BICompany report
Baidu, Inc. BIDU
The bet you're really making is that Baidu's search business, the box a billion people have typed questions into for twenty years, keeps shrinking as AI chatbots hand back an answer instead of a list of links, and that Baidu's newer business, renting out AI computing power, grows fast enough to fill the hole. You're betting the cloud arm, up 50% last quarter with the GPU-rental slice nearly quadrupling, becomes the engine before the old one stalls. Right now it is going the wrong way: full-year profit fell 75% in 2025 to $0.8 billion, the operating business swung to a loss, and last quarter's earnings landed 30% below what analysts expected. You pay about 15 times next year's expected earnings and less than the company's net worth on paper, near the cheapest the stock has been in the twelve years it has traded.
Key data
BIDU · price with moving averages
Source: market data.
The business
Baidu runs China's largest search engine, and for two decades the ads beside those blue links paid for everything. That business is now flat to shrinking: total revenue has sat between $17.9B and $19.5B for five straight years while AI chatbots teach people to expect an answer, not a list. Around the stalled core Baidu is building two things. The first is AI Cloud Infra, renting computing power and GPUs to companies training their own models: RMB7.3B in the second quarter, up 50% year over year, with the GPU-rental slice up 283%, the fastest growth in the company. The second is Apollo Go, the robotaxi fleet already carrying paying riders in Chinese cities, which the market values at roughly zero. Founder Robin Li Yanhong reportedly controls the vote through Class B shares, so the direction is his to set. The whole case is whether the new engines spin up before the old one runs dry.
The numbers
The earnings line tells the turn. Baidu beat expectations comfortably through late 2025, then missed in 2026, and last quarter missed hard.
| Quarter | Revenue | Net income | Diluted EPS |
|---|---|---|---|
| Q2 2025 | RMB32.7B | RMB7.32B | RMB20.32 |
| Q3 2025 | RMB31.2B | -RMB11.2B | -RMB33.84 |
| Q4 2025 | RMB32.7B | RMB1.78B | RMB2.48 |
| Q1 2026 | RMB32.1B | RMB3.44B | RMB9.92 |
| Q2 2026 | RMB31.2B | RMB2.31B | RMB5.76 |
The $1.06 print was the worst miss in the set, and it was not a one-off. The full-year record underneath shows where the profit engine broke.
| Fiscal year | Revenue | Net income | Diluted EPS |
|---|---|---|---|
| 2021 | RMB124.5B | RMB10.2B | RMB28.08 |
| 2022 | RMB123.7B | RMB7.56B | RMB19.84 |
| 2023 | RMB134.6B | RMB20.3B | RMB55.12 |
| 2024 | RMB133.1B | RMB23.8B | RMB65.92 |
| 2025 | RMB129.1B | RMB5.59B | RMB11.76 |
| 2026, 1H to Jun | RMB63.3B | RMB5.76B | RMB15.68 |
Revenue is dead flat across five years. What changed is the middle column: operating income climbed to $3.1B in 2023, then collapsed to an $833M loss in 2025, and net income fell 75% to $799M. Operating cash flow went with it, from positive $2.9B in 2024 to negative $431M in 2025. Flat revenue plus a swing to operating loss means the damage is cost and impairment, not just the top line, and that is the fact this cheap-on-paper stock explains least well. Q2 2026 was a small step back into the black on a reported basis, RMB2.3B ($342M) of net income at a 10% operating margin, but the trailing twelve months still carry a loss. Half-year revenue and operating income were not disclosed.
The variant is simple. The market prices Baidu as a melting search asset: below its own book value, about 15 times next year's earnings, near its twelve-year floor. What it pays almost nothing for is a cloud business now annualizing toward $4B and growing 50%. If that arm keeps compounding and core margins stop bleeding, the parts are worth more than book. The print that settles it is two numbers moving together: AI Cloud growth holding above 40% and Baidu Core operating margin turning back up. Until both do, cheap is a claim, not a fact.
Management
The record is mixed and the capital allocation is honest. Insiders bought nothing in the last year and sold once, a $2.0M disposal in May by an insider named Foo Jixun, plan status not disclosed, too small to read either way. The better signal is the buyback: Baidu has returned $259M to holders since the start of 2026 and repurchased $792M of stock in 2025, buying below book, which is accretive if the book is real. Pay is tied to margin, EPS and total shareholder return, none of which cooperated last year. The share structure carries dual-class voting rights that concentrate control in the founder's hands, for better and worse.
How it fails or surprises you
Search keeps bleeding. Chinese users are moving to AI chat for the queries that used to carry ad load, and Baidu's core online marketing revenue has been flat to down for years. If core ad revenue keeps falling faster than cloud fills in, the flat top line turns to outright decline. Watch Baidu Core ad revenue year over year; another down quarter confirms the melt.
The cloud inflection (right tail). AI Cloud Infra grew 50% and GPU Cloud 283% off a base near $4B annualized, and the market gives it almost no credit. If it holds above 40% growth for several more quarters and reaches a quarter of Core revenue at positive incremental margin, the sum-of-parts re-rates well above today's below-book price. Watch cloud revenue and its margin.
Cash burn from AI capex. Operating cash flow turned negative in 2025 and free cash flow yield is about -9%, with capital spending running near three times operating cash flow as Baidu buys chips. If the spend keeps outrunning the cash the cloud throws off, debt rises and the buyback shrinks. Watch quarterly operating cash flow against capex.
Closing thoughts
A specific print settles this one, not survival. The company is buying its own shares below book, so Baidu is not going away. The question is whether two lines cross in time: cloud growth staying high while core margin recovers. A clean quarter shows both moving up together; an ambiguous one shows cloud strong but margin still sinking, which is where you wait rather than add. The fatter tail today is the downside. Flat revenue, an operating loss, and negative free cash flow are facts, while the cloud re-rating is a possibility, and possibilities are worth less than facts even when they are large. What is at risk if search bleeds and capex burns is years of dead money below book; what the upside is worth is a double if the parts get repriced.
The bet is still that search keeps shrinking as chatbots answer the questions people used to type into the box, and that Baidu's business renting AI computing power grows fast enough to fill the hole before it does. What breaks it is the two numbers that matter drifting apart: AI Cloud growth slipping below 40% while Baidu Core operating margin stays negative. Watch those two together, quarter by quarter. If cloud decelerates and the core stays in the red, the discount to book is not a bargain, it is the market being right.
Methodology
The year-to-date row is the sum of the 2 reported quarters of the current fiscal year, diluted EPS included; the five-quarter and five-year tables are the vendor income statements.
Baidu reports in RMB; filed figures here are the company's own USD conversions (about RMB7.2 to the dollar), forward consensus is in RMB per ADS converted at the same rate. Trailing P/E is not meaningful on a trailing-twelve-month loss; valuation is framed on forward earnings and price to book (0.82x, near the 12-year low).
Data gaps: the pack carries per-quarter EPS actuals but not per-quarter revenue or Q1 2026 GAAP net income, so 1H 2026 revenue and operating income are not disclosed in the filing; AI Cloud figures are a single Q2 2026 point (RMB7.3B, +50% YoY; GPU Cloud +283%), not a sequence; Apollo Go paid-ride counts not disclosed at the release level.
Segment structure is Baidu Core and iQIYI, with AI Cloud and Apollo Go disclosed inside Core; AI Cloud revenue is reported (RMB7.3B Q2), online marketing detail is not.
Sources: Baidu Q2 2026 6-K (results filed 2026-08-18) and subsequent 6-K filings through 2026-09-04; FMP market, consensus, insider and as-filed XBRL series, pull dated 2026-09-06.
Fact check: Corrected three items: removed erroneous "net cash" claim (company carries net debt per 2025 balance sheet: $5.9B long-term debt vs $3.5B cash); softened unverified governance claim to "reportedly controls"; changed "officer" to "insider" for Foo Jixun (role not specified in evidence). All numerical claims reconciled to filed XBRL (annual revenue/income series, operating cash flow, capex, buybacks) and Q2 2026 earnings release (cloud growth, GPU growth, quarterly results). Final analysis verified as of Sep 6, 2026.
Bid Cap
Daily ideas, a 390-name database, and a model long/short book from an investor who mostly covers financials. $70 a month or $700 a year.
Subscribe on Substack


