Summary of Key Points
Companies specializing in pure large-scale AI models, such as Zhipu and MiniMax, may have modest revenues and significant losses (for example, Zhipu had revenue of 724 million yuan in 2025 but a loss of 4.7 billion yuan). Nevertheless, they have achieved market valuations hundreds of times their sales and market capitalizations in the Hong Kong stock market and have begun the process of returning to the A-share market for listing. The underlying rationale is not solely commercial profitability but rather a combination of national strategic needs for “sovereign AI” (in the context of a global AI competition) and the “asset shortage” resulting from the decline in the real estate sector (with funds seeking new investment opportunities). However, these companies face three major challenges: their value being eroded by upstream computing power providers, technological convergence, and limitations in access to computing resources. In the long run, their valuations must ultimately reflect the true value they create for users.
1. Why a Loss of 4.7 Billion Yuan Results in a Valuation of 700 Billion Yuan? It’s About “National Security”
You might wonder: How can a company that earns 700 million yuan but incurs a loss of 4.7 billion yuan be valued at 700 billion yuan? This is not a typical commercial calculation; these companies represent China’s sovereign AI efforts. The AI competition between major nations is akin to the rivalry in nuclear technology and space exploration—it determines a country’s future competitiveness. Without its own large-scale AI models, China could become vulnerable and fall behind. This concern for security outweighs the potential for profit; even if it means incurring substantial losses, having independent technology is deemed worthwhile. What the country needs is not just one champion but a group of promising startups (such as Zhipu, MiniMax, and DeepSeek) competing together. Even if most fail, at least one could succeed. This is similar to buying insurance: you pay a fee for peace of mind in the event that your technology is not constrained by others. The high valuations of these pure AI models essentially represent the “insurance premiums” paid by the state through the capital market.
2. Where Does the Money Come From? Trillions of Yuan Seeking New Investment Opportunities After the Real Estate Decline
Over the past 20 years, most Chinese households have invested their savings in real estate (accounting for nearly 70% of their assets), as it was seen as a safe and profitable investment. But with the stagnation in the real estate market, funds are looking for new opportunities. Interest rates on deposits are declining, and returns from financial products are low. AI has become an attractive option: it offers the potential for breakthroughs, is backed by national policies, and holds vast possibilities for development. Moreover, there are very few companies specializing in pure AI (few can be counted on one’s fingers), so a large amount of capital is flowing into this field, driving up their valuations.
3. Can Pure AI Companies Sustain Their Profitability When Industry Giants Are Spending Heavily?
Tech giants like Tencent, Alibaba, and ByteDance have invested billions without finding a profitable model:
- Alibaba’s capital expenditure in 2025 was 103.9 billion yuan (compared to 24.4 billion yuan three years ago), with almost all of its profits going towards AI initiatives;
- Tencent’s AI division consumed 8.8 billion yuan of its profits in the first quarter;
- ByteDance plans to invest 500 billion yuan in 2026 (almost using up its annual profits).
These giants have access to large user bases and relevant applications but still cannot turn a profit. For companies like Zhipu, which rely on selling AI models, things are even more difficult; they lack their own user scenarios and can only sell APIs or models, with much of the profit going to upstream providers of computing resources (chips and rental services). For instance, when a pure AI company raises prices for its services, the cost of computing power also increases, leaving its profit margins unchanged. If even these giants cannot achieve profitability, how can pure AI companies?
4. Three Critical Challenges Facing Pure AI Companies
Despite their current high valuations, pure AI companies face three insurmountable hurdles:
1. Value Displacement by Upstream Providers: The demand generated by AI is often captured by chip manufacturers and computing power rental services, leaving model companies with minimal profits;
2. Technological Convergence: The capabilities of leading domestic AI models are becoming increasingly similar, shifting the competition from technology to access to channels and ecosystems—something pure AI companies lack (since platforms like WeChat and Taobao have established user bases);
3. Limited Computing Power: There is a shortage of domestically produced chips, with some providers only able to supply a few hundred cards per month, and imports are restricted. AI development requires substantial computing power, which these companies often cannot afford.
5. From “Storytelling” to “Real Value”: The Ultimate Determinant of Valuations
In the short term, national strategic needs and the asset shortage may support high valuations. However, in the long run, only when AI truly creates value for users (e.g., by reducing costs or improving efficiency for businesses or making life easier for individuals) will valuations remain sustainable. The current high market valuations are more like promissory notes; whether they can be realized depends on whether pure AI companies can overcome these challenges and meet real user needs.
(The entire analysis is written in plain language to make the underlying logic understandable to non-finance professionals.)