strategy. Going forward, we plan to accelerate the launch and adoption of new product features such as multi-model generative search results, multi [indiscernible] interaction in search, and more recently distribution of ERNIE agents.

We are at the forefront globally of this unique technological change and we are confident in our abilities to innovate. By definition, we are operating in uncharted territory. As always, we want to be flexible to make timely adjustments with evolving consumer needs and how users incorporate new product features in their day-to-day lives. Now let me review the key highlights for each business for the first quarter. In the first quarter, AI Cloud revenue reached RMB 4.7 billion, up 12% year-over-year, and continues to generate operating profit on a non-GAAP basis. The revenue growth was mainly driven by gen-AI and foundation models. In the first quarter, such revenue accounted for 6.9% of total AI Cloud revenue. Currently the majority of this revenue is from model training, but revenue from model inferencing has been growing quickly.

We believe revenue from gen-AI and foundation models will continue to rise as customer adoption improves. For instance, within our internal cloud revenue, Baidu Core, other [indiscernible] groups such as mobile ecosystem groups and intelligent driving groups are increasingly leveraging the power of ERNIE. As a result, 15% of their payments to the AI Cloud group are allocated to gen-AI and foundation models. Enterprises choose Baidu AI Cloud to train and host their models because they believe we have the most powerful and efficient AI infrastructure for model training and inferencing in China. Compared to our peers, we help enterprises to train models of all sizes on our AI Cloud while also reducing model inferencing cost. This is primarily attributable to two reasons: number one, our self-developed four layer AI architecture has allowed us to innovate and optimize at each layer, enabling continuous efficiency gains; and number two, we have superior capabilities and insights in GPU cluster management.

Leveraging our technical expertise, we can now integrate GPUs from various vendors into a unified computing cluster to train an LLM. Our platform has demonstrated very high efficiency with this set-up on a GPU cluster that is composed of hundreds, even thousands of GPUs. This is an important breakthrough because of the limited availability of imported GPUs. Another growth driver for AI Cloud is cross-selling of our CPU cloud services to our GPU cloud customers. With high recognition of our GPU cloud among existing and new customers, we have seen customers increasingly switch more and more of their CPU cloud usage to Baidu. [Indiscernible] earlier on the mass side, we took many initiatives to make the ERNIE family of models increasingly affordable and more efficient than open source models.

Here are some highlights for this quarter. We have expanded and enhanced our ERNIE model portfolio offering a total of three lightweight LLMs and two task-specific LLMs on ModelBuilder. These models help enterprises and professional developers balance model performance with cost using ERNIE to reach a broader audience for model development. In addition, our mixture of experts, or MOE approach can partition a user query into distinct tasks, assigning the most suitable models to handle each task and use ERNIE 3.5 or 4.0 only for the most complex tasks. This approach allows for faster responses and lower inferencing cost while maintaining similar performance levels to using more advanced models. Last quarter, we introduced AppBuilder to developers.

Throughout the quarter, we continued enriching and refining the tools for AppBuilder, enabling developers to easily create AI-native apps in just three steps on our platform. With the launch of AgentBuilder in April, anyone can create an AI agent with just a few sentences on Baidu. Overall, we remain confident in the [indiscernible] for our AI Cloud revenue and we aim to continue generating operating profit on a non-GAAP basis. Mobile ecosystem has continued to deliver healthy margins and strong free cash flow. In the first quarter, our online marketing revenue grew by 3% year-over-year. Revenue growth was impacted by a challenging macro environment. At the same time, we have been pushing hard to transform the user experience from a traditional mobile product to a generative-AI experience.

This transition is ongoing and monetization has not yet started. [Indiscernible] to leverage earnings to reconstruct our monetization system for better conversion and efficiency gains. During the quarter, we further enhanced our ad targeting capabilities and scaled up real-time ad generation. This effort resulted in an improvement in conversion and generated incremental revenue. ERNIE agents stand for a long-term opportunity for monetization upgrade, too. Recently we have seen not only brand advertisers but also SMEs gradually adopting ERNIE agents. We have designed these agents for SMEs as virtual storefront and service desks, serving consumers around the clock. We believe that the use of agents can improve SMEs’ sell-through rate, enhance their proximity and expand their reach among users.

This will be an important step for us to transform our traditional CPC model to a significantly more efficient CPS model, and meanwhile enhancing user experience on Baidu. Going forward, I believe greater opportunities will arise from AI-native apps, particularly for gen-AI enabled search. Gen-AI complements traditional search, expanding the total addressable market. Since Q2 last year, we have been reconstructing Baidu Search with ERNIE. Now more and more search results are generated by ERNIE in a growing variety of home apps like text, image, third party links, point of interest, and citation. These results are [indiscernible] produced in real time to directly address users’ questions and problems. By doing so, we have improved and will continue to enhance the search experience, which is crucial for increasing the usage of Baidu Search.

While user feedback on this product and feature renovations has been encouraging, it is important to note that we are still in the early stages of reconstructing Baidu Search with ERNIE. This process will likely take time given that Baidu Search has a history spanning over 20 years and user behavior will evolve gradually. Overall, I believe that Search will be one most likely killer apps in the gen-AI era, and we are on the right trajectory to capitalize on this potential. I mentioned Ernie agents as an important opportunity for monetization. With the newly introduced AgentBuilder, creators, publishers and service providers will find it increasingly easy to build agents on Baidu. This is key to enhancing Baidu’s content offerings and ultimately providing an AI-native user experience on our platform.

Moving onto intelligent driving, we believe Apollo Go stands as the largest autonomous ride hailing service provider globally, measured by the rides provided to the public. In the first quarter, Apollo Go offered about 826,000 rides to the public, marking a 25% year-over-year increase. In April, the total number of rides surpassed 6 million. We are continuing to move towards achieving unit economics breakeven for Apollo Go. To make this happen, our strategy is to reach UE breakeven in key regions and then replicate the success in other regions. To reach the regional breakeven point, we are focusing on scaling up [indiscernible] a fully driverless ride hailing service and enhance the utilization of each vehicle. Wuhan, Apollo Go’s largest regional operation is progressing towards this goal.

In Wuhan, Apollo Go is gradually becoming an integral part of the city’s transportation network. Apollo Go more than doubled its operational area from a quarter ago, serving a population of over 7 million and achieving the remarkable milestone of crossing Yangtze River with fully driverless vehicles as part of its expansion. Moreover, our vehicles started to operate 24/7 in Wuhan in early March, further expanding Apollo Go’s reach and improving the vehicle utilization. All this progress has led to the rapid growth of fully driverless rides. In Q1, the rides provided by fully driverless vehicles accounted for over 55% of total rides in Wuhan, which is up from 45% in the fourth quarter last year. This figure continues to rise, exceeding 70% in April with expectations of sustained rapid growth ahead and reaching 100% in the coming quarters.

Looking ahead, we plan to deploy our RT6, our generation robotaxi in our Wuhan Apollo Go operation this year, which will significantly reduce hardware depreciation costs. With the scaling of driverless operations and continuous improvement of cost structure, we believe Apollo Go will achieve operational UE breakeven in Wuhan in the near future. As Apollo Go continues to progress, we will closely monitor efficiency and persist in optimizing the operation of our overall intelligent driving business. On auto solutions, our Apollo self-driving or ASD technology continues to evolve. I mentioned in our last earnings call that Apollo is a global pioneer in the use of visual foundation models in autonomous driving. Now our state-of-the-art autonomous driving solution solely reliant on vision is made available to OEMs. ASD can effectively navigate complex urban environments across over 100 cities in China, with plans to expand into hundreds of cities in the coming months.

This allows us to make advanced autonomous driving attainable across a broad spectrum of passenger vehicles from high end to economy models priced as low as RMB 150,000, and serves as another proof of our technology leadership. 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 first quarter financial results. Total revenues were RMB 31.5 billion, increasing 1% year-over-year. Revenue from Baidu Core was RMB 23.8 billion, increasing 4% year-over-year. Baidu Core’s online marketing revenue was RMB 17 billion, increasing 3% year-over-year. Baidu Core’s non-online marketing revenue was RMB 6.8 billion, up 6% year-over-year mainly driven by the AI Cloud business. Revenue from [indiscernible] was RMB 7.9 billion, decreasing 5% year-over-year. Cost of revenue was RMB 15.3 billion, increasing 1% year-over-year primarily due to an increase in traffic acquisition costs and costs related to the AI Cloud business, partially offset by the decrease in content costs.

Operating expenses were RMB 10.7 billion, decreasing 2% year-over-year primarily due to a decrease in personnel-related expenses and other R&D expenditures, partially offset by the increase in server depreciation expenses and lower custody fees which support gen-AI research and development inputs. Baidu Core operating expenses were RMB 9.4 billion, decreasing 1% year-over-year. Baidu Core SG&A expenses were RMB 4.5 billion, decreasing 1% year-over-year. SG&A accounting for 19% of Baidu Core’s revenue in this quarter compared to 20% in the same period last year. Baidu Core R&D expenses were RMB 4.9 billion, decreasing 1% year-over-year. R&D accounting for 21% of Baidu Core’s revenue in this quarter compared to 22% in the same period last year.

Operating income was RMB 5.5 billion. Baidu Core’s operating income was RMB 4.5 billion, and Baidu Core’s operating margin was 19%, and non-GAAP operating income was RMB 6.7 billion. Non-GAAP Baidu Core operating income was RMB 5.6 billion, and non-GAAP Baidu Core operating margin was 23.5%. Total other income net was RMB 2 billion, decreasing 52% year-over-year primarily due to a decrease in favorable gain from long term investments, partially offset by the increase in net foreign exchange gain. Income tax expenses was RMB 883 million compared to RMB 1.2 billion in the same period last year. Net income attributable to Baidu was RMB 5.4 billion, and diluted earnings per ADS were RMB 14.9. Net income attributable to Baidu Core RMB 5.2 billion and net margin for Baidu Core was 22%.

Non-GAAP net income attributable to Baidu was RMB 7 billion. Non-GAAP diluted earnings per ADS was RMB 19.91. Non-GAAP net income attributable to Baidu Core was RMB 6.6 billion, and non-GAAP net margin for Baidu Core was 28%. As of March 31, 2024, cash, cash equivalents, restricted cash and short term investments was RMB 191.8 billion, and cash, cash equivalents, restricted cash and short term investments excluding iQIYI were RMB 185.8 billion. Free cash flow was RMB 4.2 billion. Free cash flow excluding iQIYI was RMB 3.3 billion. Finally, Baidu Core had approximately 36,000 employees as of March 31, 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 Alicia Yap with Citigroup.

Alicia Yap: Hi, thank you. Good evening Robin, Rong, and also–sorry, so thank you for taking my questions. I wanted to ask, are you able to quantify [indiscernible] AI technology been helping Baidu to improve app monetization rate, can management share some feedback on those advertisers who have adopted this system? In what kind of areas do they see the largest improvements, and are there any areas that can be further enhanced? Thank you.

Robin Li: Hi Alicia, this is Robin. As you know, our monetization system was the first to benefit from gen-AI, generating several hundred million RMB per quarter in incremental revenue. Since the second half of last year, we have been utilizing ERNIE to upgrade our monetization system, enhancing various aspects of ad technology. This includes adding capabilities, refining the [indiscernible] system, automating creative generation, and add strategy formation for our advertisers. Advertisers have been seeing better conversion and more sales leads. This improvement has motivated advertisers to increase their spending with Baidu. In the first quarter, AI-related incremental advertising revenue grew on a quarter-over-quarter basis and we expect this trend to continue.

The incremental revenue has helped us mitigate the broader macro weakness and also bought us some time to reconstruct our user product with ERNIE. As I emphasized in my opening remarks, we believe ERNIE offers a significant long term opportunity for our online marketing business. Agents function as virtual [indiscernible] for domains. Advertisers can establish their online presence and interact with potential customers using natural language through multi [indiscernible] conversations. With the introduction of AgentBuilder, advertisers can easily create customized ERNIE agents. When advertisers express their intentions to these agents, they can more effectively achieve their goals, whether it’s helping potential customers understand their products or improving customer service quality.

Agents can also help enrich our content offering and improve user experience on Baidu. While still in early stage, ERNIE agents have helped some advertisers achieve better LOIs. For example, we have a customer in online education. They used AgentBuilder to create its AI agent, injecting it with key insights like product introduction and subject matter expertise, while continually providing feedback for refinement. This agent has significantly enhanced the company’s online customer service by offering round the clock high quality consultations. The adoption of ERNIE agent led to a 20% increase in ad conversion rate for this online education company. I think this is just the beginning. We believe Agent will be a major form of content and services in the new AI era.

We will continue to improve the capabilities of ERNIE Agent. Agents will not only elevate user experience, both conversion and ROI for advertisers, but also over time foster increase in transactions directly generated on our platform. This shift should help us to transform our tradition CPC model into a more efficient CPS model.

Operator: Your next question comes from Gary Yu with Morgan Stanley.

Gary Yu: Hi. Thank you, management, for the opportunity to ask questions. I have a question regarding AI Cloud business. How has the price cut initiative by some of your peer companies affected Baidu AI Cloud business, and how should we think about the cloud revenue profitability as competition heats up, and also how should we–what should be the sustainable level of cloud growth outlook going forward? Thank you.

Dou Shen: Yes Gary, this is Dou. As we have already mentioned, gen-AI and foundation models are transforming the cloud industry from general purpose computing to AI computing. This shift is re-shaping the competitive landscape within the cloud industry, presenting with valuable opportunity to establish ourselves as a leader in AI Cloud. We believe we offer China’s most efficient AI infrastructure and most advanced mass platform for model training and inference. As a result, more and more enterprises are choosing us for model training, fine-tuning and AI-native app development on our public cloud. This increasing demand has significantly boosted our AI Cloud revenue. Since the second quarter, the second half of last year, our AI Cloud revenue growth has started to accelerate from a year-over-year decline in the third quarter to an 11% increase in the fourth quarter last year, and then further accelerated to 12% in the first quarter this year.

The revenue acceleration was supported by two main factors: the incremental revenue directly generated by gen-AI and foundation models, and also the new opportunities they provide to our legacy cloud business. As Robin just mentioned, in the first quarter of this year, revenue from gen-AI and foundation models accounts for 6.9 of our AI Cloud revenue, and then our traditional CPU cloud businesses were capitalizing on the opportunities presented by gen-AI and foundation models. Both factors were important growth drivers of our AI Cloud. On a separate note, while the smart transportation business remained subdued, its impact on overall AI Cloud business in Q1 was substantially smaller than the previous quarters, so overall we expect our AI Cloud to continue benefiting from the mega trend to maintain strong revenue growth momentum in the upcoming quarters.

On the profit side, Baidu AI Cloud continued to generate non-GAAP operating profit as we have seen in previous quarters. We are committed to sustainable and healthy revenue growth. During the quarter, we maintained our focus on achieving high quality revenue growth by scaling down low margin business. Regarding the business of gen-AI and foundation models, the market is still at a very early stage, so our focus remains on educating the market and broadening our footprint across more enterprises. Looking to the long run, we expect the normalized margin for gen-AI and foundational models related business to improve further and to be higher than our traditional cloud business. Regarding the change of pricing policies of some competitors that you just mentioned, it is actually pretty common for cloud vendors to adjust their pricing for certain products.

This is a trend we have observed multiple times in the past. Given that our cloud offering has expanded beyond traditional CPU cloud to a high value AI product and services, the industry change on CPU cloud pricing has a minimum impact of the development of our AI Cloud. Actually for cloud platform services, by leveraging our unique proprietary four-layer AI architecture and our strong ability in end-to-end automation, we have lowered ERNIE’s inference cost to only 1% over the variant in March last year. This May, ERNIE handles about 200 million API calls, or approximately 250 billion tokens daily. We are confident that ERNIE’s expanding adoption will continue to enhance its performance, boosting efficiency and then further reduce costs.

It’s also important to evaluate price and performance ratio for different models at different workloads, rather than just focusing solely on some superficial prices. We believe our state-of-the-art AI infrastructure and advanced mass platform delivers the best value to our customers. Looking forward, leveraging our strong AI capabilities, we aim to continuously attract new customers and encourage existing ones to increase their spending on Baidu AI Cloud. At the same time, we aim to consistently generate positive non-GAAP operating profit. Thank you Gary.

Operator: Your next question comes from Alex Yao with JP Morgan.

Alex Yao: Hi, good evening management, and thank you for taking the question. Given the chip shortage in China, how does Baidu maintain its commitment to deliver differentiated value while enhancing its leading advantage in [indiscernible] technology in China? Thank you.

Robin Li: Hi Alex. We do think very differently in this aspect. We are taking an application driven approach for our AI push. For example, solving all the high school math problems using the most advanced LLM may not be the highest priority at this time, while generating convincing reasons for users to buy the right products is more important in a lot of applications. With that in mind, we are taking advantage of our unique four-layer AI tech stack to optimize the cost and performance of ERNIE models, and we make sure customers and developers can easily build applications using tools like AgentBuilder, AppBuilder and ModelBuilder. For the AI infrastructure layer, a key factor contributing to our high efficiency in model training and inference is our superior capability in GPU cluster management.

We have recently made a breakthrough by integrating GPUs from different vendors into one large scale, unified computing cluster, allowing us to use less advanced chips for highly effective model training and inference. Our deep learning framework, PaddlePaddle, has through continuous innovation and enhancement helped reduce the cost of model training and inferencing on a constant basis. PaddlePaddle is compatible with over 50 different chips, many domestically designed, and the developer community has grown to 13 million. With ERNIE 3.5 and ERNIE 4.0 remain to be the flagship models for sophisticated tasks, we’re making ERNIE more accessible and affordable by launching lightweight LLMs, introducing a toolkit for model development and app development, applying an MOE approach for model inference for better performance and lower cost.

With our application driven mindset, we have used ERNIE extensively to renovate our own products and we have gained experience and insights in training and using ERNIE, as well as developing AI-native applications on it. We’re making all these capabilities available to our customers and developers. With all these efforts, we are fostering a vibrant and healthy ecosystem around ERNIE. We can see that we are actually taking a holistic approach to developing gen-AI and LLM which is very different from some of our competitors. Our reserve and access to the chips on the market should be sufficient for us to support millions of AI applications in the future. In the long run, I think China will form an ecosystem of its own, probably with less powerful chips but most efficient home grown software stack.

There is ample room for innovation in the application layer, model layer, and framework layer. With ourselves developing a four-layer AI infrastructure as well as our strong R&D team, our dedication to AI and our application-driven approach in building an ecosystem around ERNIE, I’m quite confident that Baidu will stand as a leader in China’s AI ecosystem in the long term. Thank you.

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

Lincoln Kong: Thank you management for taking my question. My question is about the advertising business. What’s the major drag for the advertising growth if we compare it with our peers? What are the trends we have seen for advertising budgets and advertiser sentiment, especially into second quarter, April and May? How should we think about normalized advertising growth for 2024? Thank you.

Robin Li: Yes, you know, our online marketing revenue grew by 3% year-over-year in the first quarter. While traditional search is maturing, we are working hard to renovate the user experience with gen-AI. Right now, about 11% of our search result pages are filled with generated results. These results provide more accurate, better organized and direct answers to user’s questions, and in some cases enable users to do things they could not do before. We have not started the monetization of those gen-AI results, so it will take some time for revenue to catch up. The weak macro also contributed to the softness of our ad business. Our advertisers come from a wide range of industries with the majority being SMEs. This makes our advertising revenue highly sensitive to the macro environment, particularly to the offline economy.

In the first quarter, advertiser sentiment in some verticals, such as real estate and franchising, remained weak. Specifically in the real estate industry, not only was ad spending from developers and agencies muted, but the impact also extended to both upstream and downstream sectors. For example, energy, chemicals, machinery, building materials for upstream and home renovation, furniture, which is kind of downstream, all saw constrained ad spending on our platform. Also, many SMEs in offline sectors need more time to recover as they have been hit hard in the past few years. As we enter the second quarter, we have not seen improvement in advertiser sentiment. Given the limited visibility for sentiment improvement and paired with tough comps in Q2, our online marketing revenue should remain fundamentally solid but from a growth perspective soft over the next few quarters.

Beyond the near term challenges, we expect online marketing to remain a bread-and-butter business for Baidu for the foreseeable future. Search is one of the most popular apps in the internet age, and Baidu remains the largest search engine in China with close to 700 million MAUs. Search will likely be one of the killer apps in the age of gen-AI. Technology innovation will enable us to better engage users with developers and merchants, directly connect users’ intentions with the most relevant product and service offering in more natural ways. Thank you.

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

Thomas Chong: Hi, good evening. Thanks management for taking my question. My question is about how much more room do we see in cost cutting? Previously, you mentioned there will be a lag between the investment and AI revenue contribution. How should we look at margin trends in 2024 if you intend to expand your AI offering? Thank you.

Rong Luo: Hi Thomas, thanks so much for your question. I think our macro challenges are still weighing on our [indiscernible] business, but we are confident that there still will be some ways for us to continue optimizing the operational efficiency. We will stringently manage our cost [indiscernible] for each business and we will take some further steps as necessary, including we try to streamline the [indiscernible] structures to enhance the agility and support strategy flexibility, and we also will reallocate resources to prioritize the key structure areas. If I take look into our businesses, for mobile ecosystem, I think we can still manage the costs and expenses, so mobile ecosystem grew and continues to remain strongly profitable and cash flow positive.

For AI Cloud, as Dou just mentioned, we will continue to phase out low margin businesses and products so we can continue generating the operating profits and margins on a non-GAAP basis. If we’re looking to longer term, the normalized margin for the gen-AI related cloud businesses should be higher than the legacy cloud businesses. For other businesses, we aim to reduce our loss, as we talked about in the past few quarters, particularly our intelligent driving business with more operational efficiency gains and UI improvements for robotaxi. Many investors ask me how our investments in ERNIE will impact net margins. In fact, our reinvestments are mainly related to capex for model training and inference. In 2023, we made a large purchase which has arrived at different times, and the different depreciation [indiscernible].

While the annualized depreciation expenses will be available in the whole year 2024, the impact on our overall cost expenses and quarterly earnings is quite predictable and manageable. The depreciation expenses book under R&D expenses for computing power use and training, and the cost of revenue for computing power reduced for model inference and business expenses. In the first quarter, we can see that Baidu Core’s non-GAAP R&D and the cost of revenue both increased very slightly, while non-GAAP operating margin actually we have expanded to 23.5% due to [indiscernible] spending on other items such as SG&A. We believe that the chips we have in hand are sufficient to support [indiscernible] of earnings for the next one or two years, and because of an ability of high performance chips in China in this year, 2024, we expect our capex to be smaller versus last year.

In conclusion, the investments in gen-AI and large language models will have a manageable impact on the near term margins, and with our monetization of earnings already taking off, we expect that more and more revenue and profit will be generated from that kind of business, for both mobile ecosystem and AI Cloud. Overall, we believe that high quality growth and investment should be very balanced. For many years now, we have delivered a solid track record of top line growth with very resolute cost discipline, and we intend to continue building our future on this kind of business. Thank you so much.

Operator: Your next question comes from Billie Wan [ph] with UBS.

Billie Wan: Good evening. Thanks management for taking my question. My question is about shareholder return. How should we think about the execution pace of the current RMB 5 billion buyback program in addition to your current buyback program? Should we expect other diversified approach to improve shareholder return? Thank you.

Robin Li: Yes, thanks so much for your questions. I think we highly value our shareholders and we have been making efforts to increase the shareholder returns. We have consistently repurchased our shares on the market over the past four years, averaging around $1 billion annually. In total, we have allocated around 37% of our free cash flow towards the share buyback progress. I think the RMB [indiscernible] and going forward, we will continue to buy back more shares from the market as we believe in our long term growth opportunities, and we are very committed to the shareholder returns. In addition, we have utilized our current share repurchase program to prevent any significant increase in the total outstanding share count, aiming to reduce the potential dilution of our shareholders’ economic stake in Baidu.

In the year 2023, you can see that our total number of shares outstanding was flat year-over-year compared to a 1.2% increase in the year 2022 and a 3.2% increase in the year 2021. In this quarter, the total shares outstanding began to decrease – we can see that it has been declining by 0.5% as compared to the prior quarters. We are adopting a strategy of sustainable and recurring share buyback programs for the open market. At the same time, we are taking into consideration the opportunities ahead of us. Now we are facing a huge opportunity in gen-AI and foundation models, and we have structured a [indiscernible] plan to capitalize it, so we want to have the flexibility to invest as we consider necessary and in the best interests of long term value to the shareholders.

Furthermore, we believe that the most effective way to create value for shareholders is by building strong base fundamentals. Our core marketing business remains steady and we believe that AI will help us to build another growth engine over time. Thank you so much for your question.

Operator: Your next question comes from Miranda Zhuang with BofA Securities.

Miranda Zhuang: Thank you, good evening management. Thanks for taking my question. My question is about robotaxi. Can management share more updates on the robotaxi initiative and the geographic coverage for this year? I think previously you’d mentioned that Apollo Go will achieve operating UE breakeven in Wuhan in the near future. I want to understand what’s the logic behind the efforts to continue to improve the UE. What’s the projected size of the vehicle fleet for this year, and how will it potentially impact the cost? Thanks.

Robin Li: Sure Miranda, let me give you some more color on the robotaxi business. In 2023, Apollo Go made significant progress in improving the regional unit economics in key cities. Let me use Wuhan, Apollo Go’s largest operation to explain how we achieved that. We launched Apollo Go’s commercial operations in Wuhan back in 2022. Since then, we’ve witnessed a consistent enhancement in operational UE which can be attributed to the expanding scale of driverless operations and decreasing cost per vehicle. Regarding scale expansion, our vehicle fleet has been steadily growing. Compared to one year ago, our fully driverless fleet in Wuhan has grown threefold, reaching about 300 vehicles today. Meanwhile, both the operational area and service hours for our fully driverless operations have been consistently expanding thanks to the increasing recognition of our autonomous driving technology by the local government.

The operational coverage area for our fully driverless ride hailing service increased by eightfold from a year ago, now covering over 7 million people in Wuhan. Apollo Go’s operational hours also gradually expanded from only the off-peak hours in the beginning to adding peak hours to its operations and eventually extending to 24/7 in March of this year. The scale expansion resulted in a consistent improvement of UE in terms of revenue. Both daily rides per vehicle and distance per ride have been growing. When it comes to cost, the majority is labor cost and hardware expenditures. We have been demonstrating consistent track record of safe operations, which helps us to increase deployment of fully driverless ride hailing operations. In April, the proportion of fully driverless orders rose to 70% – that’s up from only 10% in August 2022, and 45% Q4 of last year.

We expect this figure to reach 100% in the coming quarters, thereby enabling us to minimize the cost related to safety officers. In addition to lowering the labor cost, we are steadfast in driving down hardware costs. The mass production and timeline of RT6, our sixth generation robotaxi, remains on track. Adopting a battery swapping solution, the mass production price for RT6 excluding battery is below US $30,000. We will use RT6 as the primary vehicle in the future fleet expansion and it should help to significantly reduce the hardware depreciation cost for each vehicle, further improving our UE and bringing us closer to profitability. Looking to this quarter, we plan to expand the fully driverless fleet of our Wuhan operation to 1,000 vehicles by the end of the year, more than tripling from the end of last year.

Our focus remains on improving regional UE and narrowing the losses for Apollo Go business. With continued improvement in operational efficiency and cost reduction, we believe Apollo Go will achieve operational UE breakeven in Wuhan first, and once that’s achieved, we can scale up the operation quickly. Thank you.

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