Just as the Chinese AI circle has sorted out its pecking order, the capital has changed the rules again.

In May 2026, a figure went viral: the combined valuation of the "Four AI Dragons" in China exceeded 1 trillion RMB. (Source: 36Kr, May 8, 2026)

Most reports interpreted it like this: After Zhipu listed on the Hong Kong stock market, its value increased by 7 times. MiniMax's value nearly quadrupled. Dark Side of the Moon is about to complete a new round of financing of $2 billion. DeepSeek was led by a major national fund, and its valuation is set at $45 billion. This is great. AI has really taken off.

But I think what's most notable this time isn't the trillion figure. Instead, it's that behind this figure, capital has overturned its pricing logic for large - model companies three times. Each time, it was convinced that "this time we've got it right," but was slapped in the face by the next event.

Some might say: Capital makes mistakes. So what?

Actually, it's not that simple.

There's a common underlying structure behind these three overthrows, which can be refined into a tool to help you determine which stage you're in when the next wave of hot money floods in.

First: Pricing Based on Technological Narrative

Around 2022, when the first batch of large - model companies in China received financing, the pricing logic of capital was based on technological capabilities. How many parameters did the model have? How high was its benchmark score? Could it compete with GPT - 4?

A typical conversation at that time went like this: "What's the ranking of your model's ability evaluation?" "It's in the top three on the Chinese list." "Okay, we'll give you a valuation."

This logic was effective before 2024. Since large AI models were still in the stage of capability verification, technological ability itself was the scarcest asset. Achieving good results meant the possibility of building a moat.

DeepSeek R1 shattered this logic in early 2025.

R1 proved that the same - level capabilities don't require the same - level training costs. Building a model with the same effect at one - tenth of the cost of American competitors directly undermined the pricing premise of "top - level parameters = top - level moat." Technological capabilities began to become a homogeneous asset rather than a scarce one. (Source: ChinaEntrepreneur, Report on the Six Large - Model Dragons in 2026)

First: Capital priced based on "technological scores," and DeepSeek R1 broke through the myth of technological barriers.

Second: Pricing Based on the Listing Window

After the impact of DeepSeek, the pricing logic of capital had its first shift: from "technological ability" to "certainty of listing."

The Hong Kong Stock Exchange relaxed the market - value threshold for listing of specialized technology companies. The threshold for commercialized companies dropped to HK$4 billion. In 2025, the scale of IPO fundraising on the Hong Kong stock market was about $36 billion, reaching a four - year high. The window period was obvious. Zhipu and MiniMax were the first to submit listing applications and go public, and were hyped in the secondary market. (Source: Titanium Media, May 2026)

The logic in the primary market also shifted accordingly: it was no longer about betting on which model was the strongest, but on which company could go public and cash out the fastest.

The flaws in this logic were quickly exposed in the secondary market. As of March 2026, the ARR of Zhipu's core MaaS (Model as a Service) API platform was about 1.7 billion RMB. In contrast, Anthropic's ARR was about $9 billion at the end of 2025 and exceeded $30 billion in just a few months. The gap was not on the same order of magnitude. (Source: The Paper, May 8, 2026)

Listing solved the liquidity problem but not the fundamental problem.

Zhipu's market value on the Hong Kong stock market (May 2026): about HK$434.7 billion, an increase of about 16 times compared to the pre - IPO valuation (Source: The Paper)

MiniMax's market value on the Hong Kong stock market (May 2026): about HK$257.3 billion, nearly quadrupling compared to the listing price (Source: The Paper)

Dark Side of the Moon's primary - market valuation: exceeding $20 billion, with cumulative financing exceeding 37.6 billion RMB (Source: LatePost)

DeepSeek's valuation (negotiated price in the primary market): about $45 billion, led by a major national fund (Source: 36Kr)

Third: Pricing Based on National Strategy

In the first half of 2026, the pricing logic of capital for large - model companies had its second shift. This time, the shift was faster and on a larger scale.

A major national fund led the investment in DeepSeek. This was the first time a national fund directly invested in a large AI model company. The signal was very clear: In a situation where computing power might be restricted, China needed independent model companies that didn't rely on NVIDIA or the US cloud. Large models were elevated to a strategic level comparable to chip manufacturing. (Source: 36Kr, May 8, 2026)

Meanwhile, Silicon Valley was simultaneously raising the ceiling of the reference framework: OpenAI's post - investment valuation has reached $850 billion, and Anthropic is conducting a new round of financing with a target valuation of $900 billion. Against this reference framework, the combined valuation of the Four Dragons is only 1 trillion RMB, equivalent to about $140 billion, less than one - sixth of OpenAI's single - company valuation.

This dual support of "strategic value + international reference framework" made it possible for three large - model companies to jointly raise over $10 billion in financing within three days. (Source: 36Kr, May 18, 2026)

Third: Capital priced based on "national strategy scores," but whether this score is supported by business fundamentals remains a question mark.

Putting These Three Overthrows into a Framework

I call this pattern the "Pricing Shift Model for the AI Track." Each shift in the pricing logic follows the same structure:

The old pricing logic is broken by a technological event (DeepSeek R1/Hong Kong stock listing/entry of a major national fund)

The new pricing logic is quickly established. The primary market rushes to lock in, and the secondary market follows and chases

The flaws in the new pricing logic gradually emerge in the business fundamental data

Waiting for the next technological event to trigger the next shift

This framework has two important inferences:

First, each shift is real, not a scam. The decline of technological barriers is real, the Hong Kong stock listing window is real, and the support of the national strategy is real. These are all variables with substantial content. It's just that each time, capital prices based on the "final state" rather than the "intermediate state."

Second, the real watershed is ARR. The valuation of large - model companies ultimately depends on the annual recurring revenue. Anthropic's ARR soared from $9 billion to $30 billion within a few months, which is the basis for a valuation in the tens of billions of dollars. Among the Four Dragons in China, the largest ARR is still less than one - thirtieth of Anthropic's.

Then, It's the Turn of the "Top Five" This Round

The term "Four Dragons" originally referred to Zhipu, MiniMax, Dark Side of the Moon, and Baichuan. DeepSeek was occasionally included in the early days but remained on the periphery due to its non - commercialization route. Now, the term has started to change to the "Top Five" because DeepSeek has been officially endorsed by a major national fund, and its valuation benchmark is of great significance.

The change from the "Four Dragons" to the "Top Five" is not just about the number. After the third shift in the pricing logic, the ranking method in this track has also changed: it's no longer just about the model's ability score; the "national strategy weight" needs to be added.

Where will this new pricing system be challenged next?

In ARR. When the story of national - strategy support is told, capital's focus will shift to the quarterly ARR growth rate. At that time, those with more solid fundamentals will be able to maintain their valuations, while those floating up due to the shift in pricing logic will face a reversion to the mean.

The fourth pricing shift for large models in China is on the way.

This article is only for information sharing and industry analysis and does not constitute any investment advice, investment analysis, or trading invitation. The market is risky, and investment should be made with caution. Any investment decision made based on the content of this article shall be at the investor's own risk for gains and losses. The author and the publishing platform shall not bear any legal liability.