Yahoo Finance

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 be audited most directly comparable GAAP measures and is available on the IR website at ir.baidu.com. As a reminder, this conference will be recorded.

In addition, the 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 for the fourth quarter saw a slight year-over-year increase, reaching RMB 27.7 billion. Our AI cloud business demonstrated robust momentum with revenue growth accelerating to 26% year over year, offsetting the softness in our online marketing business. The growth was primarily driven by broad market recognition of our AI capabilities in which we will continue to invest to make sure we stay at the forefront of this technology megatrend.

2024 marked a pivotal year as Gen-AI demonstrated strong market demand. We see this momentum reflected in the genuine value we created for our customers through our AI products and solutions, as well as AI cloud services. This growth shows how enterprises have awakened to AI's transformative potential, catalyzing broader adoption across diverse scenarios, while affirming the market's trust in Baidu's technical capabilities and AI expertise. As we continue to cement our leadership position as an AI innovator, our vision is gradually coming to fruition on multiple fronts.

Our strategic AI investments are showing meaningful progress across our key businesses. Our cloud business demonstrated strengthening momentum toward the end of the year with both robust revenue growth and margin expansion as we continued to rebuild the entire product portfolio of our mobile ecosystem with AI, most notably in Baidu search. Meanwhile, Baidu's autonomous ride-hailing service, Apollo Go has proven the viability of its business model, strengthening our confidence in pursuing global expansion and exploring asset-light business models. ERNIE has demonstrated industry-leading capabilities in instruction following and state-of-the-art iRAG technology that minimizes hallucination.

This best-in-class capabilities have driven widespread adoption across various scenarios, leading to significantly growing API costs. In December, ERNIE handled approximately 1.65 billion API costs daily. We witnessed a particularly strong external demand with external API costs growing 178% quarter over quarter. We saw rapid growth across various business sectors such as education, e-commerce, entertainment, and recruitment.

The robust growth in external API cost is driven by two reasons. First, businesses are finding enough value from integrating ERNIE into their daily operations. Second, price matters. As the inference cost goes down, adoption goes up quite quickly.

We recently made a significant strategic decision to open source our upcoming ERNIE 4.5 series and make ERNIE Bot free for end users, which we believe will further expand market awareness of ERNIE's capabilities and enable broader adoption across an increasing array of industries. Reflecting this growing market recognition, our AI cloud business delivered a robust performance. In Q4, we have partnered with leading enterprises across diverse sectors, helping them address specific operational challenges using our AI products and solutions. We collaborated with State Grid Corporation of China to develop a tailored AI solution based on ERNIE and our MaaS platform, [Inaudible].

We have developed comprehensive AI solutions that permeate key operational processes. From power dispatching and equipment condition diagnosis to procurement, safety supervision, marketing, and office systems. Our solutions have demonstrated remarkable effectiveness in boosting operational reliability. Impressed by these results, State Grid has increased its spending on our AI cloud services, further deepening its collaboration with us.

We also collaborated with [Inaudible], a leading recruitment company to adopt [Inaudible] ERNIE-powered outbound calling product for campus recruitment, featuring highly humanlike voice with natural conversational abilities. The product enables sophisticated multi-round interactions. ERNIE's advanced capabilities ensured accurate gauging of candidates' interest in job opportunities while consistently keeping conversations on track. With this product, [Inaudible] achieved a 50% higher acceptance rate for job fair invitations and 70% lower labor costs.

The growing market recognition and adoption reinforced our confidence, and we are optimistic that the growth momentum we witnessed today is just a start. As enterprises continue to discover novel ways to harness ERNIE and our AI cloud potential, we foresee an ever-expanding horizon of possibilities unfolding before us. On the consumer-facing product, Baidu Wenku is a shining example. In Q4, Wenku's subscription revenue grew 21% year over year.

In December, Wenku's monthly active user of AI-enabled features reached 94 million, almost doubling quarter over quarter, making Wenku one of the world's largest AI products by the number of paying users. This quarter, we integrated Baidu Wenku with Baidu Drive, enhancing Wenku as an AI-powered creativity platform. This integration allows Wenku to combine personal cloud storage with AI-powered search and document-creation capabilities, while enabling users to streamline their workflow from searching, accessing files to creating and sharing, all within a unified seamless platform. Furthermore, Wenku introduced Free Canvas in January, which transforms content creation by enabling users to organize diverse formats, text, slides, audio, videos, and even website links on a unified canvas.

Free Canvas also integrates with Wenku's existing AI capabilities, including AI editor and AI-powered PowerPoint features, allowing users to access our full suite of AI editing tools and save or share their work in any desired format via Baidu Drive. Leveraging ERNIE, Wenku is transforming into a next-generation intelligent creativity platform that redefines how users create, collaborate, and bring their ideas to life. Now, turning to search, our largest consumer-facing product, which stands as another compelling testament to our AI transformation. Over the past several quarters, we have aggressively leveraged ERNIE to transform search.

This has been a challenging but deeply meaningful journey like pioneers crossing unfamiliar waters, we embraced the uncertainties that come with starting new territories. After extensive exploration, we have defined our path forward in leveraging the power of language models to present the multimedia content in search results. Currently, 22% of our search result pages contain AI-generated content. During the quarter, we mainly focused on redefining and enhancing the quality of our AI-generated search results.

Through ERNIE's advanced capabilities in content generation, we can deliver unlimited high-quality content on demand, vastly expanding the depth and breadth of our search results. This quarter, we expanded beyond text generation to generate diverse content formats, including short videos, nodes, agents, digital humans and live streaming. Through this more dynamic and varied content combinations, users can now discover a broader range of information that was previously inaccessible, presented in more engaging ways that's aligned with evolving user preferences. As a result, we have achieved substantial improvements across three dimensions of our search results in quality, quantity, and format diversity.

Alongside this, we further enhanced the search experience by optimizing the matching between queries and search results. By analyzing the nature and intent of each query, the Gen-AI-enabled search can determine the optimal content format and presentation style such as delivering AI-generated short videos for how to queries, generating comparative charts for analysis-focused searches, and dynamically organizing content sequences. This intelligent matching ensures that users receive not just the most relevant results but in the most effective and easy-to-understand format. The effectiveness of this transformation is reflected in our improved user metrics.

On Baidu App, users exposed to AI-generated search results demonstrated higher engagement levels, conducted more search queries, and showed higher retention rates and longer time spent. We are encouraged to see daily search queries per user increased by 2% year over year in December. During our AI transformation of search, we placed strong emphasis on agents as we envision agents will become one of the most important forms of AI-native applications. Our agents have now transcended their initial chatbot origins through convergence with digital human technology, evolving into embodied intelligent entities with vivid visual presence, thanks to our industry-leading digital human technology.

This represents a powerful innovation path we pioneered, breaking new ground in how agents interact with users. Taking our newly introduced lawyer agents as an example. These embodied lawyer agents can transform legal expertise and industry know-how into engaging multimodal content, featuring lawyers' actual appearance in live streaming and AI-generated short videos. Through these agents, we can efficiently generate high-quality authoritative content at scale for Baidu search and feed while pioneering innovative formats of content and user engagement.

Moreover, these agents also create significant value for lawyers by helping them build personal brands, enhance engagement with potential clients, and improve lead conversion. Such agents are becoming a compelling new addition to Baidu's content ecosystem, creating new possibilities while further diversifying our content landscape. In the field of intelligent driving, our continued investments in AI are also propelling significant progress in our Robotaxi business. In Q4, Apollo Go delivered remarkable results that reinforced our position as the world's top autonomous ride-hailing service provider.

Our domestic operations continue to show robust momentum, and we have successfully validated our pioneering business model through years of operations and established a path to profitability in the autonomous driving industry. Importantly, our unrelenting focus on safety remains at the core of our efforts as we expand our services. To date, Apollo Go fleets have accumulated more than 130 million autonomous kilometers with an outstanding safety record, underscoring our industry-leading standards. This operational and safety track record has paved the way for further scaling and operational enhancement.

Building on this achievement, we are confident in expanding beyond Mainland China as we enter the next phase of growth. In November 2024, Apollo Go secured permits to conduct testing on open roads in Hong Kong, making us the only company to receive Robotaxi testing authorization in the region. Notably, Hong Kong represents our first entry into a right-hand drive left-hand traffic market, showcasing our ability to adopt our autonomous driving technology to different traffic systems while maintaining rigorous safety standards. This achievement recognizes our deep expertise in autonomous driving operations as we are committed to expanding our footprint and bringing safe, reliable, and convenient autonomous drive hailing services to more cities and regions worldwide.

Now, let me review the key highlights for each business for the fourth quarter. AI cloud revenue reached RMB 7.1 billion in Q4, delivering a strong year-over-year increase of 26% with expanding non-GAAP operating margins. This quarter, we doubled the scale of our unified GPU cluster, yet still achieved a 99% valid training time, a remarkable achievement enabled by our faster and more advanced network architecture. Our network has proven the capability to enable GPUs across different geographical locations to work together seamlessly with minimized performance loss and automatically resolve for the frequent network issues inherent in large-scale computing with almost unperceivable impact to our clients, demonstrating our ability to deliver peak performance at a massive scale.

We continue to enhance [Inaudible], our MaaS platform, which improves the model and app development experience for our customers and partners. Our platform features the early family of models and offers access to a selection of high-quality models across the globe with DeepSeek being one of our latest additions. We offer post-train, SFT, data labeling and preparation for our models hosted on [Inaudible], enabling us to deliver the most suitable models and tools to enterprise clients at all scales. The increasing market recognition of our AI expertise continues to drive strong momentum in our enterprise client pipeline.

We have established deep partnerships with leading enterprises, including Xiaomi and [Inaudible], further highlighting the growing trust in Baidu AI Cloud. Besides, we are increasingly becoming the preferred choice for mid-tier businesses. This reflects the strengthening competitiveness of our AI infrastructure and marks a healthy evolution and diversification of our customer portfolio. For our mobile ecosystem, our current priority is to further enhance the user experience and refine our product features.

We will explore monetization opportunities after our Gen-AI-enabled search product features and framework become sufficiently refined. Additionally, we continue to see agents emerging as a promising avenue for monetization, which represents meaningful growth potential for our advertising business. In December, over 27,000 advertisers have been generating ad spending through ERNIE agents on a daily basis with rapid growth in healthcare, legal, education, and B2B sectors. This rapid adoption stems from agents' unique ability to address their specific pain points.

Agents can quickly master complex non-standardized service portfolios, a task that traditionally requires lengthy training for human customer service people, while efficiently qualifying these from standard customer inquiries through intelligent multi-round interactions. For intelligent driving, in Q4, Apollo Go provided approximately 1.1 million rides to the public nationwide, representing a year-over-year growth of 36%. In January, the cumulative rides provided to the public exceeded 9 million. Also, we reached another milestone recently with the successful transition to 100% fully collaborative operations nationwide.

Building on this strong momentum and with an unwavering commitment to safety as our top priority, we expect to see even faster growth in our ride volumes, with a significant acceleration anticipated in 2025 as we further scale our operations. We continue to improve the technology as we scale up, combined with our validated business model and initial steps in expanding global footprint. We believe we are well positioned to create substantial value and reshape the future of mobility. Driven by our deep-rooted belief in technology and conviction in AI long-term value creation potential, we will continue to invest to make sure we stand at the forefront of cutting-edge AI technology and deploy that to all our products across the board.

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. Now, let me walk through the details of our fourth quarter and full year 2024 financial results. Total revenues in the fourth quarter were RMB 34.1 billion, decreasing 2% year over year. Total revenues for the full year 2024 were RMB 133.1 billion, decreasing 1% year over year.

Baidu Core's Q4 revenues were RMB 27.7 billion, increasing 1% year over year. In 2024, Baidu Core generated RMB 104.7 billion in revenue, increasing 1% year over year. Baidu Core online marketing revenue decreased 7% year over year to RMB 17.9 billion in Q4, accounting for 65% of Baidu Core's total revenues. Baidu Core online marketing revenue decreased by 3% year over year in 2024.

Baidu Core's non-online marketing revenue was RMB 9.8 billion, up 18% year over year. For the full year 2024, non-online marketing business increased 12% year over year. The increase in non-online marketing business was mainly driven by AI cloud business. Revenue from iQIYI was RMB 6.6 billion in Q4, decreasing 14% year over year.

Revenue from iQIYI was RMB 29.2 billion in 2024, decreasing 8% year over year. Cost of revenues was RMB 18 billion in Q4, increasing 3% year over year, primarily due to the increase in traffic acquisition costs, costs related to AI cloud business, and a one-time write-down of inventories, partially offset by a decrease in personnel-related expenses and content costs. Cost of revenue was RMB 66.1 billion in 2024, increasing 2% year over year, primarily due to an increase in traffic acquisition costs, bandwidth costs, and server custody fees, partially offset by a decrease in personnel-related expenses and content costs. Operating expenses were RMB 12.2 billion in Q4, which remained flat compared to the same period last year, primarily due to an increase in expected credit losses, server depreciation expenses, and server custody fees, which is for Gen-AI research and development input and the channel spending and promotional marketing expenses, partially offset by a decrease in personnel-related expenses.

RMB561 million of the increase in expected credit losses was pertaining to a one-time accrual. Operating expense was RMB 45.8 billion in 2024, decreasing 4% year over year, primarily due to a decrease in personnel-related expenses. Operating income was RMB 3.9 billion in Q4, compared to RMB 5.4 billion for the same period last year. Baidu Core operating income was RMB 3.6 billion, and Baidu Core operating margin was 13% in Q4, compared to RMB 4.7 billion and 17% for the same period last year.

The decrease was due to one-time losses of RMB 1 billion, including accrual of credit losses, write-down of inventories, and others. Operating income was RMB 21.3 billion in 2024. Baidu Core operating income was RMB 19.5 billion, and Baidu Core operating margin was 19% in 2024. Non-GAAP operating income was RMB 5 billion in Q3.

Non-GAAP Baidu Core operating income was RMB 4.6 billion, and non-GAAP Baidu Core operating margin was 17% in Q4. Non-GAAP operating income was RMB 26.2 billion in 2024. Non-GAAP due Baidu Core operating income was RMB 23.9 billion, and non-GAAP Baidu Core operating margin was 23% in 2024. In Q4, total other income net was RMB 2.7 billion, compared to total other loss net of RMB 2.5 billion for the same period last year, mainly due to an increase in net foreign exchange gain arising from exchange rate fluctuation between Renminbi and U.S.

dollar and a decrease in pickup of losses from an equity method investment, which modified certain terms of its preferred shares and resulted in significant loss pickup in 2023. Income tax expense was RMB1.6 billion. In 2024, total other income net was RMB 7.4 billion, increasing 120% year over year, primarily due to a decrease in pickup losses. Income tax expense was RMB 4.4 billion.

In Q4, net income attributable to Baidu was RMB 5.2 billion, and diluted earnings per ADS was RMB 14.26. Net income attributable to Baidu Core are RMB 5.3 billion, and net margin for Baidu Core was 19%. Non-GAAP net income attributable to Baidu was RMB 6.7 billion. Non-GAAP diluted earnings per ADS was RMB 19.19.

Non-GAAP net income attributable to Baidu Core was RMB 3.7 billion, and non-GAAP net margin for Baidu Core was 24%. In 2024, net income to Baidu is up 23.8 billion, and diluted earnings per ADS was RMB 65.91. Net income attributable to Baidu Core was RMB 23.4 billion, and net margin for Baidu Core is 22%. Net non-GAAP net income attributable to Baidu was RMB 27 billion.

Non-GAAP diluted earnings per ADS was RMB 76.85. Non-GAAP net income attributable to Baidu Core was RMB 26.3 billion. And non-GAAP net margin for Baidu Core was 25%. As of December 31st, 2024, cash, cash equivalents, residual cash, and short-term investments were RMB 139.1 billion, and cash, cash equivalents, restricted cash, and short-term investments, excluding iQIYI, were RMB134.7 billion.

Free cash flow was RMB 23million, and free cash flow, excluding iQIYI, was RMB 11.1 billion in 2024. We define net cash position as cash and cash equivalents plus total restricted cash, deposits, and others, less total loans, convertible senior notes, and notes payable. As of December 31st, 2024, net cash position for Baidu was RMB 170.5 billion. Baidu Core had approximately 31,000 employees as of December 31st, 2024.

With that, operator, let's now open the call to questions.

Questions & Answers:

Operator

Ladies and gentlemen, we will now begin the question-and-answer session. [Operator instructions] Your first question comes from Alicia Yap with Citigroup. Please go ahead.

Alicia Yap -- Analyst

Hey, thank you. Good evening, management. Thanks for taking my questions. My question is on the latest hot topic.

So, we have seen DeepSeek gaining a lot of attention lately and Baidu recent announcement for the open sourcing, the upcoming ERNIE 4.5-series and also making ERNIE Bot for free. So, could management share with us the strategic thinking behind this move? And also, how do we view the current competitive landscape of the foundation models? How does Baidu plan to stay ahead of this evolving market? Thank you.

Robin Li -- Co-Founder and Chief Executive Officer

Hi, Alicia. Let me answer your questions. The Gen-AI and foundation model market is still nascent but moving at a very fast pace. The DeepSeek success story will certainly accelerate the adoption rate of foundation models.

As foundation models become more accessible and affordable, I believe we are entering a truly transformative phase. We are seeing an explosion of new AI applications and use cases that will open up incredible opportunities and push the boundaries of what's possible with AI. One thing we learned from DeepSeek is that open sourcing the best models can greatly help adoption. When the model is open source, people naturally want to try it out of curiosity, which helps drive broader adoption.

ERNIE 4.5 will be our best models ever. And we would like to have our users and customers try them out more easily than before. So, our decision to open source ERNIE 4.5 series is also backed by our deep confidence in our technology leadership. This conviction stems from our decade-long commitment to R&D investment, our continuous technological innovation, and our unique position as one of the few companies worldwide with full stack AI capabilities.

ERNIE has already demonstrated strong market traction with daily API costs surging from 50 million to 1.65 billion in just one year. So, through open sourcing, we believe more developers and users will recognize ERNIE's true value, driving even broader adoption and expanding its impact across more scenarios. Similarly, making ERNIE Bot free will enable more users to compare our foundation models with other models on an apple-to-apple basis because they don't charge. Last time we had a major upgrade of ERNIE was in October 2023.

So, it's a long time ago. So, stay tuned for ERNIE 4.5 in the next couple of months. Having said that, I want to emphasize one crucial point. No matter open source or close source, foundation models become truly valuable only when they can effectively address real-world problems at scale.

We have been committed to an application-driven approach in continuously iterating ERNIE. With this mindset, since launching, we have been using ERNIE to renovate our products while serving enterprise customers. Internally, we've leveraged ERNIE to successfully transform our consumer-facing products like Baidu Search and Baidu Wenku. Externally through our MaaS platform, [Inaudible], we are enhancing the model and application development experience for enterprise clients.

ERNIE's industry-leading capabilities in instruction following and state-of-the-art retrieval augmented generation technology minimizes hallucinations, have enabled wide adoption across various scenarios. And [Inaudible] comprehensive tool chain allows our customers to customize any model for their own application scenarios. Going forward, we plan to accelerate ERNIE's iteration based on the performance improvement and cost-cutting potential. We will continue evolving it in areas with the greatest potential for real-world impact.

We are excited about the new chapter of AI development, and we look forward to seeing AI technology bring more groundbreaking applications that have lasting value for the society. Thank you.

Operator

Thank you. Your next question comes from Lincoln Kong from Goldman Sachs. Please go ahead.

Lincoln Kong -- Goldman Sachs -- Analyst

Thank you for taking my question. So, my question is about the search business. So, following recent quarters of search revamp, so how should we think about the future adjustment or changes? So, what's the current mix of the Gen-AI result, and what's our target mix? And also, as AI integration continued to deepen, could you share more color around the user metrics? Given that AI chatbot such as DeepSeek and [Inaudible] also offering search capability, so how do you view the competitiveness or the competition dynamics between Baidu search and those AI chatbots? Thank you.

Rong Luo -- Executive Vice President

Hi, Lincoln. Let me answer questions. Currently, around 22% of all search result pages have contained AI-generated content. However, there is much more premium listing.

We have fundamentally transformed search and making more powerful and efficient than ever before. And then as Robin said in the prepared remarks, with ERNIE, we have expanding the search beyond text, links to offer diverse content formats, including short videos, live streaming, agents, nodes, and short plays. And these different formats can be dynamically combined to create a personalized search experiences. As we continue to transform search, we are also developing capabilities which allow for the deep personalization adapting to each users' habits and preference.

And our efforts have led to the better user results. In January, we can see the amount monthly active users who conduct a search on Baidu App, 83% engaged with the generative AI content. What's even more encouraging is that in December, the search queries per user on the Baidu App grew by 2% year over year. We focus on continuous improvement in user experiences, expanding AI's role as we see increasingly positive user engagements.

Building on this, let me share a small case of how ERNIE agents can create values for our advertisers. During this past Chinese New Year holiday, while many small and medium enterprises are closed for one week and customer demand in certain industry actually peak, our ERNIE agents effectively bridge this gap. For example, hearing a company should increase inquiries as a family reunited and chat on elderly parents' health needs. But their customers -- but their own customer service team actually were away because they were on holiday.

Through ERNIE agents, we have the company effectively identify and qualify the highly relevant sales leads, enabling them to efficiently flow up and serve their customers seamlessly even with a reduced holiday. And regarding your question about AI chatbots and search, we believe that we are still in the very early stage of the AI revolution driven by the foundation models. Whether it's AI-native chatbot like DeepSeek or our AI-driven products, these kind of efforts, this all represent different ways to explore AI potential. As the leader in China's search market with hundreds of millions of users, we maintain agility and holistic market view by placing the best innovations and incorporating the truly innovative AI features.

We continue to pioneer AI transformations and enhancing the search experiences for our massive user base. Also, search by its nature is deeply rooted in language and text understanding, which aligns perfectly with the large language model capabilities, which position us well to lead in the AI-powered search transformation. We believe search is evolving into an integrated platform that go beyond the AI chatbots and one that not only provides the smart answers but also guides users through their in journey from finding answers to providing in-depth analysis to completing tasks and ultimately to delivering the comprehensive services and solutions. While AI chatbot represents an important early phase AI application development, the [Inaudible] today has not yet emerged.

And for us, the key is to maintain this rapid and determined pace of the AI transformation, focusing on discovering what users truly need and one from the next generation of search. Thank you, Lincoln.

Operator

Thank you. Your next question comes from Alex Yao from JPMorgan. Please go ahead.

Alex Yao -- Analyst

Hey, management, thank you for taking my question. So, could you comment on Core ad growth outlook for first quarter '25 and full year 2025? What are the macro assumptions you made underlying this outlook as the advertising business truly bottomed out in 4Q '24 and started to recover? Lastly, what are the potential opportunities for monetizing Gen-AI search, and what is the expected timing? Thank you.

Rong Luo -- Executive Vice President

Thank you, Alex. This is Julius. Let me answer your question. Yes, we are seeing the different kinds of seamless policies rolling out in the past few months like the monetary easing of fiscal stimulus and trading policies -- while we believe that these measures will eventually will benefit the economy, and they need some time to flow through.

Given that our advertising business is highly correlated with the small and medium enterprises who are particularly sensitive to macro conditions, plus is persistently challenging competitive landscape. And we haven't really seen any improvements in Q4 yet. Despite the near-term pressures, our efforts to transform search with financial models are progressing steadily. And we believe that this AI transformations will continue to enhance the user experiences and create new opportunities.

This quarter, we further deepened AI transformation of search, and we are seeing encouraging improvements in user metrics as we talked about in the prepared remarks. And we believe this will help drive revenue growth and open new monetization opportunities in the longer term. Also, we haven't started to monetize our search results at scale yet, which is around 22 % of the total queries. Once our AI-powered search features are sufficiently refined, we will move forward with monetization.

With higher quality user products, we are seeing opportunities for future growth. Based on these factors, we think that we are reaching the point of our advertisement businesses. We expect our app businesses to gradually improve with the first half of the year performing better than Q4 and the second half of the year showing further improvement over the first half. Thank you, Alex.

Operator

Thank you. Your next question comes from Gary Yu from Morgan Stanley. Please go ahead.

Gary Yu -- Morgan Stanley -- Analyst

Thank you for the opportunity. So, my question is regarding AI cloud. Given the strong growth of your AI cloud business this quarter, can we expect a continued positive trajection going into 2025 and both in terms of revenue and also profitability? What are the main drivers behind this momentum? And additionally, could management share your outlook on the AI cloud market in 2025? How should we approach enterprise cloud demand, and what incremental opportunities does AI bring to this segment? Thank you.

Dou Shen -- Executive Vice President and Head, AI Cloud

Thank you, Gary, for your question. So, as Robin mentioned, our AI cloud revenue growth accelerated to 26% year over year in Q4, contributing to a full-year revenue growth of 17% in '24. Notably, Gen-AI-related revenue nearly tripled year over year in 2024. So, this growth was fueled by rising demand for ERNIE and our AI infrastructure and scoring the market's strong recognition of our technological leadership.

So, as a result, we have successfully attracted a diverse customer base and built a robust pipeline of opportunities. So, for the AI infrastructure, we have established partnerships across diverse sectors, including Internet, automotive, smart devices, manufacturing, energy, finance, utilities, and lots of AI GC start-ups. Our customer base continues to grow healthily with strong gains among both top-tier and mid-tier customers, demonstrating our expanding market share in China's cloud market. On the MaaS platform, [Inaudible], it provides comprehensive foundation models with the industry-leading price and performance.

So, we offered the ERNIE family of models from flagship to lightweight variants designed to meet diverse needs. And besides ERNIE, we also provide access to a wide range of high-quality third-party foundation models, including DeepSeek's V3 and R1 models. Thanks to our full stack AI capabilities and end-to-end optimization, we can ensure optimal performance and stability for any model hosted on our platform, while maintaining highly competitive presence. So, in addition to that, [Inaudible] provides a full sight of tools for fine-tuning models and building AI-native apps so that our customers can easily work out solutions catering for their own specific needs.

So, in terms of the AI cloud market, we believe it will grow rapidly in 2024 -- 2025. On one hand, during the recent Chinese New Year holiday, large language models became a widely discussed topic. This not only further increased public awareness in foundation models, but also encourage more people to think simply about how to leverage them to improve their own businesses. On the other hand, we are confident that the performance of foundation models will continue to improve while the cost will steadily decline.

Surely, now, this will further lower the barrier to utilize these models. So, as a result, we believe more enterprises will integrate foundation models into every aspect of their business operations from R&D to production and to sales, which will drive explosive growth in API cost. Thereafter, we anticipate increased spending on Baidu AI Cloud as we have observed a clear trend in the past that the consumers who adopted foundation models through API costs on our [Inaudible] MaaS platform also tended to increase their spending on our AI cloud services. On the profit side, our non-GAAP operating margin continued to expand year over year in Q4 as we focused on high-value opportunities that align with our strategic priorities, which drives sustainable, robust growth.

So, looking into 2025, we are confident that our AI cloud revenue growth will maintain strong momentum while continuing to generate positive operating profit. Thank you, Gary.

Operator

Thank you. Your next question comes from Xiong Wei from UBS. Please go ahead.

Wei Xiong -- Analyst

Hi, management. Good evening. Thank you for taking my question. I want to ask about our margin trend.

So, given the near-term headwinds in core ads and the increasing mix of cloud revenue, how should we think about the core margin levels in the first quarter and also full year 2025? Is there any room to further optimize our operating efficiency? And also, are there any investment plans for the full year that we should be aware of? Thank you.

Junjie He -- Interim Chief Financial Officer

Thank you, Wei, for your question. This is Jackson. So, despite the near-term pressure, so we remain optimistic that our strategic AI investment will deliver more sustainable impact. So, we stay focused on improving resilience across all aspects of our business operations and see encouraging progress across our businesses.

First, for our online marketing business. We expect our advertising revenue to gradually improve, driven by our plan to monetize our AI-generated search results at scale and also our revenues to capture growth opportunities when the macro improves. Secondly, for our AI cloud business, it has already demonstrated robust growth with continued margin improvement, driven by our growing market presence and the competitive advantage in our unique self-developed four-layer AI architecture and full stack AI capabilities. So, we remain confident in maintaining this strong momentum going forward.

Third, for our entire driving business, particularly Robotaxi, Apollo Go, we remain focused on narrowing losses with better operational efficiency and improved UE. We are also exploring innovative operational models, including asset-light models. As you asked for the investments and optimization in 2025, we will maintain optimal resource allocation to high-growth opportunities while stay aligned with our long-term strategy. Our investments will continue to focus on initiatives that offer both tremendous future opportunities and strong ROI potential, including further advancing ERNIE's capabilities, deepening the AI transformation search, enhancing our AI cloud offerings, and expanding our autonomous driving initiatives.

As we move forward with these initiatives, we remain committed to enhancing operational efficiency and fostering synergies across business group to maximize the impact of these strategic investments. Thank you.

Operator

Thank you. Your next question comes from Miranda Zhuang from Bank of America Securities. Please go ahead.

Miranda Zhuang -- Bank of America Merrill Lynch -- Analyst

Good evening. Thanks, management, for taking my question. My question is about the Robotaxi business. Can management provide updates to the progress of the Robotaxi business heading into 2025? What other targets for fleet size unit economies and value proposition for our Robotaxi business? What do you think the industry is approaching? Are we reaching an inflection point, and what's your view about the competition dynamics as we are seeing that China's Robotaxi competitors leveraging ecosystem strategies with automakers and also ride-hailing platforms? So, wondering what is Baidu's approach to compete and scale? Thank you.

Robin Li -- Co-Founder and Chief Executive Officer

Yeah, this is Robin. We've invested in autonomous driving technology for over a decade. We have turned our big vision into reality. Through Apollo Go, we established ourselves as a leading global player in autonomous driving technology.

We operate in one of the most challenging environments globally. China's traffic conditions are among the world's most complex due to its massive population, diverse road conditions, dynamic traffic scenarios, and complicated urban layouts. The price for ride hailing in China it's about like one-seventh of that in the U.S. So, our successful operation here demonstrates our exceptional technological and operational capabilities.

We designed the RT6, our sixth-generation autonomous vehicle, from the ground up. It is the most cost-effective Robotaxi vehicle the world has ever seen. With this strength, we've successfully validated our business model and laid a strong foundation for further scaling and global expansion. In Q4, Apollo Go provided approximately 1.1 million rides nationwide.

That's a 36% increase year over year. And in January, the cumulative rides provided to the public exceeded 9 million. Furthermore, as I mentioned, we've achieved 100% fully driverless operations in China, which means that we don't have any safety drivers on the vehicles anymore. So, this is a new industry benchmark, and it reinforces our leadership in this area.

As I mentioned earlier, in November of last year, Apollo Go got the green light to start open road testing in Hong Kong. And this is an important step because Hong Kong is our first right-hand drive and right-hand traffic market. It shows our ability to adapt our autonomous driving technology to different traffic systems, opening doors for expansion into other markets with similar driving setups. This year will be a paramount year for our expansion, we expect to grow our fleet size and ride volumes faster than ever.

As we push forward, we're also actively seeking cooperation opportunities. We have identified a variety of potential partners, including mobility service providers, local taxi companies, third-party fleet operators and other potential partners. This asset-light approach will allow us to scale up efficiently while maintaining flexibility. By working with a diverse range of partners, we aim to strengthen our position in the market and bring autonomous driving to even more people.

Looking at the broader market. As I mentioned in our last earnings call, since the market is still in its infancy, competition actually helps accelerate market development and foster a more friendly pro-innovation regulatory environment. And in this growing market, our mission remains clear: delivering safer, more convenient, and more comfortable mobility experience for more users. Thank you.

Operator

Thank you. Your next question comes from Thomas Chong from Jefferies. Please go ahead.

Thomas Chong -- Analyst

Hi, good evening. Thanks, management, for taking my questions. Looking into 2025, what will be Baidu's strategic focus for business investments. Specifically, how will resources be allocated across search, autonomous driving, cloud services, and foundation models? Additionally, what is management's outlook on full-year 2025 capital expenditures as well as shareholders' returns.

Thank you.

Junjie He -- Interim Chief Financial Officer

Thank you, Thomas. This is Jackson. So, as we evaluate our capital allocation opportunities in 2025, so our approach are guided by the following strategic considerations. So, we will continue to invest in advancing our AI capabilities as a long-term strategic priority.

So, with that, we will further deepen the AI transformation across our product portfolio, particularly search. For AI cloud, we aim to drive the adoption of our AI infrastructure, our MaaS platform, [Inaudible], and the ERNIE among enterprise customers, meeting their growing demand for our AI products and solutions, as well as AI cloud services. In autonomous driving, we are focusing on scaling domestic operations, exploring innovative operational model, and expanding our international presence, while ensuring disciplined ROI management and effective loss control. For capex, two key priorities drive our decisions: to strengthening our technology leadership and accelerating the adoption of our AI products and AI cloud services across different sectors.

As you asked for shareholder returns, we repurchased over $1 billion since the beginning of 2024, which is significantly higher than our total buybacks in 2023. So, in total, we have repurchased $1.7 billion under our $5 billion purchase program, which runs through December 2025. So, going forward, we plan to accelerate our buyback program as part of our ongoing efforts to return value to our shareholders for their enduring trust and support. Thank you.

Operator

Ladies and gentlemen, that is all the time we have for questions. That does conclude our conference for today. [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

Lincoln Kong -- Goldman Sachs -- Analyst

Rong Luo -- Executive Vice President

Alex Yao -- Analyst

Gary Yu -- Morgan Stanley -- Analyst

Dou Shen -- Executive Vice President and Head, AI Cloud

Wei Xiong -- Analyst

Miranda Zhuang -- Bank of America Merrill Lynch -- Analyst

Thomas Chong -- Analyst

More BIDU analysis

All earnings call transcripts

This article is a transcript of this conference call produced for The Motley Fool. While we strive for our Foolish Best, there may be errors, omissions, or inaccuracies in this transcript. As with all our articles, The Motley Fool does not assume any responsibility for your use of this content, and we strongly encourage you to do your own research, including listening to the call yourself and reading the company's SEC filings. Please see our Terms and Conditions for additional details, including our Obligatory Capitalized Disclaimers of Liability.

The Motley Fool has positions in and recommends Baidu. The Motley Fool has a disclosure policy.

Baidu (BIDU) Q4 2024 Earnings Call Transcript was originally published by The Motley Fool