Baidu's AI revenue grew 25%. Its ad business lost more than AI gained.

Baidu reported second-quarter revenue of 31.3 billion yuan ($4.62 billion) on Tuesday, August 18, down 4% year on year and below the roughly 31.96 billion yuan analysts at LSEG had expected ^{[1] [2]}. The miss itself is small, about 2% ^[7]; the composition is the story, because AI revenue rose 25% to 12.5 billion yuan while online marketing fell 19% to 13.1 billion yuan ^{[3] [4]}.

Do the subtraction in absolute terms rather than percentages. A 19% decline to 13.1 billion yuan implies a prior-year base near 16.2 billion, so the advertising line shed roughly 3.1 billion yuan of quarterly revenue ^[2]. A 25% rise to 12.5 billion implies a base of 10.0 billion, so AI added roughly 2.5 billion ^[3]. The new business therefore covered about 81% of what the old one lost ^[4], and the residual is the 1.3 billion yuan of total revenue that went missing ^[1]. AI is now close to 40% of the company ^[5], which is a real number, and it is still not a large enough base for 25% growth to carry a legacy line falling at 19%.

Inside the AI bucket, the growth is concentrated in one place. AI Cloud Infra revenue rose 50% to 7.3 billion yuan ^[5], an increase of about 2.4 billion, which is roughly 97% of the entire AI gain ^[6]. GPU Cloud within that grew 283% year on year, accelerating from 184% the prior quarter, according to Baidu's earnings statement ^[6]. Everything else was static: AI application revenue grew 3% to 2.5 billion yuan and AI marketing services were flat at 2.6 billion ^{[7] [8]}. Note also that Baidu's definition of AI revenue includes marketing services ^[9], so the AI and advertising disclosures are not cleanly independent of each other.

The margin cost of buying that growth is visible. Net income fell 68% to 2.3 billion yuan ($324 million) ^[10], diluted earnings per ADS dropped almost 72% to $0.85 ^[11], and operating income was $446 million ^[12]. Bank of America's Miranda Zhuang kept a Buy rating but cut her price target to $165 from $180, citing falling advertising revenue and higher AI infrastructure spending ^[13]. Chief executive Robin Li said the momentum in the AI business "reaffirms Baidu's transition from an internet-centric company to an AI-first company" ^[14]. The transition is happening; it is currently a swap of high-margin search yuan for capital-intensive compute yuan.

The forward arithmetic is more favourable than the trailing arithmetic. Hold both rates constant for another year and AI adds about 3.1 billion yuan while advertising loses about 2.5 billion ^[8], which flips the sign. That depends on GPU Cloud compounding at triple digits off a base that is now 50% larger, and on the ad decline not steepening. Neither is guaranteed: management attributed the marketing weakness to China's property slump, soft consumer demand, and e-commerce platforms redirecting 618 festival budgets into user subsidies rather than search and feed traffic ^[15].

Watch three things. Operating cash flow stayed positive for a fourth consecutive quarter at 3.4 billion yuan, per CFO Haijian He, and the company expects to complete a dual-primary Hong Kong listing this year ^{[16] [17]}; that listing is the funding context for continued infrastructure spend, alongside $259 million of buybacks since the start of the first quarter ^[18]. Apollo Go added open-road testing in London with Uber and Lyft, driverless commercial rides in Dubai, Hong Kong's first driverless testing permits, and Swiss tests with PostBus ^[19], none of which yet appears as revenue. And the market reaction was mild relative to the earnings drop: shares traded near $103.67 before the release, down about 28% for the year, then fell between 3.5% and 4.35% pre-market ^{[20] [21]}.