Summary of Key Points
The two leading large-scale model companies in the Hong Kong stock market, Zhipu and MiniMax, have recently seen the lifting of restrictions on their restricted shares. Their combined market value has evaporated by over HK$900 billion compared to their peak. This release of restricted shares is not just a simple pressure of stock sales; it represents the market's first large-scale test of the value of the domestic general large-scale model industry. The high valuations that were previously sustained by "long-term AGI expectations, scarcity, and small circulating shares" are gradually returning to reality. In the future, the industry will shift from focusing on concepts to evaluating actual capabilities, with technical prowess and commercialization (especially in the B2B segment) becoming the core criteria for success.
I. Pressure from the Release of Restricted Shares: Zhipu Stabilized, MiniMax More Vulnerable
The situation regarding the release of restricted shares varies significantly between the two companies:
- Zhipu: This time, the shares released are from 11 cornerstone investors (accounting for 5.76% of the total share capital). However, nearly 70% of the institutions involved in the release (such as funds with state-owned background) have stated their intention to hold the shares in the long term. Additionally, the founder team and early-stage VC investors' shares are still locked until 2027, so the actual pressure from sales is minimal. JPMorgan Chase has even raised Zhipu's target price, believing that its GLM-5.2 model is globally competitive, and its open-weight strategy could lead to scaled growth.
- MiniMax: The proportion of shares released is as high as 60% (accounting for 63% of the Hong Kong stock market capital), and most of these shareholders are financial investors (not strategic partners), which means they have a strong incentive to sell. Although Alibaba, MiHaoYou, and the founder team have indicated they will not reduce their holdings, the market was still panicked—before the release, the company's stock price dropped by more than 3% to a six-month low before rebounding. Furthermore, MiniMax's new model, M3, has performed averagely, and its cost-effectiveness is inferior to that of DeepSeek. JPMorgan Chase has downgraded its rating from "overweight" to "neutral," adding to the challenges.
II. How Did the Valuation Bubble Form?
The previous skyrocketing stock prices of large-scale model companies were mainly driven by two factors:
1. Scarcity Premium: With only two general large-scale model companies in the Hong Kong market, investors had no choice, leading to exceptionally high valuations.
2. Liquidity Premium: Due to a large number of restricted shares and a small circulating pool, a small amount of capital could drive up stock prices significantly (for example, Zhipu's issue price was HK$116, but it reached over HK$1,600 at its peak).
These two factors are now diminishing: more shares are becoming available after the restrictions are lifted, eliminating the liquidity premium; moreover, many AI companies both domestically and internationally are planning to go public (such as OpenAI and Anthropic), reducing scarcity. Additionally, the enthusiasm for AI in the US stock market has cooled down, causing the AI bubble in the Hong Kong market to burst.
III. Commercialization: B2B Is the Real Cash Cow
If large-scale models want to make money, the B2B segment seems more promising than the C2C one:
- C2C Challenges: For example, companies using Zhipu's services (like DouBao) incur higher costs with more users, as each AI call incurs expenses. The internet-era model of "traffic in exchange for advertising" does not apply here; the larger the traffic, the higher the costs.
- B2B Opportunities: Zhipu's core revenue comes from enterprise privatization deployments (expected to account for 50.4% of total revenue by 2025), along with its enterprise intelligence services, which together account for 73.3% of its business. Enterprises are willing to pay for stable services that provide continuous cash flow. In the future, they will value cost-effectiveness more, focusing on models that can solve practical problems (such as improving coding efficiency and enhancing collaborative work scenarios).
IV. Future Trends: Moving Beyond "Technology-Only" to Focus on "Technology + Profitability"
The large-scale model industry has moved from a focus on the number of parameters to a deeper emphasis on commercial implementation:
- Technical Directions: The focus is shifting towards more efficient technologies (such as the MoE sparse architecture and long-range agents), with inference costs decreasing by 175 times in the past two years, enabling the models to handle more use cases.
- Implementation Focus: The focus is on real collaborative work scenarios (such as managing local files and handling office tasks). Coding is currently the area where the most significant efficiency improvements can be seen (high standardization and low verification costs).
- Survival Rules: In the next 1-2 years, only companies that can reduce losses, expand high-value B2B revenue, and successfully deploy large-scale intelligent applications will survive. Those that rely solely on technology but cannot generate profits will be eliminated by the market.
In Summary
The "myth of wealth creation" driven by large-scale models has temporarily come to an end. Now is the time to test whether a company's technology can truly transform into stable revenue, as only those capable of doing so will withstand valuation fluctuations and realize the long-term value of AGI.