Under-the-radar transactions of old shares: Who is quietly cashing out from Chinese large model companies?
As time has passed to the present, Zhipu has been listed for over four months, and its stock price has risen from the listing price of HK$116.20 to HK$1,001. Against this backdrop, the story gradually unfolds.
One afternoon in April 2024, an investor saw the news on their WeChat Moments and then put their phone face - down on the table and remained silent for a long time.
The news stated that Yang Zhilin, the founder of Dark Side of the Moon, cashed out tens of millions of US dollars by selling his personal shares after completing a US$1 billion financing round. "The company was only founded a year ago," the investor later said. "He got out before his investors."
Dark Side of the Moon immediately denied the news, but the denial itself amplified the controversy.
The controversy lies in several key words: old shares, value, and wealth.
Several months later, Zhu Xiaohu filed an arbitration in Hong Kong as an old shareholder of Circular Intelligence. At the end of 2024, Yang Zhilin was forced to publicly respond in writing. A dispute over old shares turned into the longest farce in the entire venture capital circle.
No one is wrong - because there are simply no rules.
This is the topic that is least formally discussed but almost unavoidable in private by every practitioner in the Chinese large - model industry in 2026: When the "Six AI Dragons" collectively rushed to the Hong Kong Stock Exchange, what exactly is happening in the old - share market hidden behind the financing rounds?
Who is selling, who is buying, who is trapped, and who left early?
The Same Card Table, Different Hands
In the spring of 2023, when Yang Zhilin spun off from Circular Intelligence to found Dark Side of the Moon, the entire large - model track was still in a state of chaos. At that time, no one knew which company would succeed, and no one expected that just two years later, these companies would stand at their respective destiny crossroads in such different postures.
Some investors had rushed in first, but a large number of them were still cautiously observing.
At that time, the "Six Dragons" - Zhipu, MiniMax, Dark Side of the Moon, Step Star, Baichuan Intelligence, and Zero One Universe - had raised a cumulative total of over 6 billion yuan in financing, accounting for more than half of the total early - stage financing in the domestic large - model field. They shared the same group of investors, competed for the same group of computing resources, and vied for the same group of top engineers.
In the eyes of outsiders, they were a whole; in the investment portfolios of GPs, they were a group of highly fate - related bets.
In this context, competition is a good thing. Only a market with sufficient competition can be heated up. At the beginning of 2026, the differentiation of this group of bets was faster than everyone expected.
Zhipu and MiniMax have been listed on the Hong Kong Stock Exchange. Zhipu's market value soared seven - fold within a month after its listing, and MiniMax also advanced vigorously. The price - to - sales ratio of the two companies once reached as high as 550 times. The pricing given by the Hong Kong stock market far exceeded anyone's expectations in the primary market.
Dark Side of the Moon just completed a US$500 million Series C financing on New Year's Eve. Yang Zhilin said in an all - staff letter that "there is no rush to go public," but then news came that the company was quietly accelerating its IPO process. Recently, Dark Side of the Moon is about to complete a new round of US$2 billion (approximately 14 billion yuan) in financing, and its post - investment valuation will exceed US$20 billion (approximately 140 billion yuan). Step Star set a new record for a single - round large - model financing with 5 billion yuan, and its Pre - IPO financing is in progress. Yin Qi has been appointed as the chairman, and the door of the Hong Kong Stock Exchange is in sight.
Baichuan Intelligence announced that it would abandon the general large - model and fully transform into AI healthcare; Zero One Universe disbanded its pre - training team and turned to vertical - scenario services. The financing records of the two companies have since fallen silent.
At the same card table, players who entered the game at the same time are in completely different situations two years later.
The Investors' Bills
To understand this differentiation, we first need to understand the people sitting on the other side of the card table - those institutions that invested their money around 2022 and are now staring at the numbers on the screen.
Legend Capital and the social security Zhongguancun special fund it manages are the largest external investors in Zhipu. They invested in Zhipu multiple times in 2022, 2023, 2024, and 2025 and became the largest institutional investor in Zhipu. They invested a total of 620 million yuan in Zhipu's multiple financing rounds and also obtained more shares through the transfer of old shares, ultimately holding 27.109 million shares.
This is a simple arithmetic problem: After Zhipu was listed in January 2026, the book value of this part of the equity exceeded HK$3.15 billion. The book return was more than four times.
However, the solution to this arithmetic problem is far more complicated than the formula itself.
"We saw this direction at the beginning of 2022," said an institutional person who participated in Zhipu's early - stage financing. "But at that time, ChatGPT hadn't emerged yet, and many LPs didn't understand what a large model was at all. You had to get approval internally, persuade the committee, and deal with the question of 'when will this company make money' - just this process was already a test."
The test didn't end after the investment. At the end of 2022, ChatGPT suddenly emerged, and Zhipu's valuation began to soar rapidly. At this time, the first real decision - making pressure appeared: Should you participate in the transfer of old shares in a certain round to lock in some profits in advance?
According to industry practice, when a primary - market project experiences significant appreciation, some LPs will achieve a phased exit through the transfer of old shares. However, if you transfer, you will have nothing to do with the subsequent appreciation after the listing. From Zhipu's valuation of US$500 million in the Series B2 round to its market value exceeding HK$400 billion after listing, selling at each time node means a permanent farewell to the subsequent price increases.
Legend's choice was to continue to increase its investment and not make an early exit. The final bill proves that this judgment was correct, but at each moment when this judgment was made, no one knew the outcome.
The social security Zhongguancun special fund managed by Legend also participated in Zhipu's investment. This detail is usually overlooked, but its significance is worth pausing for: The participants in Zhipu's wealth story include social security funds. This long - term capital with the most public nature finally shared the dividends of China's AI rise together with market - oriented institutions.
In the entire old - share map of the Six Dragons, the least - questioned investment is Mihoyo's investment in MiniMax.
In 2021, the game company Mihoyo entered MiniMax at a valuation of US$200 million as its angel investor. That year, Mihoyo had just achieved global popularity with "Genshin Impact" and was at the peak of its corporate history.
Why did a game company enter the large - model field as an angel - round investor at that time? This question has never received a formal answer. Was it a defensive bet judging that AI - generated content would disrupt the game industry? Was it personal trust in MiniMax's founder, Yan Junjie, who was once the vice - president of SenseTime and the intersection between games and AI is real? Or was it a strategic foresight earlier than the industry consensus?
In January 2026, MiniMax was listed, and the investment earned about 10 billion yuan on paper. As MiniMax's stock price rises, this figure continues to climb.
This result makes this question even more worth asking, because behind a successful investment lies the decision - making logic - and this logic can be reused.
A 50% Difference in Cost for the Same Round of Financing
If the previous stories are history that has already happened, Step Star's Pre - IPO is a real - time game and the most naked cross - section of the Chinese large - model old - share market.
At the beginning of 2026, Step Star's Pre - IPO financing was settled in two batches: The pre - investment valuation of the first batch was about US$4 billion, and that of the second batch had risen to US$5 - 6 billion. For the same round of financing, the entry costs of the two batches of buyers differed by nearly 50%.
This scenario is playing out almost every day not only in the large - model field but also in the current embodied - intelligence track.
This structure hardly ever appears in a normal financing market. Its appearance indicates that two things are true at the same time: First, there are so many buyers interested in this target that they can drive up the valuation in a short period; second, the later - entering buyers are willing to pay a premium for this price increase.
The only reason they are willing to pay the premium is the performance of Zhipu and MiniMax after their listings - a price - to - sales ratio of 550 times. That multiple has become the most powerful self - persuasion tool for later entrants: The Hong Kong stock market will give a higher premium, and I'm not the last one to pay the bill.
A joke in the industry is that someone asked an investor, "What do you think of a certain embodied - intelligence project?" The investor replied, "He/She is like an ordinary - looking classmate, but one day, suddenly the whole school is chasing him/her."
Everyone participating in large - model financing is caught in the same logic loop.
At this time, the flow of old shares is even more delicate. Some people are afraid and want to exit early, while others are firmly optimistic about the future. Who is right and who is wrong? There is no conclusion.
The Old Shares without a Market
Among the Six Dragons, the least - mentioned cases are precisely the most indispensable part for understanding the entire market.
Baichuan Intelligence and Zero One Universe chose strategic contractions. The former abandoned the general large - model and bet on AI healthcare; the latter disbanded its pre - training team and turned to vertical - industry services. The financing announcements of the two companies have almost disappeared since then.
This means that the early investors in these two companies hold a batch of old shares without a pricing anchor.
The old shares of Zhipu can refer to the market value on the Hong Kong stock market; Dark Side of the Moon has a valuation benchmark of US$20 billion in the latest round; Step Star's Pre - IPO is being priced in real - time. But how much is the market willing to pay for the old shares of Baichuan and Zero One? The industry status has changed, the track narrative has changed, and the valuation reference has disappeared.
An even harsher reality is liquidity. The buyers in the large - model track are already concentrated. When the two companies abandoned the most attractive label of "general large - model," the potential buyers of old shares will further shrink. How and at what price these equities will find an exit is an unresolved question in the entire track.
For the same group of institutions and the investment portfolios that entered the market in the same year, the old shares of the winners are queued up for purchase, while the old shares of the losers can hardly find a market - this extremely differentiated liquidity structure is the most real background color of the Chinese large - model investment cycle and needs to be remembered far more than those eye - catching return numbers.
The Unrealized Money and the Debt of Rules
From institutions to individuals, there is an overlooked hidden line in the old - share map of the Six Dragons: employees.
Among the 883 employees of Zhipu, 452 hold company shares. This 51.2% shareholding ratio is a rare high level among Chinese technology companies and is interpreted by the outside world as Zhipu's emphasis on talent incentives. After the listing, the wealth of these employees on paper has achieved a qualitative leap.
However, in the same system, there is also a silent group: former employees.
According to the common option agreement in the industry, employees usually only have a 90 - day window to exercise their options after leaving the company, and they will expire if not exercised. Exercising options means paying out of their own pockets to buy old shares at the agreed exercise price and then waiting for the listing to unlock. During the window period when the company's valuation is soaring and the IPO time is uncertain, this threshold is difficult for most people to cross - either take out a large amount of cash to bet on an unknown time point or give up permanently.
This is the most fundamental structural problem of the option mechanism in Chinese AI startups: Options are widely issued as a tool to retain employees, but the design of the monetization channel is always in the hands of the company unilaterally. For those who leave the company during its high - growth period, they participated in the creation but may never have the chance to share.
In the United States, there are relatively mature solutions to this problem. Secondary - market platforms such as Forge Global and CartaX provide quotes and matching services. The right - of - first - refusal clause is clearly defined in the shareholder agreement, and there are also industry practices for compensating the old - shareholder rights of serial entrepreneurs. The corresponding valuation of Anthropic's recent secondary - market listing exceeded one trillion US dollars, and the information is public and verifiable.
China doesn't have these. The old - share trading of large - model companies is still privately negotiated, with no public quotes, no pricing standards, and no liquidity infrastructure. Someone once commented, and it is still the most honest description of the entire market: "There is currently no common practice in the industry on whether serial entrepreneurs have cleaned up their previous ventures and how to calculate the interests of old investors."
This is not a problem of a single company, nor a moral problem of a certain founder or institution. This is the institutional debt accumulated by the entire primary market in this wave of the AI boom - the market has run too fast, and the rules haven't kept up.
The Six Dragons each have their own fates. Some have rung the bell at the Hong Kong Stock Exchange, some are still waiting for the window, some have their old shares queued up for purchase, and some can't find buyers for their old shares. Wealth is flowing in various ways at all levels: The book returns of institutions range from four - fold to nothing, employees go from being multimillionaires to having their options become worthless, and founders are walking a tightrope between cashing out and arbitration.
This differentiation is not over yet. The stories of winners are easy to tell, and there are new ones every day.
The unrealized money, the unclear rules, and the unasked decisions are the real parts of this feast that are worth recording.
This article is from the WeChat official account “Rongzhong Finance” (ID: thecapital). Author: Abu, Editor: Wuren. Republished by 36Kr with permission.