Baidu is entering a critical phase of its artificial intelligence transition, with CEO Robin Li promising to restore the company’s Ernie model to the frontier of AI even as the deterioration of its traditional advertising business weighs heavily on revenue and profit.
The Chinese search and technology giant reported second-quarter revenue of 31.33 billion yuan ($4.7 billion), down 4% from a year earlier and below the 31.96 billion yuan expected by analysts, according to LSEG data. Its U.S.-listed shares fell sharply after the results.
The numbers expose the central tension in Baidu’s strategy: the company’s AI businesses are growing rapidly, but not yet fast enough to compensate for the decline of the advertising franchise that built its business.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Online marketing revenue fell 19% to 13.1 billion yuan in the quarter as weak consumer spending and the prolonged property downturn encouraged companies to reduce marketing budgets. The weakness shows how exposed Baidu remains to China’s broader economic slowdown, particularly because advertising demand tends to weaken when businesses become more cautious about spending.
At the same time, Baidu’s AI-powered businesses are becoming an increasingly important part of the company. Revenue from Baidu Core AI-powered Business reached 12.5 billion yuan, up 25% from a year earlier and equivalent to roughly half of Baidu General Business revenue. AI Cloud Infrastructure revenue rose 50% to 7.3 billion yuan, while GPU Cloud revenue surged 283%, accelerating from 184% growth in the first quarter.
That GPU growth is notable because it indicates that Baidu is benefiting from the broader AI infrastructure spending cycle even while its consumer-facing businesses struggle. Companies are now purchasing computing capacity for model training and inference, creating a potentially significant new revenue stream for Baidu’s cloud operations.
The shift also changes the nature of Baidu’s AI opportunity. The company does not necessarily have to win the chatbot market outright for its AI strategy to succeed. It can monetize AI through cloud computing, model services, enterprise applications, advertising products and other infrastructure.
But Ernie remains strategically important because the strength of Baidu’s foundation models can determine how much of that ecosystem the company controls.
That is where the company faces a growing problem.
Ernie has gone months without a major upgrade, while competitors such as Alibaba and Moonshot AI have continued to introduce newer models. China’s AI market has moved rapidly toward open-weight models, capable reasoning systems and AI agents, raising the standard that Baidu must meet to remain a leading model provider.
Li’s response was unusually direct. He told analysts that Baidu intends to bring Ernie back to the AI frontier and continue investing in leading talent and technology.
“In a market like this, we believe long-term competitiveness ultimately comes down to sustained technology investment, application-driven approach, and patience,” Li said.
That statement points to a potentially expensive second phase of Baidu’s AI strategy.
The company has already been investing heavily in computing infrastructure and personnel. Its capital expenditure increased sharply in the quarter, with spending on AI chips and data centers adding to the financial burden of the transition.
But it comes at a difficult time. Baidu is being asked to spend more on AI precisely when its legacy business is producing less cash and its profitability is deteriorating. Net income fell to about 2.3 billion yuan from 7.3 billion yuan a year earlier. The decline illustrates the cost of maintaining an expensive AI infrastructure push while the advertising engine contracts.
Still, Baidu has financial resources to sustain the investment. The company reported 283.1 billion yuan in cash and investments at the end of June and generated 3.4 billion yuan in operating cash flow during the quarter.
The more important question is therefore not whether Baidu can afford to invest in AI. It is whether those investments can generate returns quickly enough to offset the structural deterioration of its older businesses.
This matters because China’s AI competition is becoming less forgiving.
Alibaba has been expanding its Qwen model family and pushing open-weight AI, while DeepSeek, Moonshot AI and other Chinese developers have been releasing models that compete aggressively on cost, coding, reasoning and agentic capabilities. The competitive landscape means Baidu cannot rely on its early position in China’s generative AI market.
Its challenge is also broader than simply improving benchmark scores.
Baidu needs Ernie to become commercially useful across an ecosystem that includes cloud computing, search, enterprise software and AI applications. A stronger model could improve the company’s ability to monetize search queries, attract developers, increase cloud demand and create new enterprise products.
That makes the model upgrade promised by Li strategically important well beyond the chatbot itself.
There is also a potentially important change occurring inside Baidu’s revenue mix. The company’s AI-powered business has grown from an emerging segment into roughly half of its general business revenue. Yet AI applications revenue was only about 2.5 billion yuan in the second quarter, broadly flat from a year earlier, while AI Cloud Infrastructure provided much of the growth.
That suggests Baidu’s AI monetization is currently being driven more by infrastructure demand than by rapidly expanding consumer AI applications.
The distinction carries implications for the company’s future margins. Cloud infrastructure can produce substantial revenue, but it requires expensive computing hardware and data-center capacity. A successful foundation model, by contrast, could potentially generate higher-margin revenue through software, subscriptions, advertising and enterprise applications once the underlying technology has been developed.
Baidu therefore needs to move from selling the infrastructure required for the AI boom to capturing more of the value created by the applications built on top of it.
Its advertising business adds urgency to that transition.
Baidu’s traditional search model is vulnerable to changes in how consumers obtain information. If users increasingly turn to AI assistants for answers rather than conventional search results, generative AI could disrupt both the technology and economics of the company’s historic core. That creates a paradox for Baidu. AI is simultaneously the technology threatening to reshape its search business, and the technology it hopes will create its next major growth engine.
The company is consequently attempting to replace one business model while defending another, all during an unusually intense period of competition.
The second-quarter figures show that the transition is already underway. AI-related revenue is growing at double-digit rates, GPU Cloud demand is accelerating, and AI now accounts for a substantial portion of Baidu’s core business. Yet total revenue continues to fall because the legacy advertising business is contracting faster than the new businesses can expand.
That makes the next phase of Baidu’s AI strategy a test of execution rather than ambition.
Li has made clear that Baidu intends to spend the money and accept the patience required to regain technological ground. The investment could eventually strengthen the company’s position across cloud, search and enterprise AI.
But the company must demonstrate that Ernie can once again compete with China’s fastest-moving models and that Baidu can turn that technological capability into profitable products.