This conference call transcript was computer generated and almost certianly contains errors. This transcript is provided for information purposes only.EarningsCall, LLC makes no representation about the accuracy of the aforementioned transcript, and you are cautioned not to place undue reliance on the information provided by the transcript.
Operator
Hello and thank you for standing by for Baidu's second quarter 2021 earnings conference call. At this all participants are in the 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. Now I'd like to turn the meeting over to your host for today's conference, Ms. Joan Lin. Baidu's Director of Investor Relations.
Joan Lin
Hello, everyone, and welcome to Baidu's Second Quarter 2021 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 Lee, our co-founder and CEO, Herman Lee, our CFO, and Ado Shin, our EVP, in charge of session fee. After our prepared remarks, we will hold a Q&A session. Note that the discussion today will include forward-looking statements made under the safe harbor provisions of the U.S. Private Securities Delegation Reform Act of 1995. Forward-looking statements are subject to risks and uncertainties that may cause actual results to differ materially from our current expectations. For detailed discussions of these risks and uncertainties, please refer to our latest annual report and other documents filed with the ICC and Hong Kong Exchange. Baidu does not undertake any obligation to update any forward-looking statements except as required under applicable law. Our earnings press release and these calls, discussions of certain unaudited non-GAAP financial measures. Our press release contains a submission, 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 also be available by this IR website. I will now turn the call over to our CEO, Robin.
Robin Lee
Hello, everyone. We delivered another solid quarter, with FIDO core revenue growing 27% year-over-year, accelerated by non-advertising revenues, with AI Cloud growing 71% year-over-year in the second quarter. We are entering a new era where technology is becoming more powerful and we are benefiting from the deployment of technology to upgrade China's industrial competitiveness, digitize urban cities, and improve mobility. As China plans for the next stage of growth, it rests upon us to recognize the need to align our BIN strategy with the environment that we operate in and the future that we want to build with our stakeholders. ESG highlights a shared future for humanity to support China's pledge to achieve carbon neutrality by the year 2060. Baidu aims to reach carbon neutrality by 2030 through the use of green data centers, smart office buildings, and other means. In addition, we hope to improve urban traffic flow and lower carbon emissions for 100 metropolitan areas across China with our build-out of smart transportation and accelerate the switch to EVs for the mass through our investment in water taxi and smart EVs. Leveraging technology to serve a larger population to achieve inclusiveness is part of China's growth plan. In June, we introduced Apollo Moon, fifth generation of Apollo robot taxi vehicles that once again saw a 60% drop in cost per mile. With the rapid pace of autonomous driving technology advancement and cost efficiency improvement, Apollo robot taxi stands to become much cheaper than driver-for-hire ride-hailing in the coming years, which will make robot taxi ride-hailing much more accessible. Baidu Health is another example of expanding resources to a broader population for improved living. We are enabling closed-loop telehealth, online prescription, and home delivery to free up hospital resources for critical treatments. Concurrently, we are helping hospitals find patients with specific illness that match their treatment specialties by leveraging our AI-powered search and smart mini-program. We see great opportunities to use technology to advance sustainability and inclusiveness. As we go through our quarterly highlights, you will see other examples of how our AI-powered is naturally aligned with this goal. This is due in part to our belief in using technology to build a better world. Turning to Q2 operational highlights. Our AI cloud continues to see strong growth. IDC once again grants Baidu AI Cloud the number one AI cloud provider in China in their 2020 report on public cloud. Over the past decade, ASK has been the growth driver for China's cloud market, which sets up a strong foundation to implement AI solutions that solve industry-specific problems and provide urban living upgrades. By differentiating with AI solutions, we have the opportunity to provide repeat business with expanded applications and become a one-stop shop across cross-selling apps and other services. In the past, we discussed empowering financial services, utilities, and internet media sectors with AI. Let me give you an example of how our AI cloud empowers the auto sector. We are helping Geely, a leading automotive manufacturer, to develop a multi-year digitization plan which entails one, setting up their private cloud infrastructure, two, moving Geely's business onto cloud, three, leveraging Baidu AI solution to advance Geely's high-tech manufacturing capabilities such as IOV security and risk management, and four, enabling Geely to provide cloud services to their automotive suppliers and customers. Baidu AI is deployed into multifaceted environments. For example, Quanzhou, Fujian, a city of population of 8 million, implemented Baidu AI solution at its water treatment plant. Our smart monitoring capability was used to provide early protection of malfunctioned equipment, replacing labor-intensive patrolling, and a customized machine learning model was developed to predict water usage and dynamically adjust the water pressure of different water pipelines, cutting electricity needs by 8%. Our employees feel a sense of pride when natural resource is optimally managed and sustainability is indirectly advanced. In Healthcare Cloud, we have expanded into over 400 hospitals in 29 provinces and over 1,500 primary care clinics in 11 provinces with leading solutions like our CDSS resulting in rapid revenue growth at 245% year-over-year for the first half. According to IDC's June report, Baidu deep learning platform became the most widely used in China, surpassing Google's. Our PedoPedo developer community reached 3.6 million, up 62% year-on-year, and is adopted by 130,000 businesses. Baidu's large development community helps improve our machine learning models and individualize our AI tools and technologies. Moving on to ACE Smart Transportation. As of June, Apollo has signed with 20 cities to provide ACE Smart Transportation, up from five cities a year ago, based on a contract amount of RMB 10 million and above. Apollo also renewed agreements with nine cities to roll out their next phase of ACE Smart Transportation. Beyond urban roads, Apollo signed with Xiong'an Hebei, to deploy V2X Smart Highway, enabling the city to remotely identify the need for highway maintenance, assess road closure for poor weather conditions, and identify travelers who dodge tall terminals. Turning to autonomous driving, Apollo continues to see strong momentum. Level 4 testing on public roads has reached 12 million kilometers, or 7.5 million miles. Apollo has received 278 AD permits, reflecting AD piloting in dispersed geographic region and under wide-ranging test scenarios. Apollo Monetization. In Q2, Apollo GO ride-hailing expanded into Guangzhou, our fourth city of operation, and provided 47,000 rides to the public, up 200% sequentially. Customer satisfaction for ApolloGo has been high, with user rating of 4.9 out of 5.0. As pointed out earlier, our fifth generation robot taxi vehicles saw another 60% drop in cost per mile, compared to an average decrease of 62% in the first four generations of AD vehicles. We plan to roll out Apollo GO ride-hailing across 30 cities over the next two to three years. On AFD, Great Wall Motors become the latest automaker to announce plans for AFD. Apollo automated valet parking will be installed in their WeMotors SUV flagship. due out later this year. On infotainment, dual OS for auto has been installed in 1.8 million vehicles, up 265% year on year. 12 auto OEMs, the likes of Ford, GM, Toyota, and Hyundai have signed with dual OS for auto for installation in 17 May. on dual OS. Xiaodu continues to innovate the smart home device market. For example, the always-on dual OS smart assistant allows one to conveniently order fresh home delivery flowers and other commonly used goods in a matter of seconds through conversational AI. China's elderly population is slated to exceed 300 million. XiaoDu smart displays connect elders with immediate senior community assistance and serve as a virtual companionship around the clock through conversational AI, improving their quality of life while allowing family members to monitor their loved ones' health and safety from remote. During the June 18th e-commerce festival, Xiao Du Smart Display and Xiao Du Smart Pass for Education Market topped the sales volume list in their respective categories on leading e-commerce platforms like JD.com. Xiao Du is receiving high customer satisfaction, selling well on higher price point product lines, and topping the list on units sold, demonstrating how hardware can be differentiated with innovative AI. such as new input modalities and services. Turning to mobile ecosystem, in June, Baidu app MAUs reached 580 million, up 9% year-on-year, and daily login users reached 77%. As an open in-app search app, Baidu app offers users instant replies on various topics. For example, daily telehealth consultation on vital health surpassed 2 million in second quarter, up 47% year-on-year. In addition, more than 8,000 industry experts spanning 14 other verticals, such as legal, finance, and fashion, are able to instantaneously reply to users' search inquiries through our AI building blocks. When users seek specific information of service, Baidu Search has a decent advantage. Managed Page now accounts for 40% of Baidu Core advertising, reflecting merchants' opening storefronts on Baidu, as opposed to operating their own sites. For example, a local moving company, Qianxi Moving, changed their search landing page to its managed page and saw its daily orders doubled five weeks after adoption. Through managed page, users are able to browse its services, interact with the merchant, schedule a move, and pay a deposit all seamlessly on Baidu. Managed page also monitors merchant's activities on Baidu with AI for added personal safety For example, a moving service entails strangers coming to your house. Before I turn the call to Herman, I'd like to congratulate him for his new role as Chief Strategy Officer of Baidu. Herman will continue to act as CFO until we find a qualified replacement. He will spend more time on corporate strategy and development afterwards. Over the past four years, Permanent has helped Baidu transform from a mostly search company to a leading diversified AI company. This demonstrated the business leadership, a holistic view of our business, and a keen sense of the capital market. I'm confident in Permanent's ability to help chart a course for our long-term growth and success in the age of AI. With that, let me turn the call over to Herman to go through our financial highlights. Thank you, Robin. Hello, everyone. Welcome to Baidu's second quarter 2021 call. All monetary amounts used in my discussion are in renminbi unless stated otherwise. Baidu's second quarter revenue was $31.9 billion U.S. dollars, up 20% year-over-year, driven by the growth of Baidu core revenues. which reached $24 billion, or $3.7 billion, of 27% year-over-year. Non-advertising for Baidu Core reached $5 billion, or 21% of Baidu Core's revenue. AI Cloud revenue was $3.3 billion, or up 71% year-over-year, which is an acceleration from the 55% growth last quarter. Our cloud growth is benefiting from the demand of enterprise customers looking to use AI to transform their industry and strengthen their leadership position. For example, customers from the internet, media, from financial services, from energy, and from manufacturing sectors. We're also benefiting from the urban cities seeking to adopt Apollo 8's smart transportation to modernize and digitize their transportation network. AI solutions and smart transportation are showing fast growth, serving as catalyst for our cloud to grow faster than the overall cloud market, with us being a smaller part of our cloud business. On intelligent driving and other growth initiatives, we continue to make advances in product development by leveraging Baidu AI capabilities such as speech recognition, NLP, and computer vision, and our strong internet foundation, Xiaoduo has become the leader in smart display globally. based on shipment. Xiaodu speakers entered the smart home device market in 2018, competing on sub-100 product lines. Home devices equipped with AI features can be sold at much higher price points. For example, in May, we introduced Xiaodu TianTian T10, a 10-inch smart display with karaoke features, at an MSRP of $16.99. Through our innovation, Xiaoduo services revenue, such as advertising and membership, has grown five-fold from last year, now accounting for over one-tenth of Xiaoduo's revenues. Xiaoduo, powered by DuoOS, is transforming the smart speaker market from selling hardware to selling AI-powered features and services. Operating system for the automotive industry is a much larger market. Apollo has an opportunity to leverage its leadership in autonomous driving and infotainment operating system to enable every passenger vehicle to be a smart vehicle. Just as Shoudu transformed the home device market, Apollo hopes to leverage these capabilities, ASD, and existing OEM channels to boost computing intelligence and transform the automated industry. Moving to online marketing. Q2 online marketing revenue was $19 billion, up 18% year-over-year. In-app advertising was solid, growing 26% year-over-year, partially offset by the slow growth of union and PC ad revenues. Most of our top ad verticals continue to perform well. CPM saw double-digit growth, especially for in-app online marketing. IT revenue was $7.6 billion, up 3% year-over-year. IT subscribers reached 106 million in June, which supports the large in-house production of entertainment blockbusters. Cost of revenue was $15.9 billion, up 21% year-over-year, primarily resulting from an increase in tax and cost of sales associated with new AI business. Operating expenses were $12 billion, up 30% year-over-year. SG&A for Baidu Core was up 40%, primarily due to an increase in channel spending and promotional activities, as well as an increase in sales hiring. particularly as we look out for the next 6 to 12 months to grow our cloud and also our intelligent driving business. Non-GAAP operating income for Baidu Core was $6.5 billion, or $1 billion. And Non-GAAP operating margin for Baidu Core was 27%. Adjusted EBITDA for Baidu Core was $8 billion, or $1.2 billion, and adjusted EBITDA margin for Baidu Core was 33%. Cash and short-term investments for Baidu Core as of June 30, 2021, was $167.7 billion, or $26 billion U.S. Free cash flow for Baidu, excluding IHU, was $6.9 billion, or $1.1 billion U.S. Baidu Core had approximately 36,000 full-time employees as of June 30th, up 24% from last year. Turning to Q3 guidance. For the third quarter, Baidu expects revenues to be between $30.6 billion and $33.5 billion, representing a growth rate of 8% to 19% year-over-year, which assumes that Baidu Core revenue will grow between 9% year-over-year and 20% year-over-year. The above forecast takes into consideration the current COVID-19 situation in China, which is evolving, and business visibility is limited. The above forecast reflects our current and preliminary view, which is subject to substantial uncertainty. Before I turn the call back to the operator, let me summarize our second quarter results. Baidu mobile ecosystem continues to be strong. Baidu core online marketing revenue was up 18% year-over-year in the second quarter, with in-app marketing revenue growing 26% year-over-year. Baidu app MAUs reached 580 million, up 9% year-over-year, and daily user login was 77%. Four years ago, we committed to strengthening our mobile ecosystem and lead with AI. Our search and feed is stronger now, with AI building blocks and marketing cloud, and non-marketing has become sizable and growing quite fast. We have delivered on our goals and we are very proud of our team for their strong execution. Many investors have inquired about the recent regulatory landscape. Governments generally favor fair competition and maybe even so when mobile internet is well penetrated. We see this theme play across Europe and in the US.
Healthcare Cloud
Search fundamentally is open.
Robin Lee
As exclusivity is forbidden for the largest players, this will allow more players in each internet sector to exist, which gives consumers more choices and makes search more useful. Some investors have asked about government incentives and special tax treatment. It's important that we elevate this to see the overall direction of China's development plan. Consumer internet has been in China for two decades. And the country is promoting new growth sectors such as industrial internet, V2X autonomous driving, and the modernization of city governments. Thus, it would not be surprising if certain incentives for the older industries gradually decrease while the new economy benefits from government incentives. We believe government policy will also be adjusted to support these new growth areas. Our new AI business is well positioned in sectors of the new economy such as AI cloud, smart transportation, autonomous driving, smart devices, and AI chips. In the second quarter, I'm sorry, Midea returned $566 million to our shareholders under the 2020 share repurchase program this year, bringing the cumulative repurchase from last year to $2.5 billion U.S. dollars. We are excited about Baidu's future, excited about Baidu's durable search and feed business, and our new AI business to support China's growth while advancing sustainability and exclusiveness. On August 18th, starting at 9 a.m., 9.30 a.m. Beijing time, we will be hosting Baidu World on CCTV. Please join us to follow the latest on our product development. And lastly, It's truly been a privilege to serve as Baidu's CFO and witness our growth initiatives, like AI cloud solutions, smart transportation, self-driving solutions, and smart devices, develop into a fifth of Baidu core business, using the steady profitability of a search entity to fund and accelerate Baidu's growth. I'm excited about my new role as CSO, which will allow me to spend more time thinking about technology opportunities. Operated with that, it's now open to call for two questions.
Operator
Thank you. Ladies and gentlemen, we will now begin the question and answer session. If you wish to ask a question, please press star 1 on your telephone and white for your name to be announced. If you wish to cancel your request, please press the pound or hash key. Participants are requested to restrict to one question at a time. If you have follow-up questions, please request to rejoin. Once again, that's dial 1 for questions. Our first question comes from the line of Alicia Young from CT Group. Please ask your question.
Alicia Young
Hi. Thank you. Good evening, Robin, Herman, and Dren. Thanks for taking my questions. Congratulations, Herman, for your new role as the CFO. To follow up your kind of like brief remarks about the regulation, I actually have a question on this data security. What could be the impact on Baidu, anything that we need to do or the changes that we need to do to ensure in compliance by September 1st for this data security law? And then very quickly, as we just finished the summer Olympics and with the Winter Olympics coming up. Can you provide some update in terms of the latest robot taxi or even the robot bus commercialization service? Any expectations on this right healing demand in this upcoming winter game? Do you think Baidu's driverless car service could leverage this big event to draw more commercial usage and enhance the branding? Thank you.
Robin Lee
Hi Alex. Data security is very important to us and we have consistently improved on the management of our internal data security system over the past few years. We have a data privacy committee among our management team that oversees the practices of the whole company. We are also active in assisting with the setting of industry standards, and we share our experiences and dialogues with others to ensure that we are on top of industry best practices. If you look at MSCI, by just reading on data privacy and data protection, it's among the highest compared to our peers. So I'm quite confident we will be able to cope with the new regulatory environment well in terms of data security and use of privacy. On Olympics side, yes, we do plan to do something with our robot taxi during the Winter Olympics. But what's important is that we just launched this fourth study of robot taxi operation which is Guangzhou, I think we will continue to roll out our robust taxi services in many different cities, different regions. And the number of rides that we are providing have been growing very quickly, as I mentioned during the prepared remark. We delivered 47,000 rides for our customers. that's 200% growth over the first quarter of this year. So we believe that the scale will grow very quickly and we will be able to learn all kinds of corner cases and quickly improve our technology. I'm quite optimistic that Robotech will commercially be available in more than 30 cities in two to three years. Yeah, and I just wanted to add a point from what Robin just said. You asked specifically about Winter Olympics. If you look at today in Sogang Park in Beijing, we believe we're the only second autonomous driving car company in the world where you actually have a car driving without someone at the driver's seat. So I think that shows the number of years experience that we've been doing in autonomous driving. It also shows the fact that we've been doing operation for autonomous driving for a while. That's why, you know, we can get the permit and actually start testing that. So, you know, it's open to the public. You guys are free to come and look at how autonomous driving is being done without someone at the driver's seat.
Operator
All right, thank you. Our next question comes from the line of Piyush Mubai from Goldman Sachs. Please go ahead.
Piyush Mubai
Thank you, Robin Herman. When I look at the way your ride-hailing business is expanding, I can't help but think that you'll soon be running a business without the cost of a driver to factor in. If you could just take us through the math as you see it evolve to getting to 20 cities, at what point in time does it get to be a revenue line that you'll spell out separately and talk about the profitability of it all? Thanks.
Robin Lee
Yeah, we continue to work very hard to drive down the cost of the rubber taxi vehicles. It has just gone down 60% for our fifth generation vehicle. In the meantime, we are running more testing miles and taking more orders to learn from all kinds of corner cases and quickly improve our technology. Based on our The current projection, I think by the year of 2025, we will cross the line, which means that the total cost of robot taxi or ride-hailing will be lower than manned vehicle ride-hailing. And after that, I think the scale will be able to grow much larger than than it is today, and I think around that time, we should be able to report in a separate line. Yeah, and just to add to that, Apuge, if you look at the economics, Robin talked about how from fourth generation to fifth generation, we reduced by 60% in cost per mile. If you look at the prior four generations, we average around 62%. So we've been consistently decreasing costs for the last five generations and at around 60% each time. And if you look at what Robin just said, you know, back to 2025, what you're seeing in the economics is that in a ride-hailing with actual drivers, look, the cost of a labor cost only goes up. It doesn't go down over time. But you're competing with technologies. We're competing with the fact that the more miles that we have, the more data that we have, our operational experience, that this thing will continue to go down. So I think that's the effectiveness of this business model. If you look beyond 2025, if you look at the next five years, what we have internally, we can drive that cost down a lot, lot more based on what we have seen in the first five generations. So this thing will be very interesting once you pass 2025. You have technology. efficiency technology operational improvement compared to labor costs.
Operator
Thank you. And next question comes from Alex Yeo from JPMorgan. Please ask a question.
Healthcare Cloud
I have a follow-up question to Piyush's question. So in 2025 the CPM of your Apollo solution could be cheaper than human driver. My follow up question is when do you expect the safety ratio for your Apollo solution to be safer than the human driver and then for these two condition to materialize? What does it take? Is it the cumulative mileage to reach a certain level or is it something else? Thank you.
Robin Lee
Yeah, that's a good question. We take safety very seriously. I think we've been testing for like more than 10 million kilometers. We haven't had a serious injury yet. So that demonstrates that once we decide to operate a certain route for our robot taxi by saving services, we have the confidence that the service has a higher safety level than human drivers. That would be a minimum bar for us to roll out services without human drivers. But my expectation is really that we should be 10 times more safer than the human drive. So that kind of benefit is quite clear and it's absolutely assured.
spk02
What's the second part of the question?
Healthcare Cloud
The second part is what does it take for these two conditions to play out, i.e.,
Jerry Liu
Yeah, that's right.
Healthcare Cloud
It would be cheaper, yeah.
Robin Lee
Right. There are 500 million, there are 5 million kilometers of roads in China, and the road conditions vary greatly. The approach we're taking is a very gradual one. We carefully choose roads that we have a high confidence that can be operated without a human driver. That's why you're saying that we're pacing ourselves, we're carefully picking areas and routes so that we have a high confidence that we can operate this robot taxi.
Operator
All right, thank you. Our next question comes from Gary Yu from Morgan Stanley. Please ask your questions.
Gary Yu
Thank you, management, for the opportunity to ask questions. I actually have two questions, both of which are follow-up of the previous questions. First is on regulation. I appreciate the management comment about data piracy. How about on data security issue? Have we heard anything from kind of regulator or government level where the officials are comfortable for a private company, tech company like us, to have kind of full control of these sensitive road data traffic on our system? Or is there any discussion on, you know, potential formation of state-owned joint venture company having full access of these data? Is that fully on the hands of private company? So that's our first question related to data security. My second question is also related to RoboTaxi. So we talk about the cost, you know, opportunity. How about, you know, revenue? You know, after 2025 or even longer term, you know, how should we project in terms of penetration, you know, mobility penetration or share mobility penetration that we think will be kind of served by, you know, driverless cars? The reason I'm asking this question is seems like that currently, driverless car is still limited to certain usage scenario with specific locations. So how should we look at kind of penetration going forward? Thank you.
Robin Lee
Yeah, on the regulation with the data security, we maintain a constant dialogue with the regulators. And as I mentioned, sometimes we participate in the setting of industry standards. So I would say that we have a very high standard and we haven't heard anything that's very abrupt or adverse that's gonna have adverse impact on our business operation yet. On the penetration of Robotaxi, Right, like I mentioned, we're taking a gradual approach. There are certain roads that are suitable for robotaxi. There are certain roads that are not suitable to robotaxi. We are carefully selecting what road to operate on. But if you look at the overall market, I think right-hitting has like probably 50 million orders per day. If the cost can become one-fifth of the current cost, I think that the number of orders could easily double. So it's a huge market. It just depends on how fast we can improve our technology and scale up our services.
Operator
All right, thank you. Next question comes from Jerry Liu from UBS. Please, that's a question.
Jerry Liu
Hi, thanks for taking my question. Yeah, maybe let me ask a question on the advertising business. Mainly, if I look at the third quarter guidance, we talk about some of the, you know, COVID-19 resurgence potential impact. I just want to ask, first of all, is this something we're already seeing or is it more preemptive? And then secondarily, we're also seeing some potential COVID or regulatory impact across different sectors in the whole industry. So I wanted to get a sense on the strengths and weaknesses of the different verticals at the moment. Are we seeing further impact, for example, online education, e-commerce, or other verticals. Thank you.
Robin Lee
Okay. Thanks for the question, Jerry.
Healthcare Cloud
You're right. So we already see the impact of COVID-19 for the Q3. So especially for tribal.
Robin Lee
We see very good dose trend in the early few weeks for Q3. But certainly, you know, the COVID-19 just stopped that trend, actually. It stopped pretty quickly. On the other hand, you know, we do see pretty good progress on the media in entertainment, vocational education, and those verticals. So talking about the regulation, I mean, so far we see pretty clear variations on the K-12 education and some other verticals. The good thing for us is that the K-12 education itself is a pretty big vertical force, but K-12 takes only up more partition in our evidence. And also for the K-12 innovation, we are pretty optimistic about it because we see two potential good things. Number one is that we are expecting more such colleagues along the K-12 because students need to work on themselves. And second one, as far as I know, some of the resources are moving to vocational education, which is a very good vehicle. We are pretty strong at. That's it.
Operator
Thank you. Our next question comes from Eddie Lowe from Bank of America, Maryland. Please have the question.
spk02
Good evening. I would like to have a question on your cloud business. It seems like a strong growth this couple of quarters, potentially driven by new projects. So just wondering, could you give us some rough timeline that when you could get into the development stage that the growth of your cloud business would be more driven by your existing clients upgrading the services and then buying more, kind of like more from the usage of existing clients rather than opening up a new project? Thank you.
Robin Lee
Yeah, as you know that our cloud is still growing at a very high rate, high double-digit growth rate. So by nature, you're right that there are a lot of new projects from new customers. And what I can say is that our customer retention has been very good, and it's also improving. As we expand our customer base, more of them will stay with us and give us new businesses. If we want to maintain a very high growth rate, it's inevitable that we will have a large percentage of so-called new projects. So at this stage, I wouldn't manage the percentage of new customers or existing customers. try to grasp the opportunities that are presented to us as AI-driven solutions or cloud solutions. That is our strength to try to do as much as we can. Let me add to that, Eddie. Robin summarizes pretty well. When you're growing at 71% year-over-year, clearly you cannot be relying on one driver. What you're seeing actually is two drivers. Number one is you are seeing us getting repeat solutions. We talked about, for example, smart transportation. Look, we're in 20 cities. If you look at contracts over 10 million, nine of them have renewed. If you look at, for example, we talked about the GLEAP. We didn't just, our first phase was building that private cloud for them and then we talked about a three year plan and then we talked about the other things that we're gonna do and Robin talked about those four steps. So clearly that's not a one solution, one time. That is a repeat solution. Join us for a call for Baidu World. We're actually gonna go through AI solutions for many different industries and you'll see what I'm talking about. Basically our model is We've served them with the AI path. That's the AI engine. And if they want to do different things, they would then come back for different phases to add applications. So those you can see, it's because that the original engine is on Baidu Cloud, the applications that would come back to us to add on that. Very similar to then with smart transportation, you have V2X, and then you can expand the different applications of V2X. So you can see our business model enables us to do repeat businesses. At the same time, if we don't expand to different industries, if we don't add different customers, we cannot keep up with the kind of growth that we have. Look, you look at Q2, our annualized revenue based on Q2 is $2 billion U.S. and growing at 70%. So you have a very exciting business in order to contain that robust growth. You really have to accelerate from products, from different industries and adding different customers.
Operator
Thank you. Next question comes from James Lee from Mizuho. Please go ahead.
Robin Lee
Great. Thanks for taking my question. Just want to follow up Eddie's question on cloud. Maybe can you guys talk about the improvements that you guys are making to make your offering more scalable? I think last quarter you talked about developing a standardized solution layer that will enable third-party developers and integrators to customize for their clients. Can you give us an update on that? And also, secondly, the adoption for machine learning and AI for cloud in the U.S. is much smaller, much slower in the U.S. Can you help us understand why we're seeing a faster adoption cycle in China? Thanks. Right. On the cloud solution. As we do more projects and we serve more customers, we are able to more or less standardize our solutions. In the beginning, we can only take more of system integrator role to get all the deals, but these days, we are able to provide in a lot of cases standard packages or parts to the other people who are system integrators who get a large part of those total solution projects. That demonstrates that our capability to continue to improve the the standardization of R. And on the adoption of machine learning, I think the difference between U.S. and China is that U.S. tend to separate the product or solution on a more horizontal fashion, mainly that they have at each layer you have large players that serve all kinds of different industries. But in China, you can easily see more of those vertically integrated solution providers. For example, for Baidu, we provide smart transportation solutions for a lot of cities. That sometimes includes hardware, software, you know, cloud, and on-premise, all kinds of things. But that doesn't mean it's not standardizable. It's actually quite similar from city to city and from customer to customer. And by vertically integrated or by providing a solution, end-to-end solution, we can actually innovate a lot more to really improve the efficiency and serve the customer better. Therefore, the customers are better served. And when you can vertically integrate a lot of things, AI or machine learning naturally plays a much more important role. And let me just give more granularity of what Robin just said. When you look at, for example, smart transportation, we started with V2X, right? You're trying to synchronize the traffic light. We first did the project back in Bao Ding. So obviously, we had the product, but you're going into a city, you're trying to understand their systems and so forth, and a lot of times you've got to then connect to the system, their legacy system, in order to make it work. But as you're going to different cities in China, once you have certain tools that are set up to deal with the same legacy system and so forth, you get, number one, that you perceive from the operational experience, and secondly, a lot of tools that we've built up. So when you think about us doing V2X specifically to synchronize our traffic lights because we've been doing this for several years, you're actually seeing a very high scalability. You're actually seeing better margins and so forth because we've been doing this. So now if you see us start expanding into different scenarios, for example, into highways, into mass and so forth, and what I would predict based on our past experience is that these margins will improve as we find more familiar environments. Same thing as we're going into enterprises. We talked about automated call centers. The first one we did with Unicom very profitable. And then they start expanding into different call centers throughout the country. So all of a sudden, we're familiar with the backend system. You just have to buy another, you know, a service, you know, put the software in there. All of a sudden, you're up and running because you've already trained the machine learning model for a specific client. And then once you start expanding to different industries, you get one or two leading customers. Again, you're now repeating a lot of things that you're doing. All of a sudden, you get that inertia. So I think AI solution is very similar to, you know, the enterprise solution that you have, you know, for softwares and so forth in the U.S. It takes time whenever you're getting into a new industry, into new customers. But once that environment connecting to the legacy system is very familiar, you get that inertia in the pages after that.
Operator
Great. Thank you. Our next question comes from Natalie Rue from Hytale International. Please go ahead.
Alicia Young
Hi, good evening. Thanks for taking my question. We noticed recently from a press release that you are expanding your campus recruitment this year, scaling up to nearly 20% of the total headcount. Just wondering if management can share with us the rationale behind that extensive expansion at this point, and how should we see the margin chance of by-date call ahead? Also, if we simply look at the marketing-related business in Baidu Corp, just wondering what's the margin profile of that segment alone? Thank you.
Robin Lee
Yeah, the headcount growth, if you look at our growth rate of Smart Cloud, it's very high. And this kind of... industry solutions require a lot of headcount addition. If you compare our cloud team with other industry peers, I think that the headcount for Baidu is not that high. It's actually quite reasonable and we do need a lot more people to come up to deliver our services. Yeah. And then on the margins for by the core, if we're looking at a non-GAAP basis, you look at Q3, my estimate right now would be we're probably gonna increase cost of sales plus OPEX, maybe a billion to a billion and a half on a sequential basis. So that's how I would look at this. And you guys have the last few quarters, I would look at the non-GAAP for by the core And I think you guys have a very clear track.
Operator
Thank you. Our next question comes from Tian He from TH Capital. Please go ahead.
spk04
Hi Robin, Herman. The question is related to your AI cloud. So there are normal clouds and there are AI clouds. So I wonder if we measure this to two type of clouds. So from a financial point of view, can I say AI cloud is much more higher margin? So can I treat that as a difference? That's number one. What kind of clients need AI clouds? What kind of clients need regular clouds? So that's the first one question. The second one is managed page has already increased to a bigger portion of the online marketing. So what's the outlook for this managed page going forward? Thank you.
Robin Lee
Yeah, the margin for AI Cloud is higher, significantly higher than the more general cloud, especially app. I think you all know that app and CDN generally has a lower margin. But I would say clients need AI cloud and general cloud. They need all kinds of technology and solutions. It's just our strength is that we We are very advanced in AI, and we find that by integrating our AI technology, AI solution to really solve our customers' problems, we are able to do a much better job. We do provide channel cloud services, but I think the market is very large. It's large enough and flowing very quickly, and we are gaining much share in this area. attracted space. And let me just add, you asked about who needs AI. If you look at our customer set, you know, we see several specific, one is, for example, internet media, right? When you have a media, for example, some of the media companies, it's very data heavy. When you look at, for example, our industrial sectors, you know, different areas of manufacturing, when you look at utilities, energy, and so forth, they all have a lot of data sets. You look at, for example, smart transportation, the government has a lot of data sets. So our AI solution basically says, if you are a customer, you have a lot of data sets, you need to improve operational efficiency and so forth. Our machine learning models, when you go in there, are by your brain. We can help you. A lot of things that you have to do manually before, we can automate that. We can do it much faster with AI computing. So companies or industries that have a lot of data sets, we can automate. For example, you know, you look at financial services, rather than computing and proving every loan, you have examples of a few million customers, we can, you know, automate that whole computation where you just ask by do-blame rather than having someone manually compute all the different ratios and all the numbers on that particular customer. So you can see that usage of very, helpful, especially in China's economy. Yeah, Tian, I'll take the second one. You're right. So the Minds page now accounts for like 40% of the Baidu core advertising. So actually, Minds page, it is not alone. We treat it as part of the marketing call that we have been discussing a few times, including the Minds page and also the tools we provide to the merchants to make it easy for them to run campaigns on Baidu. So as we always say, as we always say, so managed page together with by Yahoo and the small middle programs, they all together, you know, all the building blocks for the mobile ecosystem. Actually with that, because we can provide more close to loop experience in which increase the thickness or the engagement between the user and the by app. So as a result, we can see So the conversion rate with the building blocks are much higher than before. That's why we can see this COVID-19 outbreak in Q1 last year. So our online marketing business actually rebounded consistently in the last five quarters, especially for the in-house market because of the three building blocks. And it grows as much as 26 yearly here. So that side, we'll keep working on the main stages because it will improve the user experience and also the monetization capability.
Operator
All right. Thank you very much. So ladies and gentlemen, we have reached the end of the question and answer session. So with that, we conclude our conference for today. Thank you very much for participating. You may all disconnect.
Disclaimer