Summary of Key Points
MoonFace and DeepSeek are the hottest AI large-model companies in China's primary market, with their valuations rising at an astonishing pace (MoonFace’s valuation increased by 7 times in half a year, while DeepSeek’s increased by 36% in just one month), approaching or even surpassing some of their counterparts in the secondary market. However, the reasons behind their high valuations differ: MoonFace relies on model performance (with a large number of parameters in its Kimi K3 and strong capabilities) and expected revenue growth; DeepSeek, on the other hand, benefits from leading global usage volumes, high computational efficiency, and the adaptation of domestic computing power—advantages that are not purely financial in nature. Nevertheless, they will face challenges such as proofing their growth rates during future IPOs, pressure from share lock-ups expiring, and having to compete with their valuation benchmarks. There is also a potential risk that the gap in model capabilities between China and the United States could widen again.
I. MoonFace and DeepSeek: Different Foundations for Their Soaring Valuations
MoonFace: Leveraging “Hard Power” and “Revenue Expectations”
- Strong Model Performance: The newly released Kimi K3 has 2.8 trillion parameters (the more parameters, the smarter the model), can handle contexts of up to 1 million tokens (able to process hundreds of thousands of words at once), and can also understand images. In terms of AI capabilities, it is on par with top models like OpenAI and Anthropic, and even ranks first in certain code-related tasks. The company plans to open-source the model weights, which will attract more developers and further expand its influence.
- Positive Revenue Growth Expectations: The market expects MoonFace’s annualized recurring revenue (ARR) to reach $1 billion by the end of the year. Its current valuation of $31.5 billion corresponds to a P/ARR ratio of about 70 (the lower this ratio, the more realistic the valuation). Investors are betting that it can follow the path of Zhipu—Zhipu’s valuation in its primary market round was $13.7 billion, and its stock price soared to $10 billion after going public, representing a multiple increase after listing.
DeepSeek: Relying on “Usage Volumes,” “Efficiency,” and “Strategic Scarcity”
- Leading Global Usage Volumes: On the OpenRouter (a global model invocation platform), DeepSeek has more usage volumes than Anthropic. High usage indicates user recognition, which holds great potential for converting into revenue in the future.
- High Computational Efficiency: The new model requires only 1/20 of the cost of its predecessor to process the same amount of content. Although its API fees are much lower than OpenAI’s, it still maintains a gross margin of over 50% (low costs without sacrificing profitability).
- Special Strategic Positioning: DeepSeek is adapted to domestic computing powers such as Huawei’s Ascend and Cambricon, and it is backed by a national AI fund, making it a “national-supported benchmark for domestic models.” Additionally, the founder has invested personally with $20 billion, which adds to investors’ confidence.
II. The Primary Market Is Willing to Offer High Valuations, but the Secondary Market Is More Realistic?
The valuation logic in the primary market (for unlisted companies) differs significantly from that in the secondary market (for listed companies like Zhipu and MiniMax):
- Zhipu: Achieving High Valuations Through Rapid Growth: Its stock price increased by 10 times since its listing in January this year, as its ARR grew from $67 million to $1 billion in just 7 months—faster than Anthropic’s growth rate that same year. The market is willing to pay for such rapid growth.
- MiniMax: Valuation Drops Due to Unclear Narrative: Although it had a high valuation at listing, the industry shifted focus to “code and intelligent assistants,” while MiniMax’s approach of targeting both text and image tasks in a multi-modal manner, as well as serving both corporate and individual markets, led to doubts about its focus, resulting in a 75% drop in its market value.
- **The Primary Market’s “Fuzzy Benefits”: MoonFace and DeepSeek are still unlisted, so information is less transparent, allowing investors to pay a premium for their potential. However, the secondary market relies on real data daily; if performance fails to meet expectations, valuations will decline immediately.
III. Three Challenges They Will Face During Future IPOs
If these companies go public, they will face more stringent tests:
1. Quarterly Growth Rate Verification: The secondary market will regularly assess their revenue growth. For example, although DeepSeek currently leads in usage volumes, the price of its services has plummeted (from $1.2 per million tokens in March to $0.058 in June). Whether high usage volumes can translate into revenue growth is a question that must be answered in each quarterly report.
2. Pressure from Share Lock-ups Expiring: Both Zhipu and MiniMax saw declines in their stock prices after share lock-ups were lifted, and they also had to issue additional shares to raise funds. Similar situations could impact these companies’ stock prices if early investors sell their shares after listing.
3. Valuation Benchmarking Must Be Realistic: After going public, their valuations will be compared with those of their peers based on the “monetization rate per token” (how much money is earned per token processed). They can no longer rely on attractive narratives; for instance, Anthropic’s current P/ARR ratio is only 20 times, while DeepSeek’s is 148 times—this large gap will likely shrink after listing.
IV. Are Our Valuations Excessively High Compared to American Companies Like Anthropics?
Anthropics, a leading U.S. AI company, had a valuation of $965 billion after its May funding round, with an ARR of $47 billion, resulting in a P/ARR ratio of about 20 times. In contrast, DeepSeek’s and MoonFace’s ratios are 148 times and 105 times, respectively. This reflects an “anticipation of future growth”:
- Anthropics’ ratio has decreased from 184 times at the early stages to 20 times now, indicating that as the company grows, the ratio will likely decrease. The same will apply to our companies; the key is whether their revenue growth can outpace this reduction.
- Zhipu’s current ARR of $1 billion has already reduced its valuation ratio significantly, showing that it’s possible for high valuations to be justified if growth continues.
V. Could the Gap in Model Capabilities Between China and the U.S. Widen Again?
Current valuations for Chinese models are based on the assumption that the gap between the two countries is narrowing. However, a Jefferies report suggests that the U.S. may gain access to newer generations of computing power (such as more advanced GPUs) in the second half of the year and could use “anti-distillation” techniques to prevent China from replicating their models, potentially widening the gap again.
Although the release of MoonFace’s Kimi K3 has temporarily alleviated these concerns, if U.S. models make faster progress, our high valuations could be challenged, as model capabilities are fundamental.
Conclusion
The current high valuations for MoonFace and DeepSeek reflect the market’s optimism about the future of Chinese AI large models. However, after their IPOs, they will need to prove their worth with actual revenue growth and stable monetization capabilities. The direction of technological competition between China and the U.S. will also significantly impact their valuation prospects. For ordinary investors, while the excitement is palpable, it’s important to understand that high valuations carry significant risks and involve bets on future performance.