MiniMax and Zhipu have become the cleanest public-market test yet of China's AI startup boom, and the real exam may arrive when locked shares begin to loosen in July.
The strongest AI trade in Hong Kong is starting to look less like a verdict on model quality and more like a lesson in market plumbing. MiniMax Group and Knowledge Atlas Technology, better known as Zhipu, have raced higher since their January debuts, but only a tiny slice of each company is actually available for ordinary trading. That matters because a thin float can make enthusiasm look deeper than it is.
According to a Bloomberg report republished by The Edge, MiniMax has climbed around 400% since listing, while Zhipu has surged roughly 860%. Those are the kinds of moves that make a new sector feel inevitable. Yet HSBC's analysis cited in the same report puts MiniMax's freely tradable stock at only about 5% of total shares, with Zhipu's at about 4%. In a market that wants a China answer to OpenAI, scarcity has become part of the story.
The shortage cuts both ways. Investors who want exposure to Chinese foundation models have few listed options. Short sellers who think the numbers have run too far face an unusually expensive trade. S3 Partners and market data cited in the report put annualized borrow costs near 39% for MiniMax and 27% for Zhipu. When betting against a stock costs that much, price discovery gets distorted because skepticism has to clear a high hurdle before it can show up in the share price.
This is why the July lockup dates matter. HSBC estimates that Zhipu's cornerstone investor lockup expires on July 7, while a larger MiniMax block tied to cornerstone investors and employees remains unavailable for trading until July 8. Zhipu also has another employee and related-holder block that HSBC expects to be released from next January. None of that automatically means a flood of selling, but it changes the equation. A stock supported by a small tradable base can behave very differently once more holders have the option to cash out, rebalance, or simply test demand at richer prices.
MiniMax and Zhipu came to market at a moment when investors were looking for listed AI scarcity. Zhipu began trading in Hong Kong on January 8, followed by MiniMax on January 9. Earlier reporting from Reuters and other outlets described the pair as among China's leading AI model developers, with Zhipu rooted in enterprise deployments and MiniMax leaning more heavily into consumer AI applications. That gave public investors two different ways to buy into the same national technology narrative.
The problem is that the operating story is still expensive. Reuters reported in March that Zhipu's 2025 revenue rose 132%, helped by on-premise deployments and API services, but the company still posted a net loss of 4.72 billion yuan. The same report said MiniMax posted a net loss of $1.87 billion for 2025. Growth is real, but so is the cash burn. Public markets can tolerate losses when revenue is compounding fast, but they usually become less patient when valuations are being carried by scarcity as much as fundamentals.
That is the tension behind these stocks. China has produced serious AI labs, and the market is right to pay attention to them. Zhipu has pushed its GLM models into enterprise and government-linked work, while MiniMax has built consumer products that show real user appetite for AI-native applications. But neither company has yet proved that model competition, infrastructure spending, customer acquisition, and price pressure can add up to durable profits.
July could reset the next AI IPO
The bigger issue is what these two listings signal to the next wave of Chinese AI companies. A spectacular debut tells founders and bankers that Hong Kong can still finance ambitious technology stories. A sharp reset after lockups expire would send a different message: investors may love the AI theme, but they will still ask how much revenue, margin, and liquidity sit underneath it.
This matters beyond MiniMax and Zhipu. China's AI sector is trying to prove it can compete with OpenAI, Anthropic, Google, DeepSeek, Alibaba and ByteDance while also navigating chip limits, fierce domestic pricing, and rising compute demand. Public investors are now being asked to fund that race. They will want evidence that these companies can do more than release impressive models and collect strategic capital.
For now, the thin floats have created a market where momentum looks clean and conviction looks high. July will make that picture more honest. If more supply arrives and the shares hold their ground, it will strengthen the case that investors are buying future earnings power rather than just a scarce ticket into Chinese AI. If prices buckle, the lesson will be just as useful. Startup valuations do not become stronger simply because they move from private rounds to public exchanges. They only become more visible.
Also read: TSMC says AI will push the chip market past $1.5 trillion by 2030 • Huang Foundation turns AI compute into a new kind of influence • OpenAI shows why coding agents need real Windows sandboxes
