Baidu: Advertising under short-term pressure, relying on AI transformation for search (1Q24 conference call minutes)

The following is the summary of the 1Q24 performance conference call for $ Baidu(BIDU.US) . For financial report analysis, please refer to "Can Baidu's Search Rely on AI to Defend the City" article.

I. Financial Performance Overview

II. Management Report

In the first quarter, our business continued to grow. Baidu's core total revenue increased by 4% year-on-year to RMB 23.8 billion, and the Non-GAAP operating profit margin improved to 23.5% compared to the same period last year. In particular, revenue from Baidu AI Cloud accelerated to 12% year-on-year this quarter, while continuing to achieve positive Non-GAAP operating profit.

2024 is the second year of our journey on the Gen AI road. As we consolidate our leadership position in the field of large models, we are transforming the company from an internet-centric business to an AI-first business.

Given that ERNIE is China's most powerful LLM, we are actively promoting the adoption of ERNIE in our 2C and 2B businesses to provide a better user experience, increase return on investment for advertisers, and enable developers to write agents and applications for more effective and efficient patterns. While we operate traditional businesses in a challenging environment and experience lower revenue growth in the short term, we still believe that in the long run, AI will bring continuous revenue and profit growth to us.

1. ERNIE Recent Developments

（1）To B

In April, ERNIE processed approximately 200 million API calls per day, a significant increase from around 50 million calls in December last year. This remarkable growth indicates the increasing adoption of ERNIE and suggests the huge revenue potential for model inference in the future. To accelerate the adoption of ERNIE, we are building a vibrant and healthy ecosystem around it.

We believe that over time, the ERNIE ecosystem will include millions of applications, especially agents developed by a diverse community of cross-industry enterprises and individual developers, meeting a wide range of needs in people's daily lives and work. Our large user base in the mobile and desktop fields will allow us to distribute these agents and applications to anyone in need at the right timeThe backbone of this ecosystem is the ERNIE series of large models, including our standard versions ERNIE 3.5 and ERNIE 4.0, as well as the lightweight version we launched in the first quarter. Throughout the quarter, we have been continuously improving the efficiency of ERNIE using our unique proprietary four-layer AI architecture and powerful end-to-end optimization capabilities. For example, compared to the version in March 2023, ERNIE's training efficiency has increased by 5.1 times, with an inference cost of only about 1%. In order to make ERNIE more user-friendly and affordable, we now offer 3 sets of tools on the MaaS platform for customers to choose from.

Last quarter, we introduced AppBuilder and ModelBuilder for enterprises and individual developers to develop applications and build models. In April, we further launched AgentBuilder, a platform that includes tools for easily creating AI assistants. This is because we anticipate that AI assistants will become one of the most important application forms supported by Gen AI and foundational large models.

With the ability to use natural language as a programming language, developers will be able to build AI agents without writing any code. Currently, new ERNIE agents are created on our platform every day, with daily distributions reaching millions of times, serving a wide range of verticals such as education, law, B2B, and tourism.

All of these initiatives stem from our rich experience and insights in building and operating ERNIE and developing AI-native applications. We believe that the true value of ERNIE can only be realized when the many applications built on it are widely used by users and customers.

I am pleased to note that ERNIE is expanding its influence to smart devices through APIs. Last quarter, we announced partnerships with well-known smartphone brands such as Samsung China and Honor to help them enhance their native app experiences using ERNIE. This quarter, we are excited to expand our cooperation with more leading smartphone manufacturers such as Oppo, Vivo, and Xiaomi, who are using the ERNIE API to enhance user experiences.

Furthermore, our business scope has now expanded from smartphones to personal computers and electric vehicles. The ERNIE API is now being used by top PC brand Lenovo to support its AI assistant in the default browser of its PCs. Leading Chinese smart electric vehicle manufacturer Nio has started using the ERNIE API to enhance the in-car experience of its vehicles. We are expanding partnerships to various smart devices, providing ample opportunities for widespread user adoption and paving the way for ERNIE-supported applications to become the entry point into the generative AI world（2）To C

We have been rebuilding all consumer-facing products, aiming to build proprietary artificial intelligence applications that could be killer applications in the ERNIE ecosystem. By doing so, we should be able to create new growth opportunities. For example, after the reconstruction with Gen AI and LLM, our document creation one-stop shop has led to double-digit year-on-year growth in paid users for Baidu Wenku in the first quarter. ERNIE's penetration time in Baidu search and feed is longer than expected, as the user base has reached hundreds of millions, and use cases are usually very sensitive to cost and response time.

We need ERNIE models of various sizes optimized for different scenarios to achieve the best cost-effectiveness. After several quarters of repeated experiments, we are firming up our strategy.

Looking ahead, we plan to accelerate the launch and adoption of new product features, such as multi-mode generated search results, multi-round interactions in search, and recent ERNIE agent distribution. We are at the forefront of this unique technological revolution, and we are confident in our innovation capabilities. By definition, we are conducting business in unknown territories. As always, we hope to be able to adjust flexibly according to the ever-changing consumer demands and how users integrate new product features into their daily lives in a timely manner.

2. AI Cloud Business

Now let me review the key highlights of each business in the first quarter. In the first quarter, revenue from artificial intelligence cloud reached RMB 4.7 billion, a year-on-year increase of 12%, and continued to generate non-GAAP operating profit. Revenue growth was mainly driven by Gen AI and basic models.

Revenue from Q1 large models and Gen AI accounted for 6.9% of total smart cloud revenue. This part of the revenue mainly comes from model training, but model inference revenue is growing rapidly. We believe that with the increasing adoption rate by customers, revenue from Gen AI and large models will continue to grow. For example, within our internal cloud revenue Baidu Core, other business groups, such as Mobile Ecosystem Group and Intelligent Driving Group, are increasingly leveraging the power of ERNIE.

Therefore, the 15% payment made by the Mobile Ecosystem and Intelligent Driving businesses to the AI Cloud section is used for the use of Gen AI and large models. Enterprises use Baidu's artificial intelligence cloud to train and host their models because they believe we have the most powerful and efficient model training and inference artificial intelligence infrastructure in China. Compared to peers, we help enterprises train models of various scales on the artificial intelligence cloud while also reducing model inference costs. This is mainly due to two reasons.

First, our independently developed four-layer AI architecture allows us to innovate and optimize at each layer, resulting in continuous efficiency improvements. Second, we have outstanding capabilities and insights in GPU cluster management. Leveraging our technical expertise, we can now integrate GPUs from different vendors into a unified computing cluster for model trainingOur platform has demonstrated very high efficiency on GPU clusters consisting of hundreds or even thousands of GPUs. This is a significant breakthrough due to the limited availability of imported GPUs.

Another growth driver for the AI cloud is cross-selling our CPU cloud services to GPU cloud customers. With the increasing recognition of our GPU cloud among existing and new customers, we see more customers shifting their CPU cloud usage to Baidu. Earlier, in terms of MaaS, we took many measures to make the ERNIE series models more cost-effective and efficient than open-source models.

In the first quarter, we expanded and enhanced the ERNIE model portfolio, offering a total of 3 lightweight LLMs and 2 test-specific LLM ModelBuilders. These models help enterprises and professional developers balance model performance and cost in ERNIE to attract a wider audience of model developers.

Additionally, our expert or MOE hybrid approach can direct user queries to different tasks, allocate the most suitable model to handle each task, and only use ERNIE 3.5 or 4.0 for the most complex tasks. This approach can achieve faster response times and lower inference costs while maintaining performance levels similar to using more models.

Last quarter, we introduced AppBuilder and continued to improve it, allowing developers to easily create AI-native applications on our platform in just 3 steps. With the launch of AgentBuilder in April, anyone can create an AI assistant on Baidu in a few sentences. Overall, we remain confident in the strong revenue of our artificial intelligence cloud, and our goal is to continue generating operating profits based on non-GAAP principles.

3. Advertising Business

The mobile ecosystem continues to provide healthy profit margins and strong free cash flow. In the first quarter, our online marketing revenue increased by 3% year-on-year. Revenue growth was affected by the challenging macro environment. Meanwhile, we have been working to transition user experience from traditional mobile products to generative artificial intelligence experiences. This transformation is ongoing, and monetization has not truly begun.

We all need to use ERNIE to rebuild our monetization system for better conversion and efficiency gains. This quarter, we further enhanced ad targeting capabilities and expanded the scale of real-time ad generation. These efforts have improved conversion rates and increased revenue. ERNIE agents also represent long-term opportunities for monetization upgrades.

Recently, not only have we seen brand advertisers, but even small and medium-sized enterprises are gradually adopting ERNIE agents. We designed this agent for small and medium-sized enterprises as a virtual storefront and service counter, serving consumers around the clock. We believe that using agents can increase and alleviate sales rates, improve their productivity, and expand their influence among users. This will be an important step for us to transform the traditional CPC model into the significantly more efficient CPS model while enhancing the Baidu user experienceOverall, I believe that search will become the most likely killer application in the new generation of artificial intelligence era, and we are on the right track to harness this potential. I mentioned that the ERNIE agent is an important monetization opportunity. With the newly launched AgentBuilder, creators, publishers, and service providers will find it increasingly easy to build on Baidu. It is key to enhancing Baidu's content products and ultimately providing an AI-native user experience on our platform.

4. Intelligent Driving

Measured by the ride-hailing services provided to the public, Apollo Go is the world's largest autonomous ride-hailing service provider. In the first quarter, Apollo Go provided approximately 826,000 ride-hailing services to the public, a year-on-year increase of 25%. In April, the total number of trips exceeded 6 million. We are continuing to move forward - striving to achieve the unit economic breakeven point of Apollo Go. To achieve this goal, our strategy is to achieve user breakeven in key regions and then replicate success in other regions.

To reach the regional breakeven point, our focus is on expanding fully autonomous ride-hailing services and increasing utilization in various areas. In Wuhan, Apollo Go's largest regional operation is moving towards this goal. In Wuhan, Apollo Go is gradually becoming an indispensable part of the city's transportation network.

The operating area of Apollo Go has more than doubled from a quarter ago, serving a population of over 7 million, and has achieved the extraordinary milestone of using fully autonomous vehicles to cross the Yangtze River during its expansion. In addition, our vehicles started operating in Wuhan in early March, further expanding the coverage of Apollo Go and increasing vehicle utilization. All of these advancements are driving the rapid growth of autonomous driving. In the first quarter, the proportion of trips provided by autonomous vehicles accounted for over 55% of the total trips in Wuhan, higher than the 45% in the fourth quarter of last year. This number continues to rise, exceeding 70% in April, and is expected to continue to grow rapidly in the future, reaching 100% in the coming quarters.

In terms of automotive solutions, our Apollo autonomous driving ASD technology is continuously evolving. As I mentioned in the last earnings conference call, Apollo is a global pioneer in using vision-based models in autonomous driving. Now, we provide OEMs with the most advanced vision-based autonomous driving solution.

ASD can effectively handle the complex urban environments of over 100 cities in China and is planned to expand to hundreds of cities in the coming months. This enables us to achieve advanced autonomous driving on a variety of passenger vehicles. From high-end to economy models, priced as low as 150,000 RMB, this is another proof of our technological leadership.

5. Business Outlook

(1) We expect our cloud business to accelerate in the remaining time of this year, while our Robotaxi business will gradually reduce losses.(2) We expect the mobile advertising business to be weak in the short term, and it will start to recover when Gen AI becomes the new core of our existing products in 2025.

(3) Looking ahead, Gen AI and foundational models will bring us huge opportunities, ushering in a new cycle of innovation. Both enterprises and individual developers have rapidly shifted from worrying about missing this opportunity to utilizing foundational models like ERNIE to build artificial intelligence applications. Baidu is well prepared to benefit significantly from this technological transformation. We believe that one of the most important long-term opportunities is model inference, which will be a key driver of growth for our future artificial intelligence cloud revenue.

(4) Looking ahead, we plan to deploy our sixth-generation robotaxi RT6 in Wuhan's Apollo Go operation this year, which will significantly reduce hardware depreciation costs. With the expansion of autonomous driving operations and continuous improvement in cost structure, we believe that Apollo Go will achieve operational breakeven in Wuhan UE in the near future. As Apollo Go progresses, we will closely monitor efficiency and continuously optimize the overall intelligent driving business operations.

III. Analyst Q&A

Q1: Can you quantify whether AI technology has helped Baidu improve its advertising monetization rate? Could management share some feedback from advertisers using this system? In which areas have they seen the most progress? What aspects can be further strengthened?

A: As we all know, our monetization system is the first to benefit from Gen AI, generating billions of incremental revenue each quarter. Since the second half of last year, we have been using ERNIE to upgrade our monetization system, enhancing various aspects of advertising technology. This includes improving conversion capabilities for our advertisers, refining the bidding system, automating creative generation, and forming advertising strategies.

Advertisers have seen better conversions and more sales leads. This improvement has led advertisers to increase their spending on Baidu. Therefore, AI-related incremental advertising revenue in the first quarter increased month-on-month, and we expect this trend to continue. Incremental revenue has helped us alleviate broader macro weakness and has given us some time to rebuild our user products with ERNIE.

As I emphasized in the opening remarks, we believe ERNIE provides important long-term capabilities for our online marketing business. Advertisers can build online virtual images and interact with potential customers through natural language in multi-round conversations. With the launch 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 also help enrich our content and enhance Baidu's user experience.

Although still in the early stages, ERNIE Assistant has already helped some advertisers achieve better return on investment. For example, we have an online education client. They used AgentBuilder to create an AI agent, injecting key insights such as product introductions and subject matter expertise, and continuously improving based on feedbackThis AI agent significantly enhances the company's online customer service by providing high-quality consultations around the clock. After adopting the ERNIE agent, the advertising conversion rate of this online education company increased by 20%. We believe this is just the beginning. We are confident that agents will become the primary form of content and services in the new AI era. We will continue to enhance the capabilities of the ERNIE agent.

The agent will not only improve the user experience, conversion rate, and return on investment for advertisers, but over time, it will also increase the volume of transactions directly generated on our platform. This shift should help us transition from the traditional CPC model to a more efficient CPS model.

Q2: I have a question about the AI Cloud business. How do price reductions by some peer companies affect Baidu's AI Cloud business? How should we view the profitability of cloud revenue as competition intensifies? What is the level of future sustained cloud growth?

A: As we have mentioned, Gen AI and foundational models are transforming the cloud industry from general computing to artificial intelligence computing. Therefore, this transformation is reshaping the competitive landscape in the cloud industry, providing us with a valuable opportunity to establish ourselves as leaders in the artificial intelligence cloud. We believe we offer the most efficient artificial intelligence infrastructure and advanced MaaS platform for model training and inference in China. As a result, more and more enterprises are choosing us for model training, fine-tuning, and native artificial intelligence application development on the public cloud. The growing demand has significantly increased our artificial intelligence cloud revenue.

In fact, starting from the second half of last year, the growth of our AI cloud revenue began to accelerate, from a year-on-year decline in the third quarter to an 11% growth in the fourth quarter of last year, and further accelerated to 12% in the first quarter of this year. Therefore, revenue acceleration is supported by two main factors:

(1) Incremental revenue directly generated by Gen AI and foundational models, as well as the new opportunities they bring to our traditional cloud business. As Robin just mentioned, in the first quarter of this year, revenue from Gen AI and foundational models accounted for 6.9% of our AI Cloud revenue. Our traditional CPU cloud business is leveraging the opportunities brought by Gen AI and foundational models, both of which are significant growth drivers for our AI cloud.

(2) On the other hand, although the intelligent transportation business remains sluggish, its impact on the overall cloud business in the first quarter is much smaller than in previous quarters. Therefore, overall, we expect our intelligent cloud to continue benefiting from the AI trend and maintain strong revenue growth momentum in the coming quarters.

In terms of profit, Baidu's artificial intelligence cloud continues to generate Non-GAAP operating profit. As we have seen in the past few quarters, we are committed to achieving sustainable, healthy revenue growth. This quarter, we continue to focus on achieving high-quality growth by trimming low-profit businessesFor Gen AI and large model businesses, the market is still in a very early stage. Therefore, our focus remains on educating the market and expanding our penetration into more enterprises.

Looking ahead, we expect the standardized profit margin of Gen AI and basic model-related businesses to further improve and be higher than our traditional cloud business.

Regarding the changes in pricing policies of some competitors you mentioned earlier, yes, it is actually quite common for our cloud providers to adjust pricing for certain products. This is a trend we have observed multiple times in the past. Given that our cloud products have expanded from traditional CPU cloud to high-value AI products and services, industry changes in CPU cloud pricing have minimal impact on the development of our AI cloud.

In fact, for our cloud platform services, by leveraging our unique proprietary four-layer AI architecture and powerful end-to-end optimization capabilities, we have reduced the inference cost of ERNIE to 1% of what it was in March last year. In May this year, ERNIE processed approximately 200 million API calls per day or around 250 billion tokens. We believe that the widespread adoption of ERNIE will continue to enhance its performance, improve efficiency, and further reduce costs. It is equally important to evaluate the current performance of different models under different workloads rather than just focusing on some annual surface prices. Therefore, we believe that our state-of-the-art AI infrastructure and advanced MaaS platform can bring the maximum value to our customers.

Looking ahead, leveraging our strong AI capabilities, our goal is to continuously attract new customers and encourage existing customers to increase their spending on Baidu AI Cloud. At the same time, our goal is to continue generating positive non-GAAP operating profit.

Q3: In the context of chip shortages in China, how does Baidu maintain its differentiation while enhancing the leading edge of large model technology in China?

A: Our approach in this regard is indeed very different. We are adopting an application-driven approach to drive artificial intelligence. For example, using state-of-the-art large models to solve all MaaS problems may not be the optimal choice at the moment. While providing compelling reasons for users to purchase the right products is important, it is more crucial in many applications. Therefore, considering this, we are leveraging our unique four-layer AI technology stack to optimize the cost and performance of the ERNIE model. We ensure that customers and developers can easily build applications using tools like AgentBuilder, AppBuilder, and ModelBuilder.

For the AI infrastructure layer, we have achieved high efficiency in model training and inference through outstanding GPU cluster management capabilities. We recently made a breakthrough by integrating GPUs from different vendors into a large-scale unified computing cluster, allowing us to efficiently train and infer models using less advanced chips.

Our deep learning framework PaddlePaddle continuously reduces the cost of model training and inference through innovation and enhancement. PaddlePaddle is compatible with over 50 different chips, many of which are domestically designed, and the developer community has grown to 13 millionERNIE 3.5 and ERNIE 4.0 are still the flagship versions for handling complex tasks. By introducing lightweight large models, model development and application development toolkits, applying the MOE method for model inference to achieve better performance and lower costs, ERNIE has become easier to access and more affordable. Adhering to the application-driven concept, we extensively use ERNIE to transform our own products, accumulating experience and insights in training and using ERNIE and developing AI native applications on it. We are providing all these capabilities to our customers and developers. Through all these efforts, we are nurturing a vibrant and healthy ecosystem around ERNIE. As you can see, we are actually taking a holistic approach to developing Gen AI and LLM, which is very different from some of our competitors.

Our reserves and access to chips in the market should be sufficient to support millions of AI applications in the future. In the long run, I believe China may form its own ecosystem with functionally weaker chips but the most efficient local software stack. There is great innovation potential at the application layer, model layer, and framework layer. With our independently developed four-layer AI infrastructure, a strong R&D team, our focus on AI, and our application-driven approach to building an ecosystem around ERNIE, I am very confident that Baidu will become the leader in China's AI ecosystem in the long term.

Q4: My question is about the advertising business. So, compared with peers, what are the main obstacles to Baidu's advertising growth? Especially in the second quarter, April and May, how should we view the normal growth of advertising in 2024?

A: Yes. In the first quarter, our online marketing revenue increased by 3% year-on-year. At the same time, traditional search is maturing. We are working hard to innovate user experience through Gen AI. Currently, about 11% of our search result pages are generated by Gen AI. This result will provide users with more accurate, more organized, and more direct answers to their questions, and in some cases, enable users to do things they couldn't do before.

We have not yet started monetizing these Gen AI. Therefore, it will take some time for revenue to catch up. Weak macroeconomic conditions have also led to the weakness in our advertising business.

Our advertisers come from various industries, with most being small and medium-sized enterprises. This makes our advertising revenue highly sensitive to the macro environment, especially the offline economy. In the first quarter, advertisers in vertical industries such as real estate still had weak sentiment, with reduced advertising spending from developers and institutions, affecting upstream and downstream industries.

For example, upstream industries like energy, chemicals, machinery, building materials, as well as home decoration and furniture, all see advertising spending on our platform being restricted downstream. In addition, many offline small and medium-sized enterprises need more time to recover due to efforts in the past few yearsWhen we entered the second quarter, we did not see an improvement in advertiser sentiment. Given the limited visibility of sentiment improvement, coupled with a high base in the second quarter, our online marketing revenue should remain relatively stable. However, from a growth perspective, the next few quarters will be weaker. Despite recent challenges, we expect online marketing to remain an important business for Baidu in the foreseeable future. Of course, it is one of the most popular applications in the Internet era, and Baidu remains China's largest search engine with nearly 700 million monthly active users.

Search may become one of the killer applications in the era of artificial intelligence. Technological innovations will enable us to better engage users with developers and merchants in a more natural way, connecting user intent with the most relevant products and services directly.

Q5: My question is how much room do we have for cost reduction? Previously, we mentioned that the revenue contribution from AI lags behind AI investment. If you plan to expand AI products, how should we view the profit trend in 2024?

A: I believe macro challenges continue to impact the advertising business. However, we believe there are still ways for us to continue optimizing operational efficiency. We will rigorously manage the cost expenditures of each business and take further measures as needed, including trying to streamline the organizational structure to enhance agility and support strategic flexibility. We will also reallocate resources, prioritizing key strategic areas. When we look at our mobile ecosystem business, I believe we can still optimize costs and expenses. Therefore, the mobile ecosystem group continues to maintain strong profitability and positive cash flow.

For AI Cloud, we will continue to phase out low-profit businesses and products. Therefore, we can continue to generate operating profit and profit margins on a non-GAAP basis. Looking ahead in the long term, the standardized profit margin of the new generation of AI-related cloud businesses should be higher than traditional cloud businesses.

For other businesses, our goal is to reduce the losses we have discussed over the past three quarters, especially in our smart driving business, where operational efficiency has been greatly improved, and in the enhancement of the user experience of robotaxis. Many investors ask me how our investment in ERNIE will affect my profit. In fact, our investment is mainly related to capital expenditures for model training and inference.

In 2023, we made a large purchase, which arrived at the best time - at different times and different start dates for depreciation. Although the annual depreciation expense will be calculated for the full year in 2024, the impact on our overall costs, expenses, and quarterly earnings is quite predictable and controllable. Depreciation expenses are included in R&D expenses for computing power training, as well as income costs for model inference and business expansion audits.

In the first quarter, we can see that Baidu's core non-GAAP R&D and income costs have both slightly increased. Due to our strict spending on SG&A and other items, the non-GAAP operating profit margin has actually expanded to 23.5%We believe that the chips we currently have are sufficient to support ERNIE's training for the next 1-2 years. Due to the limited supply of high-performance chips domestically in 2024, we expect our capital expenditures to be lower than last year.

In short, investments in Gen AI and large language models will have a controllable impact on short-term profits. With the monetization of our ERNIE already underway, we expect that businesses such as the mobile ecosystem and AI cloud will generate increasing revenue and profits.

Overall, we believe that high-quality growth and investments should be able to achieve a good balance. Over the years, we have achieved a good performance record in revenue growth through strict cost control, and we intend to continue building our future in such businesses.

Q6: My question is about shareholder returns. So let's consider the execution speed of the current 5 billion RMB share repurchase plan. Apart from the current repurchase plan, should we expect more diversified ways to enhance shareholder returns?

A: Yes. Thank you very much for your question. I believe we value shareholders greatly and have been working hard to enhance shareholder returns. Over the past 4 years, we have been repurchasing stocks in the market, with an average annual repurchase amount of around $1 billion. Overall, we allocate approximately 37% of our free cash flow to repurchases. I think during these periods and in the future, we will continue to repurchase more shares from the market as we believe in our long-term growth opportunities and are very committed to shareholder returns.

In 2023, you can see that our total outstanding shares remained flat year-over-year, while in 2022 it increased by 1.2%, and in 2021 it increased by 3.2%. This quarter, the total outstanding shares have started to decrease, down by 0.5% compared to the previous quarter.

We are adopting a strategy of sustainable and regular stock repurchases on the open market. At the same time, we will also consider the opportunities in front of us. We are currently facing significant opportunities in the Gen AI foundational model, and we have developed a specific plan to leverage it. Therefore, we hope to flexibly invest when we deem it necessary and in the best long-term interests of shareholders.

Furthermore, we believe that the most effective way to create value for shareholders is to build a strong business foundation. Our core marketing business remains stable, and we believe that over time, artificial intelligence will help us establish another growth engine.

Q7: My question is about Robotaxi. Can management share more latest information about this year's Robotaxi plan and geographical coverage? I recall you mentioned that Apollo Go will achieve operational breakeven in Wuhan in the near future. I would like to understand the logic behind the efforts to continuously improve UE. What is the expected fleet size this year? How will it impact costs?

A: In 2023, Apollo Go has made significant progress in improving the UE model in key cities. Let me explain how we have achieved this goal using the changes and actions of Apollo Go in WuhanWe launched the commercial operation of Apollo Go in Wuhan as early as 2022. Since then, we have witnessed the continuous enhancement of the operational unit's economic model, which can be attributed to the expansion of the scale of autonomous driving operations and the decrease in the cost per vehicle.

In terms of scale expansion, our fleet size has been steadily growing. Compared to a year ago, the number of autonomous driving vehicles in our fleet in Wuhan has tripled to about 300. At the same time, the operational area and service hours of fully autonomous operations continue to expand. We appreciate the increasing recognition of our autonomous driving technology by the local government.

The coverage area of autonomous ride-hailing services has expanded 8 times compared to a year ago, now covering over 7 million people in Wuhan. The operating hours of Apollo Go have also expanded from initially operating only during off-peak hours to adding peak-hour operations, ultimately extending from March to July this year.

The scale expansion has led to a continuous increase in UE's revenue, with both daily average ride volume per vehicle and remote ride volume continuing to grow. In terms of costs, the majority are labor costs and hardware expenses. We have consistently demonstrated a strong safety record, which helps us in deploying fully autonomous ride-hailing operations.

In April, the proportion of fully autonomous orders increased to 70%. This is a significant increase from just 10% in August 2022 and 45% in the fourth quarter of last year. We expect this number to reach 100% in the coming quarters, allowing us to maximize the reduction in costs associated with safety personnel.

In addition to reducing labor costs, we are also steadfast in reducing hardware costs. The mass production of our sixth-generation robotaxi RT6 is proceeding as planned, using a battery-swapping scheme. The mass production price of RT6 (excluding the battery) is below $30,000. We will use RT6 as the main vehicle for future fleet expansion, which will significantly reduce the hardware depreciation costs per vehicle and further improve our UE, bringing us closer to profitability.

Looking ahead to this quarter, we plan to expand the fully autonomous fleet in Wuhan to 1,000 vehicles by the end of the year, more than doubling from the end of last year. Our focus remains on improving regional UE and narrowing the losses of the Apollo Go business. With continued improvement in operational efficiency and cost reduction, we believe Apollo Go will first achieve operational breakeven in UE in Wuhan. Once this is achieved, we can rapidly expand our operations.

Risk Disclosure and Statement for this article: Dolphin Research Disclaimer and General Disclosure