From the disclosure in the third-quarter report, AI's contribution to the performance of technology giants is gradually significant, continuously verifying an industry trend: AI has entered the second half of the application battle, with commercialization accelerating. For example, this is intuitively reflected in overseas AI application companies such as Applovin and Palantir collectively showing performance growth exceeding expectations.

In this trend, capital markets' confidence in AI applications continues to strengthen, considering it as one of the main themes of technology investment, driving a surge in stock prices of some related companies at home and abroad.

On the other hand, from a broader perspective, the AI application field contains rich investment opportunities, with some companies yet to be fully explored, recognized, or facing opportunities for value reassessment. For example, Zhihu, which is investing in AI search.

1. Substantial reduction in losses + rapid growth in AI search, reflecting correctness and foresight of the strategy.

Firstly, the latest financial report released by Zhihu sends an important signal.

The financial report shows that in the third quarter of 2024, Zhihu achieved revenue of 0.845 billion yuan, with a net loss decreasing by 96.8% year-on-year to 9 million yuan, once again creating the largest single-season record of loss reduction and the smallest net loss since its listing; the gross margin reached 63.9%, setting a new record since going public.

This means that Zhihu's loss reduction process far exceeds market expectations, with profitability accelerating, moving towards the goal of quarterly profit and long-term profitability growth.

Breaking down into specific businesses, Zhihu continues to advance its multi-engine strategy and fully benefit from it, with the paid reading business contributing 0.459 billion yuan in revenue, the revenue proportion exceeding half for the first time, reaching 54%; marketing services and professional education businesses generated revenue of 0.257 billion yuan and 105 million yuan respectively, with self-operated professional education showing positive growth.

Meanwhile, zhihu's AI search product 'Zhihu Direct Answers' continues to maintain high growth, achieving a dual leap in both traffic and industry reputation.

Combining these two clues, it can be found that zhihu's rapid development of AI search business did not come with profit margin pressure or a significant increase in capital expenditure, indicating that zhihu chose not to invest in costly large models training, but instead focused on specific applications that align with user needs. Similarly, the correctness and high quality of this approach have been further validated, showing that zhihu can drive the expansion of AI search business at relatively low cost, finding a more suitable, longer-term, and healthier strategy.

Moreover, looking at it now, this choice of direction is undoubtedly very forward-looking.

As mentioned earlier, AI has entered the second half of the application battle, where the competition key has shifted from model development to application development. Ultimately, large models themselves do not directly create value; applications developed based on large models are what hold value. Moreover, there are a large number of domestically developed large models, with the industry's weak point lying in applications rather than large models. This determines that technology companies that can truly land AI applications with AI technology can outperform in fierce competition.

Zhihu's path is rooted in specific applications, essentially choosing a path aimed at winning.

Furthermore, zhihu's recent trends reveal its early planning. For example, zhihu's founder, director, and CEO Zhou Yuan stated a year ago that the main focus is on applications, allowing the technology of large models to be enhanced and innovated in scenarios such as discussions, memberships, and education. This year, zhihu successively released 'Discovery · AI Search' and 'Zhihu Direct Answers,' officially productizing the AI search functionalities.

Therefore, 'Zhihu Direct Answers' is the fruit of zhihu's forward-looking vision in the AI field, and the long-term accumulation in the content field, rapidly unleashing its potential energy.

It is also worth noting that, firstly, in late October, Zhihu launched a new 'Professional Search' feature under 'Zhihu Direct Answer', introducing professional content sources such as Weipu, Zhihu Featured, entering a more professional and practical deep search capability expansion phase; secondly, Zhihu's community ecosystem continued to optimize in this quarter, including the cumulative content creation on the platform, the cumulative content creators maintaining a double-digit year-on-year growth, and monthly active users showing a month-on-month growth, reaching 81.1 million. These provide support for Zhihu to deepen its development path, and it is expected to further drive growth flywheel in the future.

Secondly, looking at Zhihu's advantages and potential in AI applications from an industrial perspective.

Furthermore, from an industrial perspective, further analysis can help better understand Zhihu's advantages and potential.

1) AI application competition shifts focus to more in-depth scenario demands.

From an intuitive point of view, burning money on large models is no longer the optimal solution; on the contrary, there is greater potential on the scenario side.

The competition for large models in China has heated up, truly becoming a 'Battle of Big Models,' and the half-life of large model computing power is very short, often completing iterations in half a year. However, is this rapid computing power iteration really in line with market demands? Can it continue to evolve? The answer may not be so certain.

For most users, what matters more is whether AI applications can solve their actual problems, provide stable, reliable, and efficient services, rather than how powerful the big model behind it is. In fact, to meet a certain core user need dimension, such rapid computing power iteration may not necessarily be required; for example, the AI question-answering engine Perplexity, recognized to have the best user experience and highest precision results, uses third-party models and does not rely solely on advanced models for advancement.

For market participants, large models often bring astonishing computing power costs. Taking OpenAI as an example, according to earlier foreign media reports, it is expected to incur a loss of 5 billion USD this year, with computing power costs as high as 7 billion USD, meaning that even if OpenAI's model is advanced and has many users, it is still difficult to profit easily.

This type of competitive approach does not have a high cost-performance ratio, especially for some companies with weaker capital strength, it may even lead to earlier eliminations, and it is destined that only a few players can stay at this table.

At the level of scenario application, large models have not gone deeper or faster, especially focusing on a practical user demand, therefore having more development possibilities.

Previously, Zhou Yuan also publicly stated, "We need to seize those most fundamental needs, such as using search engines to find information, continuously dig deep into it, and then seize every opportunity for media upgrades and technological advancements to make ourselves better and stronger. This is what Zhihu is doing."

Under the above-mentioned premise, let's take a look back at "Zhihu Direct Answers", which fully caters to the user demands in the Zhihu community and achieves deep-level iteration, from meeting the general needs of search engines to meeting professional search demands, entering a phase of expanding more professional and practical deep search capabilities. This also represents Zhihu Direct Answers as the first domestic product to provide an AI search and genuine academic paper library one-stop solution.

It is equivalent to saying that Zhihu has taken the next step in the competition of AI applications, and has done so very solidly.

2) Zhihu has a natural commercial model advantage in AI applications.

On one hand, Zhihu started with Q&A, which is also one of the core business scenes for Zhihu's large model applications. It can be described as the natural and high-quality "soil" for the growth of AI search.

This is clearly shown in the product "Zhihu Direct Answers". Users in the Q&A community have core needs for searching, asking questions, and seeking answers. Community content is also one of the sources of answers, which makes one side demand, the other side supply, with users providing positive feedback to the large model through "Zhihu Direct Answers", forming an efficient closed-loop system. This explains why Zhihu's development path rooted in specific applications can be successful.

At the same time, this is also something that other application products do not have.

Even when compared to 'Perplexity', you can still see the differentiated advantages of zhihu. For example, 'Zhihu Straight Answer' is based on the real Q&A data of zhihu creators, with high reliability and reference value, not just relying on AI itself. At the same time, it also supports 'Find People', amplifying the circulation effect of community creators and their content, playing a role that seems to be more than just an 'answer engine'.

Zhou Yuan also mentioned before that the AI search + community done by zhihu should be a brand new product, where AI search is a productivity tool and a connector for discovering the world.

On the other hand, AI search is one of the most likely scenarios to give rise to super applications in the AI era, representing a new 'internet entry point', revealing the greater value hidden in zhihu's business model.

Simply put, AI search can extract information from a wider range of data sources, and its usage scenario is reversed - results are generated in a short time, users are interested in further follow-up questions, or studying around the sources, thus solving users' problems at a lower cost and higher efficiency, responding to users' urgent need for efficient information retrieval, and is expected to continue to be an important traffic entry point.

Furthermore, AI search has more possibilities. We should not define it as a market represented by traditional search engines, including its features that are somewhat different from the 'use and leave' traditional search. For example, in Perplexity, users often spend more time reading content, asking further questions, and then reading more content.

Looking ahead, with the continuous advancement of zhihu's AI search business, its natural advantages in business model are expected to be better demonstrated.

In conclusion, the results achieved in the first quarter demonstrate that AI capabilities have brought new opportunities to the company. With the continuous increase in the penetration rate of large models, continuous enhancement of product performance, diversification of landing scenarios, and further expansion of overseas business, Cheetah Mobile is expected to welcome a broader development space.

The above clues ultimately also point to a reevaluation of the value of zhihu.

Whether it is the continuous improvement in profitability, the iterative evolution of the business model, or the positive progress of new businesses, all reveal that zhihu is undergoing a profound transformation. The market undoubtedly needs to rethink the new potential and value of zhihu from these perspectives.

At the very least, an unprecedented imaginative zhihu is gradually taking shape.