<p>Baidu shifts gears: AI carries half the load, but the growing pains of transformation amid slowing ad revenue will take time to digest</p>

On August 18, 2026, $BIDU-SW(09888.HK) released its Q2 2026 earnings report. The core data presents a clear fact: $Baidu(BIDU.US) is transitioning from a company driven by a single engine of search advertising to one powered by dual engines of AI infrastructure and advertising. However, this gear shift is experiencing significant growing pains.

Following the earnings release, Baidu's US and Hong Kong stocks fell by 12.73% and 11.03%, respectively, reflecting a negative market reaction. However, setting aside short-term sentiment, the information revealed in this report is not unilaterally bearish; rather, it highlights the structural friction that inevitably arises when a company undergoing technological transformation alternates between old and new growth drivers.

Achievements Cannot Be Ignored: AI Business Has Moved From "Story" to "Ledger"

First, we must acknowledge that Baidu's AI business is no longer just conceptual investment but has become a substantial contributor, accounting for half of its core revenue.

Baidu executives disclosed that in Q2 2026, core AI new business revenue reached RMB 12.5 billion (same unit below), representing 49.6% of Baidu's general business revenue (RMB 25.2 billion). Among these, smart cloud infrastructure performed most prominently, with year-on-year revenue growth of 50% to RMB 7.3 billion, surpassing industry average growth rates for multiple consecutive quarters. Notably, GPU cloud revenue surged 283% YoY, accelerating further from last quarter's high base of 184%, indicating that Baidu's commercialization at the compute layer is fully benefiting from the explosion in current AI training and inference demand.

From a profitability perspective, management revealed during the earnings call that both profit and margins for AI cloud infrastructure improved YoY. This is not a "burn cash for growth" story—GPU clouds inherently possess higher margins than traditional CPU clouds, and with customer structure optimization and improved resource utilization efficiency, management explicitly stated there is "still significant room for improvement." Additionally, external client Token call volume revenue on the Qianfan MaaS platform grew over nine times YoY, signaling that Model-as-a-Service (MaaS) commercialization is crossing from zero to one.

In other words, Baidu's investments in the AI infrastructure layer may be converting into real revenue and profit contributions at a pace exceeding market expectations. This aspect was somewhat overlooked in the earnings report.

The Unavoidable Reality: Advertising Business Undergoing Deep Adjustment

Contrasting with the bright spots in AI, Baidu's traditional online marketing revenue declined 19% YoY to RMB 13.1 billion, serving as the primary drag on overall performance.

This decline stems from multiple factors and cannot be simply summarized as "declining competitiveness."

First, Baidu is proactively advancing the strategic transformation of AI search. Management explicitly stated that the current priority is refining the product experience and user satisfaction of AI search, deliberately controlling the commercialization pace of traditional search ads, and pausing monetization actions related to AI search.

Second, users' methods of acquiring information are shifting from traditional search to diversified forms such as AI chatbots and vertical content platforms. This is a structural change facing the entire industry, and Baidu's traditional search traffic pool has indeed been affected by diversion.

Third, under the macroeconomic environment, brand advertising budgets have contracted overall, exerting certain pressure on the performance of the traditional search advertising market.

Management expects continued pressure on the advertising business in H2. This pressure includes: proactive strategic choices (pausing AI search monetization to prioritize product refinement); and passive industry competition (user traffic migrating to diversified information platforms). The market needs to distinguish between these two: the former implies that once the AI search product matures, there will be room for advertising revenue release; the latter requires Baidu to continuously iterate its AI search products to prove long-term competitiveness to the market.

Competitive Landscape: Crowded Large Model Track, But Far From Finalized

At the foundational model level, Baidu's Ernie Bot faces fierce competition from Alibaba (09988.HK)'s Tongyi Qianwen, ByteDance's Doubao, Moonshot AI's Kimi, and other products. Multiple Chinese large models lead in specific capability dimensions, and the entire industry remains in a rapidly iterating dynamic development stage.

Robin Li's response during the earnings call is noteworthy: "Foundational models are still developing rapidly. Approximately every few months, different models take the lead in certain capabilities. In such a dynamic market, long-term competitiveness depends on sustained technology investment, an application-driven approach, and patience." This judgment aligns with industry reality—today's leadership does not equal final victory, nor does today's lag mean complete elimination. Baidu's recent optimization of organizational structure and introduction of top AI talent are precisely attempts to accelerate catching up in model capabilities.

More importantly, Baidu has adopted a differentiated path: leveraging its own deployed applications like AI search, digital humans, Miaoda, and Wenxin Assistant to feed back real-scenario application data into model training. This positive closed loop of "Application → Data → Model" differs from development paths that merely compete on model parameter scale, and its long-term technical iteration effects warrant continued observation. Early commercial progress of Baidu's enterprise decision-making agent "Famou" in solving complex tasks, along with public third-party data showing Miaoda holding a 33.4% market share as the No.1 domestic AI-native no-code platform by monthly active users according to Frost & Sullivan's H1 2026 report, suggest Baidu has achieved solid landing results in the AI application layer.

Kunlun Xin: Potential Catalyst for Value Reassessment, But Viewed Rationally

The spin-off listing of Kunlun Xin is another focal point of market attention. Reports indicate Kunlun Xin is advancing its main board listing on the HKEX while simultaneously proceeding with STAR Market counseling filing (A+H layout). Market rumors suggest a target valuation of up to USD 50 billion. With Baidu's current market cap at approximately USD 30.9 billion, this implies potential upside for Baidu's valuation—though still dependent on the actual value upon Kunlun Xin's listing.

Strategically, after over a decade of technology investment, Kunlun Xin has completed the R&D and commercial deployment of three generations of AI chips. Its third-generation product, P800, is fully compatible with the latest versions of mainstream domestic large models including Kimi K3, GLM 5.2, and Minimax M3, gaining recognition from leading clients across industries such as internet, finance, energy, government, and telecom operators. Against the backdrop of supply-demand imbalance in domestic AI compute, Kunlun Xin, as one of the few domestically capable fully stack self-reliant and controllable AI chip suppliers, indeed possesses scarce strategic value.

Commercially, Kunlun Xin is still in the early stages of rapid growth. There is a significant gap between its current revenue scale and the USD 50 billion target valuation, but capital market premiums for the AI chip track are inherently forward-looking. The core key to judging the rationality of its valuation lies in whether Kunlun Xin can continuously prove its ability to independently serve external markets, rather than just being a chip design department serving Baidu's internal needs. Progress in customer expansion within core tracks such as telecommunications and leading internet enterprises will be the core basis for the market to validate this valuation logic.

For Baidu as a whole, the greatest significance of Kunlun Xin's listing may lie in allowing the market to re-examine the implicit value of various AI assets within the Baidu ecosystem—if Kunlun Xin achieves a reasonable high valuation in the public market, then the valuation logic for Baidu's other core AI assets, such as AI cloud infrastructure, autonomous driving, and large models, may also be repriced by the market. This could be the variable with the largest elasticity in Baidu's current valuation system.

Autonomous Driving: Short-Term Twists and Long-Term Certainty

Apollo Go completed approximately 1 million autonomous driving trips in Q2 2026. Affected by operational adjustments in some cities, order volumes faced temporary pressure, but management emphasized that affected cities are gradually resuming operations, and order volumes are expected to return to a growth trajectory in subsequent quarters.

Baidu is further expanding its global footprint: In Dubai, Apollo Go has launched fully driverless commercial operations, allowing passengers to book rides via its own app and Uber; in London, Apollo Go partnered with Uber and Lyft to launch open-road testing; in Hong Kong, Apollo Go obtained the local batch of fully driverless test licenses and initiated testing on Airport Island, becoming the first platform globally to conduct fully driverless testing in a "right-hand drive, left-side traffic" system. By Q2 2026, Apollo Go had covered 28 cities worldwide.

From a broader time scale, the standardization process in the autonomous driving industry is continuously advancing. As a leading enterprise in China's autonomous driving sector, Baidu deeply participates in the formulation of relevant industry technical standards, demonstrating its voice in the industry.

Financial Health: Foundation Remains, But Safety Margin Is Narrowing

Based on our estimates, as of June 30, 2026, Baidu's total of cash, long-term deposits, held-to-maturity investments, and short-term investments amounted to RMB 240.562 billion (approximately USD 35.454 billion), exceeding its current market capitalization. This means that from an asset liquidation perspective, the market's pricing of Baidu's entire operating business is close to zero or even negative—reflecting both extreme pessimism in the market and the possibility of undervaluation.

However, declining profitability is an indisputable fact. Non-GAAP net profit attributable to parents in Q2 2026 dropped 46.34% YoY, mainly impacted by shrinking advertising revenue, rising AI cloud costs, and exchange rate fluctuations.

In the current AI investment cycle, Baidu needs to find a better balance between "continuing investment" and "maintaining profits." We also note that its CapEx in Q2 2026 surged to RMB 11.39 billion, far higher than RMB 3.8 billion in the same period last year and RMB 5.916 billion in the previous quarter, with free cash outflow widening further to RMB 7.954 billion. Management emphasized that "investments revolve around clear customer needs and can be recovered quickly," but the continuous growth in capital expenditure (such as GPU procurement) will still pressure profits in the short term.

The dual primary listing process is underway, expected to be completed within the year. If included in Stock Connect thereafter, it is expected to attract more Southbound capital, improving shareholder structure and liquidity. Although this does not directly change fundamentals, it may help raise the valuation floor.

Baidu is merely a microcosm of tech companies transforming in the AI era. How should one evaluate the value of a listed company during this critical cycle? The 13th edition of the Hong Kong Stock Exchange Top 100 Strong evaluation is about to commence grandly. Anchored by objective fundamentals, it systematically verifies the growth quality and asset value of listed enterprises, helping the market identify Hong Kong-listed targets with genuine long-term competitiveness. Whether Baidu will be selected is worth continued attention.

Wu Yan