Sharon Ng - Director of IR Xinzhe Li - CFO Yanhong Li - Co-Founder, Chairman and CEO Qi Lu - Vice Chairman, Group President and COO

Alan Hellawell - Deutsche Bank Alex Yao - JP Morgan Chase & Co Alicia Yap - Citigroup Inc Chi Tsang - HSBC Eddie Leung - BofA Merrill Lynch Grace Chen - Morgan Stanley Juan Lin - 86Research Limited Karen Chan - Jefferies LLC Ming Xu - UBS Investment Bank Piyush Mubayi - Goldman Sachs Wendy Huang - Macquarie Research Xiaoguang Zhao - Barclays PLC Yue Wu - China International Capital Corporation Limited Zoe Zhao - Credit Suisse

Hello, and thank you for standing by for Baidu's Second Quarter 2017 Earnings Conference Call. (Operator Instructions) 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, Sharon Ng, Baidu's Director of Investor Relations. Thank you. Please go ahead. Sharon Ng^ Hello, everyone, and welcome to Baidu's Second Quarter 2017 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. Today, you will hear from Robin Li, Baidu's Chief Executive Officer; Qi Lu, Baidu's Chief Operating Officer; and Jennifer Li, Baidu's Chief Financial Officer. After their prepared remarks, Robin, Qi and Jennifer will answer your questions. Before we continue, 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. Potential risks and uncertainties include, but are not limited to, those outlined in our public filings with the SEC, including our annual report on Form 20-F. 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 also be available on Baidu's IR website. I will now turn the call over to Baidu's CEO, Robin Li.

Hello, everybody, and thank you for joining today's call. We delivered solid financial performance and made some significant announcements and developments in the second quarter. The arrival of the AI era creates new exciting opportunities for Baidu. With the work that we're doing with our core business in mobile and the leading AI capabilities that we are applying to autonomous driving, DuerOS, AI cloud and financial services, we have a broader mission than we set out to achieve 17 years ago, when the company was founded. In May, we announced a new mission statement to make a complex world simpler through technology. In order to achieve our mission, we will execute on 2 strategic pillars: to strengthen our mobile foundation; and to lead in AI. As we reflect on the first half of 2017, we are very pleased with our overall progress. This quarter, in strengthening our mobile foundation, we made significant improvement in our flagship Mobile Baidu app with version 9.0, which makes the overall Baidu Search experience more mobile native-app-like and is the best way to experience the dual search products of query based traditional search and feed, which we think of as "query-less search." We've seen strong user adoption for Mobile Baidu, with DAU growing the fastest among the top 10 apps in China from March to June, according to QuestMobile. Feed user adoption has been very strong. Feed daily active users surpassed 100 million this month and the key metrics of volume of content distributed and monthly time spent grew rapidly, reaching new highs. Through voice search, we make it more natural for users to express queries. In search, the large majority of search landing pages now deliver a native-app-like experience, having more than doubled in proportion from the beginning of the year. We also dramatically increased the coverage of richer content such as video. Overall, we are laser-focused on user experience. Our mission in the AI age is very clear, we are open and we are enabling. We attract the top human talent to our platform and forge ahead in the AI age. We held our first AI developer conference earlier this month and shared our principles of open platform and win-win ecosystem with our partners and developers. We've made significant progress in autonomous driving by officially launching project Apollo, the first open platform for autonomous driving, attracting over 50 strong partners. In DuerOS, we continue to lead the industry in technology and with our large and growing ecosystem. In ABC cloud, we are seeing strong momentum in customer additions and revenue, and Baidu FSG, our FinTech platform, is healthy and gaining industry recognition, having signed a strategic alliance with the Agricultural Bank of China. Each initiative is available to us because of our AI leadership and each represents substantial opportunity and has the potential to become a sizable core business for Baidu. I'll now turn the call over to Qi to go through our business progress.

Thank you, Robin. Let me now walk you through our operating highlights and some additional details. On the current core business front, we are making sustained progress in our product, in particular Mobile Baidu, which is the mobile app that represents our flagship product, with the twin-engine user experience of search and feed. To us, feed is query-less search and it is a great way to extend our offering to users on mobile devices. There are strong synergies between search and feed to drive further experience improvement, usage and engagement. Earlier this month, we launched Mobile Baidu version 9.0, a major update that delivers a much improved user experience across a number of key areas, such as faster load speed, smoother navigation, a cleaner interface and a more mobile native app-like experience for consuming content such as video, literature and music. At the same time, we drove strong user adoption of Mobile Baidu in the second quarter. Mobile Baidu DAU grew the fastest among the top 10 mobile apps in China from March to June according to QuestMobile. On search, we continue to make good progress in improving our overall search experience by making the experience much more mobile-native. For example, in Mobile Baidu, we shipped a new search UI and a new navigation framework which substantially increased the percentage of search result landing pages with a native app-like experience as a portion of the total search landing pages to 85% this month, from 40% at the beginning of the year, through intelligent filtering and advanced presentation technologies. Overall, we're focused on using AI as the fundamental driver to deliver sustained improvement of search, enabling a better and a more natural search experience via voice and images as inputs, and increasing the coverage of richer content such as video. On feed, our product quality and user experience improves at a rapid pace, particularly as we leverage our strengths in AI technologies and our search data assets. Such improvements underpin the strong growth of overall daily users. Our volume of aggregated content consumption and distribution as well as user time spent, are reaching historical highs this month. Our feed DAU surpassed 100 million this month, up from 83 million DAU in April, outpacing our industry peers. On the monetization front, because our search plus feed twin-engine and the usage growth in feed, our value propositions to advertisers becomes richer, more diversified and overall, more compelling for both performance and brand advertising. As a result, our customer base is growing at a healthy pace. Over the quarter, we added nearly 20,000 online active customers sequentially across a range of industries, including education, auto, logistics and the local services. In particular, we continue to receive good feedback from our advertisers, especially those who benefit from the comprehensive combination of direct response search ads and branded feed ads and other branded ad formats. This is because search and feed ads satisfy distinct advertising needs, and we see feed monetization as fundamentally incremental to search ad-spend of our core search product. Advertisers recognize the value of our platform and we have seen strong momentum in feed monetization, with feed revenue doubling between March and June. IQiyi plays an increasingly important role in strengthening our overall mobile foundation. Its rich content and strong user base are key pieces to Baidu's overall ecosystem. We have made a lot of progress in establishing operational synergies in data and content across iQiyi and other Baidu products to spur long-term growth. In Q2, iQiyi continued its excellent performance and leads the online video industry across multiple key metrics. In June, according to iResearch, iQiyi's PC and mobile app daily active users reached 75 million and 146 million, respectively, reaching new highs. Monthly time spent on PC and mobile continue to grow, with users spending 124 billion minutes on PC and 383 billion minutes on the mobile app, second only to time spent on Tencent's Weixin app. Next, I'll talk about our progress on the second strategic pillar, leading in the AI era. Our focus is to accelerate the commercialization of AI technologies. Our best strategic path forward is to platformize our AI technologies, build vibrant ecosystems based on healthy business models and sustained win-win with developers and partners. We held a very successful AI developer conference on July 5, where we clearly articulated our AI platform and ecosystem strategy blueprint. The foundation of our AI platform is Baidu Brain plus Baidu ABC Cloud, where Baidu Brain offers the most complete and leading AI technology suite to developers, coupled with Baidu ABC's growing set of AI-based industry solutions. With nearly 5,000 developers and partners in attendance and wide press coverage, our AI platform and strategy have been positively received by the broad developer and partner community. The strong support from partners and developers is key to cementing Baidu's industry-leading position in AI. At the conference, we officially launched Apollo 1.0 and unveiled Apollo's detailed technology road map. Apollo is the Android for the auto industry, more open and even more powerful. It enables our partners to go from 0 to 1, to quickly assemble their own autonomous vehicles and start their product R&D. The Apollo platform and its ecosystem is architected with a powerful set of win-win principles, with open capabilities and increasingly shared resources such as precious data resources. Our partners can have access to Apollo's growing powerful capabilities. They can also contribute to Apollo. The key is the more a partner contributes, the even more outsized benefit the partner will receive from the Apollo ecosystem. As a result, the pace of innovation will continue to accelerate. We have seen the industry buy in to our approach. Within the short span of just a few months, we have assembled the world's most powerful autonomous vehicle ecosystem, with over 50 strong partners, including 13 OEMs, 2 of them are world-leading big brands in Daimler and Ford; and world-leading Tier 1 provider such as Bosch and Continental, as well as leading AI technology providers such as NVIDIA, Intel and Microsoft. Currently, project Apollo is #1 on Github in monthly ranking. At the same time, Apollo creates compelling long-term economic opportunities for Baidu, as we will focus our business model on providing a series of high-value services, including HD maps, simulation engine, security as well as the connected car experience. Also at the conference, we showcased the strong progress we are making with DuerOS. It is the natural language-based, human-computer interaction platform of the AI era. It is embeddable to every single device on earth and enables them to hear, to understand and to fulfill user's needs. It will become the future universal gateway to the digital world. DuerOS is leading China's market because it has developed a skill library of 10 major domains and over 100 subdomains. Even more importantly, DuerOS has established the leading ecosystem where we have built close partnerships with over 100 branded major appliances, including Vivo, Xiaomi, HTC, Midea, Haier, TCL, Lenovo and many others and provide them with end-to-end solutions by working together with chipset providers such as MTK, ARM and Qualcomm, and content providers such as iQiyi and Himalaya. We've also announced the acquisition of KITT.AI, a Seattle based AI startup that provides industry-leading development tools for natural language understanding and voice-based hotwords. In the AI era, the future of the cloud industry will be increasingly driven by integrating AI in the big data. We were among the first to spot that trend and seize the opportunity to pursue our ABC - AI, big data cloud - strategy by delivering a growing suite of powerful industry vertical AI-based solutions to partners and developers, including marketing, media, IoT, video, et cetera. By fully leveraging Baidu's AI technology strengths and the vast data assets, our ABC business is growing very rapidly, securing a growing number of customers. For example, we are working with one of China's top mobile phone manufacturers to provide them with a 360-degree solution from marketing, to deep learning, to intelligent CDN. One of the most promising areas for commercialization of AI technology is financial services. At the Baidu FSG, our focus is AI-enabled fin-tech, by leveraging Baidu's technology strengths and data assets to build a suite of leading AI-based FinTech capabilities. Our long-term objective is to build out a platform that enables Baidu and our partners to provide much more compelling financial services. Towards this, we are making very good progress in developing core capabilities and growing our businesses, last month, we announced a strategic partnership with the Agricultural Bank of China, one of China's largest banks, to collaborate on building an intelligent bank using AI. At the same time, in order to further drive our FSG business, we are beginning the process of working out a future operating structure that allows FSG to operate more independently, to expand into areas that may require domestic licenses and enable stronger, long-term growth. There's no specific timetable and structure determined at this point. Overall, we are very pleased with our progress in the first half of the year. We look forward to continuing our momentum and strong execution on our 2-pillared strategy in mobile and AI, to better achieve our mission of making a complex world simpler through technology. With that, I will turn the call over to Jennifer to go through the financials.

Thank you, Qi. Hello, everyone. We delivered a solid second quarter and further drove operational efficiencies on our platform. We further scaled back spend for Nuomi and shifted our resources to invest in AI, where R&D is a key investment area. iQiyi, a strong partner in our ecosystem, performed well and we will work more closely together to leverage the strengths of our platforms. We'll continue to extract efficiencies and allocate resources with deliberation and discipline. As Qi mentioned earlier, FSG operates very well with its positioning as a FinTech service provider. To build the capabilities, we have taken on some credit originations and wealth management assets to gain customer and business insights. It is our plan that we will manage financial services-related asset exposure as a portion of our total assets around the current range and we'll also contemplate an appropriate structure that enables FSG to operate with independence and flexibility. And finally, about a month ago, we raised $1.5 billion in a public note offering, with cash coming in at the beginning of July. We intend to use the net proceeds from this offering to repay existing indebtedness and for general corporate purposes. Now moving to the financials. All monetary amounts are in RMB, unless otherwise noted. For the second quarter, total revenue were CNY 20.9 billion, representing a 14% increase from the corresponding period in 2016. During the second quarter, Baidu had approximately 470,000 active online marketing customers, representing a 4% increase compared to the previous quarter. Revenue for online marketing customers was CNY 37,500, a 16.5% increase compared to the first quarter of 2017. Traffic acquisition cost, as a component of cost of revenue, in Q2 was CNY 2.5 billion, representing 11.9% of total revenues compared to 15.9% in the corresponding period in 2016, and 12.9% in the first quarter. Bandwidth costs, as a component of cost of revenue, in Q2 were CNY 1.4 billion, representing 6.6% of total revenue compared to 6.3% in the corresponding period in 2016. Depreciation costs, as a component of cost of revenue in Q2, were CNY 828 million, 4% of total revenue compared to 4.1% last year, same period. Operational costs, as a component of cost of revenue in Q2, were CNY 1.2 billion, representing 5.9% of revenue compared to 5.4% last year in the same period. Content costs, as a component of cost of revenue in Q2, were CNY 3.1 billion, 14.9% of total revenue compared to 9.3% in the corresponding period in 2016. This increase was mainly due to iQiyi's increased content cost. SG&A expenses in Q2 were CNY 2.9 billion, representing a decrease of 30% from the corresponding period in 2016. The year-over-year decrease was primarily due to decreased promotional spending for transaction services. R&D expenses in Q2 were CNY 3.1 billion, a 27.7% increase over the corresponding period last year. The increase was mainly due to the growth of R&D personnel-related expenses. Share-based compensation expenses, which were allocated to related operating cost and expense line items, increased in aggregate to CNY 796 million in Q2 from CNY 401 million in the corresponding period last year. This year-over-year increase was a result of increased share grants to employees. Operating profits for Q2 were CNY 4.2 billion, representing a 46.9% increase from the corresponding period in 2016. Non-GAAP operating profit was CNY 5 billion, a 53% increase over the corresponding period in 2016. Income tax expense was CNY 564 million for the second quarter. The effective tax rate was 11.3% compared to 24.8% last Q2. The decrease in the effective tax rate was due to preferential tax status that was granted to certain PRC subsidiaries in this quarter of 2017. Net income attributable to Baidu for Q2 was CNY 4.4 billion, an 83% increase from the corresponding period in 2016. Basic and diluted earnings per ADS for the second quarter amounted to CNY 11.36 and CNY 11.31, respectively. Non-GAAP net income attributable to Baidu for Q2 was CNY 5.6 billion, a 98% increase year-over-year. Non-GAAP diluted earnings per ADS for Q2 was CNY 16.0. As of Q2, the company had cash, cash equivalents and short-term investments of CNY 92.2 billion. Net operating cash inflow and CapEx for the second quarter were CNY 6.6 billion and CNY 1.1 billion, respectively. Total headcount on a consolidated basis, including invested entities, was about 42,200 as of the end of the second quarter 2017. This represents a decrease of 2.6% as compared to last quarter. Now let me provide you with our top line guidance for the third quarter of 2017. We currently expect our total revenues for the third quarter to be between CNY 23.13 billion and CNY 23.75 billion, representing an annual increase of 26.7% to 30.1%. On an apples-to-apples basis, excluding mobile games from Baidu's financials, the guidance represents a 21 -- 29.1% to 32.6% year-over-year increase. Please note this forecast reflects Baidu's current and preliminary view and is subject to change. I will now open the call to questions. Operator, please go ahead with questions.

[Operator Instructions] The first question comes from the line of Eddie Leung of Merrill Lynch.

Just quickly on 2 things. The first one is we noticed that other revenue, other services growing pretty nicely in the quarter. So just wondering what are the key growth drivers for that business line. Is this more from cloud or financial services? And then secondly, quickly on Qi's comment on AI commercialization. I'm curious on the longer-term economic opportunities about IoT. Are we thinking more along the line of a direct monetization approach, like, for example, charging our brand partners a fee per device? Or are we talking about more indirect approach? And if so, what could be some of the commercialization opportunities that we are thinking down the road?

Hi, Eddie. I'll take the first question. Other revenues, yes, you're correct. Other revenues, sequentially have stepped up sizably. The components in there includes, of course, the subscription income that we generate from iQiyi's paid membership. There is also financial services income and also cloud service income and also other fee service income related to, for example, personal cloud services. So it's a combination of matters that other revenues are growing, and we're very pleased with the progress that we’re making on multiple fronts.

So let me take the second part of the AI questions. With regard to AI as it applies to IoT as you asked, our approach is using our DuerOS platform. The DuerOS platform will enable each IoT device, particularly intelligent home devices, to be able to listen, to talk, to understand users to provide services. And economically, there's 2 future functions. One is these IoT devices that are enabled by DuerOS, they become the future information entry points. The economics will be very similar to today's search business because as long as the people are seeking information, interacting with the information that represent services, content, knowledge, there will be natural expansion of economics like today's search engine business. That's one. The second is because DuerOS is a platform that enables devices, there's typical what's called platform economics. You mentioned per-device licensing, that's one way to do that. Windows, for example, can do that. Or there's channel distribution economics that's in App Store, there's channel distribution economics. And the third part is future VIGs you can take. For example, like App Store, when you sell content through app stores like our iOS, Apple takes 30% profit. So DuerOS, in the long run, represents very compelling economic opportunities for Baidu. Our focus is really execution, building out our products, our platforms, a very healthy ecosystem. And there's many, many important execution paths in front of us over the coming years. We are fully prepared, when these economic opportunities become more material, we will be able to share future operating metrics when we are ready.

[Operator Instructions] The next question comes from the line of Chi Tsang of HSBC.

I wanted to ask you about 2 topics. Firstly, newsfeed and secondly, search. In terms of newsfeed, I was wondering if you can give us a little bit more color as it relates to the synergies between search and newsfeed in terms of measures of engagement, app opening times, paid click growth. And secondly, as it relates to search, and, I guess, this is related, paid clicks grew 11% in 2016. I'm wondering what your expectation might be for paid click growth over the medium to long term. And sort of how much more can click-through rates improve from the current levels?

Let me take the first part. Jennifer, maybe you'd take the second part. So with regards to synergies between search and feed, we see plenty of opportunities to drive those user experience synergies. As a matter of principle, for each of the vertical segments, whether it's commerce, travel, retail services, often human needs can be satisfied by a combination of search and the feeds together. For example, in terms of, let's say, travel as potential verticals. When you're looking for picking leisure locations, you go through a planning phase. And in that planning phase, you will do a bunch of searches to find the information. And you also want to receive a set of feeds, with regards to pricing and new offers that are available. But this applies to pretty much every key verticals where that information matters to our users. So as I said, as a matter of principle, there are systematic synergies to be explored. And our Mobile Baidu is really the flagship contingent product that enables us to systematically to explore all those synergies. Particularly I'll emphasize our version 9.0. It's a great app. If you haven't used it, I highly recommend that you download and use it. There's a lot of improvements and all the key improvements is enabled from engineering product perspective to drive those synergies across all those verticals. So the fruits will come out over the coming quarters and years as we get into more and more product innovations on the synergy front.

And for the search clicks, Chi,last year, our search click grew double-digit. And for the past quarter, we continue to see double-digit growth on paid clicks. So search is very solid. And feed is creating additional advertising inventory for an additional revenue source for us. So on both the twin-engine, on the search and feed side, it's progressing well.

Your next question comes from the line of Alicia Yap of Citigroup.

Relating to your 3Q guidance, can you help us to think about what will be the assumptions for the numbers of online marketing customer versus the ARPU growth? How should we think about the absolute addition of the online marketing customer on a sequential basis from the second quarter to the third quarter? And then on the year-over-year basis, will the customer additions still experience some year-over-year decline given it is still a relatively high base last year? And related to the traffic acquisition cost, as a percentage of sales as you continued to decline, what was the reasons and will this trend continue?

Alicia, thank you for the question. You did notice that we had pretty solid additions on the customer front in Q2, and seasonally, we do carry out our national -- nationwide campaign to promote our search and overall data services to our customers. As you note earlier in our prepared remarks, there is a rich combination of services that we're building for our customers, and we hope to service our customers with a variety of -- with their variety needs -- varied needs, that we can meet their desires. And so we do see last year towards year-end, we implemented stricter rules to have higher standards for our customers to qualify to do business with us. And after that exercise, we're building our customer base again. And so we do anticipate our customer base should grow. And because of the rich product portfolio that we offer, we aim to grow our ARPU as well. With regards to TAC, the traffic acquisition cost, you will note that year-over-year, it is a decrease. There are basically 3 factors driving this change. One is last year, you will recall that we had gone through some significant changes, and some lost revenue last year was related to organic revenues that made our base of last year artificially high. That's one. And at the same time, we are deliberately moving some of the TAC resources to support the feed business, and that will gradually ramp up. And the third reason would be a growing contribution on the iQiyi side, which normally would not carry a TAC. And so we do anticipate that the contractual and the union network will continue to be very important for us and the TAC, we do manage TAC proactively.

The next question is from Alex Yao of JP Morgan.

I want to dig a little deeper into the near-term trends of paid clicks and CPCs. I think in your most recent 424 filings, you disclosed that 1Q paid clicks declined by 6% on a year-over-year basis, which is a bit different compared to the trends in the past few years. What is the reason for paid clicks to decline 6% in 1Q? And then on the other hand, this implies a CPC to be roughly flat on a year-over-year basis, which I think is a quite achievement given the first quarter last year, you had not had the regulatory impact and health care clean-up. And can you talk about why with a different basis the CPC is flat. Does it indicate it -- indicating faster-than-expected recovery of the ARPU, particularly from health care sector?

Alex, if you recall last year, Q1 was a normal quarter, and Q2 was a little bit of noises because of the incident that happened in the middle of the quarter. So as we have gone through our customer base and kind of start to rebuild, our Q1 is still in the ramp-up stage. And compared to last Q1, which is a high rate, I think both on the -- especially on the paid clicks side, our year-on-year number is just as what you noted. But as we continue to build up our business, recovering from Q4, Q3 of last year, we do see the whole business is on a good trajectory. And the paid clicks is recovering very nicely and so that's growing. And overall, I think the paid -- I think the CPC is also solid and healthy. And we wanted to really deliver the kind of traffic and good ROI to our customers. And the CPC is just a natural fallout of the customer dynamics right now and also the kind of key words and the bidding activities that's going on.

The next question comes from the line of Juan Lin of 86Research.

My question is related to the core search. Firstly, could you please share with us the current user metrics and trends of user metrics for Mobile Baidu App? And secondly, I'm wondering what is the current CPC and RPM for mobile search as compared to PC pricing.

So let me talk about the user trend, particularly on Mobile Baidu, our flagship product. The trend is users are increasingly using more and more AI-like search queries; for example, using voice to search queries seek information; for example, using their cameras to look for information based on the images the camera's taking, so that's an important usage trend. The second important usage trend is more and more richer type of content is being showcased and presented as compelling product search results. In the past quarter, we talked about richer video content. But going forward, our focus is to use AI technology as a foundation to continue to drive more compelling and natural search experience so that making much, much easier for people to use mobile devices or the capability of the device, voice, camera, location, other ways. And then the result will be presented in a way that combination of richer content and also much native app-like quality of experiences. That's the overarching usage trend. And we fully expect our user engagements will grow as we continue to improve the experience, and that will also continue over the coming years.

Your next question comes from the line of Karen Chan of Jefferies.

I've got 2 questions. One is how should we think about revenue contribution from mobile newsfeed ads in a full year context? What's the budget allocation from advertisers between search and newsfeed? And can management share more color on the tie up with PayPal announced yesterday?

As we mentioned earlier, feed is, on the user front, making great progress, and user engagement is also ramping up nicely. And also, on the advertising and revenue side, it's contributing more. Sequentially, it grew sizably. And we anticipate feed will continue to contribute increasingly in the overall ad spend from our customer. It is -- we do focus very much so this year on the user experience and not very reflexively pushing toward the newsfeed. But it's growing nicely and the customers' acceptance are pretty good. On the PayPal, yes, we entered into a strategic partnership with PayPal. And under the agreement, the Chinese consumers will be able to use Baidu Wallet to shop and pay at PayPal's network. So it offers basically a service solution for many of the Chinese customers traveling overseas, and it's providing a greater choice for international shopping opportunities. It's just one of the features that we offer on Wallet.

Next question comes from the line of Piyush Mubayi from Goldman Sachs.

On the call, you spoke about the strong adoption of apps and time spent on search. Could you also take us through potentially the rising relevance of search itself? And if possible, comment on the sort of paid click growth rate or CPC trends we can expect and the impact of AI there? I know that's been asked in a different form earlier on the call.

Yes, yes. Let me take that question. So with regard to our core search product, particularly using AI technology to increase the core aspects of the product which is relevancy. There's indeed plenty of opportunities to increase the overall relevancy of a search experience, particularly as we increasingly personalize our search processes. One of the important areas we see we can grow, not only the search quality in terms of relevancy but also use that to expand the scope of what search can offer. So search can become more and more personalized over time by using AI as technological foundation to drive those growths. On the monetization side, indeed, we see headroom by using AI technologies to continue to improve the economic yield. Particularly by leveraging more conversion data, using deep learning. For example, over the last 2 quarters, we actually are able to deliver a meaningful improvement on conversion rates. As we all know, search advertising is really conversion economics. Any time you use data and technology to improve conversion, over time, you will be able to deliver more economics. And we do believe that opportunity will continue to exist as we continue to drive deeper utilization of AI technology and more data access to improve search relevancy and also advertising yield over a period of time.

The next question is from Natalie Wu of CICC.

Seems that there is a significant step-up in the loan balance, if you look at your balance sheet, also the item of amounts due to the third-party investors. So just wondering what specifically the item is. And what kind of impact does it have for the income statement? Also, for the financial-related business, about the -- like the cost, which line of the corresponding COGS is that classified into?

Within our financial services business, we have some credit business as well as wealth management business. So these are basically balance sheet that you would see more material existence. P&L-wise, FSG, the financial services, does not really represent significant P&L items, line items. And their revenue, financial ad revenue and the cost of funding goes through the operating income. So it's above the -- it's not the interest income or interest expense that you see in other income. It's part of the operational expense and operational income.

Next question is from Alan Hellawell of Deutsche Bank.

I just wanted to touch on iQiyi as it relates to comments we made earlier in the year. I think Jennifer reflected on the belief that maybe margins should improve at iQiyi. And then we also talked about content growing by as much as 100%, admittedly some of that would go into newsfeed. Content spend seems to be tracking well below that guideline. And I'm wondering, Jennifer, whether you can give us an update about both the bottom and top lines of iQiyi. And then just on content, how much might be now focused on newsfeed?

Yes. Hi, Alan. With regards to iQiyi's -- the revenue line as well as the content cost line, both are tracking very well. iQiyi's performance in delivering revenue, the 2-engine revenue, both advertising as well as subscription revenue, is tracking very well, and is very, very solid and strong. And on the content cost, our guidance provided earlier in the year continued to hold. We do anticipate that we invest significantly on the content cost-wise. Quarter-on-quarter, you might see variations because content cost kicking in depends on when you have the popular show starts to get into the audience. So quarter-on-quarter, there might be variances but the whole year's spend is tracking as planned.

The next question comes from the line of Ming Xu of UBS.

So my question is could you help us understand the comparison of the pricing of the newsfeed ad between you and your major competitors, in terms of both RPM and also like CPC, CTR, et cetera. And also, very quickly, you mentioned a lot about the rising contribution of the native landing pages. I think recently you also changed the algorithm of your search engine, which caused some -- which increased the weighting of native landing page requirement. So what kind of impact -- while this is very good for consumer experience, what kind of impact do you think this will have to the recovery of your number of advertisers? And do you have any target for the recovery in the coming quarters?

Regarding to the pricing for the feed product, it's not that directly comparable. Because a lot of our advertisers come from the Phoenix Nest search advertisers, they would use a combination of search and feed to decide how much they spend and what kind price they pay. So this is quite different from the competitors outside of Baidu. The native experience, I think, overall, we are still driving a lot of traffic to third-party websites. And in order to provide the best user experience for search and therefore, for most of the websites, designed for mobile phones, we need to kind of regulate the format and user experience for the third-party websites. I think that's good for the overall ecosystem. Even after the new rules and regulations, I think this still gives website owners much more liberty and freedom to do things that they cannot do in other closed ecosystems for mobile phones.

Your next question comes from the line of Grace Chen of Morgan Stanley.

The question's about the DuerOS. I can see that this is a great opportunity for the company and also for the industry as well. And can you -- and I can see that you've been building up partnerships with platform vendors, home appliance brands. Can you talk about the competitive landscape? Definitely, it's a brand-new opportunity not just by doing other -- and also other competitors as well. So it would be great if you share your thoughts about the competitive landscape and also Baidu's competitive edge.

We'll try to -- make sure we do understand the question but sorry to -- would it be possible to ask you to maybe just say that again make sure that we give you the right answer.

Yes, but what kind of competitor you are asking?

Yes, in terms of DuerOS, the product, -- yes, this is a great opportunity, it's a great start for the industry and also for Baidu as well, right? So I can see that we are very aggressive but also other competitors, or domestically and foreign competitors, they're also very aggressive too. So I'm just wondering can you share your thoughts about the competitive landscape in this operating system in AI, in an AI age? And also what's Baidu's competitive edge in order to succeed in this new game?

Got it, got it. Let me take that question. So with regards to DuerOS, the competitive landscape is forming. First there's the -- on the domestic front, there are more traditional voice recognition technology-based companies such as iFlyTek. Those are one set of competitors. Baidu's the message is, in many ways, very apparent because we have not only voice speech recognition technologies, but we have the entire stack of all the content integrations. And as well as the cloud platforms that delivers the end-to-end, full packaged solution to our partners. And then there's online companies. They have some ability to offer similar type of services. On those fronts, DuerOS' advantage is also truly substantial because as I mentioned in my prepared remarks, there's a few important dimensions we are leading strongly. First is the platform understands much more domains than any other competitors. We have 10 major domains and 100 subdomains. And this is a long-term strength because it's one thing to understand the voice to text tokens, it's not -- another thing to understand the semantics. In order to understand the semantics of the user sentence, you need the knowledge graph, knowledge, and Baidu has by far the largest knowledge graph in China. We have 100 millions of knowledge nodes for knowledge entities, and then we have more than 10 billion knowledge relationships in our knowledge graph, and it's growing at a very rapid pace. And that's, by the way, through accumulation of 17 years operating a search engine to build that knowledge graph. We believe this is one of the very important sustained advantages that Baidu DuerOS platform enjoys over the competition. And also our technology is by far leading strongly. As I mentioned, we have more than 100 branded appliances, partners, using our technology, put into their appliances or devices. So overall, we enjoy very large positions against domestic competitors. And then international competitors, for example, let's say Amazon or Apple or Google, there, the advantage is also very apparent because to local -- services have to be localized. You have to have a strong local language, local content. Again, Baidu enjoys overwhelming advantage in understanding Chinese conversations and integrating all the Chinese services content. So overall, in summary, DuerOS is very strongly positioned competitively, but we take nothing for granted. Our focus is to stay absolutely laser-focused accelerating the pace of innovation, deliver to more and more value to our customers and build up the full ecosystem with more partners on board.

Your next question is from Zoe Zhao of Credit Suisse.

Just 2 quick ones. One is what is the normalized tax rate that we should expect going forward? And second one is can management give us some color on iQiyi's current subscriber base?

The tax rate, I did give an indication earlier in the year. It should be in the low 20s. They should still hold. We again have a preferential tax treatment that came through this quarter but the normalized tax rate should be in 20s, low 20s.

Yes, on the iQiyi subscriber base, it's more than 30 million. And the revenue from subscriptions is -- it's much larger than any of our competitors.

The next question comes from Wendy Huang of Macquarie.

Your margin recovered nicely in the quarter. I can see that R&D cost and also the SG&A cost both actually decreased significantly year-over-year. I just wonder, is this kind of the cost level change is structural or sustainable. How should we detect a margin trend going forward? And also, I think in the past you provide the margin drag from iQiyi. Can you provide this quarter's number as well?

Okay. The margin, we did deliver a pretty solid margin this quarter. And you -- I mentioned earlier in the prepared remarks that we are really driving further operational efficiency, having a keen focus to direct our resources to AI and having high efficiency in other parts of the business. And we dialed back again in Nuomi. And so hence, the savings on SG&A related expense. Our focus is to invest in AI capabilities and a lot of that is incremental and is R&D focused. So you will see we have an increase in R&D expenses but we do have a year-on-year decrease in SG&A expenses. We'll continue to exercise discipline as we go forward. We do have resources allocated for -- to support our feed product, the feed user coverage. And so second half of the year, maybe slightly more aggressive on the SG&A expenses because we do want to have an installation coverage for that. R&D expenses will be a steady, incremental, consistent with the prior practice. For iQiyi's drag, it's in the earnings release.

We will take our last question from Gregory Zhao of Barclays.

I have 2 follow-up questions. The first one is about our Q3 guidance. I just wanted to understand what's the driver of the strong Q3 revenue guidance? Shall we expect continued strong growth in other revenue? Or I mean the growth is from the marketing services that will drive the overall growth. And another follow-up question, still about voice. So in China and too globally, we see more and more young users, especially some teenage users. They prefer to use audio and video functions than text. So given your advantage in voice technology versus other vertical apps such as online travel or O2O, do you think the voice can be a new engine to attract users from other verticals back to our search platform? And also I want to understand what the percentage of traffic or search contribution from voice.

The Q3 guidance reflects our current best estimate. We mentioned earlier, we're very pleased with our progress on the core search side as well as iQiyi's performance. So both on search, feed and iQiyi, they're very solid. And of course, we're also growing our business in the financial services and the cloud business. So that reflects the combined force of all these different fronts that is reflected in our Q3 guidance.

I can take the voice question. Terrific question. The simple answer is there's huge opportunity ahead. The fundamental reason isn't just voice as a modality, it’s natural language conversation. Because for the first time in our history, because of AI, every device you do not have to learn. They learn you. You can talk to every device using human language. It understands you. So it's highly disruptive as you mentioned, to all vertical apps, vertical domains. Once our users get used to using Xiaodu as their assistant, DuerOS is a constant companion that understands users and is able to fulfill users' needs. It can change the landscape. There's a huge set of economic long-term opportunities for Baidu. I mentioned the natural economics that would be search advertising like economics but it can go to all of these verticals as well. It can be human tasks or it can be particular verticals. So to summarize it, that's why we are so focused on investing in building out substantially leading positions for DuerOS and driving accelerated expansion of our ecosystem and growing our installed base and growing user active usage so that over time, we can deliver really compelling experience to our users and economically, there's a huge disruptive opportunities for Baidu over the coming several years.

We are now approaching the end of the conference call. Thank you for your participation in today's conference. You may now disconnect. Good day.