Summary of Key Points
As one of the first large-model unicorns to go public, MiniMax experienced a surge in its stock price at the beginning of its listing (with a market value reaching as high as HK$410 billion), but within half a year, its market value plummeted to HK$68.4 billion, showing a significant gap compared to Zhipu, which went public during the same period and was valued at HK$575.9 billion. The core issue was the failure to establish a positive cycle of "foundation model capability → commercialization results → infrastructure investment": the new generation foundation model M3 underperformed, leading to insufficient commercialization barriers and a lack of capital willing to invest on a large scale, ultimately resulting in its rejection by the market. Although its fundamentals (such as revenue growth and margin improvement) showed signs of improvement, the weakness in its foundation models hindered its development.
I. Stock Price Volatility: The Market Focuses on a Positive Cycle, Not Short-Term Growth
MiniMax was highly favored at the time of its listing due to its broad business model covering both consumer (C-side) and enterprise (B-side) markets, as well as international operations, with revenue growth exceeding expectations. However, the subsequent drop in stock price reflected the market's realization that it had not formed a self-reinforcing positive cycle.
- Initial Price Rise: The C-side offerings, including Talkie (AI companionship) and Xingye (AI creation), along with an open platform for the B-side, and international revenue accounting for 73%, suggested significant potential to investors.
- Subsequent Price Drop: The high cost of acquiring customers on the C-side (international traffic settled in US dollars, with promotion expenses accounting for 92% of sales costs) and a low gross margin of only 4.7% were major factors. More critically, the underperformance of the new foundation model M3 made its products less competitive, preventing further investment in infrastructure (such as purchasing computing power), thus breaking the cycle. Investors, being astute, voted with their money when they saw this deadlock.
II. Business Model: The "All-Around" Approach Hides Weaknesses on the C-Side
MiniMax's business model appeared comprehensive, covering all three segments (C, B, and international), but the C-side was unprofitable:
- C-Side Challenges: Products like Talkie had high customer acquisition costs due to expensive international marketing efforts, and user retention was poor because the AI companionship experience was one-way, with no emotional connection between users and the AI (e.g., data deletion by the platform could result in the loss of user relationships). The C-side contributed 67% of revenue but had a mere 4.7% gross margin, meaning for every HK$100 in sales, only HK$4.7 was earned, with significant additional expenses required to acquire new users.
- B-Side Strength: The B-side experienced faster revenue growth (197.8%) and higher margins, which helped boost overall profitability. However, the B-side also faced competition from companies like Zhipu, which had stronger barriers due to government and enterprise clients.
III. Profit Improvement: Limited by C-Side Constraints and Efficiency Gains
MiniMax's gross margin increased by 437% in 2025, rising from 12.2% to 25.4%, but the adjusted loss was still HK$1.756 billion (similar to the previous year). The improvement in profitability was mainly due to cost-cutting measures:
- Cost Cuts: Reduction of C-side promotion expenses by 22.3%
- Efficiency Enhancements: Optimizing model efficiency (e.g., reducing memory usage and improving GPU utilization) lowered computing costs significantly.
- Open Source Strategy: An increase in B-side revenue contributed to higher overall margins. However, the net profit margin remained negative at -356%, indicating that for every HK$1 earned, HK$3.5 was lost.
IV. M3 Foundation Model Fails: The Core of the Positive Cycle
The foundation model is the heart of a large-model company (equivalent to the brain of an AI system). The poor performance of M3 after its release disrupted the positive cycle:
- Underperformance: High expectations before its launch were not met; the usage rate of OpenRouter dropped from third place to eighth, compared to DeepSeek V4.
- Pricing Issues: The model was forced to be sold at half-price permanently, indicating that the market did not recognize its original pricing as competitive.
- Infrastructure Weakness: Most research and development (R&D) expenses were spent on renting computing power from Alibaba (cloud computing accounted for 79% of R&D costs in the first nine months of 2025), with no investment in own hardware (fixed assets amounted to only HK$11 million). This slow iteration pace hindered its ability to catch up with competitors.
V. Industry Divergence: Accelerated Segregation into Two Groups
The large-model industry is now divided into two distinct camps:
- Advancing Group: Companies with strong foundation models have better commercialization (high usage rates, stable clients) and are willing to invest in infrastructure (purchasing computing power, building data centers), leading to even stronger foundation models, such as Zhipu.
- Regressing Group: Companies with weaker foundation models struggle with commercialization and are reluctant to invest in infrastructure, resulting in further decline, like MiniMax.
The rapid divergence is due to open-source models (like DeepSeek V4) making 85%-90% of the technology accessible to everyone. The technological gap has narrowed from several years to just two to three months, and companies without core foundation model capabilities are quickly eliminated.
MiniMax's strengths include strong international operations and efficient technology, but its weak foundation models are a major setback. It also faces pressure from convertible bonds that need to be repaid upon maturity. Whether it can turn things around depends on its ability to overcome the hurdle of investing in infrastructure first and then waiting for the foundation model to improve. In the large-model race, sometimes you must "believe first" to see results.
Conclusion
MiniMax's story illustrates the polarization in the large-model industry: short-term growth alone is not enough; a positive cycle of "foundation model → commercialization → infrastructure" must be established. For investors, evaluating large-model companies should not focus solely on revenue growth but also on the competitiveness of their foundation models and whether they have their own infrastructure—these factors are crucial for determining long-term value.