On September 6, 2026, Zhihu (NYSE: ZH, HKEX: 02390) announced on the Hong Kong Stock Exchange its plan to commit 1.5 billion yuan of its own funds to subscribe for limited partnership interests in an AI fund managed by Monolith Management 鈥?Tianjin Lisi Xingshen Equity Investment Partnership (Limited Partnership) 鈥?with an expected stake of no more than 30%. Based on Zhihu's total assets of 5.053 billion yuan as reported in its 2026 interim results, this commitment accounts for nearly one-third of its total assets.
More critically, Lisi Xingshen is a blind-pool fund: when Zhihu hands over the money, which companies the fund will invest in and at what prices are entirely decided by Cao Xi and his team, with Zhihu as an LP having no say in specific investment decisions. This is not an ordinary industrial investment. A content community that had just achieved its first full-year Non-GAAP profit in 2025, yet saw revenue decline 7.2% year-on-year in the first half of 2026 and adjusted net profit plunge from 98.3 million yuan in the same period last year to just 6.9 million yuan, is locking roughly one-third of its cash into a blind-pool fund with a seven-year lifespan and a four-year investment period. The market voted with its feet the day after the announcement, with Zhihu's Hong Kong shares closing down 2%. But the real analytical value of this deal lies not in the short-term stock price reaction, but in the deep and awkward predicament it exposes for a content platform navigating its transformation in the AI era: when its own path to AI monetization remains unclear, it chooses to use capital to buy a ticket held in someone else's hands.
Pricing Trust: Why Cao Xi, Why a Blind Pool
To understand why Zhihu is willing to accept a blind-pool structure, one must first understand Cao Xi's position in the 2026 primary market. Monolith Management was founded in 2021, and founding partner Cao Xi was once one of Sequoia China's youngest partners, having invested in Kuaishou, Douyu, and Tencent Music. After leaving Sequoia, he shifted his focus to AI hard tech, building positions in projects such as MetaX, Moonshot AI, Unitree Robotics, and Infinigence AI, and securing approximately 3 billion yuan in allocation during DeepSeek's first financing round. In other words, what Zhihu is buying is not a specific portfolio of AI assets, but Cao Xi's personal track record of sector judgment. The directions covered by Lisi Xingshen 鈥?foundation models, AI infrastructure, humanoid robots, and AI applications 鈥?are precisely the AI strategic directions Zhihu has repeatedly emphasized in its own business. However, Zhihu's own ability and speed in capturing these projects in the primary market falls far short of the network and deal record Cao Xi has already established. From this perspective, the blind pool is not an oversight but a deliberately paid trust premium: Zhihu acknowledges its information disadvantage and judgment disadvantage in accessing AI projects and is willing to exchange the right to decide on targets for the qualification to enter Cao Xi's project circle. But the cost of this trust is amplified by a key financial fact. Zhihu had cash and equivalents, deposits, and short-term investments totaling 4.45 billion yuan at the end of 2025. Although total assets stood at 5.053 billion yuan in the first half of 2026, the company had already recorded an operating loss during the same period. After the 1.5 billion yuan is deployed, the remaining available liquidity is in the range of roughly 2.9 billion yuan. For a company whose revenue is still contracting and which needs to continuously invest in AI product iteration, this means strategic maneuvering space has been substantially compressed. The blind-pool fund's seven-year lock-up period further intensifies this rigidity: over a time span during which AI technology roadmaps may undergo several iterations, this money cannot be used for acquisitions, cannot be used to increase R&D spending, and cannot serve as a buffer when competition deteriorates. Zhihu has exchanged its scarcest resource 鈥?cash that preserves optionality during a transformation period 鈥?for a black box it cannot influence.
Of course, when Lisi Xingshen ultimately completed fundraising at a scale of 7.52 billion yuan, Zhihu's 1.5 billion yuan was diluted to approximately 20% after all partners fully paid in their commitments, rather than the initially estimated 30%. This alleviates concerns about "excessive exposure of a single LP" to some extent and shows that industrial capital willing to bet on Cao Xi extends far beyond Zhihu: listed companies including Perfect World, Jinko Power Technology, Litong Electronic, Century Huatong, By-health, and Andon Health have all appeared on the LP lists of various Monolith funds. But when we shift the reference frame from "who else is investing" back to "Zhihu's own financial scale," the nature of the question changes: for Jinko Power Technology or Perfect World, with market capitalizations of 300 billion yuan, a fund commitment is asset allocation; for Zhihu, with total assets of 5 billion yuan and having just crossed the profitability threshold, this is a bet that must succeed 鈥?there is no middle ground.
The Ambiguity of "Synergy" and the AI Dilemma Facing Content Platforms
The investment rationale Zhihu provided in its announcement centers on "industrial synergy": leveraging Monolith's project resources to create linkages with its own knowledge large model, intelligent Q&A, and other businesses. But how this synergy logic would actually materialize under a blind-pool structure finds no clear path in publicly available information. Zhihu has had successful industrial investment precedents before. In 2023, it led ModelBest's angel round, and the two sides subsequently deepened collaboration on the "Zhihai Tu AI" large model. ModelBest has now initiated its A-share IPO and is one of the highest-valued unicorns in the on-device AI sector. That investment had distinctive features: clear targets, controllable amounts, and specific post-investment synergies 鈥?ModelBest's CEO Li Dahai was himself a former CTO of Zhihu, and the model capabilities could be directly embedded into Zhihu's product system. In contrast, for the AI projects Zhihu participates in indirectly through Lisi Xingshen, Zhihu can neither see the investment decision list nor has any contractual guarantee that any business connection will occur between portfolio companies and Zhihu. The so-called synergy is at best a "possibility": among the companies Cao Xi's team invests in, perhaps one might in the future intersect with Zhihu's data, scenarios, or expert network. But "perhaps" and "one company" are insufficient to constitute a strategic rationale for a listed company to allocate one-third of its assets to a blind-pool fund.
The deeper problem is that Zhihu is using capital instruments to respond to a problem that capital instruments cannot solve. Zhihu's AI-related professional content grew approximately 37% year-on-year in the first half of 2026, and its 81.5 million creators and nearly 1 billion pieces of content constitute a scarce high-quality corpus in the Chinese internet. Zhihu's真正的 AI assets are not in Cao Xi's fund but on its own servers: the expert network, authentic discussions, and traceable professional judgments. Zhou Yuan and his team are already doing things in the right direction 鈥?Zhihu Zhida was upgraded to "AI Kanshan," Data Lab serves frontier model training, and short-drama adaptations of Yanyan Story IPs have produced hits exceeding 3 billion views. But the common characteristic of these businesses is that they require sustained, flexible R&D investment and operational patience that matches the pace of product iteration 鈥?not a seven-year, non-exitable private equity allocation. When a content platform hands one-third of its assets to an external GP to "understand AI," it is effectively sending a signal: its own organization has not yet established sufficient certainty in its judgment and execution regarding AI technology evolution and commercial opportunities. The blind-pool structure of Lisi Xingshen amplifies this uncertainty 鈥?Zhihu has not only entrusted its funds but also delegated the judgment of "what to look at, what to invest in, and when to enter." If Cao Xi's judgment proves correct, Zhihu gains financial returns and a thin layer of "participation." If that judgment proves wrong, Zhihu loses its most precious strategic freedom during the transformation window. And regardless of the fund's performance, whether Zhihu's own AI products can retain users and whether paid reading and advertising can continue to hold up in the era of AI search and intelligent agents remains a question that Cao Xi's fund ultimately cannot answer on Zhihu's behalf.
This deal may ultimately prove to be a successful financial investment. Cao Xi's track record is impressive enough, dozens of industrial capital and market-oriented institutions stand behind Lisi Xingshen, and the actual fundraising scale of 7.52 billion yuan demonstrates the market's recognition of this fund. But "a good investment" and "the right strategy" are not always the same thing. For Zhihu, the real question that needs answering is not "can Cao Xi help us make money," but rather: when a content company stakes one-third of its assets on a blind-pool fund it cannot influence, is it positioning itself for the AI era, or is it paying for its own anxiety about the AI era? The answer will not be written in the fund's NAV curve, but in the user retention and revenue structure of Zhihu's core products over the next two to three years.
Comments