BIDU earnings call for the period ending September 30, 2024.
Image source: The Motley Fool.
Baidu (BIDU -5.90%)
Q3 2024 Earnings Call
Nov 21, 2024, 7:30 a.m. ET
Contents:
- Prepared Remarks
- Questions and Answers
- Call Participants
Prepared Remarks:
Operator
Hello and thank you for standing by for Baidu’s third 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 Lin, Baidu’s director of investor relations.
Juan Lin — Director, Investor Relations
Hello, everyone, and welcome to Baidu’s third quarter 2024 earnings conference call. Baidu’s earnings release was distributed earlier today, and you can find a copy on our website, as well as on newswire services. On the call today, we have Robin Li, our co-founder and CEO; Rong Luo; our EVP in charge of Baidu Mobile Ecosystem Group, MEG; Dou Shen, our EVP in charge of Baidu AI Cloud Group, ACG; and Junjie He, our interim CFO. After our prepared remarks, we will hold a Q&A session.
Please note that the discussion today will contain forward-looking statements made under the safe harbor provisions of the U.S. Private Securities Litigation Reform Act of 1995. Forward-looking statements are subject to risks and uncertainties that may cause actual results to differ materially from our current expectations. For detailed discussions of these risks and uncertainties, please refer to our latest annual report and other documents filed with the SEC and Hong Kong Stock Exchange.
Baidu does not undertake any obligation to update any forward-looking statements except as required under applicable law. Our earnings press release and this call include discussions of certain unaudited non-GAAP financial measures. Our press release contains a reconciliation of the unaudited non-GAAP measures to the unaudited most directly comparable GAAP measures and is available on our IR website at ir.baidu.com. As a reminder, this conference is being recorded.
In addition, a webcast of this conference call will be available on Baidu’s IR website. I will now turn the call over to our CEO, Robin.
Robin Li — Co-Founder and Chief Executive Officer
Hello, everyone. Baidu Core total revenue was RMB 26.5 billion. That’s roughly flat for the third quarter on a year-over-year basis. Revenue growth from AI Cloud was 11%, continuing its double-digit growth trajectory, thanks to the sustained momentum in gen AI-related revenue.
Our non-GAAP operating profit and non-GAAP operating margin remained stable, which demonstrates the resilience of our business. While navigating the ongoing macroeconomic weakness, we remain patient on our strategic focus of AI-driven innovation, with a particular emphasis on transforming our existing products and businesses, as well as fostering a new ecosystem for ERNIE. Despite near-term headwinds, we are deeply convinced that this AI-focused strategy will position us well to capture significant growth opportunities in the long run. Underpinning our long-term strategic focus is the advancement of our AI capabilities, which serves as the cornerstone of our AI-driven transformation.
In the third quarter, we continued to improve ERNIE through pre- and post-training, optimizing model efficiency while tailoring to the needs of our diverse application scenarios. Our flagship model, ERNIE 4.0 Turbo, has achieved notable improvements in inference efficiency, with throughput increasing by 48% compared to its debut in June. The efficiency gains were driven by the optimization of our self-developed four-layer AI infrastructure, and we expect such improvements to further reduce model inference costs going forward. We have also expanded our lightweight model offerings with the introduction of Speed Pro and Lite Pro in the third quarter.
As enhanced versions of their predecessors, Speed Pro and Lite Pro feature lower latency, higher throughput, improved stability, and superior accuracy. Over the past 24 months, we have focused on resolving LLM hallucinations through RAG, retrieval-augmented generation. However, we observed that image generation still faces widespread hallucination issues, leading to inaccurate and logically inconsistent outputs, particularly in e-commerce and marketing scenarios where sellers need to generate images that precisely match their text descriptions. Aiming at mitigating hallucinations in text-to-image generation, we launched ERNIE iRAG at Baidu World 2024 to revolutionize image generation, delivering more precise and demand-driven results with system — with substantially enhanced accuracy.
These additions have made our offerings more comprehensive, extending the reach and effectiveness of ERNIE across diverse scenarios. Thanks to ERNIE’s enhanced capabilities, wide-ranging applications, and improved cost efficiencies, we’ve seen a substantial increase in API calls. In November, ERNIE handled approximately 1.5 billion API calls daily, a significant jump from 600 million in August. Meanwhile, over 1.7 trillion tokens were generated on a daily basis.
The surging API call volume and token generation demonstrate strong market adoption of ERNIE, validating its value in real-world applications. Powered by the enhanced ERNIE models, we are achieving comprehensive penetration of AI capabilities across our entire product line, from consumer-facing products to enterprise solutions. Our continuous efforts to transform search have yielded promising progress that is encouraging user-centric outcomes emerging this quarter. Currently, over 20% of all search result pages contain AI-generated content, increasing from 18% in August.
On Baidu App, where the majority of our search queries were conducted, an increasing number of users are engaging with generative content, reaching nearly 70% of its monthly active users. Users exposed to AI-generated search results spend longer time on Baidu, used more complicated queries, and were more likely to return the next day. These positive changes become increasingly pronounced from quarter to quarter, indicating that our gen AI-enabled search is better serving users’ needs and enhancing user satisfaction. We have also observed changes in user behavior patterns, suggesting Baidu is evolving beyond the traditional search in users’ minds, serving as a more comprehensive destination for their needs.
Users are staying longer to interact with our AI features after initial queries. They are proactively engaging in more complex multiround conversations with growing interaction frequency, reaching tens of millions of daily interactions. Users are also moving beyond pure search to content creation on Baidu. For example, during this year’s Mid-Autumn Festival, we saw a surge in both image creation and text generation, indicating changes in how users interact with our platform and their growing exploration of ERNIE-powered functions.
These changes inspired us with new ideas for future enhancements. Going forward, we will enrich our content diversity, explore more comprehensive and unified ways of content presentation, and promote chat-based interaction. We envision agents as the most prominent form of AI-native application and a transformative force in driving meaningful innovation. This quarter, we continued to enhance ERNIE agents’ capabilities and expand their applications across various scenarios.
We have witnessed significant growth in the conversation rounds with ERNIE agents, reaching an average volume of 15 million daily. The recently released top 100 agent list at Baidu World covers a wide array of scenarios and industries, showcasing the breadth and variety of our agents. The ERNIE agent we created for BYD demonstrates how agents can enhance brand engagement and create significant value. Rather than navigating through traditional websites, users can now discover BYD through dynamic conversations with its agent on Baidu.
This innovative approach has naturally inspired users to explore the brand more proactively, leading to higher engagement and stronger purchase intent. The effectiveness of the agent is further validated by the fact that around 70% of users who engaged with the BYD agent returned to Baidu within seven days for BYD-related searches, highlighting how the agent’s compelling content sparked genuine interest and motivated further brand exploration. Beyond BYD, we also partnered with leading enterprises such as Samsung and Yanghe to build their own agents on Baidu. While these examples demonstrate the versatility of ERNIE agents, we believe their transformative potential is just beginning to unfold.
We envision agents redefining information in the AI era, much like how websites shifted the internet age or apps and social media accounts transformed the mobile internet era. We believe agents are emerging as the fundamental carrier of content, information, and services. What sets agents apart is their human-like capabilities. They can serve as your sales representative, customer service agent, or personal assistant, delivering content and services in a more intelligent and interactive way.
As we continue to advance our AI technology, we expect agents to unlock unprecedented value for our users. Another standout in our product transformation journey is Baidu Wenku, our one-stop shop for document creation. Since the official ramp-up of Wenku’s AI renovation late last year, the MAU of AI-enabled features has grown rapidly, exceeding 50 million in September, representing a year-over-year increase of over 300%. Users of AI-enabled features demonstrated a stronger willingness to pay, contributing to accelerated growth in our paying user base.
In the third quarter, Wenku’s subscription revenue continued its robust growth trajectory, reaching a year-over-year growth of 23%. We continue to roll out new AI-powered features to facilitate users in their thinking, learning, and creative process. After enabling natural language instructions for PowerPoint slide building and editing last quarter, we introduced a new feature to transform even the most basic text slides into visually stunning, sophisticated, and professional presentations, making presentation creation easier than ever and significantly boosting productivity. With a 14-year-long history at massive user base, Wenku is now experiencing a remarkable transformation as AI technology rejuvenates this product and with unprecedented capabilities.
Building on this momentum, we expect such revolution to ripple through our entire product ecosystem, unleashing extraordinary potential and creating breakthrough innovations. Our vision extends beyond our consumer-facing products. By collaborating with our clients across businesses and public sectors, we can unlock greater potential of our advanced AI technologies. This quarter, we partnered with Yum China, a leading Chinese restaurant company, to deploy Keyue, our ready-to-use customer service product, powered by ERNIE.
Keyue’s key advantage lies in its superior intelligence and ability to significantly reduce the need for human intervention in customer service as its AI capabilities can resolve most user inquiries independently. This simplified solution has enhanced customer service efficiency and reduced operational costs, driving substantial growth in Keyue’s API call volume, with peak API calls reaching several millions a day. The partnership serves as a strong testament to the reliability and excellence of our product. Another example that highlights the expanding use scenarios of ERNIE is [Inaudible] an online exam preparation platform that leverages ERNIE 4.0 Turbo and ERNIE Speed Pro through public cloud API calls.
Initially, [Inaudible] question bank was limited by missing answers and explanations. ERNIE tackled this by automatically generating answers and explanations for existing questions while also creating new answer-explanation pairs, making the question bank far more effective. This resulted in a 150% increase in effective questions in [Inaudible] question bank. By replacing manual explanation writing, the cost of creating explanations dropped by 99.8%.
Additionally, ERNIE powers real-time tutoring services, where users receive instant explanations from AI tutors by simply uploading their questions. With this improvement, users showed a stronger willingness to pay as the paying user ratio increased dramatically by over 100%, leading to a 246% increase in revenue. Our strategic focus on AI innovation continues to drive transformation in consumer-facing applications, as well as industrywide solutions. And we are seeing great results.
Now, let me review the key highlights for each business for the third quarter. AI Cloud revenue reached RMB 4.9 billion in the third quarter, maintaining double-digit year-over-year increase at 11% while sustaining non-GAAP operating profitability. Gen AI-related revenue maintained strong growth momentum and remained a key growth driver, accounting for about 11% of our total AI Cloud revenue in the third quarter, up from 9% in the previous quarter. This trend reflected increasing recognition of ERNIE’s value among enterprise customers.
Our strong technological advantages in AI infrastructure served as a key enabler for this momentum. During the quarter, we advanced our AI infrastructure management across our GPU cluster that is composed of tens of thousands of GPUs, achieving 99.5% valid LLM training time. Also, we further improved our capabilities in combining GPUs from different vendors for training and hosting models by minimizing throughput loss to within 5%. With this advancement, we further reinforced our position as a leading provider of the most powerful and cost-effective AI infrastructure.
Our MaaS platform continued to evolve, with a strong focus on developing comprehensive toolkits that enhance the model and app development experience for our customers and partners. In Q3, we improved ModelBuilder to make lightweight model fine-tuning easier. With ModelBuilder, lightweight models can be fine-tuned into purpose-specific models, achieving performance comparable to ERNIE 4.0 while offering faster inference speed at lower costs. We are leveraging our AI capabilities to improve internal R&D efficiency and facilitate enterprise application development.
Internally, we have widely adopted AI coding through Comate, our coding assistant for developers. AI now contributes to generating over one-third of our new code, with our developers providing oversight and approval. We are also lowering the barrier for enterprise in AI-native application development. Enterprises can build sophisticated applications through simpler approaches on our MaaS platform, for example, using natural language instructions with minimal coding efforts to develop the most part of a business application.
We introduced Miaoda at Baidu World 2024, marking our first step in delivering null code capabilities. We are excited about its potential to democratize AI as we believe expanding AI accessibility beyond developers will drive innovation at a speed we’ve never seen before. Meanwhile, the growing recognition of ERNIE’s capabilities continues to attract new clients, with notable partnerships established this quarter with FlashEx [Inaudible] These partnerships with leading enterprises reflect the strength and credibility of Baidu AI Cloud. We have also seen significant growth in adoption from mid-tier enterprise customers as reflected in their increasing investments in GPU public cloud services.
In Q3, the incremental revenue from the mid-tier enterprise customers on our public cloud increased by 170% year — quarter over quarter. With these positive trends and developments, we reinforce our long-term optimism about the growth trajectory of our AI Cloud revenue stream. For our mobile ecosystem, Baidu Core’s online marketing revenue declined by 4% year over year in the third quarter due to challenging macroeconomic conditions, our ongoing AI-driven search renovation, and an evolving internet content ecosystem. In the third quarter, we saw continued growth in incremental ad revenue from gen AI and LLM enhancements to our advertising system.
This reflects our ongoing efforts to refine our monetization system and marketing platform, particularly in improving ad-targeting capabilities and scaling up real-time ad generation. We believe agents opened up a lot of new possibilities for our advertising business. This quarter has seen significant improvements in ERNIE agents for advertisers in terms of response accuracy and intent recognition capabilities. As these agents become increasingly sophisticated and effective, we’ve observed growing recognition of their value among advertisers.
As of September, approximately 20,000 advertisers have been generating ad spending through our ERNIE agents on a daily basis. A case in point is one of our healthcare clients. After building its own agent, the client was impressed with its performance. They then proactively contributed their proprietary medical insights to further enhance the agent’s capabilities.
Through continuous fine-tuning, the agent demonstrated good intelligence and service quality, particularly in terms of dialogue accuracy and content — and context comprehension. This enables the agent to deliver more professional and accurate responses to customer inquiries around the clock. Through extended conversations, the agent can better identify customers’ underlying needs and provide tailored service recommendations, significantly enhancing the customer experience. As a result, the client has seen a notable increase in customer engagement, with customers’ willingness to initiate the conversations rising by 13.6% and lead conversion rate increasing by 50% from July to September.
Overall, while our online marketing business is going through a transitional period, we remain steadfast to aggressively drive forward our AI transformation. We are confident this will deliver long-term value for our users, customers, and shareholders. Turning now to intelligent driving. We’ve reached another significant milestone, the sixth generation of our autonomous vehicle, RT6, is now operating on public roads in multiple cities in China.
This not only expands our vehicle lineup but also reaffirms our commitment to scaling operations and providing users with safer, more affordable, and comfortable mobility experiences. Following our achievements of 100% fully driverless operations in Wuhan last quarter, the proportion of fully driverless operations nationwide surpassed 70% in the third quarter and 80% in October. Recently, we have taken another step forward in expanding fully driverless operations. We are delighted to share that in October, we achieved 100% fully driverless operations in Chongqing, where we currently operate a growing fleet of autonomous vehicles.
We continue to scale up our services in third quarter with Apollo Go providing about 988,000 rides to the public nationwide, representing year-over-year increase of 20%. The cumulative rides provided to the public surpassed 8 million in October, further solidifying our leadership in smart mobility. We’re fully confident that our autonomous driving technology has achieved technical maturity with proven safety and reliability through extensive testing and real-world operations. While our technology is ready for wider deployment, safe and responsible autonomous driving requires a solid foundation of a harmonized regulatory framework.
We remain patient as we continue to work alongside the evolving regulatory framework, standing ready to rapidly scale up our operations when the time is right. Meanwhile, we are also actively exploring expansion into additional cities and new operational models such as asset-light strategies to enhance operational resilience and unlock new growth potential. In summary, while our overall revenue growth remains measured in the near term, we have strong conviction in our strategic direction, supported by encouraging progress across multiple fronts, from increased user metrics in search to the growing adoption of our large language models and cloud solutions. Looking ahead, we’re confident that our strategic focus and execution will yield meaningful returns.
Before I hand over the call, I’m pleased to welcome our interim CFO, Junjie He or Jackson. With that, let me turn the call over to Jackson to go through the financial results.
Junjie He — Interim Chief Financial Officer
Thank you, Robin. Hi, everyone. I’m honored to step into the role of interim CFO. Following my initial conference call, I look forward to engaging with analysts and shareholders in the coming period.
Now, let me walk through the details of our third quarter financial results. Total revenues were RMB 33.6 billion, decreasing 3% year over year. Revenue from Baidu Core was RMB 26.5 billion, which was basically flat from last year. Baidu Core’s online marketing revenue was RMB 18.8 billion, decreasing 4% year over year.
Baidu Core’s non-online marketing revenue was RMB 7.7 billion, up 12% year over year, mainly driven by AI Cloud business. Revenue from iQIYI was RMB 7.2 billion, decreasing 10% year over year. Cost of revenues were RMB 16.4 billion, increasing 1% year over year, primarily due to an increase in traffic acquisition costs and costs related to AI Cloud business, partially offset by a decrease in personnel-related expenses and cost of goods sold. Operating expenses were RMB 11.2 billion, decreasing 5% year over year, primarily due to a decrease in personnel-related expenses and partially offset by an increase in channel spending and promotional marketing expenses.
Baidu Core’s operating expenses were RMB 9.9 billion, decreasing 5% year over year. Baidu Core SG&A expenses were RMB 5 billion, increasing 4% year over year. SG&A accounted for 19% of Baidu Core’s revenue in the quarter, compared to 18% in the same period last year. Baidu Core R&D expenses were RMB 4.9 billion, decreasing 13% year over year.
R&D accounted for 19% of Baidu Core’s revenue in the quarter, compared to 21% in the same period last year. Operating income was RMB 5.9 billion. Baidu Core’s operating income was RMB 5.7 billion, and Baidu Core’s operating margin was 21%. Non-GAAP operating income was RMB 7 billion.
Non-GAAP Baidu Core operating income was RMB 6.7 billion, and non-GAAP Baidu Core operating margin was 25%. Total other income, net was RMB 2.7 billion, increasing 40% year over year, primarily due to an increase in fair value gain from long-term investments and disposal gain, partially offset by an increase in net foreign exchange loss arising from exchange rate fluctuations between renminbi and the U.S. dollar. Income tax expense was RMB 814 million, compared to RMB 1.3 billion in the same period last year.
Net income attributable to Baidu was RMB 7.6 billion, and diluted earnings per ADS was RMB 21.6. Net income attributable to Baidu Core was RMB 7.5 billion, and net margin for Baidu Core was 28%. Non-GAAP net income attributable to Baidu was RMB 5.9 billion. Non-GAAP diluted earnings per ADS was RMB 16.6.
Non-GAAP net income attributable to Baidu Core was RMB 5.7 billion, and non-GAAP net margin for Baidu Core was 21%. As of September 30, 2024, cash, cash equivalents, restricted cash, and short-term investments were RMB 144.5 billion and cash, cash equivalents, restricted cash, and short-term investments excluding iQIYI were RMB 140.3 billion. Free cash flow was RMB 2.6 billion, and free cash flow excluding iQIYI was RMB 2.4 billion. Finally, Baidu Core had approximately 31,000 employees as of September 30, 2024.
With that, operator, let’s now open the call to questions.
Questions & Answers:
Operator
Thank you. 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 — Analyst
Hi. Thank you. Good evening, management. Thanks for taking my questions.
My question is related to AI and search. So, how should we think about the ramp-up pace for the generative AI results penetration rate and what does management think of the optimal level, and how long do you plan for the search product transformation period to last? And could management also give more color on the recent developments in the monetization of the AI search? So, when does management expect to commercialize the AI search results? Thank you.
Rong Luo — Chief Financial Officer
Hi, Alicia. This is Julius. Let me take your question. We have been continuously transforming our core search with AI technology.
And same as what Robin has said just now, today, over 20% of pages and over 70% of the monthly active users, they have engaged with gen AI contents already. However, I think the number merely represents our initial progress in one aspect of our AI-driven transformation of search. It should not be taken as a comprehensive indicator of our AI-generated content, nor should it be used as the only benchmark for our progress. What really matters goes beyond it.
In fact, we aimed to revolutionize the search through ERNIE and transform every aspect of search experiences, providing the users with unlimited, on-demand, and personalized content through the combination of various formats whenever they need. Leveraging ERNIE, we gain deeper insights into the user intent behind increasingly complicated queries, which can enable us to customize the content generation to better fulfill their needs. While continuously enhancing the AI-generating content quality, we’re also trying to diversify our content formats such as AI summaries, as you can see in the top page, and images and videos, agents, posts, and even digital humans. All of these different formats can be dynamically combined to create some personalized search experiences.
With ERNIE, we can tailor both the gen AI content and its presentation format, which can best match what users want to see and how they prefer to consume this kind of information. This can allow us to optimize the user experiences and drive higher engagement. The AI-driven transformation of our search has already shown initial progress across certain user metrics, which gives us some kind of confidence to keep pushing forward with our efforts. Through ongoing explorations, we are charting the new frontiers of possibility, unlocking unprecedented potentials with the power of AI.
And quarter by quarter, we will continue to see an update of achievement across various aspects of AI-driven search transformations. Specific about your question in the monetization side, we are seeing some encouraging early results in certain verticals, including legal, education, and B2B services. For example, our ERNIE agents for advertisers today have demonstrated increase in effective sales leads, which can bring the value to advertisers, while enhancing the overall user experience. As our ERNIE foundation models continue to evolve, we expect agents to unlock greater revenue opportunities in the future.
That said, we are still in the early stages of exploring monetization opportunities for AI search and we are taking a measured approach at our own pace. In the near term, we will prioritize the user experience enhancement instead of rushing into monetization. In fact, our product transformation journey strongly reflects our strategy direction, which is we are making deliberate choices to prioritize the long-term value creation over short-term gains. It’s true that we experience some short-term pressures, but these near-term trade-offs are something we are willing to accept as we work toward our long-term vision.
Beyond such, we’ll — we also have cultivate a comprehensive mobile ecosystem with multiple consumer-facing products that serve the massive user bases such as Speed, Wenku, and [Inaudible] And we also are leveraging ERNIE to revolutionize the entire product lines across our mobile ecosystem, enhancing the content production and try to optimize the distribution mechanism and try to strengthen the monetization capabilities. We remain fully committed to this transformation and focusing primarily on enhancing the user experiences and engagement. With this kind of encouraging early results have reinforced our confidence in the strategic direction, and we believe that following this period of adjustment, we expect to start seeing improvements in our advertising business next year. Thank you for your question.
Operator
Your next question comes from Alex Yao with J.P. Morgan.
Alex Yao — Analyst
Thank you, management, for taking my question. My question is about the ERNIE’s API. I think the API calls have been growing fairly rapidly in the recent months. Could you guys share more details on the key drivers behind this growth? In addition, has any potential killer app emerged in the 2B or 2C markets? Finally, what is management’s outlook of ERNIE adoption over the medium to longer term? Thank you.
Robin Li — Co-Founder and Chief Executive Officer
Hi, Alex. This is Robin. Yes, we’ve seen remarkable growth in API calls. I think a number of factors are driving this growth.
First, ERNIE gets smarter every day and hallucinations have been reduced significantly. So, more and more applications can use it to add value to the users. And second is the inference cost. It gets lower and lower, so more customers can afford it.
And third, it’s the toolchain we provide to our customers. They can tailor the model to address their own specific needs easily in each application scenario. The substantial volume of ERNIE API calls is driven by both internal and external demand. We began our product transformation in Q2 of last year.
We first upgraded our monetization system, delivering higher advertiser ROI and generating hundreds of millions of incremental revenues each quarter. And then we extended the AI transformation across our major consumer-facing products that includes those products over, you know, 100 million MAUs: Baidu Search, Wenku, Baidu Drive, Baidu Keyboard, and Baidu Map. We are encouraged by how AI could thoroughly revolutionize and rejuvenate these long-standing products with massive user bases. And this year, we accelerated the renovation of search, showing more and more generated content on our search result page.
We believe the new Baidu Search has the potential and is well-positioned to become a killer app in the age of gen AI. This is because search is, by nature, deeply rooted in language and text understanding, which aligns perfectly with our LLM capabilities. We are fortunate to have hundreds of millions of search users every day and the most advanced foundation models. And along this line, millions of agents are being built to answer users’ and customers’ questions.
This also contributed to the growth of ERNIE API calls. While most API calls currently come from the renovation of our 2C products, ERNIE has gained strong recognition in helping businesses tackle challenges effectively. An external API calls have grown quite quickly, roughly, like, 240% year — quarter over quarter in the last week of Q3. We see strong adoption across sectors, including online education, social media, restaurant and food services, healthcare, legal consultation, and recruiting.
The widespread adoption shows that enterprises recognize the value of our powerful models and are willing to invest in them. We are thrilled with our progress and committed to accelerating ERNIE’s adoption across all our product lines as we transform from an internet-centric business to an AI-first business. With gen AI becoming central to our entire product lineup, we are confident it will unlock new revenue streams and strengthen our market leadership. Thank you.
Operator
Your next question comes from Gary Yu with Morgan Stanley.
Yuko Oku — Analyst
Hi. Thank you for the opportunity to ask question. I have a macro-related question. Can management comment on the underlying app demand trend? And also, have you seen any ad spend sentiment improvement, especially after the latest stimulus policy? And what’s management view on the macro outlook heading into 2025? Thank you.
Rong Luo — Chief Financial Officer
Hi, Gary. Let me take your question. I think our advertising business is highly correlated to the macro environment, especially with the offline small and medium enterprises across a wide range of industries, which make up the majority of our advertising spaces. As these small and medium enterprises are both deeply connected to the domestic consumption and highly sensitive to macro conditions, their vitality and recovery pace serves as more relevant indicators for our business.
In Q3, we have seen that continued weakness across the verticals such as the real estate, franchising, and healthcare. And so far, in Q4, we have not observed the notable improvement in advertiser spending patterns, and consumer spending remains subdued. And having said that, we are particularly encouraged by the strength and the timeliness of the recent stimulus policies, which continue to be rolled out, as well as the constructive initiatives we are seeing across the market. And while we think it will still take some time for these measures to reach offline small and medium enterprises and boost their confidence in advertising spending, so we remain conservatively optimistic about the recovery trend ahead.
We believe that once the macro environment improves and domestic consumption can pick up, small and medium enterprises’ confidence will quickly rebound. And Baidu has long been the primary platform of choice for these advertisers. With proven effectiveness in driving their customer acquisition and growth, we expect them to return for advertisement spending, which can help us to drive a meaningful recovery in the advertising business. With our extensive reach and unique value propositions to offline small and medium enterprises, we believe that our advertising business will resume with strong recovery momentum once the market confidence comes back.
Thank you, Gary.
Operator
Your next question comes from Lincoln Kong with Goldman Sachs.
Lincoln Kong — Goldman Sachs — Analyst
Thank you, management, for taking my question. So, we — given that ongoing ad pricing pressure and our plan to accelerate and broaden our AI transformation, so could management share more color around our business focus at the moment, our priority in terms of our investment, resources spending, and capital allocation, and how should we think about the margin trend in 4Q as well? Thank you.
Junjie He — Interim Chief Financial Officer
Hi, Lincoln. This is Jackson. I will take your call. Yeah, currently, we will remain committed to our AI-focused strategy, as opposed to our near-term priority and the long-term strategic emphasis.
Given our ongoing AI-driven renovation of search and the fact that we are not heavily monetized AI-generated search results soon, we expect the circumstances will remain under pressure for our online marketing business in the near term. However, we remain steadfast in this strategic direction as we see strong long-term value ahead. In line with this strategy, we are going through a period of continued initiatives where we further intensify our product renovations, with a strong emphasis on AI search, and continue to invest in advancing ERNIE’s capabilities while enhancing our AI Cloud offerings to maintain a healthy margin and expanding our autonomous driving initiatives, where we view has a viable path to profitability. We believe these initiatives are essential to sustain our position as a leading technology innovator in China.
As a result, the near-term margins will be in a period of adjustment. While looking to 2025, we will focus on optimal resource allocation to high-growth opportunities while staying aligned with our long-term strategy. Thank you.
Operator
Your next question comes from James Lee with Mizuho.
James Lee — Analyst
Great. Thanks for taking my questions. I have several follow-up questions on cloud. In terms of revenue growth, it has moderated maybe compared to the last couple of quarters, and maybe can you unpack the performance of both personal and enterprise cloud? And in addition, you know, can you give us an update in terms of competitive landscape in the market? And lastly, can you share the outlook for cloud business, particularly contribution from gen AI, and how should we think about the long-term margin assumptions? Thank you.
Dou Shen — Executive Vice President and Head, AI Cloud
Thank you, James, for your question. This is Dou. Since last year, we have observed an accelerating shift in China’s cloud industry toward AI computing, driven by the development of gen AI and foundation models. We have seen more customers choosing Baidu’s AI Cloud infrastructure and foundation models.
The growing adoption is evidenced by our gen AI-related revenue. It started at about 5% of our total AI Cloud revenue when we first reported it in Q4 last year, and has now more than doubled, reaching 11% in the third quarter. So, we are confident that this upward trend will continue. And so, as you know, our AI Cloud business consists of two main parts: personal cloud and cloud service for enterprise and public sector.
This quarter, total AI Cloud revenue growth slightly moderated, primarily due to a temporary impact on personal cloud revenue brought by our short-term business adjustment. However, we believe ERNIE’s ongoing renovation of the personal cloud will help mitigate this short-term impact and position it well for greater long-term growth opportunities. Meanwhile, our cloud service for enterprise and public sector, which makes up the majority of our AI Cloud revenue, actually maintained a strong momentum this quarter, continuing to outpace our overall AI Cloud business. So, the growth in enterprise cloud revenue has been driven by strong demand for model training and inference across various sectors, including internet education, finance, and so on.
This reflects our customers’ strong recognition of our AI infrastructure and MaaS platform capabilities. In particular, the major customers in industries like internet, tech, and automotive increased their spending on our GPU public cloud. So, we are also seeing a rise in both the number and spending of mid-tier enterprises, especially in sectors like marketing software and other growing industries. In today’s cloud market, foundation model capabilities have become more and more essential.
Given the significant requirement for powerful AI infrastructure, specialized AI expertise, and substantial capital investments, only a select few foundation models will survive, both in China and overseas. And for sure, we will be one of them standing as the front-runner. So, we are proud that ERNIE has already demonstrated its market leadership through widespread adoption and advanced technology. So, these [Inaudible] position ERNIE to stay at the forefront, attracting more cloud customers and satisfying our market leadership.
So, on the profit side, our AI Cloud is healthier than ever, with non-GAAP operating profit margin expanding year over year. So, this improvement is driven by a continued margin enhancement from our gen AI-related revenue and our commitment to achieving high-quality revenue growth. So, looking forward, we are confident that our AI Cloud business will maintain strong revenue growth momentum over time while continuing to deliver a healthy operating profit. Thank you, James.
Operator
Your next question comes from Thomas Chong with Jefferies.
Thomas Chong — Analyst
Hi. Good evening. Thanks, management, for taking my question. My question is related to robotaxi.
Given the recent IPO wave of autonomous driving-related companies that have partnerships with big automakers and ride-hailing platforms, what are your thoughts about how the competitive dynamic going forward? Looking ahead, how do you see the Apollo Go robotaxi fleet and business strategy development? Can management share any details about the future plans for expanding and growing Apollo Go? Thank you.
Robin Li — Co-Founder and Chief Executive Officer
Yeah. Autonomous driving has a high bar. It requires best-in-class technical expertise, top talent, as well as strong commitment to long-term investment and significant capital reserve. We have built an unparalleled foundation in all these areas, giving us a strong competitive edge and setting a high bar for others to follow.
As one of the earliest players in the market, we’ve consistently demonstrated our unwavering commitment through over a decade of investment. This long-term dedication has fueled our technological advancements, positioning us as the global leader in autonomous driving technology. As mentioned earlier, Apollo Go hit a significant milestone of 8 million total rides, making us the world’s top autonomous ride-hailing service provider. In addition to our cutting-edge technology and strong operational capabilities, we’ve achieved significant cost-effectiveness in hardware.
RT6 stands out with a competitive price of under $30,000 for mass production, making it the most competitive option in the market. All the strengths mentioned above are driving us forward, paving the way to validate our business model. The overall market is still in its infancy, so competition should help accelerate the growth of the market and foster a more friendly, pro-innovation regulatory environment. We believe such regulatory environment is essential for the healthy growth of this industry, and we remain prepared for expansion when the time is right.
Meanwhile, we are actively seeking new opportunities for international expansion. We see significant potential in cities that enable large-scale, fully driverless operations, similar to Wuhan and Chongqing. Additionally, we are open to exploring different business model options, focusing on asset-light strategy to keep our operations flexible and efficient as we grow. Thank you.
Operator
Your next question comes from Xiong Wei with UBS.
Wei Xiong — Analyst
Good evening, management. Thank you for taking my question. My question is regarding capital allocation. I was wondering, could management provide any updates on capital return plans for shareholders such as share buybacks and dividend? Thank you.
Junjie He — Interim Chief Financial Officer
Hi, Wei. I will take a question. So, since becoming a public company in 2005, we have made consistent efforts to generate long-term value to our shareholders through our growth as a leading technology company in China, complemented by share buyback programs. So, over the past few years, we have maintained a steady pace of share repurchases, averaging about $1 billion annually.
So, I’m pleased to share that through our consistent share repurchases efforts, we have seen a decline in our total number of shares outstanding this year. While the pace of share buybacks may vary from quarter to quarter according to our established execution rhythm, we remain firmly committed to our share buyback programs as a key component of delivering value to our shareholders. While we are currently focused on executing our sustainable and recurrent share buyback programs, we remain open to evaluating various options for returning value to shareholders. As always, we firmly believe in Baidu’s long-term and sustainable growth potential, and we’ll ensure that our shareholders will be rewarded for their trust in us.
The ultimate way of serving shareholders’ interests is to have a solid business foundation and capitalize on growth opportunities. We are in the midst of a paradigm-shifting technology transformation based on gen AI and foundation models. At Baidu, we have the AI infrastructure, technical expertise, as well as financial and human resources, to execute on our plans. So, we want to have the flexibility to invest, as needed, to propel our new business forward and, with that, bring long-term value to our shareholders.
Thank you.
Operator
Your next question comes from Miranda Zhuang with Bank of America Securities.
Miranda Zhuang — Bank of America Merrill Lynch — Analyst
Thank you, management, for taking my questions. So, can management provide us an overview of ERNIE technology road map for the future development? And what are the upcoming milestones for the company’s AI model? And then when can we expect to achieve them? Thank you for my question about ERNIE. Thank you.
Robin Li — Co-Founder and Chief Executive Officer
Well, we launched ERNIE Bot in March of last year. Since then, we’ve been continuously enhancing our foundation model capabilities, particularly our flagship models. In October last year, we introduced the ERNIE 4.0, China’s first GPT-4-type model, with world-leading capabilities. And then we launched ERNIE 4.0 Turbo in June of this year and delivering even greater performance.
Building on these milestones, we are committed to continuously advancing our flagship models to deliver better performance, accuracy, and broader support for diverse user needs. We expect to launch a new version of ERNIE early next year to further strengthen our leadership position in foundation models. Compared to overseas companies, we distinctly take an application-driven approach. We believe foundation models hold real value only when they are widely used practical applications that meet user needs.
Over the past 18 months, our model development has centered on addressing real-world problems according to market demand. With this approach, we’ve made significant progress in resolving hallucinations and improving accuracy while also introducing a series of models tailored to diverse needs. To make learning models more accessible and affordable, we boosted the performance, lowered inference costs, and enhanced response speed. What sets us apart is that we really have a clear sense of our strengths and where we are heading.
We prioritize those technologies that align most closely with our business and create the most value. Earlier this year, I mentioned that we have expanded our visual foundation model capabilities to our autonomous driving business, aiming to further enhance our leading position in this space. Now, we are also proactively exploring multimodal capabilities and applications based on our strengths in language models, speaking to create more synergistic value and unlock new possibilities. Throughout all areas, we are very thoughtful about how we allocate our resources, optimizing our foundation models in directions that maximize impact across our business and help maintain our market leadership.
Thank you.
Operator
[Operator signoff]
Duration: 0 minutes
Call participants:
Juan Lin — Director, Investor Relations
Robin Li — Co-Founder and Chief Executive Officer
Junjie He — Interim Chief Financial Officer
Alicia Yap — Analyst
Rong Luo — Chief Financial Officer
Alex Yao — Analyst
Yuko Oku — Analyst
Lincoln Kong — Goldman Sachs — Analyst
James Lee — Analyst
Dou Shen — Executive Vice President and Head, AI Cloud
Thomas Chong — Analyst
Wei Xiong — Analyst
Miranda Zhuang — Bank of America Merrill Lynch — Analyst
