Summary of Key Points
On July 8-9, the six-month lock-up periods for the two major Hong Kong-based large-scale AI models, Zhipu and MiniMax, expired simultaneously, releasing a combined market value of nearly 100 billion Hong Kong dollars. The market had expected a significant drop in stock prices following the release of this restricted capital, but the outcomes were mixed:
- Zhipu: Due to the majority of its cornerstone investors (mainly state-owned assets and strategic capital) expressing their intention to hold the shares in the long term, the company's stock price rose by more than 10%.
- MiniMax: With a higher proportion of financial investors, there was significant selling pressure, causing the stock price to fall by over 10%, erasing the gains from the previous day.
At the same time, the large-scale AI model industry is transitioning from a phase focused on "burning money and telling stories" to one that emphasizes cost control and profitability. Financing is becoming more concentrated in the hands of a few leading companies (a small number of firms securing 30% of the total funding), with competition centered around cost management and profitability. The logic of valuing companies based on scarcity is no longer effective, and fundamental metrics such as revenue and profit have become the core determinants of their value.
I. The Divergent Trends on the Release Day: Why Such a Difference?
The difference in outcomes stems from who is selling the shares:
- Zhipu: Most of its cornerstone investors chose not to sell, resulting in little pressure on the stock price.
Eleven cornerstone investors (accounting for 5.76% of the total share capital, worth over 40 billion Hong Kong dollars), including state-owned funds and strategic investors like Lingyun Guang, publicly stated their intention to hold the shares for the long term. This message sent to the market was that they were not looking to cash out immediately. As a result, although the stock price opened lower, it quickly recovered and ultimately rose by more than 10%.
- MiniMax: Financial investors, seeking to realize profits, exerted significant selling pressure.
The scale of the release was larger (63% of the total share capital, worth over 40 billion Hong Kong dollars), with more than a third being financial investors who were primarily interested in making short-term gains. Despite efforts by the founding team, Alibaba, and MiHao to persuade them to hold on, the selling pressure was too strong. The company had previously used 600 million Hong Kong dollars for employee incentives, but this was insignificant compared to the total amount of shares released. Consequently, MiniMax's stock price first rose and then fell, ending up down by more than 10%.
II. The Shift in Valuation Logic: Why Zhipu Outperformed?
Half a year ago, MiniMax was more popular (its stock price soared 109% on the first day of trading, and its market value briefly surpassed Baidu's). Now, Zhipu's market value is five times that of MiniMax. The reason lies in whether the company's narrative resonated with the market:
- Zhipu: It tapped into the trend of "domestic infrastructure development."
Zhipu focuses on developing domestic foundational models and providing enterprise services, which are supported by government policies promoting local alternatives to imported technologies. This alignment with market trends led to a higher valuation. For example, its stock price soared from 500 million Hong Kong dollars at listing to over one trillion Hong Kong dollars, driven by the popularity of the "domestic large-scale AI model" narrative.
- MiniMax: Its strengths in multimodal capabilities (text, images, video) and consumer-facing products (such as AI chatbots) did not translate into substantial revenue. The market saw little potential for profit generation, leading to a 72% drop in its valuation from its peak.
III. Concentration of Financing in the Industry: Money Goes to the Few
Although recent financing activities in the large-scale AI model sector seem intense (with companies like DeepSeek and Yuezhiànmen securing over $10 billion), the reality is a "Matthew effect":
- A few companies dominate funding: In the first half of 2026, AI firms raised a total of 300 billion yuan, with DeepSeek and others accounting for 30% (about 93 billion yuan), leaving the remaining companies to compete for the remaining 70%.
- Rapid industry consolidation: 2023 was a year of intense competition, but in 2024, funding activities halved. Now, only the top players are able to secure funds. This indicates that smaller companies are either being acquired or going out of business, leading to a consolidation of the industry.
IV. From "Telling Stories" to "Calculating Profitability": The Era of Burning Money is Over
In the past, the focus in the large-scale AI model industry was on who could raise the most money and create the largest models. Now, the emphasis has shifted to profitability and cost-effectiveness:
- Profitability becomes a critical metric: Baidu's AI revenue surpassed its online marketing business for the first time, and Alibaba Cloud's AI revenue has doubled for ten consecutive quarters, demonstrating that these companies can generate profits.
- Cost and pricing are key factors: Zhipu increased its prices by 80% but saw a four-fold increase in usage, indicating customer willingness to pay. DeepSeek reduced API prices by half, achieving cost savings through scale. Whether raising prices or lowering them, the goal is to optimize profitability rather than relying on massive capital expenditure.
V. The Future Challenge: Changing Logic and New Requirements
With the release of restricted shares, the trading volumes for both companies have increased significantly, eliminating the scarcity-based rationale that previously drove high stock prices. All large-scale AI model companies will now face three critical questions:
1. How will they generate profits? Through enterprise services (like Zhipu) or consumer-facing products (like MiniMax)?
2. When will they start making a profit? They can't continue to rely on heavy capital investment; there must be a clear timeline for profitability.
3. How much profit can they earn? What are their revenue growth and profit margins?
Failing to answer these questions well may result in elimination, even if the companies survive the current period. After three years of hype around large-scale AI models, it's time for them to prove their worth with tangible financial results.
In Conclusion: The contrasting performances of Zhipu and MiniMax after the lock-up expiration reflect the industry's transition from a bubble phase to a more mature stage. In the future, only those companies that can effectively convert their technological advantages into stable profits will remain competitive. Investors should focus on more concrete indicators such as revenue, profit margins, and customer retention when evaluating large-scale AI model companies, rather than just listening to their promotional narratives.