AI continues to account for more than half of revenue, and Baidu (BIDU.US) valuation logic shift drives value return

On the evening of August 18, Baidu (BIDU.US) released its second-quarter earnings report, showing total quarterly revenue of 31.3 billion yuan and AI business revenue of 12.5 billion yuan, accounting for 50% of Baidu's general business revenue. The AI business accounted for more than half of the two consecutive quarters, leaving the market forced to ignore a situation where the world's top institutions and asset managers are facing an iconic collective shift in their attitude towards Baidu.

According to the 13F documents for the second quarter of 2026, the Duquesne family office under legendary Wall Street investor Stanley Druckenmiller proposed shares for the first time in two and a half years after clearing Ali in the fourth quarter of 2023. 88,000 Baidu ADR is the only one that won the bid in its current holdings.

This move is not an exception: during the same period, Morgan Stanley opened a new position with Baidu ADR of about 6.5 million shares; UBS Group increased its holdings by 917,700 shares, +43% month-on-month; Bank of America increased its holdings by 1.087 million shares, +305% month-on-month.

David Tepper, who has the most vocal voice in the hedge fund field, has an even more extreme attitude towards Baidu. In the context of the overall contraction of China's Internet assets in the second quarter, its subsidiary Appaloosa Management was the only company that bucked Baidu's trend and increased its ADR position by more than 600,000 shares, an increase of 87% over the previous quarter.

From Druckenmiller's return to China for the first time after a lapse of two and a half years, to the simultaneous expansion of the three leading global banks, Morgan Stanley, UBS, and Bank of America, to Tepper's ultimate merit selection — multi-dimensional, cross-strategic, and cross-capital institutional resonance, clearly conveyed a key signal: Baidu's AI investment value is being collectively recognized by the world's top asset managers and leading banks.

Behind this, thanks to Baidu's AI commercialization, it has already entered the strong implementation stage of large-scale replication. The company's AI business revenue accounted for more than half for two consecutive quarters, which means that for every 2 yuan of Baidu's general business revenue, 1 yuan comes from the AI business.

Obviously, AI has become Baidu's true revenue base. This marks the “core, cloud, model, and body” full-stack layout that Baidu spent ten years building. As a set of reusable underlying capabilities, it has achieved accelerated commercialization on a large scale.

As one of the few leading companies in the world that has achieved a full-stack layout in the AI industry, Baidu has become a “weather vane” for observing the commercialization progress of China's AI industry. The long-term real value is expected to be systematically revalued under the continuous increase in the share of AI revenue. This is also the underlying logic of the world's top institutions using “real money” to collectively vote Baidu.

Leading the AI office circuit, “large-scale replication” consolidates the foundation for AI application growth

The AI industry has evolved until 2026, and the main line of the story has gradually changed — from “whose model is stronger” to “who can turn AI into real money”. Even OpenAI, which stands at the top of the global AI wave, is no longer enjoying technological dividends leisurely, but is speeding up commercialization in the face of huge profit pressure.

When the AI industry enters the second half of the “fight for commercialization,” whoever can take the lead in breaking through the closed loop of “technology-product-commerce” will hold the ticket to the next round of competition. And the AI office, as the superscenario with the highest certainty of commercialization of large models, has become the main battleground for giants to “fight”.

In 2026, Tencent will gather resources such as the QClaw team to WorkBuddy, focusing on the C-end office portal based on the WeChat/Enterprise WeChat ecosystem; while Ali integrates the three smart devices QoderWork, Goku, and MuleRun into “Qianwen Office” to become an enterprise-level AI productivity platform; Byte will incorporate the Feishu team into Doubao and reassign the sales side to Volcano Engine, forming a “Douba+Feishu+Volcano Engine” To-B office intelligence system.

At a time when the giants are all upgrading the AI office from an “internal horse racing” to a “group campaign,” Baidu took the lead in the super entrance to AI TOB, and its various AI application products, such as Baidu Hitzi, Kuku AI, and Secoda, performed well in the market.

In the list of AI office smart devices released on August 17, Baidu's monthly activity increased by 1063.79% over the previous month, ranking first in the AI office smart device growth rate list. On August 10, Baidu Dingzi launched a design kit and a financial suite to accelerate Baidu Dingzi's user expansion by expanding the AI office scene.

Since the 1.0 version was officially launched in April 2025, Kuku AI has rapidly iterated to version 4.0 within a year, and can be used across all domestic and international products such as Library Network and Oreate AI. At present, Kuku's AI office MAU has exceeded 25 million, ranking first in the AI office circuit industry.

In the “Office Agent Work Flow Assessment Report” released by the National Center for Industry and Information Security on August 5, Baidu Library ranked first in terms of score and occupied the first tier alone.

Furthermore, according to the “China AI Native Codeless Application Generation Platform Market Research, 2026 H1” published by Sullivan, Baidu ranked first in the industry with a market share of 33.4%, and among the two core indicators of OPC/commercial monetization users and effective application creation users, Second ranked first in the platform usage penetration rate.

The outstanding performance of various application products in the AI office field highlights Baidu's leading edge in this racetrack, and Baidu's AI applications are also expanding rapidly towards the B-side enterprise scene.

Among them, since its launch, the self-evolving intelligence Baidu Fumou, 3,000 companies have applied for trial use, covering various key industries such as ports, logistics, automobiles, and finance.

In addition, Baidu Yigang has now served more than 100,000 customers, covering more than 30 industries. The platform's average live streaming conversion rate has increased by 29%, delivery efficiency has increased by 150%, and production costs have decreased by 63%. According to reports from agencies such as Sullivan and IDC, Baidu Mirror has ranked No. 1 in China's AI digital human market share and No. 1 in overall product strength for two consecutive years.

The reason why Baidu was able to hit the C-side and B-side at the same time is no accident — this is an inevitable result of the reuse of its “core, cloud, model, and body” full-stack capabilities across customers, industries, and scenarios. From Kuku AI, Baidu's partner, and Secoda leading the way in the office scene, to scheming and the accelerated implementation of a mirror on the B-side, Baidu has clearly taken the lead in breaking through the complete closed loop of AI application “technology-product-commerce”.

Investors should be aware that when the same full-stack architecture can be quickly replicated from office scenarios to many other vertical fields, the “scale replication” effect of decreasing marginal costs and increasing customer lifetime value will continue to expand, which is expected to lay a solid foundation for the future growth of Baidu's AI application business.

Full-stack capabilities drive multi-tier monetization, and AI commercialization accelerates towards multi-scenario replication

Baidu's thorough cross-scenario restructuring of business lines with full-stack AI capabilities is rarely matched by domestic technology companies; the direct return of this strategic choice is a fundamental shift in the growth paradigm — the company is accelerating from a traditional Internet model to a high-value, sustainable growth model driven by the AI core. The multi-level commercial implementation of AI businesses such as AI Cloud, Radish Express, and Kunlun Core provided the most solid financial footnote to this transformation.

During the reporting period, Baidu's AI cloud infrastructure revenue was 7.3 billion yuan, up 50% year on year; of these, GPU cloud revenue surged 283% year on year, further accelerated from 184% in the previous quarter, and achieved three-digit growth for four consecutive quarters.

The continued explosion of GPU clouds essentially reflects that customers' demand for AI training and inference computing power based on public clouds is still expanding. GPU public clouds can simultaneously serve multiple types of customers through a unified resource pool, and customer computing power consumption continues to occur as model applications deepen. This means that Baidu AI Cloud's revenue structure is expanding from one-time project delivery to a standardized and sustainable service form, and the business model is being upgraded from “doing projects” to “making large-scale services.”

And this “large-scale service” upgrade has been directly verified in terms of market share and customer structure. According to IDC related data, in mid-2025, Baidu Smart Cloud will rank first in the financial model market, AI cloud gaming market, and embedded smart cloud market. Furthermore, in the first half of 2026, Baidu Smart Cloud ranked first in the industry with a bid amount of 1,385 billion yuan, accounting for nearly 60% of the total bid amount won by the five cloud merchants, achieving continued leadership.

Currently, Baidu Smart Cloud's new full-stack AI cloud has been cultivated on a large scale in core industries. In the financial sector, it has achieved 100% coverage of systemically important banks and more than 800 financial institutions; in the automotive sector, it supports the delivery of more than 20 million new L2 assisted driving vehicles; in the field of AI hardware, it has served more than 1,000 customers; in addition, over 80% of central state-owned enterprises have chosen Baidu Smart Cloud and helped the national grid implement smart devices on a large scale in more than 40 scenarios in the energy sector.

Leading market position and dual leadership in customer coverage clearly show that Baidu's AI cloud service is expanding in depth from Internet customers to industrial scenarios, and the “scale replication” effect is accelerating. This is also one of the core factors driving the explosive growth of GPU clouds.

As Baidu's most technologically barred business segment, Radish Express is accelerating large-scale replication across markets. Since the second quarter, Radish Express's global operations have accelerated markedly. Radish Express launched fully unmanned commercial operation in Dubai, obtained the first batch of fully unmanned test licenses in Hong Kong, and launched tests without safety personnel in cars on Airport Island, becoming the first platform in the world to carry out fully driverless testing on the “right steering and left steering” system. Meanwhile, Radish Express launched public road tests in London and Switzerland, and signed a strategic cooperation with Kazakhstan's TPH, marking the first time that Chinese unmanned vehicles have entered the Central Asian market, filling the gap in the region.

Up to now, Radish Run has covered 28 cities around the world, with a cumulative autonomous driving mileage of more than 350 million kilometers, of which no one has reached 240 million kilometers. The same set of autonomous driving technology can be promoted in parallel in multiple markets with very different regulations, road conditions, and driving habits, showing that Baidu replicates far more than a fleet of vehicles, but a combination of technology, operation, and compliance capabilities — every new scenario provides new training materials for model iteration. The accumulation of this data is an asset that is difficult for latecomers to quickly replicate, and is also the key for Radish Express to continue to consolidate its leading position in the market.

Furthermore, Kunlun Core is also in the rapid expansion phase of large-scale replication. At the hardware level, Tianchi Super Node 1.0 has been mass-produced on a large scale. The 2.0 256-card version will soon be launched, and it is expected that the next 512/1024 card supernode will be released next year. A single node can support trillion parameter model training.

At the delivery level, the P800 30,000 card cluster has been lit up. Since 2025, multiple 10,000 card clusters have been delivered, covering the Internet, finance, energy, manufacturing and other industries, and hundreds of customers, including Tencent. The delivery scale ranges from tens of cards to tens of thousands of cards.

At the software level, Kunlun Core has previously completed the adaptation of mainstream models such as Wenxin, DeepSeek, GLM, and Minimax, increasing inference efficiency by 50%. Large-scale clusters, customer groups, and software ecosystems have all together formed a complete closed loop of commercialization verification of Kunlun Core's self-developed chips.

From the explosion of the smart cloud GPU cloud, to the global operation of Radish Express, to the large-scale delivery of computing power of Kunlun Core, Baidu is transforming the “core, cloud, model, and body” full-stack capabilities into commercial results in various forms such as cloud services, travel, and computing power. When this base was implemented simultaneously on multiple tracks, Baidu's AI commercialization has clearly moved from single-point verification to multi-scenario replication.

The pricing logic switched to market consensus, and two major catalysts accelerated value return

As Baidu's AI business continued to account for more than half of its revenue, a real problem faced the capital market, namely how to make a reasonable valuation for this company with scarce AI full-stack capabilities and leading the way in AI commercialization on multiple tracks?

In fact, the true value of Baidu is an “old normal” topic, and the core factor behind this is that the market always uses traditional PE valuation methods to evaluate Baidu as a whole, but this causes the value of high-growth businesses such as AI Cloud, Kunlun Core, and Radish Express to be seriously overshadowed by the search business.

As AI commercialization continues to materialize and the AI core business revenue share exceeds 50%, it is gradually becoming a market consensus that Baidu's pricing should switch from the traditional PE valuation method to the new SOTP framework. This is also the core driving force driving Baidu's valuation center upward.

Recently, it is clear that Baidu has adopted the SOTP (Classification Plus Valuation Method) /segment valuation method for Baidu, including many top investment banks such as J.P. Morgan, UBS, Macquarie, CICC, and BOC International. Obviously, Baidu's valuation framework needs to be changed, which is already “deeply rooted in the hearts of the people” in the institutional community.

J.P. Morgan said that if Baidu's value is measured using the SOTP valuation method, then the valuation of the AI business single sector alone would reach 169 US dollars/ADS, and the reasonable value corresponding to the US stock Baidu should be 230 US dollars/ADS, which has room to double the current stock price.

Based on the change in the valuation framework, two things are expected to accelerate Baidu's value return process. First, Baidu is promoting a dual major listing transition for Hong Kong stocks. On August 19, Baidu Group Chief Financial Officer He Haijian said that the board of directors previously approved the conversion of Hong Kong's listing status from a secondary listing to a dual major listing. It is expected that the conversion process will take effect within this year, and the group is preparing to be included in the Southbound Hong Kong Stock Connect after the conversion.

If we take advantage of the Hong Kong Stock Connect adjustment window in September 2026, more mainland capital familiar with the Chinese Internet, domestic computing power, and the AI industry will directly participate in the pricing of Baidu, and the true value of the AI full-stack business is expected to be rediscovered by Southbound Capital.

Second, the Kunlun Core Hong Kong stock IPO is being actively promoted. After the listing is completed, it will provide a direct catalyst for Baidu's revaluation. As the core asset of Baidu's full-stack AI infrastructure layer, Kunlunchip's listing will provide a clear valuation anchor for its true value.

According to public market information, Morningstar (Morningstar), the world's leading investment research institution, estimates that Kunlun Core's valuation is in the range of 51.04 billion to 63.8 billion US dollars; J.P. Morgan Chase's March research report gave Kunlun Core an independent valuation of 40 billion to 49 billion US dollars. This means there is strong room for revaluation of Baidu's value in the medium to long term.

At this point, this mismatch between “underestimation” and “high AI growth” is providing a very attractive layout window for long-term capital. With the implementation of the Q2 earnings report, Baidu's current valuation reflects more of the market's phased judgment on traditional businesses, while the long-term value of businesses such as AI Cloud, Radish Express, and Kunlun Core still needs to be further unleashed.

As the expected rise in the Hong Kong Stock Connect in September and the potential value of businesses such as AI applications, Kunlun Core, and Radish Express is being re-evaluated by the market, the reshaping of Baidu's valuation anchor is expected to begin at any time. Investors waiting for catalytic implementation may reap considerable excess profits.