Baidu: A Deep Research Study

Research Summary

The hardest thing about studying Baidu today is that it has already built itself into a company straddling three valuation languages at once, rather than the question of "whether it does AI." The first language is the old world of search advertising: mature, cash-flow-driven, and visibly exposed to macro conditions and traffic migration. The second is AI cloud and the model platform: fast-growing, capital-intensive, with profit conversion lagging. The third is Robotaxi and Kunlunxin: both carry a big story, yet neither has been truly priced by capital markets on a stable, repeatable cash-flow basis. Baidu therefore looks at once like an internet platform on the defensive, like an AI infrastructure company burning cash at the front line, and like a holding company sitting on a bundle of options.

What it actually makes money on is still advertising and platform traffic, but that answer is no longer as simple as it was a few years ago. In Q1 2026, Baidu's General Business revenue was 26 billion yuan, of which online marketing services were 12.6 billion yuan, with their share down to 48%; in the same quarter, Baidu's Core AI-powered Business revenue was 13.6 billion yuan, or 52% of General Business, clearing half for the first time. Broken down, AI Cloud Infra revenue was 8.8 billion yuan for the quarter, up 79% year over year, with GPU Cloud up 184% year over year; AI Applications revenue was 2.5 billion yuan, and AI-native Marketing Services revenue was 2.3 billion yuan. In other words, what actually happened is that the old advertising base shrank while the new AI base grew, and the two curves finally crossed in their revenue share, rather than Baidu "suddenly making money from cloud."

What the market is mainly trading now is whether this crossing curve can be sustained. The bull case is the structural inflection behind "more than half of revenue is AI-related": AI cloud infrastructure is capturing enterprise and government compute demand, Ernie 5.1 still holds the top tier among Chinese models, Apollo Go has pushed order volume, city replication and overseas expansion into a new commercialization phase, and Kunlunxin is advancing a spin-off listing, meaning a previously hard-to-price compute asset inside the group is starting to list independently. The bear case is a different reality: advertising is still falling, with Q1 2026 online marketing revenue down 22% year over year; the AI cloud is high-growth, but cost pressure is now clearly rising, with the quarter's year-over-year and sequential cost pressure coming mainly from AI Cloud; and the company itself kept emphasizing improved operating cash flow through Q4 2025 and Q1 2026, an emphasis that by itself signals free cash flow is not abundant.

Baidu's share-price moves over the past few years have swung back and forth around these two main threads. The most pressured phase was when the advertising base took the hit from China's macro conditions, real estate and traffic migration before the AI story could backstop revenue, so the market treated it as a stalled platform stock; the relatively strong phase came when capital markets began to believe Baidu did not merely "talk models" but could actually turn AI into cloud orders, marketing conversion and Robotaxi operating efficiency. From Q4 2025 to Q1 2026, Baidu's AI-powered Business revenue rose from 11.3 billion yuan to 13.6 billion yuan, with its share of General Business climbing from 43% to 52%; this is the single most important fundamental fact of the recent stretch.

The most important bull-bear divide right now is "whether the revenue growth that technology brings will ultimately be eaten by depreciation, price wars and organizational complexity," rather than "whether Baidu has the technology." In China's AI cloud market, Alibaba Cloud remains the leader in infrastructure revenue on Omdia's measure, Tencent has the ecosystem and enterprise customers, and ByteDance's Volcano Engine is chasing hard on Doubao and aggressive pricing; on the model side, DeepSeek reset the Chinese market's expectations on open source, low prices and inference cost; on the Robotaxi side, Waymo has pushed weekly paid rides to the 500,000 level, and Tesla has rolled out unsupervised robotaxi service in several US cities. Baidu's strength is that it has built chips, models, cloud and autonomous driving into one vertically integrated system; its weakness is that almost every layer of that system sits in the most fiercely competitive, most price-pressured battlefield.

If I had to capture this company in one phrase, I would rather define it as a company in transition undergoing valuation re-rating. It is no longer a pure mature cash cow, because both its advertising revenue share and its growth logic are changing; nor is it yet a high-quality growth stock fully recognized by the market, because the profit conversion of AI cloud and Robotaxi has not crossed the proof line. Baidu most resembles a company in the middle ground of "old business undervalued, new business not yet fully acknowledged": the cash-flow base is still there but no longer steady; the second curve is putting on muscle but has not grown enough to carry the valuation on its own. This position means it can hardly be re-rated on sentiment alone; it can only do so by turning "AI past half" from a news item into a stable financial structure across several consecutive quarters.

The Company's Long Arc

Origins and the Listing Path

Baidu emerged in 2000, and this was not simply a matter of happening to catch the internet wave. In the management bios, Robin Li discloses that he has been chairman since the company's founding and CEO since 2004; before founding it, he had a background in search and ranking algorithms. Early legal filings also show that Baidu Netcom, the local operating entity, was held by Robin Li and co-founder Eric Yong Xu, used to satisfy China's local licensing requirements for internet content and advertising at the time, which is also the institutional source of the VIE / contractual-control structure Baidu has long used since.

The first problem it solved was the search-efficiency problem of the Chinese internet in the portal era. Baidu Japan's account of the founding story notes that the company first supplied search technology to portal sites in its early days and only launched Baidu.com Beta in 2001; this shows Baidu started from the supply side of search technology and then turned that technology into its own entry point and advertising platform, rather than "building a traffic gateway first and thinking about monetization later." A company that starts this way is naturally algorithm- and engineering-heavy, unlike the many later content platforms that did traffic distribution first and bolted on a technology layer afterward.

The listing path also says a lot about Baidu's relationship with capital markets. In August 2005, Baidu listed on Nasdaq at an ADS issue price of 27 dollars, raising about 4,040,402 ADSs publicly, of which the company issued 3,208,696 and selling shareholders sold 831,706. The IPO documents also disclose that Series B in 2000 brought in investors such as DFJ and IDG, and that in the 2004 Series C round Google also bought shares. The story Baidu told capital markets at its listing was "China's leading search engine," not "an AI company," and the market's initial understanding of it was very close to "China's Google."

In March 2021, Baidu completed a secondary listing in Hong Kong, offering 95,000,000 Class A ordinary shares at an issue price of 252 Hong Kong dollars, equivalent to about 260.01 dollars per ADS on the 8-shares-to-1-ADS conversion. This move had two layers of practical meaning: first, it gave China ADRs a backup listing venue and Asian-time-zone liquidity; second, it preserved a more complete conversion and migration channel in case the US listing environment tightened further. In January 2026, media reported that Baidu was considering upgrading its Hong Kong status to a primary listing to seek broader access to mainland capital, but as of this report's writing I have not seen a formal company announcement from Baidu on the matter, so it can only be treated as an unconfirmed capital-markets option rather than an accomplished fact.

Phases and the Business Narrative

If I compress Baidu's history into a few genuinely causal business narratives, I would split it into five phases.

The first phase was the validation and commercial formation of search technology. The key in this period was that the combination of "Chinese search + auction advertising" was proven able to make money, rather than traffic scale. Supplying technology to portals let Baidu first learn enterprise demand; after pivoting to its own search entry point, it then loaded both traffic and ad revenue onto its own statements. To capital markets, Baidu in this phase was a fast-growing Chinese search company with a clear model.

The second phase was post-listing search dominance and platform expansion. Baidu turned search traffic from a tool into a platform, with Tieba, Baike, Maps, news and other services expanding around the search entry point. The long-term effects this phase left are large: on one hand, Baidu captured the thickest advertising cash flow; on the other, it was locked in by path dependence too, since the heavier the platform, the easier it is to react half a step late in the eras of mobile internet and content distribution. Capital markets in this phase treated Baidu as a growth stock, using the story of traffic, advertising and China's internet penetration.

The third phase was repositioning under the impact of mobile internet. Robin Li has long been the company's core decision-maker in public management narratives, and Baidu also began investing heavily in AI, autonomous driving and voice in this period. The problem was that user time in the mobile-internet era was more fragmented, and social and short-video products diverted search advertising more visibly; though Baidu still had a vast traffic base, it was no longer the sole center of internet innovation. The market's label for it began sliding from "growth leader" toward "established platform."

The fourth phase was the large-model pivot around 2023. After ChatGPT broke out, Baidu was among the first big Chinese internet players to push generative AI to the front, with Ernie, Qianfan and an AI search revamp launched densely in this phase. But starting early does not mean winning immediately, and competition then intensified fast: Alibaba's Qwen, DeepSeek's open-source low-price path, and ByteDance's Doubao and Volcano Engine all meant Baidu's "first mover" did not automatically convert into a high valuation granted by the market. Capital markets swung back and forth on Baidu in this phase: they acknowledged it as one of the core players at China's AI table, but were unwilling to grant a pure-AI-company multiple before advertising and profit had stabilized.

The fifth phase is now: AI-related revenue past half, Apollo Go entering volume operations, and Kunlunxin launching a spin-off. In Q4 2025, Baidu's Core AI-powered Business revenue was 11.3 billion yuan, or 43% of General Business; by Q1 2026, that figure rose to 13.6 billion yuan, or 52%. Apollo Go completed 3.2 million fully driverless rides in Q1 2026, with cumulative rides exceeding 22 million in April and a global footprint reaching 27 cities in May. Kunlunxin, for its part, announced in January 2026 a proposed spin-off and a confidential filing with the Hong Kong Stock Exchange. The company is finally no longer just telling a story of "I will become an AI company in the future," but has made AI a visible proportion of its revenue structure.

Key Milestones, Financial Arc and Reading the Share Price

The milestones that truly changed Baidu's fate over the past decade were a few moments when "the market re-decided what framework to use to view it," rather than any single product launch.

The first key milestone was the Hong Kong secondary listing. It did not change the business itself, but it changed the geographic and institutional boundaries within which Baidu is traded. The Hong Kong secondary listing gave Baidu an extra layer of institutional insurance among China-ADR risk, ADR conversion and a potential future dual-primary listing. HKEX's official explanation is clear: a secondary-listed company is primarily regulated by its original primary listing venue, while a dual-primary-listed company must fully comply with the Hong Kong listing rules; the 2022 expansion of Southbound Stock Connect brought "foreign companies with a Hong Kong primary listing" into the eligible universe. If Baidu does upgrade to a primary listing in the future, the most direct capital-markets implication is that a new channel may open between it and the mainland capital pool, rather than "a sexier story."

The second key milestone was the 2025 annual report breaking out AI-powered Business separately. Before that, the market always felt Baidu's AI was hard to land in the statements, because "cloud, models, applications and marketing transformation" were all mixed together. The Q4 2025 and Q1 2026 announcements broke out AI Cloud Infra, AI Applications and AI-native Marketing Services into structured numbers for the first time, and the importance of this was underrated by many investors. For a platform long suppressed by "advertising decline," the biggest valuation repair has always come from spinning the new growth engine into a trackable disclosure metric, not from naming more products.

The third key milestone was Apollo Go shifting from "technology demo" to an operating system delivering "unit-economics improvement." Baidu said in its Q4 2025 earnings-call transcript that Apollo Go was the first to reach unit-economics breakeven in Wuhan at the end of 2024; the Q1 2026 earnings-call summary further said that since the Wuhan breakeven, the company's unit economics has kept improving. The key here is that it means Robotaxi at least begins to escape the fate of "forever being only an R&D expense," rather than the word "breakeven" itself. Companies like Waymo, Tesla, Pony.ai and WeRide are all advancing robotaxi, but Baidu's distinguishing feature is running the business first in lower-fare, lower-ticket Chinese cities, which gives its cost learning curve more substance.

Financially, the core change at Baidu over the past few years can be summed up in one sentence: its revenue looks more like two companies, and its costs look more like three. For full-year 2025, Baidu's total revenue was 129.1 billion yuan, down 3% year over year; costs were 72.4 billion yuan, up 10% year over year, mainly because of rising AI-powered Business costs; the same year it also recognized 16.2 billion yuan of long-term asset impairment, which produced an operating loss on a GAAP basis, but operating profit excluding the impairment was still 10.4 billion yuan, with non-GAAP operating profit of 15 billion yuan. Q1 2026 revenue was 32.1 billion yuan, down 2% sequentially, but with selling, administrative and R&D expenses down clearly sequentially, operating profit returned to 3.2 billion yuan and non-GAAP operating profit was 3.8 billion yuan. Baidu's problem, therefore, is whether the speed at which the new business grows can rebuild the predictability of the income statement under the impact of cost, depreciation and one-off items, rather than "being unable to make money at all."

If you only look at the share-price history as a chart, it is easy to reach two extreme misjudgments: one is "Baidu is a declining old China ADR," the other is "Baidu is a buried AI treasure." Both capture half of it. Baidu did experience the highs of the search era and the downward shift of its valuation center in the mobile era; but from the second half of 2025, the share price began to be reactivated by AI cloud, Robotaxi and the Kunlunxin spin-off expectations. How high it once traded does not matter; what matters is whether the market is willing to admit that today's Baidu is no longer the Baidu of 2022 that could only passively defend against advertising decline.

Business Model, Industry, Competitors and Current Fundamentals

Business Model and Moat

Viewing Baidu through today's statements, the most fitting breakdown is "old cash flow + new compute platform + long-dated options," rather than the old framing of "search, cloud, autonomous driving."

The old cash flow is online marketing and search distribution. In Q1 2026, online marketing services were 12.6 billion yuan, or 48% of General Business; this remains the company's most stable, most easily-monetized revenue source, but it fell 22% year over year, showing it has shifted completely from a growth engine to a pressured base. The new compute platform is AI Cloud Infra, AI Applications and AI-native Marketing Services. In Q1 2026, these three totaled 13.6 billion yuan, of which AI Cloud Infra alone was 8.8 billion yuan. The long-dated options are Apollo Go and Kunlunxin: the former is transitioning from operating metrics toward unit economics, the latter is advancing a spin-off, intending to capitalize the group's internal chip assets separately.

The moats I think genuinely hold for Baidu are four.

The first is the Chinese search entry point and its commercialization data loop. Statcounter shows that in May 2026, Baidu's all-device search market share in China was about 47.16%, and on a mobile-search host basis about 57.58%. This does not mean Baidu has returned to the era of "ruling everything," but it still controls China's largest pool of high-intent search traffic. For AI-native Marketing Services, this kind of traffic where "users arrive to search with a clear need" converts more easily than pure content traffic and is better suited to using models to reshape ad placement and merchant operations.

The second is vertical integration of the tech stack. Baidu does not just have a model, or just a cloud. It has stacked Ernie, AI cloud, Kunlunxin and autonomous driving into a mutually feeding system: models need compute, compute can settle into AI Cloud Infra orders, cloud customer demand in turn feeds model iteration, and autonomous driving is a high-intensity scenario test of chip, model and cloud-scheduling capability. The Kunlunxin spin-off announcement states explicitly that one goal of the spin-off is to display Kunlunxin's value independently, broaden financing channels and support the value release of Baidu's AI business; this is precisely why management itself realizes that when these internal assets were bundled together, capital markets could not see them clearly.

The third is capital and organizational endurance. As of Q1 2026, Baidu's disclosed total cash and investments were still 279.3 billion yuan, about 40.49 billion dollars. Even discounting this metric heavily, it still shows Baidu can outlast many pure-AI startups: it can advance models, cloud infrastructure, Robotaxi and a chip spin-off all at once, without being forced to retrench the moment one business hits a financing snag, as a single-business company would. This is also why I do not think Baidu will be knocked out directly by China's large-model price war.

The fourth is founder control. Under Baidu's A/B dual-class structure, Class A shares carry 1 vote and Class B shares 10; as of 2026-01-31, Robin Li's economic interest was about 18.6%, but his voting power was about 59.9%. This brings a governance discount, because ordinary shareholders have inherently weaker checks on major direction; but from another angle, for a company still in a high-investment transition and making long-term bets on chips and autonomous driving, strong control also reduces the interference of short-term market noise on strategy. There is no absolute good or bad here, only whether investors accept this governance structure.

What truly does not count as a moat is the slogan itself of "first to launch Ernie" and "Baidu is China's search leader." The former is no longer enough in China's 2026 AI market, because competition is no longer about launch timing but about price, inference cost, ecosystem and commercial conversion; the latter is no longer enough either, because the user decision entry point is being diverted to short video, social recommendation, in-app search and AI assistants. The moat is still there, but it must be validated through cloud revenue, marketing conversion and AI application subscriptions, not through past glory.

Management and Governance

Robin Li has been chairman since the company's founding in 2000 and CEO since 2004, with essentially no transition period where "the founder steps into the background." On finance, the 2026 disclosure documents and earnings calls are signed and delivered by Rong Luo as CFO. Structurally, Baidu remains a typical complex China-ADR holding structure: a Cayman parent, contractual control, a local licensing entity and dual-class shares all coexisting. The 20-F clearly warns that if the VIE contractual arrangements are deemed invalid or hard to enforce, Baidu could lose control of the relevant businesses and key assets such as the baidu.com domain.

One new external variable should be added on governance. On 2026-06-09, Baidu responded to its inclusion on the US Department of Defense's so-called CMC List. The company said it is neither a Chinese military company nor a contributor to military-civil fusion, and stressed that the list is not a sanctions list, does not restrict trading in its securities, and that the US government procurement restrictions tied to the list will not affect its business. In terms of near-term financial impact, this is not a direct sanction; but in valuation terms, it will raise the China-regulatory and geopolitical discount that some international investors demand.

Audit and internal control show no "hard defect" for now. In the 2025 annual report, Ernst & Young Hua Ming issued an effective opinion on Baidu's internal control as of year-end 2025, noting only that the assessment excluded the internal control of YY Live, acquired in 2025, which accounts for less than 1% of total assets and less than 3% of full-year revenue. My read is that Baidu currently does not show a typical China-ADR financial-trust crisis, but its very governance complexity means investors need to leave a little room for a structural discount.

Industry, Cycle and Peer Comparison

The hardest single thing about Baidu is that it does not belong to a single industry. Search belongs to a mature industry, AI cloud to a high-momentum expanding industry, and Robotaxi to an early commercialization industry driven by both regulation and technology. Mixing these three lines into one company means Baidu is simultaneously exposed to the advertising cycle, the capex cycle, the technology-iteration cycle and the policy cycle.

Start with search. China's search market has not disappeared, but it is highly mature, and the remaining competition is mainly about who can turn content entry points beyond search into their own traffic pool, rather than "who can grab another ten points of share." Statcounter's May 2026 data shows Baidu is still China's search leader, but its share is no longer an absolute overwhelming advantage. For Baidu, this means search's commercial value depends more on quality than on sheer quantity.

Next, AI cloud. The industry momentum here is real. Tencent, citing IDC data, says China's public cloud market reached 25.01 billion dollars in H1 2025, of which IaaS was 14.01 billion dollars and PaaS 4.74 billion dollars, with PaaS growing fastest and AI Agent and MaaS commercialization the core drivers; Omdia gives a more AI-focused measure, putting China's AI cloud market at 56.7 billion yuan in 2025, of which IaaS was 69% and MaaS 31%, with Alibaba Cloud leading at 38.1% share. Baidu's own figure comes from IDC: a 28% share of China's MaaS market and a 17% share of the AI-solutions market in 2024. These three sets of numbers use different definitions, but together they point to the same thing: compute, model hosting, Agents and industry solutions have entered a rapid expansion phase, but "who is first" depends heavily on whether you use the cloud-infrastructure, MaaS or industry-AI-solution lens. Baidu's competitors are a group whose very definitions differ from one another, not a single company.

Among China-market comparables, Alibaba is the most practical peer reference. Alibaba Cloud remains the leader in China's cloud infrastructure market on Omdia's measure, with a 37% share in Q4 2025, and its AI-related products have delivered triple-digit year-over-year growth for ten consecutive quarters; Alibaba Group's 2025 interim results also disclosed cloud revenue up 26% year over year to 33.4 billion yuan. Alibaba's strengths are enterprise-cloud scale, the Qwen ecosystem and a clearer cloud-platform leadership; Baidu's strengths are the search entry point, the marketing data loop, and the depth of Robotaxi and chips. To customers, Alibaba looks more like a "general-purpose AI and cloud base," while Baidu looks more like a "vertical AI system with search and autonomous-driving features." Alibaba's entry into Southbound Stock Connect after upgrading to a Hong Kong primary listing also shows that the capital-markets structure itself can bring extra liquidity to a same-type Chinese tech giant, which is a concrete reference for Baidu, not an abstract notion.

Tencent is a different kind of rival. It is not the company most like Baidu, but it exerts real pressure on Baidu in enterprise customers, ecosystem and distribution. Tencent Cloud discloses that it operates 55 data centers across 21 markets and regions globally and continues to expand its overseas infrastructure; in Omdia's tracking of China's Q3 2025 cloud-infrastructure market, Tencent Cloud's share was about 9%. Tencent's strength is binding AI into the vast distribution network of WeChat, gaming, enterprise services and cloud, rather than a single-point technology bet. Compared with Tencent, Baidu is more focused and more fragile; Tencent is more diffuse and harder to break out at a single point.

The best global reference is still Alphabet. This is because Alphabet represents how the combination of "search cash cow + cloud + Waymo" can be priced in capital markets, not because Baidu and Alphabet are alike. A March 2025 Reuters report noted that Waymo was already providing more than 200,000 paid rides per week in the US; by March 2026, Waymo publicly stated it had reached 500,000 paid robotaxi rides per week. Stock tools show Alphabet's current trailing PE at about 27x. The market is willing to give Alphabet a higher valuation because search and cloud are strong enough to "afford" this kind of long-dated option, not because Waymo is already profitable. The biggest difference between Baidu and Alphabet is that Baidu's base is far less solid than Google Search, so its AI and Robotaxi options can hardly earn the same valuation premium.

In the Robotaxi arena, Baidu faces two kinds of rivals. The first is a player like Waymo that has already gone deeper on commercial density; the second is a player like Tesla, which capital markets have already priced with extreme imagination ahead of business delivery. Reuters reported in June 2026 that Tesla had rolled out unsupervised robotaxi service in several US cities, operating about 50 vehicles in Austin; tool data put Tesla's trailing PE at about 366x. Compared with Tesla, Baidu's problem has always been that the market is unwilling to grant it such an expensive multiple before the story is delivered, rather than "the story not being big enough." Compared with Waymo, Baidu's cumulative order volume is no longer behind, but Waymo's current weekly commercial volume is higher, showing that the high-ticket, high-utilization model in mature US cities is currently still more convincing.

In terms of niche, Baidu occupies the position of "a full-stack supplier with an entry point" in China's AI value chain, neither a pure leader nor a pure follower. The gap it fills is putting search, marketing, models, cloud, chips and autonomous driving inside one company and connecting them, being neither a broad cloud-infrastructure leader like Alibaba Cloud, nor a model company like DeepSeek that hits the whole industry's pricing at a single point, nor a capital-markets myth like Tesla that pitches the vision first and runs the service later. But this is also its burden: as long as one line visibly falls behind, the group valuation gets dragged down as a whole.

The Last Four Quarters and the Current Bull-Bear Divide

Over the last four quarters, Baidu's story has been almost brutally clear. Q2 2025 total revenue was 32.71 billion yuan, down 4% year over year, with Baidu Core revenue 26.3 billion yuan, down 2% year over year, of which online marketing revenue was 16.2 billion yuan, down 15% year over year, and non-online-marketing revenue was 10 billion yuan, up 34% year over year, driven mainly by AI Cloud. Q3 2025 total revenue was about 31.2 billion yuan, down 7% year over year, with AI Cloud Infra at 4.2 billion yuan, up 33% year over year. Q4 2025 total revenue was 32.7 billion yuan, up 5% sequentially, with AI Cloud Infra at 5.8 billion yuan and about 20 billion yuan for the full year; Apollo Go did 3.4 million fully driverless rides in the quarter. By Q1 2026, total revenue was 32.1 billion yuan, with Baidu's General Business revenue flat at 26 billion yuan, AI Cloud Infra up to 8.8 billion yuan, up 79% year over year, while online marketing fell to 12.6 billion yuan. Strung together across four quarters, advertising kept falling and AI cloud kept growing; the real inflection is the business structure, not total revenue.

The signal on the profit side is more subtle. Q4 2025 non-GAAP operating profit was only 3 billion yuan, with a non-GAAP operating margin of 9%; Q1 2026 it recovered to 3.8 billion yuan, with a 12% margin. But at the same time, Q1 2026 costs rose 7% sequentially, which the company explicitly attributed to AI Cloud-related cost growth. In other words, Baidu has already proven AI cloud can indeed lift revenue, but has not yet proven AI cloud can deliver a stable margin uplift to the group after expansion. What the market is trading now is precisely this gap: it believes Baidu's AI is growing, but it is not sure the quality of that growth can cover the intensity of the investment.

The bull case has four core pieces of evidence. First, AI-powered Business is past half, a structural change already in the statements, not a vision. Second, AI Cloud Infra is growing extremely fast, with GPU Cloud up 184% year over year, showing Baidu is selling more low-level compute and platform capability, not just model APIs. Third, Apollo Go has entered a high-frequency operating phase, with 3.2 million fully driverless rides in Q1, cumulative rides over 22 million in April and expansion to 27 cities in May; commercialization is not a slide deck. Fourth, if the Kunlunxin spin-off advances smoothly, an asset inside the group that could previously only be bundled at a discount has a chance to be valued independently.

The bear case is equally solid. First, advertising is under continuous pressure, not "short-term volatility"; Q1 2026 online marketing fell 22% year over year and its share has already dropped to below half. Second, China's AI cloud competition will soon turn "high growth" into "high growth but low profit," because Alibaba, Tencent, Volcano Engine, DeepSeek and even Huawei are all pressuring price and delivery at different layers. Third, capital markets will still add a China-ADR and geopolitical discount to Baidu; the latest CMC List event, though not a sanction, reminds overseas capital that this kind of tail risk is always present. Fourth, Baidu has not yet made "cash-flow improvement" beyond dispute; management repeatedly stressed that operating cash flow turned positive in 2H25 and Q1 2026, which precisely shows this part was not stable before.

Valuation Analysis

Historical Valuation and Choice of Method

Baidu cannot be judged on a single multiple. Treating it as a pure search platform misses AI cloud, Robotaxi and Kunlunxin; treating it as a pure AI growth stock ignores the reality of advertising decay and margins. I think the most applicable framework for Baidu is "a conservative SOTP + a low-error-tolerance range valuation," rather than a grand far-vision DCF.

There are three reasons. First, the asset properties differ too much. General Business still holds mature advertising cash flow as well as high-growth, high-investment AI cloud; Robotaxi is still in the investment phase; and Kunlunxin has entered a proposed spin-off phase. Second, the balance sheet is too thick, and simply looking at PE easily lumps together large cash, long-term investments, convertible bonds and equity holdings. Third, management and the market use ever-finer metrics for the new businesses, but the long-cycle validation of cash flow and depreciation is not yet long enough. As of Q1 2026, Baidu's total cash and investments were 40.49 billion dollars; at the end of 2025 they were 42.06 billion dollars. At the same time, the 20-F discloses that the company has US dollar debt, renminbi debt and the Trip.com-share-linked 2032 exchangeable bonds, among other liabilities and capital-structure arrangements. On the surface Baidu looks like "more cash than market cap," but this does not mean all these assets can be returned to shareholders 1:1 immediately.

Cash-Flow Pass-Through and Valuation Scenarios

Strictly per the template requirement, an owner-earnings lens should take priority over accounting profit. But given the convenience of Baidu's current public disclosure, what I can verify with high confidence is the 2025 audited statements, the Q4 2025 and Q1 2026 results announcements, and management's description of operating cash flow; the company did not fully break out maintenance capex versus expansion capex in its recent quarterly announcements. The known facts are: Baidu's operating cash flow in the second half of 2025 turned positive at a combined 3.9 billion yuan, and Q1 2026 operating cash flow stayed positive at 2.7 billion yuan; meanwhile Q1 2026 costs were lifted significantly by AI Cloud. Given this disclosure boundary, I do not think it appropriate to derive Baidu's "precise fair value" with a headline PE or an aggressive DCF. The safer approach is a three-scenario framework that looks at the operating business and the balance sheet separately.

The valuation core I use here is very plain:

First, take Q1 2026 non-GAAP operating profit of 3.8 billion yuan as the current earnings-power anchor, but do not extrapolate it linearly; instead, set lower, similar and higher "sustainable operating profit" assumptions in the conservative, neutral and optimistic cases respectively. Second, do not include the balance sheet at full book value, but assign only a heavily discounted attributable value, because total cash and investments mix operating-essential cash, restricted funds, long-term investments and holdings that need discounting. Third, give Robotaxi and Kunlunxin only a small option value in the neutral case, raising the weight only in the optimistic case.

Table note: the operating-profit anchor, balance sheet, spin-off and business structure all come from BAIDU's 2025 annual report and Q4 2025 and Q1 2026 announcements; the table above is a valuation-scenario projection under a research framework, not investment advice.

Extending from this model to the price signal, the conclusion I reach is clear: Baidu is not "obviously overvalued," but it is not yet cheap enough to give new money a strong margin of safety. At the current 117.49 dollars, it sits roughly near the neutral value, more like a position of "can hold but not worth chasing." The price the market gives Baidu today essentially says: I admit your AI transition is happening, but I am only willing to pay on a "partial-delivery" standard, not on a "certain-winner" standard.

Peer Valuation, Expectation Gaps and Margin of Safety

In a cross-comparison, Baidu looks very cheap, but this "cheap" must be unpacked. Alphabet's current trailing PE is about 27x, and Tesla's trailing PE about 366x; what Baidu is currently traded at is a combination closer to "a low-multiple platform stock + a small amount of AI/Robotaxi options," rather than a pure AI narrative. As China's cloud-infrastructure leader, Alibaba Cloud's AI value is mostly wrapped inside Alibaba Group's valuation; Baidu's discount comes from a two-way tug-of-war: weak advertising, and AI that is not yet fully mature. Cheap does not automatically equal attractive; only when you believe AI cloud and the new businesses will keep raising the revenue share while not wrecking margins does this discount have room to converge.

The expectations the market currently implies, I think, are roughly these: advertising at least will not collapse faster; AI cloud will keep growing high; but Apollo Go and Kunlunxin will not contribute profit large enough to change the income statement within the next 12 months. What could truly create an expectation gap is three kinds of indicators. The first is whether AI Cloud Infra's year-over-year growth can stay above 50% for several consecutive quarters. The second is whether the online-marketing decline can narrow from -22% to single digits. The third is whether operating cash flow can stay positive rather than being dragged back by subsequent depreciation and capex. By comparison, Ernie climbing a few more spots on the rankings may not have the largest marginal impact on the share price.

After rechecking the margin of safety, my conclusion is: not obvious. There are three reasons. First, the current price is clearly above the "ideal buy zone" I give, and above the conservative-case implied value of 74 dollars; the margin of safety therefore cannot be called adequate. Second, the most fragile assumption is "AI cloud high growth not crushing margins," not Robotaxi or Kunlunxin. If I cut the neutral-case sustainable operating profit by another 30%, the neutral fair value would drop from about 116 dollars to around 90-95 dollars. Third, if earnings merely move sideways over the next three years, and the market keeps treating Baidu as a low-multiple platform stock rather than a growth stock, then the annualized return on buying today would not be very attractive, and might well be just waiting for the spin-off and a sentiment repair to unlock value. This is the classic case of "a good asset is not necessarily a good price."

Risks, Catalysts and Tracking Indicators

Main Risks

The risks most likely to cause permanent capital loss, I think, are five.

The first is the continued structural decline of the advertising base, with probability medium-to-high and impact high. The evidence is already in the last four quarters: Q2 2025 online marketing was -15% year over year, and Q1 2026 went further to -22%. If Baidu cannot steadily raise commercial conversion on "AI reshaping search," while user attention keeps migrating to short video, social content and in-app search, then the advertising business will become a negative-feedback loop that keeps devouring the group's valuation anchor. The indicators most worth watching are online-marketing revenue year over year, advertising's share within General Business, and whether Baidu App MAU stabilizes.

The second is AI cloud high growth but profit not delivering, with probability high and impact high. Q1 2026 costs rose 7% sequentially, which the company explicitly attributed to AI Cloud business cost growth; meanwhile China's AI cloud market has many competitors, with Alibaba, Tencent, Volcano Engine and Huawei all pressing at different layers. Once this risk materializes, the transmission path is very direct: revenue looks great, but costs and depreciation grow faster, margins stay stuck for the long term, and the market ends up treating Baidu again as "low-quality growth."

The third is that Robotaxi scaling does not equal profitability, with probability medium and impact medium-to-high. Apollo Go's orders, city count and overseas progress are all impressive, but this is a business naturally constrained by regulation, operations, vehicle utilization and single-city density. Waymo is strong because it built up order density in high-ticket cities first, not just because of technology; Tesla is treated with a high valuation because investors bet on the robotaxi endgame first. If Baidu cannot replicate the "Wuhan breakeven" to more cities over the next two years, Robotaxi will remain for a long time in the "story but no profit" option layer within the group valuation.

The fourth is capital-markets-structure and geopolitical risk, with probability medium and impact medium-to-high. The secondary listing gave Baidu a backup channel but did not eliminate all external risks. The latest CMC List event shows that even without a direct securities-trading ban, the geopolitical label still affects some overseas institutions' risk budgets; and if Hong Kong remains stuck at secondary-listing status, the potential liquidity from mainland Southbound capital cannot be opened with certainty. What this risk most calls for watching is: whether the company formally advances a primary-listing status change, whether US regulation takes further action, and whether the Hong Kong shares enter a broader Stock Connect capital pool.

The fifth is that the governance and structural-complexity discount is hard to remove, with probability medium and impact medium. Robin Li holds 59.9% of the voting power, and the VIE structure is still the underlying legal scaffold. For long-term investors, this means that even if Baidu's fundamentals improve, it may not earn the same governance premium as comparable US tech companies. This discount mostly will not erupt suddenly, but it will persist through every style rotation, regulatory disruption and cross-border friction.

Catalysts

Among positive catalysts, the foremost is AI Cloud Infra continuing to sustain growth above 50% and driving General Business margins to stay at least above 12%. The second catalyst is a clear narrowing of the online-marketing revenue decline, which would directly change the market's view of Baidu's "old engine stalling." The third catalyst is a more substantive milestone in the Kunlunxin spin-off, such as a hearing, a prospectus or clearer financial disclosure. The fourth catalyst is Apollo Go's overseas replication running faster than expected, especially if high-fare markets such as Dubai, London, Germany and the UK can advance smoothly, which would make the market re-estimate its unit-economics ceiling. The fifth catalyst is a change in capital-markets structure: if Baidu really advances a Hong Kong primary listing and enters a broader Southbound capital pool, the valuation discount itself could converge.

Negative catalysts are the reverse. If online marketing's decline stays below -15% over the next two quarters while AI cloud growth slips toward 30%, the market will quickly return to the pessimistic frame of "AI not fast enough, advertising falling too fast." If operating cash flow turns negative again, or new depreciation makes the non-GAAP margin clearly slip, it would also break the narrative that "the investment is entering the harvest period." If Robotaxi sees broad regulatory tightening, delayed overseas pilots or a major safety controversy, it would likewise knock out the long-dated options in the valuation. On the geopolitical front, if a more substantive escalation of US investment or procurement restrictions follows the CMC List, the discount would widen immediately.

Tracking Dashboard

Table note: the current values and threshold designs in the table are based on Q4 2025 and Q1 2026 disclosed data and company announcements; the thresholds are research monitoring lines, not mechanical trading lines.

Among these indicators, the three things I value most are not the share price or the large-model rankings: AI Cloud Infra growth, the online-marketing decline, and whether operating cash flow stays positive. The first two decide whether Baidu has truly completed the old-to-new engine switch, and the third decides whether this switch will be only a "transition in revenue" rather than a "transition in shareholder returns."

Cross and Long Arc Synthesis

Looking along the long arc, Baidu has truly proven only two kinds of capability. The first is turning Chinese search into a scalable, monetizable engine, a thing validated for many years in China's internet history. The second is engineering-implementation capability: it cannot always make the most popular consumer products, but it can often turn underlying capability into deliverable infrastructure. Today's AI Cloud, Ernie, Kunlunxin and Apollo Go are all extensions of this kind of capability. The problem lies exactly here: strong engineering capability does not automatically mean capital markets will grant a high valuation; what the market ultimately wants to see is whether these capabilities can pass through the income statement and become sustained, compoundable shareholder returns.

Baidu's past success did capture the dividend of its era. The formation period of China's internet search entry point, the restrictions on overseas giants in China, and the explosion of the local advertising market all gave it enormous tailwinds. But explaining Baidu's history solely as a reward of the era also underrates its sustained investment. It has reinvested heavily beyond search in AI, cloud, autonomous driving and chips for more than a decade, not just in the past couple of years. The most remarkable thing about this route today is that it finally made AI-related revenue more than half of Core in Q1 2026, showing the path was not walked in vain. But at the same time, one must also see that the factors that once helped Baidu succeed have not been preserved intact: the traffic entry point is no longer exclusive, the advertising industry is more cyclical, user behavior is more fragmented, and competitors are more diverse. Baidu now must re-prove itself with a new business logic.

Looking across peers, Baidu's real advantage over competitors lies in system completeness, not in single-point model rankings. Alibaba has a bigger cloud, Tencent a stronger ecosystem, ByteDance a fiercer growth drive, Waymo a lead in robotaxi commercial density, and Tesla a runaway lead in story pricing. But few companies, like Baidu, hold both the data entry point of China's search scenario and pieces placed simultaneously across the four layers of cloud, models, chips and autonomous driving. This gives it a unique position in China's AI value chain. The weakness is equally clear: this set of capabilities is too heavy, too expensive and too easy for the market to dock fifty points on every front. Advertising is weak, so the market says you are old; cloud profit is low, so the market says you are heavy; Robotaxi is not yet independently profitable, so the market says you are far off. Part of Baidu's weakness is phase-specific, such as the lag in AI cloud profit conversion; part of it is structural, such as the search-traffic share that can never again form an overwhelming advantage as it did in the early years.

The place I think the market is now most prone to misjudge is its persistent urge to understand Baidu with "one simple label." Seeing it as a pure advertising platform underrates the structural change in AI revenue; seeing it as a pure AI company overrates the speed of profit conversion. The current valuation does not over-borrow from the future; rather, it looks more like giving Baidu only half of the future: admitting AI will grow, but refusing to treat it ahead of time as a high-quality growth stock. For long-term investors, this means Baidu's odds are not bad; but for new money, today's price is also not wide enough to ignore execution risk.

Over the next one, three and five years, the key variables actually differ. The one-year view is whether AI cloud growth can be sustained, whether the advertising decline can narrow, and whether operating cash flow can stay positive. The three-year view is whether Baidu can re-couple the high growth of its AI business with the group's margins, and whether Kunlunxin and Apollo Go can form a clearer capital-markets narrative. The five-year view is the more fundamental question: whether Baidu can turn itself from "a portfolio of new businesses fed by an old entry point" into "a company whose new businesses grow their own profit core." If it cannot, it will remain for a long time between a discounted holding platform and a transitional China-tech asset; if it can, today's valuation center is a floor with another layer below, not a ceiling.

The Bull Case

• AI-powered Business already accounted for 52% of General Business in Q1 2026; Baidu has for the first time turned "AI transition" into a verifiable revenue structure rather than a concept.

• AI Cloud Infra grew 79% year over year in Q1 2026 and GPU Cloud grew 184% year over year, showing enterprise-side compute demand is ramping fast.

• Apollo Go reached 3.2 million fully driverless rides in Q1 2026, with cumulative rides over 22 million in April and expansion to 27 cities in May; commercialization has crossed the "just an experiment" stage.

• Kunlunxin's proposed spin-off and confidential filing provide a real path to value the group's internal chip assets independently.

• The balance sheet is thick, with total cash and investments of 40.49 billion dollars in Q1 2026, giving the high-investment transition a fuller buffer than most AI startups.

The Bear Case

• Online marketing revenue fell 22% year over year in Q1 2026; the advertising base is still bleeding fast, and the old engine stalling is a real problem.

• AI cloud's high growth is bringing higher costs, with Q1 2026 costs up 7% sequentially; Baidu has not yet proven the high growth can deliver steadily as high-quality profit.

• China's AI cloud market is exceptionally competitive, with Alibaba Cloud leading in infrastructure share and ByteDance, Tencent and Huawei all accelerating; Baidu can hardly enjoy high growth alone for long.

• Robotaxi has scale, but Waymo's current weekly paid rides have reached the 500,000 level and Tesla is also advancing fast; the high-ticket US market pressures Baidu.

• Dual-class shares, the VIE structure and the latest CMC List event together form a persistent valuation discount, not one-off noise.

Pre-mortem

The first script that could cost me 50% three years out is a clearer price war in China's AI cloud in 2027. Alibaba, Volcano Engine, Tencent and DeepSeek keep pushing inference and MaaS prices down, and Baidu is forced to follow on price to defend AI Cloud Infra growth; meanwhile advertising revenue falls another roughly 10% per year. By the end of 2027, Baidu's non-GAAP operating margin is squeezed from 12% in Q1 2026 to 7%-8%, the market re-rates it at 7-8x as a platform stock, and the Kunlunxin spin-off brings no adequate cash-realization dividend. The share price could well return to the 60-70 dollar zone then. This script does not require "business failure," only a continued deterioration in growth quality.

The second script is Robotaxi and the capital-markets structure catching fire at the same time. In 2027, overseas robotaxi replication runs slower than expected, and London, Germany, Dubai and other sites are dragged by regulatory or operational issues; Apollo Go's "Wuhan breakeven" cannot be replicated to more cities for a long time. Meanwhile geopolitics keeps tightening, a CMC-List-like risk escalates into more substantive capital restrictions, and Baidu's Hong Kong primary-listing status sees no substantive progress either. The market would then discount all of Baidu's long-dated options and price it only on advertising decay and a low-multiple platform stock, with the share price likewise possibly falling toward 55-65 dollars.

Final Research Conclusion

Baidu is a company that has already completed a revenue-structure inflection but has not yet completed a profit-structure inflection, deserving neither the simple label of "a cheap China AI stock" nor a simple placement in "a declining old internet platform." AI-related business past half is this company's most substantive change in recent years; advertising under continued pressure is this company's hardest-to-avoid reality in recent years. Both are true at once, so the valuation the market gives it also shows a mezzanine state: no longer viewed at extreme pessimism, yet far from viewed at certain growth.

If I look only at the next three to five years, I still think Baidu is one of the large platforms most worth tracking continuously in China's AI infrastructure and application chain. It has the entry point, the tech stack, the cash buffer, and assets like Apollo Go and Kunlunxin that could change the valuation coordinates. But if I look only at today's price near 117 dollars, I do not think it gives new buyers an adequate margin of safety. In other words, Baidu is worth studying and worth tracking for the long term, but is not necessarily worth owning heavily at this price and this moment. My main judgment is therefore to reject the urge of "must buy now," rather than to reject the company.

As long as two of three conditions hold, I would be more willing to change my view: first, a clear narrowing of the online-marketing decline; second, AI cloud growth sustaining high momentum for at least two more quarters while margins do not deteriorate; third, at least one of Kunlunxin or Apollo Go genuinely pushing the option toward value realization. If none of these three happens and the share price rises only because the market moves first on sentiment, I would instead be more cautious.

【Company Profile Scores】

• Fundamental quality: Medium

• Growth: Medium

• Moat: Medium

• Financial soundness: Strong

• Management credibility: Medium

• Valuation attractiveness: Medium

• Risk level: Medium

• Suitable investor type: long-term growth

【Investment Rating】

• Rating: Watch

• One-line investment thesis: the AI cloud is accelerating, but the advertising decline and unproven profit conversion have yet to deliver an adequate margin of safety.

• Three-tier price signals: 【Ideal Buy Price】60-74 USD Basis: corresponds to at least a 20% margin of safety against my conservative-case fair value of 74 dollars.

• Acceptable holding price: 99-133 USD

• Clearly overvalued price: 186-205 USD

• Current price classification: acceptable to hold

• Worth waiting for a better price: yes. The specific level that triggers a buy is below 74 dollars; if the online-marketing decline narrows and AI cloud sustains high growth at the same time, the 74-85 dollar range could also be reassessed. The opportunity cost of waiting is possibly missing an event-driven upgrade from Kunlunxin or Robotaxi.

• Target holding period: 3-5 years

• Expected annualized return: conservative -14% to -8%; neutral 0% to 5%; optimistic 12% to 18%

• Maximum loss risk: if advertising keeps declining double digits, AI cloud margins deteriorate, and Robotaxi and the spin-off do not deliver, the share price could fall to 55-65 USD, a drawdown of about 45%-53% from the current price

• Signals that trigger a re-assessment: Online marketing revenue year over year below -15% for two consecutive quarters

• AI Cloud Infra year-over-year growth breaking below 30%

• Non-GAAP operating margin below 10% for two consecutive quarters

• Operating cash flow turning negative again

• Kunlunxin spin-off showing no substantive progress within the next 12 months, or a more substantive US restriction escalating

【Valuation Range】

• current: 117.49 (as of the 2026-06-11 close)

• bear (conservative · ideal buy zone): [60, 74]

• base (fair · acceptable holding zone): [99, 133]

• bull (optimistic · above the clearly-overvalued line): [186, 205]

Key Data Table

Table note: data from Baidu's 2025 annual report, Q4 2025 and Q1 2026 announcements, and the spin-off announcement.

Research Uncertainties

• I have not obtained Kunlunxin's formal prospectus after the confidential filing, so I cannot fully fold its revenue, losses, external-sales share and customer structure into the SOTP; at this stage it can only be treated as "an option with limited visibility," not an asset that can already be precisely quantified.

• Baidu has not publicly broken out maintenance capex versus expansion capex in recent quarters, so the owner-earnings lens can only make conservative inferences and cannot be modeled as precisely as for an industrial company.

• The discussion of upgrading the Hong Kong shares to a primary listing currently comes mainly from media reports and lacks a formal company announcement, which limits the certainty of valuing the Southbound-capital liquidity dividend.

• The unit-economics improvement of Robotaxi has been verbally stressed by management, but city-level profit and vehicle-utilization data remain insufficient, so outside investors cannot precisely re-verify it for now.

• The CMC List is not currently a sanction, but its subsequent legal and capital-level spillover effects still carry a time lag and uncertainty.

References

The core primary materials come mainly from Baidu's investor-relations and annual-report system, including the 2025 Form 20-F, the Q4 2025 and Q1 2026 results announcements, the Kunlunxin spin-off announcement, the latest CMC List response, and management and governance documents. The capital-markets institutional and Southbound Stock Connect portions draw mainly on official explanations from HKEX and the SFC. The industry and competitor portions draw mainly on Omdia, IDC citations, Reuters, Waymo's website, Alibaba Cloud and Tencent Cloud official disclosures, and Statcounter's search-share data.

Other Tickers Mentioned in This Report

• GOOGL.US — the most important valuation reference for the global "search cash cow + Waymo option," illustrating why the market grants a different premium even when both have Robotaxi.

• BABA.US — the most direct leader reference for China's AI cloud infrastructure, and the strongest publicly comparable rival to Baidu's cloud.

• 00700.HK — Tencent Cloud constitutes long-term pressure on the enterprise-customer and ecosystem side, representing the "ecosystem-type AI cloud" path.

• TSLA.US — the company whose Robotaxi long-dated option capital markets price most aggressively, the upper-bound reference in comparing Baidu's autonomous-driving valuation.

• PONY.US — one of the listed pure-Robotaxi names, used to observe how the market prices a single autonomous-driving story.

• WRD.US — another listed pure-Robotaxi name, used to observe overseas expansion, regulatory and operational risk.

• TCOM.US — an important strategic-holding and financing-structure reference on Baidu's balance sheet, affecting the asset-discount judgment in the group SOTP.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.