Baidu (BIDU) · Deep analysis

Report previewBIDU.USReport date Last updated
← Back to reports
Report contents
01

Business model

Business model: an ad seller turning compute landlord

Baidu sells to two customers who barely resemble each other. Chinese advertisers bid for search and feed placements and pay per click or impression; that auction still pays most of the bills. Enterprises buy compute and model access, paying per GPU-hour or per token (the unit in which model output is billed): AI Cloud Infra revenue reached RMB 8.2 billion in the quarter ended September 30, 2026, up 94% from a year earlier, and GPU Cloud inside it grew 191%, Baidu reported on November 16, 2026. Consumers pay with attention, not cash. The enterprise customer now sets the growth rate; the advertiser still sets the profit.

That makes Baidu a platform in transition, with a mature advertising marketplace that is shrinking attached to an AI infrastructure business that is expanding fast. Search and feed advertising, AI Cloud Infra (Baidu's name for its cloud infrastructure line), the ERNIE model family and the applications built on it, and Apollo Go robotaxis all sit under one roof, but two of them move the numbers. The rest is optionality, and on the published record it is option value without disclosed revenue or margins.

The mix has flipped before the profit has

Baidu Core's AI-powered Business accounted for 53% of Baidu General Business revenue in the third quarter, the company said, while General Business revenue returned to growth at 1% year over year. Put those together and the remainder has to be shrinking. A segment growing 94% alongside a total growing 1% means the rest, essentially online marketing, is still contracting, and management's own reference to headwinds in online marketing says as much. The announcement does not isolate how much advertising fell, so the pace of erosion is an inference, not a disclosure.

The 53% figure needs care. "AI-powered Business" is a management category rather than an audited segment, and it plausibly includes AI-assisted advertising products. A majority of revenue touching AI does not show that Baidu has replaced its dependence on advertising; it shows that the label now stretches across the product line.

Compute sells in volume and bills in capital

What has genuinely changed is the contract. Advertising revenue arrives ahead of the cost of serving it, and the marginal cost of one more click sits near zero. GPU cloud is capacity rental: Baidu must buy and install accelerators before it can bill an hour of compute, so revenue trails capital spending by a quarter or more, and installed capacity sits idle if demand pauses. Baidu does not disclose AI Cloud Infra operating margin or the capex attributed to the segment, so the incremental return on that build cannot be calculated from reported results. The CFO's statement that AI Cloud Infra contributes more to both revenue and profit, and that capital allocation will stay disciplined, is guidance; the margin itself is not published.

Funding is where this bites. Operating cash flow was RMB 2.8 billion in the quarter, a group figure that leans on advertising. Baidu has returned $373 million to shareholders through repurchases since the start of 2026 and expects to declare a first dividend before the end of 2026. The advertising base is therefore being asked to fund both the compute build and shareholder returns. If advertising cash falls faster than cloud cash arrives, the transition gets financed from the balance sheet or from slower returns, not from operations.

Bargaining power has moved to the buyer

In a search auction Baidu sets the rules and advertisers compete for position. In capacity rental the buyer sets terms, because compute is broadly comparable across suppliers and a handful of large AI developers can negotiate hard or build their own clusters. Baidu describes demand as broadening without publishing concentration figures. Its November 2026 Token Exchange, a marketplace matching model providers with customers, is an attempt to own the distribution layer instead of only renting hardware. A deliberately neutral exchange, though, makes model access easier to compare and switch, which pulls pricing toward cost.

Consumer distribution is the older moat and the more exposed one. Baidu's search box aggregated query intent and sold access to it; AI assistants answer directly and shorten the path where ad inventory lives. Baidu's response is its own AI application portfolio, which it says it is expanding, but it has not disclosed paying users or subscription revenue for those products, so the economics of that defense remain unproven.

The judgment that matters: growth now comes from a capital-intensive, buyer-favored business replacing a capital-light, seller-favored one, and the profit pool has not been shown to survive the swap. Evidence that would overturn this is specific. If Baidu starts reporting an AI Cloud Infra segment margin that widens while capex grows more slowly than cloud revenue, and online marketing revenue turns positive, the mix shift is creating value. One quarter of 1% General Business growth does not establish that turn.

Questions worth following

  • Will AI cloud GPU rental demand fade before capacity is filled?

    AI cloud is the offset to advertising and the largest piece of DBS's sum-of-parts value. If GPU rental demand or pricing softens while Baidu adds capacity, depreciation and power costs land on a smaller revenue base. A growth story can become a return-on-capital problem.

  • Will search advertising decline outpace AI cloud growth?

    Search ads still supply most of Baidu's profit. If high-margin ad revenue falls faster than cloud revenue grows, group earnings and cash flow stay below prior levels even as AI revenue rises. That would weaken the sum-of-parts case for the whole company.

  • Are Chinese users shifting search queries to rival AI assistants?

    Baidu's ad inventory depends on owning the query. If consumers move intent to rival AI assistants, both the ad pool and the distribution Baidu uses to launch products weaken. That would turn the cloud build into a substitute for a shrinking base instead of an addition to it.

02

Financial quality

Financial quality: a solid balance sheet, weak returns

Baidu is a mature, barely growing search business paying for an early-stage AI infrastructure build. Solvency is not the issue; the return on a very large asset base is. Return on equity fell to 2.0% in FY2025 from 8.7% a year earlier, and return on assets to 1.3% from 5.7%. Estimating ROIC as EBIT taxed at 20% over equity plus debt less cash gives roughly 2%, down from about 7% in FY2024 and below any reasonable cost of capital. Equity of CNY 266 billion now supports CNY 5.6 billion of net income.

Margins compressed before the reinvestment peak

Gross margin slid from 51.7% in FY2023 to 50.3%, then 43.9% in FY2025 and 39.0% in the June 2026 quarter, 12.7 percentage points off the peak. Revenue gave no offset: CNY 129.1 billion in FY2025 was below FY2023's CNY 134.6 billion, a four-year compound rate near 1%. The cause is mix and cost. AI cloud infrastructure rental carries GPU depreciation and power that search advertising does not, and model-training spend sits above the gross line. Operating income halved to CNY 10.4 billion. DBS forecasts AI Cloud Infrastructure revenue growth of 60% in FY2026 and puts the earnings trough in FY2026, which means mix-driven dilution continues while the build runs.

Cash conversion broke in fiscal 2025

Operating cash flow was negative CNY 3.0 billion in FY2025 against positive net income of CNY 5.6 billion. The conversion ratio had run at 3.5x in FY2022 and 0.9x in FY2024, so the swing is large enough that working capital and changes in cloud billing terms both need checking in the annual statement. Capex, implied by operating cash flow less free cash flow, was CNY 13.4 billion for FY2025 and CNY 11.4 billion in the June 2026 quarter alone, about 36% of that quarter's revenue against roughly 6% in FY2024. Free cash flow was -CNY 16.4 billion and -CNY 8.0 billion respectively. Long-term debt rose from CNY 61.1 billion at end-2025 to CNY 73.5 billion at June 2026: borrowing is funding part of the build.

Leverage stays comfortable on conventional measures. Liabilities are about 37% of assets, the current ratio is 2.33 and the quick ratio 2.23, and EBIT covered interest 4.9 times in the June quarter, down from 11.2 times in FY2024. Cash of CNY 24.7 billion is a small slice of CNY 207.4 billion of current assets; the remainder is largely short-term investments and other financial assets Baidu does not itemize, which likely leaves the group near net cash. Refinancing is not the risk. Whether the capital being sunk earns anything is, and a ROIC still near 2% in two years would settle that negatively.

What needs checking

The third quarter of 2025 lost CNY 11.2 billion while operating income was positive. EBIT of -CNY 12.5 billion against operating income of CNY 1.1 billion implies a non-operating charge of roughly CNY 13.6 billion. Goodwill also rose from CNY 22.6 billion at end-2024 to CNY 37.6 billion during 2025. The filing needs to show whether the charge was an impairment of an investment or an acquisition-related item, and how much carrying value remains exposed to a repeat.

Negative operating cash flow alongside positive earnings. Receivable days held near 35, so collections are not the leak. Payables, accrued marketing, or movements in Baidu's consumer credit book would reverse naturally; a permanent change in deferred revenue would not. Another year below 0.8x conversion would settle it as structural.

Inventory is small but growing fast. The balance rose from CNY 1.4 billion at end-2023 to CNY 6.0 billion a year later and CNY 8.3 billion at June 2026, with turnover down from about 49x to 12x. Composition is not disclosed; if this is accelerator stockpiling, a demand miss would force write-downs on top of already thin margins.

Metric (CNY bn unless noted)FY2021FY2022FY2023FY2024FY2025Q2 2026 (3 mo)Assessment
Revenue124.5123.7134.6133.1129.131.2Weak
Gross margin (%)48.348.351.750.343.939.0Weak
Operating margin (%)8.412.916.216.08.09.6Weak
Net margin (%)8.26.115.117.84.37.4Weak
ROE (%)4.93.38.18.72.01.5Very weak
ROA (%)2.92.05.15.71.30.9Very weak
Operating cash flow20.126.236.621.2-3.03.4Weak
Capex, implied10.98.411.38.113.411.4Weak
Free cash flow9.217.825.313.1-16.4-8.0Very weak
OCF / net income (x)2.03.51.80.9negative1.5Weak
EBIT interest cover (x)4.24.58.811.23.44.9Adequate
Total liabilities / assets (%)41.139.235.433.735.537.2Adequate
Current ratio (x)2.862.673.012.091.762.33Strong
Metric (CNY bn unless noted)Q1 2025Q2 2025Q3 2025Q4 2025Q1 2026Q2 2026AssessmentTrend
Revenue32.532.731.232.732.131.2WeakFlat
Gross margin (%)46.143.941.244.238.939.0WeakDeteriorating
Net income7.77.3-11.21.83.52.3WeakDeteriorating
Net margin (%)23.822.4-36.05.410.77.4WeakDeteriorating
Total assets454.0450.6444.1449.2n/c470.1AdequateStable
Total liabilities163.9152.7156.5159.4n/c174.9AdequateRising
Operating cash flow-6.00-0.881.262.612.673.44WeakImproving
Capex, implied2.903.803.412.525.9211.4Very weakRising
Free cash flow-8.90-4.68-2.150.09-3.25-7.95Very weakDeteriorating
OCF / net income (x)negativenegativenegative1.50.81.5WeakMixed

Capex is implied by operating cash flow less free cash flow, since net PP&E is not disclosed for these quarters. The March 2026 balance sheet is reported in US dollars, so it is marked not comparable rather than converted at an assumed rate.

Questions worth following

  • Will Baidu's AI capex earn a return above its cost of capital?

    Baidu is spending on data-center capacity faster than its legacy business generates cash. Whether that capacity earns more than the capital tied up in it decides if returns recover or the equity base keeps growing faster than earnings.

  • Can search advertising still fund Baidu's AI build?

    Advertising remains the cash engine behind Baidu Core, and the pace of its decline sets how much of the capex program must be financed by debt or asset sales. A slower ad decline reduces external funding needs and protects the balance sheet.

  • Will Baidu's reported profit start converting into cash again?

    Cash conversion has fallen from three times earnings to negative while capex rises. Until profits turn back into cash, the AI build depends on the balance sheet, and further deterioration would raise the cost of funding it.

Generated by DeepVal AI. Data and models may contain errors or be outdated. For research only; not personalized investment advice.

Go beyond the preview

Continue reading the full report

  1. Key risksWhat could break the investment case
  2. Valuation and margin of safetyWhat the current price assumes, and how much room for error it leaves
  3. SummaryOur overall judgement and what to watch next

Free to sign up · 20 research points · unlocking this report uses 10

20 points cover 2 full reports. No card needed.

Research reflects its generation date. Preview the first two modules.

More on Baidu Inc