Summary of Key Points
DeepSeek has initiated preparations for its IPO, and if successful, it will become the third major Chinese large-model company to go public after Zhipu and MiniMax. Each of these three companies has its own focus: Zhipu emphasizes enterprise APIs, coding tools, and intelligent assistants (agents); MiniMax focuses on multi-modal consumer products and globalization; DeepSeek, on the other hand, prioritizes open-source technology and low prices. After MiniMax's listing, its stock price fluctuated due to share lock-ups being lifted and financing activities, and it also faced competition from DeepSeek's pricing strategy. Although MiniMax's diverse product portfolio has attracted users and generated revenue, it suffers from resource dispersion and a lack of a dominant core product. OpenAI's decision to streamline its product lines provides new insights for companies like MiniMax.
Detailed Analysis
1. The Three Large-Model Companies: Each with Its Own Strengths, Competition Already Underway
These three companies are like competitors in different fields, each with their unique selling points:
- Zhipu: Targets enterprise services, focusing on coding tools, intelligent assistants (agents), and APIs that meet the needs of corporate clients. Its revenue mainly comes from institutional customers.
- MiniMax: Develops consumer products such as chatbots (Talkie), video generation tools (Seinoe), and music applications (HailuoAI), with a focus on globalization. In 2025, overseas revenue accounted for 73% of its total income, and it had over 236 million users.
- DeepSeek: Adopts a strategy of low prices and open-source technology, offering its models at very competitive rates (for example, the V4-Pro model was temporarily discounted by 75%). This approach has made DeepSeek a significant player in the large-model market.
The competition between these companies has been intense from the beginning. When DeepSeek reduced its prices, Zhipu and MiniMax's stock prices dropped by more than 9%. MiniMax attempted to counter with its M3 model, but changes to its pricing packages caused dissatisfaction among users. This battle has been ongoing for some time, not just since their listings.
2. MiniMax's Stock Price Volatility: Caused by Share Lock-ups and Financing?
MiniMax's stock price soared to HK$1330 ($165 at launch) after its listing but then plummeted after the lock-ups were lifted. On July 9, the stock price fell by 17.98%, and another 9.68% the following day. The main reasons for this were:
- Pressure from Share Sales: Early shareholders could now sell their shares, leading to a decrease in market demand.
- Fears of Financing Dilution: MiniMax announced plans to issue new shares and convertible bonds, which would reduce the proportion of existing shareholders' equity. Investors were concerned about this and sold their shares.
The deeper issue is whether MiniMax's business model can sustain its growth given its high research and development (R&D) expenses, which exceed three times its revenue. If these costs cannot be converted into stable profits, the company may face future challenges.
3. Intensifying Price War: How DeepSeek's Low-Price Strategy Affects Competitors
DeepSeek's low-price strategy has directly impacted MiniMax:
- When DeepSeek released the V4 model in April, it emphasized its cost-effectiveness, causing Zhipu and MiniMax's stock prices to drop by over 9%.
- Later, DeepSeek offered a temporary discount of 75% on the V4-Pro model, and the price of cached inputs was reduced by 10 times, further driving down MiniMax's stock price by more than 14%.
MiniMax initially gained popularity with its M2.5 model, which had high usage on the OpenRouter platform (3.07 trillion tokens per week). However, DeepSeek's continuous price cuts have made cost-effectiveness less of an advantage. It's like a competition between two奶茶 shops: if one sells a cup for 10 yuan and the other for 2 yuan, customers will naturally choose the cheaper option.
4. MiniMax's Dilemma: Is a Diverse Product Portfolio a Strength or a Burden?
While MiniMax has a wide range of products (text, voice, video, music) and a large user base, this also poses challenges:
- Resource Dispersal: Its R&D expenses of $252.8 million in 2025 exceeded three times its revenue, requiring significant investment in each product line. This approach is similar to operating multiple businesses with limited resources, potentially preventing any one product from becoming a market leader.
- Lack of a Core Product: Companies like Anthropic have achieved substantial success with their core products (e.g., Claude Code for coding). Although MiniMax has developed its own coding tool, it hasn't become as profitable as Claude Code. Without a dominant product, revenue is less stable.
Yan Junjie's company-wide message (no salaries for employees, 5% of shares allocated as incentives) aims to boost morale by showing commitment to R&D. However, without addressing resource dispersion, long-term growth will be impacted.
5. OpenAI's Product Line Consolidation: A Lesson for MiniMax
OpenAI has recently restructured its products, integrating Sora (video generation) and Atlas (browser) into ChatGPT and Codex. For example, Codex's features are now integrated into ChatGPT Work, allowing users to access all functions through a single interface.
This suggests that MiniMax should focus on a core area instead of trying to cover everything. Although its products like Talkie and Seinoe have users, these capabilities could be consolidated into a central platform (e.g., the MiniMax Agent) to reduce redundancy and direct resources towards more profitable areas (such as enhanced coding tools or enterprise APIs). Otherwise, efforts will be spread thinly, and each product may not achieve excellence.
Conclusion
DeepSeek's IPO will make the competition among these companies more transparent. However, MiniMax's main challenge is finding a balance between diversifying its products and concentrating resources. OpenAI's approach and Anthropic's success with coding tools provide valuable lessons for MiniMax. In the high-cost realm of large-model development, being comprehensive may not be as effective as focusing on one area of excellence.