Summary of Key Points
In the past six months, China's large-scale model industry has experienced explosive growth: annual revenue has soared from $4 billion in December 2025 to nearly $13 billion, more than tripling in just eight months. The growth has not been driven by price wars (in fact, sales have increased even after price hikes), and overseas revenue has become an important source of growth (marking the first time large-scale software exports have been achieved). However, there is a significant divergence in industry valuations, ranging from 20 to 70 times annual recurring revenue (ARR). The main reasons for this are differences in strategic directions and operational efficiency.
1. Revenue Growth: Surpassing Expectations
The growth rate has been much faster than anticipated:
- Timeline Comparison: From $4 billion in December last year to $6 billion in March this year, to $10 billion in June, and now nearly $13 billion. The revenue increased by $600 million to $1 billion each month in the first three months, and by more than $1 billion each month in the last two months, with the growth rate accelerating.
- Case Studies of Leading Companies: The annual recurring revenue (ARR) of Zhipu's API business increased from $250 million in March to $2 billion in August, with August's revenue alone exceeding the total for the first half of 2026. MiniMax's revenue rose from $100 million in December last year to $800 million in August, and it is expected to exceed the $1 billion target by the end of the year.
In short, large-scale models have finally moved from a phase of "expensive research and development" to a phase of "profitable monetization," and the profits are growing rapidly.
2. Counterintuitive: Higher Prices Lead to Increased Sales
Contrary to previous assumptions that large-scale models would compete by lowering prices, this time the situation is the opposite:
- Zhipu: The average price of its API increased by 101% (doubling), but the usage of tokens increased by more than 40 times.
- DeepSeek: Some of its products saw price increases of 3 to 12 times (for example, the price of caching inputs increased by 12 times), but usage did not decline significantly.
Why? Because the capabilities of the models have improved—enterprise customers are more concerned with whether the models can solve their problems, not whether the prices are low. For instance, if large-scale models can handle complex data analysis or code generation effectively, they are willing to pay more. Additionally, market demand has surged, and with supply falling short of demand, price hikes do not affect sales.
3. Overseas Markets as a New Revenue Source
For the first time, Chinese AI companies are earning money from foreign customers through software exports:
- MiniMax: 60.8% of its revenue comes from overseas.
- Kimi: Overseas revenue exceeds domestic revenue.
- Zhipu: Nearly 40% of its revenue comes from overseas markets.
- Large Companies (Alibaba, Tencent): Their overseas model revenue also accounts for 5% to 10% of their total revenue.
This indicates that Chinese AI technology has finally established a foothold in the international market, with significant potential for future growth.
4. Valuations Vary by Three Times: Why Some Companies Are More Valuable
Despite being large-scale model companies, their valuations differ dramatically (ranging from 20 to 70 times ARR). There are two main reasons for this:
1. Different Strategic Directions:
- With limited resources, each company can only focus on 1-2 areas: Zhipu is focusing on MaaS (API services) and domestic computing power; Kimi is focusing on long-text processing; MiniMax is focusing on multimodal applications and overseas markets. Choosing the right direction leads to faster growth and higher valuations.
- Domestic companies do not have the same resources as OpenAI to change their strategies at will, so early strategic decisions are crucial.
2. Operational Efficiency Differences:
- DeepSeek: Its gross margin is 82.9% (earning $83 on every $100 sold), higher than that of the overseas company Anthropic.
- Zhipu: Its gross margin is only 24.6% (earning $25 on every $100 sold).
Higher gross margins indicate stronger profitability, and investors are willing to offer higher valuations. With the same revenue of $1 billion, DeepSeek can earn $800 million, while Zhipu can only earn $250 million—clearly, one company is more valuable.
Conclusion
This growth is not a bubble but the result of a combination of technology, demand, and strategy. In the future, only companies that can maintain "high growth and high profits" will become the true winners in this industry.