Baidu’s AI Transformation Is Becoming Commercially Material

Baidu’s second-quarter 2026 results show that AI has moved from a strategic promise to a meaningful commercial business. On August 18, the company reported RMB31.3 billion in total revenue, down 4% year over year, while its Core AI-powered Business generated RMB12.5 billion, up 25% and equal to half of Baidu General Business revenue for the second consecutive quarter. 4

That combination captures the central tension in Baidu’s transition: AI is scaling quickly inside the company, but it has not yet restored growth across the whole group.

The headline is a mix shift, not a completed turnaround

Baidu General Business generated RMB25.2 billion in Q2, and its AI-powered activities accounted for 50% of that figure. Because total group revenue also includes iQIYI, the RMB12.5 billion AI figure represents roughly 40% of group revenue rather than half of the consolidated total. 4

The distinction matters. Baidu can accurately say that AI is now the largest part of its general business while still reporting a shrinking overall top line. The results therefore support a more measured conclusion: AI is becoming commercially material, but it has not yet fully offset pressure in legacy businesses such as online marketing. 26

AI Cloud is the clearest evidence of customer demand

AI Cloud Infrastructure generated RMB7.3 billion in Q2, up 50% year over year. Within that segment, GPU Cloud revenue increased 283%, accelerating from 184% growth in the previous quarter. Baidu attributed the expansion to rising demand for public-cloud computing used for AI training and inference. 4

This is the strongest monetization signal in the quarter because it reflects customers paying for computing capacity rather than simply testing models. It also gives investors a more concrete set of indicators to watch: cloud utilization, project-delivery timing, customer renewals and margins.

The growth is not perfectly linear. One industry commentary reported that AI Cloud Infrastructure revenue fell 17% quarter over quarter, partly reflecting the timing of project deliveries. 6 That makes year-over-year growth impressive, but it also argues against treating one quarter of GPU Cloud acceleration as a guarantee of smooth future results.

Baidu is pursuing a full-stack AI strategy

Baidu’s approach is broader than launching a single chatbot or adding generative features to search. Its strategy connects several layers:

• Kunlun chips provide AI computing hardware.
• Baidu AI Cloud supplies infrastructure and development platforms.
• ERNIE and other models support enterprise and consumer use cases.
• Agents and applications create end-user products and potential recurring revenue.
• Apollo Go extends the strategy into physical AI and autonomous transportation.

Reporting on the quarter also described external delivery of large Kunlun P800 clusters and compatibility with several third-party models. Those claims would strengthen the case for Kunlun becoming a commercial hardware and platform business, but the available material does not independently verify every specific cluster-count or compatibility detail. 13

Baidu has also announced a planned Hong Kong spin-off for Kunlun. If executed, that could give the chip business a more distinct funding and valuation profile; it would not, by itself, prove that the unit has durable margins or competitive advantage.

Applications are growing, but they remain the next test

Baidu’s AI Applications business generated RMB2.5 billion in Q2 revenue, up 3% year over year. The reported portfolio includes products and services built around Wenku, NetDisk, office work, application development and digital agents. 18

Baidu and related reporting also highlighted products including Baidu Dazi, Miaoda, Famou and Yijing, along with higher AI-feature penetration in Wenku and NetDisk and more than 25 million monthly active users for Kuku AI. 31 38

These figures show distribution, but distribution is not the same as durable economics. The more decisive evidence will be recurring paid users, enterprise seat growth, retention, usage intensity and application gross margins. A large user base can support monetization, but it does not establish it on its own.

Search is being rebuilt around AI—with a real near-term risk

Baidu’s AI strategy also reaches into search, marketing and existing products rather than sitting beside them. AI-native answers, agents and marketing tools could help defend the company’s commercial position and make its large installed user base more valuable.

The transition also creates friction. If users receive more answers directly from AI interfaces, search-query patterns, advertising inventory, pricing and traffic-acquisition economics could change. Q2 makes that risk visible: AI revenue grew strongly even as total revenue declined. 20 26

The key question is not whether Baidu can add AI features. It is whether those features can generate enough incremental revenue and efficiency to replace what changes in the legacy search-advertising model.

Apollo Go adds a physical-AI option

Autonomous driving is another part of Baidu’s longer-term AI expansion. Management reported one million fully driverless operational rides during Q2 and more than 23 million cumulative rides. It also cited an average of one airbag deployment every 14.4 million kilometers. 29

Those operating figures provide more evidence than a purely conceptual robotaxi announcement, but they are not yet a complete commercialization case. The business still needs to demonstrate attractive fleet economics, regulatory durability, local partnerships and unit profitability. Ride volume and autonomous kilometers measure deployment; they do not by themselves measure returns.

The listing change and investor activity are secondary signals

Baidu’s conversion of its Hong Kong listing from secondary to primary status became effective on September 1, 2026, creating a dual-primary listing on the Hong Kong Stock Exchange and Nasdaq. The company said the conversion involved no new share issuance or fundraising. 39 43 46

A primary Hong Kong listing may improve market access and create a path toward Stock Connect eligibility, but eligibility or inclusion would be a market-structure catalyst rather than evidence of stronger operating fundamentals. 44

Investor filings have also drawn attention to Baidu. Duquesne Family Office disclosed an 88,000-ADR position in its second-quarter 13F filing, while other reports described buying or increased positions by large investment firms. 40 48 These filings are delayed snapshots, however, and they do not establish an investment thesis or guarantee that positions remain current.

What “concentrated harvest” means for Baidu

The phrase is useful if it describes a change in how Baidu is measuring AI: through cloud revenue, GPU demand, application usage, external chip sales and autonomous-driving operations rather than research announcements alone.

It becomes overstated if it suggests that the transformation is finished. Q2 2026 showed a strong AI-led mix shift, but consolidated revenue still fell 4%, AI Applications grew only 3%, and the company has yet to demonstrate that its newer businesses can deliver sustained group growth and superior profitability. 4 18

The most defensible reading is that Baidu has reached the commercialization phase of its AI strategy. Infrastructure is currently carrying the strongest evidence, applications are still proving their economics, and chips, search, agents and autonomous driving represent additional options. The next earnings reports will show whether those pieces add up to a self-reinforcing AI business—or remain a collection of fast-growing but uneven businesses inside a pressured legacy company.