Summary of Key Points
The two domestic large-scale models, Zhipu and MiniMax, have recently released financial reports showing impressive growth, but they also reveal issues such as ongoing losses and high pressure on gross margins. More importantly, these two companies, which once followed completely different paths—Zhipu focusing on B-side projects and MiniMax expanding overseas through C-side channels—have now both shifted to a commercialization strategy that combines B-side services with APIs. They are even competing in terms of Annual Recurring Revenue (ARR). The capital market is more optimistic about Zhipu's model capabilities, assigning it a market value 4.6 times that of MiniMax. However, both companies will face the pressure of stock lock-ups in January next year, and their future performance remains to be seen.
Detailed Analysis
1. Revenue Surges, but Profit Models Have Changed—Both Moving Towards B-Side and API Services
Both companies have seen their revenue more than double in the first half of the year: Zhipu's revenue increased by nearly 400% (to 954 million yuan), and MiniMax's by 283% (to 116.6 million US dollars). The most significant change is the shift in revenue structure:
- Zhipu: The company has moved from selling complete model solutions (for local deployment) to selling usage fees through an open platform and APIs. Revenue from project-based sales has halved, while API services have surged by 2736%, accounting for 86% of total revenue. In simple terms, instead of selling a “complete set of software” to companies, Zhipu now charges per use, providing flexibility and higher frequency of revenue generation.
- MiniMax: The focus has shifted from primarily C-side products (such as its own chat app) to a greater proportion of B-side revenue. The revenue from B-side and API services has increased from 30% last year to 63.4%, while C-side product revenue has decreased to 36.6%. This means that MiniMax is now generating more revenue from companies using its model interfaces (APIs) rather than from selling its own apps.
Why this shift? The API model is more lightweight and sustainable; companies only need to pay for usage as needed, allowing large-scale model companies to earn stable fees, which is more reliable than selling projects once or relying on C-side advertising.
2. Continuing Losses and High Gross Margin Pressure—An Inevitable Part of Expansion?
Despite rapid revenue growth, both companies are still in the red: Zhipu lost 2.072 billion yuan, and MiniMax lost 358 million US dollars. A more concerning issue is the decline in gross margins:
- Zhipu: Gross margins have dropped sharply from 50% last year to 26.4%. This is due to the new cloud-based API services being in their early stages, with high costs for servers and bandwidth, which lower the overall margin.
- MiniMax: Gross margins increased from 12.1% year-over-year to 17.9%, but decreased by 16 percentage points quarter-over-quarter (from 33.7% to 17.9%). The expansion of the B-side business has required more investment in manpower and servers, leading to increased costs.
This is a common phase for large-scale model companies during expansion. To capture the market, they must invest in infrastructure and model iteration, and only as their scale grows will gross margins gradually improve.
3. The ARR Race: Different Algorithms, but High Goals
ARR (Annual Recurring Revenue) is a key indicator of a company's future profitability. Both companies are competing with different methods to estimate this figure:
- Zhipu: Uses the formula “monthly revenue × 12” and claims that its ARR reached 1.6 billion US dollars in August, with a target of 2.4 billion by the end of the year. It also jokes that its competitor uses a method of “weekly revenue × 52,” arguing that based on this calculation, it would have already exceeded 2 billion US dollars.
- MiniMax: Uses the same formula and targets an ARR of 1 billion US dollars for the end of the year, with an August figure of 800 million US dollars.
The reason for the different methods is that the monthly × 12 approach is more conservative (assuming stable monthly revenue), while the weekly × 52 approach is more optimistic (assuming weekly growth). Regardless of the method, both companies are demonstrating their potential for stable revenue generation, which is what investors value most.
4. Different Paths Initially, but Now Both Aiming for the Same Goal—APIs as the Common Destination
Previously, the two companies were perceived as follows:
- Zhipu: A “B-side expert” that sold projects to large enterprises based on its powerful model capabilities (such as the GLM series).
- MiniMax: A company that expanded overseas through C-side products (such as its chat app) to generate revenue.
Now, both are focusing on APIs, but with different approaches:
- Zhipu: Its model capabilities drive growth. The more advanced its models (e.g., GLM5.3), the more willing companies are to use its APIs, creating a cycle of “model iteration → increased API usage → revenue growth.”
- MiniMax: It uses C-side user data to iterate its models, then sells the models to enterprises, forming a cycle of “C-side users → model optimization → B-side API revenue.”
Although their starting points were different, both are ultimately aiming for APIs as the most sustainable commercialization model.
5. A 4.6-Time Market Value Gap: Model Capability Is Key, with Future Challenges Looming
As of August 31, Zhipu’s market value was 4.6 times that of MiniMax, despite Zhipu having only 1.2 times the revenue. The reason for this gap is that the market values Zhipu’s model capabilities more:
- Zhipu’s GLM series has attracted industry attention, suggesting it has the potential to generate more revenue in the future.
- Although MiniMax’s models (e.g., M3) are strong, the market believes that the competition in the large-scale model market depends on technology, not just user numbers.
However, both companies face potential risks: A large number of initial shares will be released in January next year (40% for Zhipu and core employees for MiniMax), which could lead to stock price fluctuations. Both company founders have taken measures to stabilize investor confidence: Yan Junjie (MiniMax) has pledged not to receive a salary until AGI is achieved and is offering shares as incentives to the team; Tang Jie (Zhipu) has launched a “reach-for-the-peak” initiative to push technological boundaries without focusing on short-term profitability.
In Conclusion
Zhipu and MiniMax are both transitioning from independent strategies to competition in the API market. Although they are still in the process of burning capital, Zhipu, with its stronger model capabilities, is favored by the capital market. However, the pressure of stock lock-ups next year and the speed of model iteration will determine which company will emerge as the winner.