Baidu’s U.S.-listed shares fell Wednesday after the Chinese search giant posted a slight revenue miss and confirmed a steeper-than-expected hit in its core ads business, even as AI-driven sales rose and free cash flow swung negative. The company reported second-quarter revenue of 32.71 billion yuan, just shy of estimates, down 4% year over year, with core online advertising down 15% and now weighing on roughly 60% of its sales base. Non-ad revenue rose 34%, led by AI cloud and autonomous initiatives, but the growth failed to offset the ad shortfall. “In the second quarter, our AI cloud registered solid, healthy revenue growth,” CEO Robin Li said, citing broader AI capabilities and products. Investors cheered the AI message in premarket trading, then faded as the costs came into focus, highlighting a familiar question for Chinese tech: can AI scale fast enough to stabilize profit and revive the multiple while the economy drags on ads.
The ad slide is the headline risk. Weak consumer demand, a property market slump, and elevated youth unemployment in China continue to compress marketing budgets. For a company still anchored to search and display inventory, a 15% drop in core ad revenue is a body blow and a reminder that cyclical pressure is now structural until growth restarts. Baidu has overhauled its search experience for the first time in a decade, pushing more generative answers and AI-native results. That could defend engagement, but monetization lags: generative outputs reduce ad slots and require new ad formats and pricing to catch up. The shift is necessary, but in the near term it dilutes yield per query. If ad budgets stabilize into year-end, Baidu’s revamped search could claw back revenue; if macro deterioration persists, the core stays under pressure and AI must do more of the heavy lifting sooner.
Non-ad revenue rose 34% as Baidu leaned into AI infrastructure and services. Management cited a 10% year-over-year increase in AI Cloud revenue, helped by Qianfan, its model-as-a-service platform, and the release of PaddlePaddle 3.0. A redesigned search interface and the launch of MuseSteamer, Baidu’s AI video generator, deepen the product stack. The company also open-sourced a version of its Ernie model to seed an ecosystem and blunt the momentum of rivals like DeepSeek. The strategy is orthodox: expand developer tools, embed LLMs into enterprise workflows, and sell compute, models, and managed services. The constraint is time and margin. Training and inference are capital intensive, chip supply is tight under export controls, and China’s AI buyers are price sensitive. Alibaba Cloud and Tencent Cloud are pushing aggressively with their own models and partnerships. The result is growth that looks solid on paper but does not yet move the consolidated needle fast enough to counter a double-digit ad decline.
Free cash flow was negative 4.7 billion yuan in the quarter, primarily due to incremental investment in AI. That is both rational and uncomfortable. Rational because early winners in AI infrastructure and model platforms can lock in multi-year share via switching costs and developer gravity; uncomfortable because investors now have to underwrite lower near-term returns while visibility on unit economics remains cloudy. AI cloud gross margins tend to be thinner than software, at least at first, and model inference costs scale with usage unless offset by strong pricing or custom silicon. Baidu is optimizing stack efficiency and localizing supply, but the GPU arms race is a cost line, not a footnote. The company is effectively asking the market to trade current cash for optionality. That is a familiar pitch in U.S. big tech, but in China’s uneven recovery and with AI competition intensifying, the bar for patience is higher.
Beyond cloud, Baidu continues to lean into autonomous driving under the Apollo banner and richer AI-native media tools. These are credible adjacencies to search and maps and potentially powerful moats if regulation and urban partnerships keep lining up. Monetization, however, is staged: robotaxis and self-driving platforms offer long-dated revenue, not near-term P&L relief. The MuseSteamer video generator and AI search can open new ad formats and subscription opportunities, but they need user-scale adoption and clear pricing. Open-sourcing Ernie aims to seed community uptake and reduce friction for enterprise pilots. It also concedes that the center of gravity is shifting to ecosystems where model providers compete not just on raw benchmarks, but on tooling, cost, safety, and vertical solutions. The execution question is whether Baidu can turn these launches into recurring, high-attach revenue faster than ad headwinds erode legacy cash engines.
The macro overlay is doing real damage. China’s property malaise and soft consumer sentiment translate directly into fewer auction bids, lower cost-per-click, and advertisers delaying campaigns. On top of that, cloud customers are managing budgets tightly, pushing for value and performance guarantees. Meanwhile, Alibaba Cloud and Tencent Cloud are not waiting around; they are bundling AI with storage, security, and payments tie-ins, making it harder to pry away workloads. On the model side, homegrown challengers are iterating fast, while U.S. export controls keep pressure on the supply of advanced accelerators, raising costs and elongating deployment timelines. Baidu’s answer—optimize models, expand Qianfan, and deepen vertical offerings—is sensible. But competition compresses pricing power and makes the path to scale more execution-dependent. Against that backdrop, the 10% AI Cloud growth is a positive data point, not a thesis ender.
The stock’s intraday swing captured the split view. Pre-market strength on the AI narrative faded as negative free cash flow and the magnitude of the ad decline set in. The setup is now a tug of war between long-term AI optionality and near-term earnings pressure. Consensus wanted a cleaner beat and clearer trajectory on margins. Instead, investors got a small top-line miss, better non-ad growth, and a bigger cash burn. That combination usually compresses the multiple unless management can credibly guide to improving conversion of AI demand into higher-margin revenue. In the near term, the tape will trade on three signals: stabilization in ad trends, evidence that Qianfan bookings are accelerating, and any visibility on capex intensity peaking. Without that, rallies on product headlines may be sold as cash flow remains under strain.
Baidu does not need to win AI outright to win the stock. It needs to demonstrate operating leverage in AI services, show that generative search can monetize without cannibalizing too much ad yield, and slow the pace of cash burn. Concrete markers would include rising AI Cloud gross margin, a step-up in large enterprise contracts for Qianfan, and paid conversion for MuseSteamer or other AI tools. On the legacy side, ad decline needs to moderate from minus-15% to single-digit negatives as macro normalizes. If Baidu can deliver that while maintaining double-digit growth in AI Cloud and containing capex, the narrative shifts from defense to offense. If not, investors will continue to discount the AI story until it shows up in consolidated margins and cash.
Baidu’s quarter was a microcosm of China tech in 2025: the future is AI, but the present is ads and cash. With revenue a touch light, ads down hard, and free cash flow in the red, the company asked investors to trust the trajectory of its AI stack—Qianfan, PaddlePaddle, Ernie, and a redesigned search experience—against a tough macro and fierce competition from Alibaba and Tencent. That is a defendable strategy, but one that now requires proof of operating leverage. Until the company shows that AI growth can meaningfully offset ad cyclicality and fund itself, BIDU will trade on execution beats and misses, not on vision.
