MiniMax Hong Kong stock falls below IPO price, Yan Junjie's net worth shrinks by about HK$92.5 billion in half a year

Hong Kong tech stocks plunged collectively, and MiniMax (0100.HK), one of Hong Kong's "two big model giants," also fell below its issue price.

On October 22, MiniMax's share price opened low and moved lower, closing down 6.52% at HK$160.7 per share, with its total market value dropping to HK$56.6 billion. This was also the first time since MiniMax's listing that its share price fell below the issue price of HK$165 per share.

As one of the two AI large-model companies listed in Hong Kong, MiniMax saw a rapid rally after its listing. In March this year, the company's share price rose as high as HK$1,330 per share, more than 7 times the issue price, and its total market value also exceeded HK$410 billion.

After that, MiniMax's share price began to pull back. Compared with the high point in March, the company's latest share price has fallen by nearly 90%, and its total market value has evaporated by more than HK$350 billion.

As MiniMax's share price declined, the net worth of company founder Yan Junjie shrank accordingly.

Currently, Yan Junjie holds 79.1025 million MiniMax shares, a stake of about 22.65%. At the peak of the company's share price in March this year, Yan Junjie's net worth once exceeded HK$100 billion. As MiniMax's share price fell by nearly 90%, Yan Junjie's net worth also continued to shrink, with the latest value of his shareholding at about HK$12.7 billion.

In just seven months, Yan Junjie's net worth fell by about HK$92.5 billion.

Similar to MiniMax, Zhipu (2513.HK), another Hong Kong-listed AI large-model company, has also seen its share price fall sharply in recent months. On October 22, Zhipu's share price plunged 7.10% to HK$460.8 per share, with its total market value dropping to about HK$225 billion. Compared with the high point of HK$2,980 per share in June, the cumulative decline during the period also reached about 85%.

Model capability remains the core variable

MiniMax and Zhipu both listed on the Hong Kong stock market in January this year, but after listing, their share price performances followed different trajectories.

MiniMax was founded five years ago. Its founder, Yan Junjie, is 37 this year. He once worked at Baidu and later joined SenseTime, rising all the way to vice president and deputy head of the research institute. In 2021, Yan Junjie left SenseTime to start his own business, and MiniMax was born.

At the time of listing, MiniMax was favored by the market for its higher overseas revenue, and its share price reached a high of HK$1,330 per share in March. But as the AI sector underwent a revaluation, the market began to focus on Zhipu, which has stronger model capabilities.

After that, the two companies' share price performances crossed paths. In July, Zhipu's market value once reached 7 times that of MiniMax. Today, Zhipu's market value is still 4 times that of MiniMax. While MiniMax's share price has fallen below its issue price, Zhipu's share price is still nearly 3 times its issue price.

The difference in the two companies' market performance lies in the gap in their model capabilities. In the market's view, as the large-model sector becomes increasingly crowded, only companies with stronger model strength and continuous iteration capability can gain a competitive advantage.

In June this year, MiniMax released its flagship large model MiniMax M3. At the time, the company's official website showed that MiniMax M3 was at the industry frontier in capabilities for professional tasks such as coding and agents. On the SWE-Bench Pro leaderboard, which tests coding ability, M3's score surpassed GPT-5.5 and Gemini 3.1 Pro and approached Claude Opus 4.7.

But in the four months since then, MiniMax still has not released a new flagship large model. It only updated a Flash model, M3.1 Flash, on October 19. On the Artificial Analysis leaderboard, this model's score is slightly higher than the previously launched GLM-5.3 Flash from Zhipu, but still lower than Zhipu's flagship large model GLM-5.3.

MiniMax previously said it is developing two models, M3.1 and M3 Pro, with M3 Pro's parameter scale expected to increase to about 3T, further expanding reinforcement learning and long-horizon task training. But when these models will be launched remains unknown.

By comparison, Zhipu's model iteration is faster and even once reached the position of domestic SOTA (state of the art).

In June and August this year, Zhipu launched two models, GLM-5.2 and GLM-5.3. When GLM-5.3 was first launched, its score on the Artificial Analysis leaderboard reached as high as 60 points, tied with Kimi K3 for first among open-source models. The GLM-5.3 Flash (Niulai large model) launched by the company in August this year also attracted market attention for a time.

In a research report in July, JPMorgan said that open weights expand the scope of model distribution, but monetization capability still depends heavily on model quality, iteration speed and endpoint reliability. For models with leading capabilities, open weights can expand usage scale through external GPU compute such as cloud service providers, inference platforms and private deployments. But for models with insufficient differentiation, broader access instead accelerates price comparison and traffic substitution.

Of course, relying on its long-term accumulation in the consumer market, MiniMax has a certain competitive advantage in multimodal capabilities. In July this year, MiniMax launched the open-source video generation model MiniMax H3, which focuses on high cost-effectiveness. Within three weeks of launch, downloads exceeded 24 million, making it one of the main competitors to ByteDance's Seedance 2.0.

The recent overall revaluation of the AI sector is also one of the important factors driving MiniMax's share price downward.

Since ChatGPT set off the generative AI revolution, the AI sector has entered an unprecedented period of rapid development. Many views hold that AI will reshape a host of traditional industries while creating new ones, thereby creating the biggest industrial opportunity of the next decade.

But in the first half of this year, the market's enthusiasm for AI began to diverge from the sector's fundamentals. At MiniMax's share price high in March this year, the company's price-to-sales ratio (TTM) once exceeded 600 times, and Zhipu's price-to-sales ratio (TTM) at its June high even exceeded 1,000 times.

As one of China's leading large-model companies, MiniMax once relied on consumer-facing AI products to support revenue. As large-model capabilities improved, the business-facing open platform and other AI-based enterprise service businesses have replaced consumer products as MiniMax's main source of revenue.

In the first half of this year, revenue from MiniMax's open platform and enterprise service business reached US$74 million (about RMB 497 million), up 703.1% year on year, with its revenue share rising from 30.3% in the same period last year to 63.4%.

Driven by this business, in the first half of this year, MiniMax's total revenue rose 283.1% year on year to US$117 million (about RMB 778 million), while its adjusted net loss widened 52.7% year on year to US$293 million (about RMB 1.96 billion).

But compared with Zhipu, MiniMax's revenue scale still lags somewhat. At this year's interim results meeting, Zhipu's management mentioned that as of the end of August, the company's ARR had reached US$1.6 billion, while MiniMax's ARR was US$800 million.

To incentivize the team to deliver more results, Yan Junjie issued an internal letter in July this year, announcing that until the day the company achieves AGI, he will no longer receive any compensation from the company. And over the next four years, he will set aside shares in his own name equivalent to 4% of the company's total share capital to incentivize team members who have fought alongside the company over the long term and jointly created value; at the same time, he will set aside 1% of shares to establish a special fund to continuously support the development of relevant open-source communities.

This article comes from the WeChat public account "Damofinance" (ID: damofinance), author: Damo Finance, published by 36Kr with authorization.