Product1 distinct publisher3 min readUpdated
Revenue fell for a fifth straight quarter because a 19 percent ad decline on a bigger base beats 25 percent AI growth on a smaller one. Capex tripled meanwhile.
The Product Desk · Product desk

Compiled by The Product DeskSomething wrong?How this is made
Baidu reported second-quarter revenue of RMB31.3bn ($4.62bn), down 4 percent year on year and 2 percent on the prior quarter, its fifth consecutive quarterly decline [1]. The number worth studying is not the AI growth rate, which is high, but the subtraction underneath it: the AI segment grew 25 percent and still did not fill the hole the ad business dug [2][4].
Run it. Online marketing, the legacy mainstay, fell 19 percent to RMB13.1bn [2], which implies roughly RMB16.2bn a year earlier and about RMB3.1bn of revenue simply gone [1]. The Core AI-powered Business came in at RMB12.5bn, up 25 percent [4], implying a base near RMB10bn and a gain of about RMB2.5bn [2]. The new business therefore replaced roughly 80 percent of what the old one lost [3]. To hold revenue flat, that AI line would have needed to grow about 31 percent, not 25 [4]. Group revenue fell about RMB1.3bn [5]. Reuters attributed the ad decline to a weak Chinese property sector and cautious consumer spending, both of which have pushed companies to cut marketing budgets [3].
This is the standard shape for a portfolio funding a new bet out of a declining cash cow. Fast percentage growth on the smaller base loses to slower percentage decline on the bigger one until the two bases cross, and Baidu's have only just met: the AI segment is now half of general business revenue [4]. Until the crossover, headline growth rates flatter. GPU Cloud, the accelerator rental line, rose 283 percent, accelerating from 184 percent the quarter before [6], and Baidu renamed it this quarter from "subscription revenue from AI accelerator infrastructure" [7]. AI Cloud Infra rose 50 percent to RMB7.3bn [5], a gain of about RMB2.4bn that accounts for nearly all of the AI segment's increase [10].
The cost side is the other half of the arithmetic. Excluding iQIYI, capital expenditure tripled to RMB11.4bn from RMB3.78bn, Bloomberg reported [11]. The increase alone, about RMB7.6bn [8], is roughly three times the AI revenue gain it bought [9]. Capex ran about 3.4 times operating cash flow of RMB3.4bn [12][6] and about five times net income, which fell to RMB2.3bn from RMB7.3bn, a drop Bloomberg put at 68 percent [8][7]. Baidu has RMB283.1bn in total cash and investments to absorb that [12] and has returned $259mn through buybacks since January [13]. Tencent reported this month that its own AI compute bill outgrew its cash flow [14].
The spending has not yet bought a leading model. Ernie has gone months without a major upgrade, Reuters reported [15], and Bloomberg wrote that it now trails open-weight models such as Moonshot AI's, which match OpenAI and Anthropic on key benchmarks [16]. Chief executive Robin Li told analysts Baidu would return Ernie to the frontier, saying competitiveness "ultimately comes down to sustained technology investment, application-driven approach, and patience" [17]. The market was less patient: shares fell as much as 10 percent to $93.70 according to Bloomberg, with Reuters reporting a 7 percent early slide, against a Bloomberg-poll consensus of RMB31.6bn cited by the South China Morning Post [9][10]. Apollo Go, the robotaxi unit, is in 28 cities with more than 350mn autonomous kilometres logged [18].
Watch the two growth rates, not the segment mix. The ad decline has to decelerate or the AI line has to clear about 31 percent for group revenue to stop falling [4], and the capex-to-operating-cash-flow gap sets how long the crossing can be financed [6].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Baidu reported second-quarter revenue of RMB31.3bn ($4.62bn), down 4 percent year on year and 2 percent on the previous quarter, the fifth straight quarterly revenue decline.
Online marketing revenue, Baidu's traditional mainstay, fell 19 percent year on year to RMB13.1bn.
Reuters attributed the online marketing drop to a weak Chinese property sector and cautious consumer spending, both of which pushed companies to cut marketing budgets.
Baidu's Core AI-powered Business brought in RMB12.5bn, up 25 percent, and now accounts for half of the company's general business revenue.
Within the AI segment, AI Cloud Infra rose 50 percent to RMB7.3bn.
GPU Cloud revenue rose 283 percent year on year, accelerating from 184 percent growth the quarter before.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Company-reported figures, single-outlet aggregation
Every quantitative claim traces to Baidu's own quarterly disclosure, relayed by one publisher that attributes individual figures to Reuters, Bloomberg, SCMP and LSEG. The core numbers are specific, internally consistent and survive independent arithmetic checks (segment gains against group decline, capex against cash flow). Two weaknesses hold the score below the mid-70s: there is no second publisher in the cluster to corroborate framing, and the relayed inputs disagree with each other on the share move (10 percent vs 7 percent) and on consensus revenue (RMB31.6bn vs RMB31.96bn) without reconciliation. The most decision-relevant figure — AI segment profitability — is absent entirely.
Real revenue and fleet scale, growth rates without bases
Adoption is genuinely measurable here, not projected: AI Cloud Infra is a RMB7.3bn quarterly revenue line, the AI segment is RMB12.5bn and half of general business revenue, Apollo Go runs in 28 cities with over 350mn autonomous kilometres, and the Baidu App carries 644mn monthly actives. That is substantial deployed usage backed by paid revenue. The ceiling is set by disclosure quality: the fastest-moving line, GPU Cloud at 283 percent, has no absolute base and was renamed this quarter, so its contribution cannot be sized; and no segment margin is given, so it is unknown whether adoption is profitable or subsidised by the tripled capex.
Percentage framing outruns absolute contribution
Modestly positive. The overstatement lies in the framing Baidu itself supplies — AI as core growth driver, an identity shift, GPU Cloud at 283 percent — which is expressed in percentages off small bases while the absolute arithmetic goes the other way: RMB2.5bn of AI gain against roughly RMB3.1bn of advertising loss, covering only about 81 percent, with capex rising about RMB7.6bn to buy that RMB2.5bn. Robin Li's pledge to return Ernie to the frontier is stated without a roadmap while the model reportedly trails open-weight rivals. The gap is not larger because the cluster's own coverage performs this arithmetic explicitly, prices the shortfall, quotes an analyst doubting the pivot, and does not repeat the promotional frame uncritically.
Issuer-shaped disclosure with visible reporting changes
Incentives are documented rather than inferred. Baidu restructured reporting in late 2025 to break out AI businesses, renamed its fastest-growing line to GPU Cloud this quarter, and highlights the AI segment crossing half of core sales — all choices that flatter the pivot narrative while the group posts a fifth straight revenue decline and a 68 percent profit drop. Capital-markets pressure compounds it: a dual-primary Hong Kong listing due this year, a Kunlunxin spin-off pitched on demand for domestic Nvidia alternatives, $259mn of buybacks and a first dividend. Countervailing signals are present in the source — a named Bloomberg Intelligence skeptic, the consensus miss, and Tencent's comparable compute-versus-cash-flow squeeze — which is why this is not scored higher.
Backward-looking numbers solid, forward pivot unresolved
Confidence is moderate-to-good on what already happened: the reported quarter's revenue, segment mix, profit, capex and cash position are specific, arithmetically coherent and attributed to established wires. Confidence drops on interpretation, for three reasons the source itself names — AI segment profitability is undisclosed, the GPU Cloud growth rate has no absolute base after a line rename, and whether upgraded Ernie can close the benchmark gap is unanswered. The cluster also rests on a single publisher, so framing cannot be triangulated, and its inputs conflict on two secondary figures.
invest
Baidu's AI revenue grew 25%. Its ad business lost more than AI gained.1 distinct publisher
invest
Google Ships Flash Instead of Pro While OpenAI Loses Its Two Best Operators1 distinct publisher
product
Alibaba's 75% profit drop is what a working cloud business costs1 distinct publisher
build
China's accelerator swap makes Cambricon supply, not export policy, your ship-date risk1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 18, 2026