Behind the Thousand-Dollar Stocks: The Narrative Drive and Valuation Reconfiguration of the AI Rally in Hong Kong Stocks

In the Hong Kong stock market of Q1 2026, the pricing of AI assets has clearly entered an emotion-driven phase. The narrative surrounding large models continues to heat up, propelling companies like Zhipu AI and MiniMax, which are still in the early stages of commercialization, to be rapidly bid up by capital in a short period, even touching the "thousand-dollar stock" range at one point.

This price change does not essentially reflect a synchronous leap in corporate operations; it is more like the market casting an early vote for the "platformized future of AI" against a backdrop of ample liquidity and a strengthening narrative. In other words, what is being traded is not the present, but an amplified long-term probability distribution.

A noteworthy change is that the pricing logic of this round of AI market activity is gradually detaching from the traditional fundamental framework. In the past, investors would at least make marginal judgments based on revenue, profit, and growth quality. Now, the discussion is more about "whether they possess the potential to become a platform-level gateway." When the market begins pricing in this way, the assets themselves are already being redefined as a type of "growth target with option-like attributes."

From the perspective of capital behavior, this round of rally is closer to a typical "narrative-driven market." Discussions around AGI, Agents, and AI-native applications continue to spread, endowing AI with the expectation of reshaping all industries. The problem with this expectation is that it often advances faster than commercial reality, thereby amplifying valuation elasticity in the short term.

In such an environment, the explanatory power of traditional financial analysis frameworks is indeed declining, but it has not completely failed. Taking Zhipu as an example, its revenue still mainly relies on enterprise-level API calls, industry solutions, and government-enterprise cooperation; overall, it remains a typical B2B model company. MiniMax leans more towards C-end multimodal applications and subscription models, but commercialization is still in its early stages, with revenue scale and stability not yet verified through a full cycle. The commonality between the two lies in: revenue is still relatively early-stage, and a stable, scaled cash flow structure has not yet been formed.

At the same time, rigid constraints on the cost side remain very apparent. Whether it's the computing power investment required for large model training, the R&D expenses brought by continuous iteration, or the long-term reliance on high-end talent, all determine that this type of company will find it difficult to enter a state of stable profitability in the short term. Under this structure, using the P/E ratio to explain valuation is clearly invalid, and even discounted cash flow models become highly distorted due to the uncertainty of future paths.

The market has therefore turned to another, more "unconventional" pricing method: using probability to price the future industry landscape. That is to say, the current price is not based on the value already created by the enterprise, but on trading the possibility of them becoming the next-generation platform. This logic is closer to an option than the stock itself.

The problem is that this pricing method itself is highly dependent on assumptions: Will the models converge into an oligopolistic structure? Do the enterprises possess the ability to build an ecosystem? Can a truly AI-native application gateway be born? There are currently no clear answers to these questions, but the price is already giving the answer in advance.

Especially the point of "ecosystem capability" is gradually outweighing the model capability itself. Historical experience repeatedly proves that technological advantages can be caught up with, but once an ecosystem is formed, it creates stronger path dependence through developer systems, distribution mechanisms, and user habits. Therefore, the market's analogy of these companies to early Apple or Microsoft is not entirely a judgment on technology, but more an imagination of the platform structure.

The other side of this imagination, however, is significant fragility. On one hand, the current market lacks a mature valuation anchor, with prices determined more by a combination of capital and sentiment. On the other hand, large model technology is still evolving rapidly, and open-source models and architectural innovations could reshape the competitive landscape at any time. Coupled with the fact that commercialization paths have not yet been fully verified, the pricing at this stage is inherently unstable.

From an industrial structure perspective, the ends with relatively higher certainty are actually infrastructure and vertical applications. Computing power and cloud services have strong path dependence characteristics, belonging to the underlying resources that "whoever wins must rely on." Vertical industry AI applications, on the other hand, can more easily form a real cash flow closed loop. In contrast, general-purpose large model companies, while at the center of the narrative, also bear the greatest expectation volatility.

Overall, the current AI market activity has, to some extent, already front-loaded the long-term probability of "platform victory." What the market is pricing is not current profitability, but the possibility of becoming future industry infrastructure. And in the early stages of a technological revolution, this probability is often systematically overestimated and will be recalibrated through severe volatility in the medium term.

The industry will most likely go through an evolution of "frenzy — shakeout — reconstruction," with only a few enterprises ultimately able to establish sustainable competitive advantages.$KNOWLEDGE ATLAS(02513.HK)