The artificial intelligence premium in Hong Kong is evaporating. This week, the market faces a massive liquidity test as cornerstone investor lock-ups expire for Zhipu AI and MiniMax, two of China’s high-profile generative AI pioneers. Roughly six months after their high-flying debuts in January, the initial euphoria has been replaced by a cold reality. The lifting of trading restrictions will release billions of dollars in shares, threatening to flood an already volatile market and forcing institutional investors to decide whether to hold the line or cut their losses.
It is a classic market reckoning. A few months ago, scarcity value drove these stocks to dizzying heights, but the arrival of lock-up expirations strips away the speculative shield. Investors are shifting their focus from broad promises of general intelligence to the punishing reality of corporate income statements, expensive infrastructure, and a brutal domestic price war.
The Approaching Supply Shock
The scale of the shares hitting the market explains the nervous trading rooms across Hong Kong. Zhipu AI, listed under its corporate entity Knowledge Atlas Technology, faces the expiration of restrictions on 25.68 million cornerstone shares. This represents nearly six percent of its outstanding equity.
MiniMax faces an even more precarious hurdle. Unlike its peer, MiniMax will see almost half of its total outstanding shares unlocked as early investors, anchor backers, and cornerstones become free to sell. This sudden expansion of the tradeable float threatens to distort the supply-demand balance overnight.
Markets often price in these events ahead of schedule. The sell-off began in late May, when both companies reached peak valuations. Zhipu had rocketed to an intraday high of HK$1,993, briefly commanding a market capitalization near HK$880 billion. MiniMax peaked above HK$410 billion, drawing comparisons to traditional tech heavyweights.
Then came the correction. Within weeks, both stocks shed roughly half their value. By early July, MiniMax had tumbled over 70 percent from its historic high, wiping out billions in paper wealth. This early retreat was not merely technical profit-taking. It was an explicit acknowledgment that the incoming wave of available shares would find few willing buyers at premium prices.
Unmasking the Financial Deficit
The underlying numbers reveal why public markets are struggling to sustain these valuations. In the private venture capital ecosystem, metrics like user engagement and model parameters dictate pricing. Public markets demand revenue and path to profitability.
2025 Financial Performance (Yuan)
Company Revenue Adjusted Net Loss
Zhipu AI 724 Million 3.18 Billion
MiniMax 543 Million 1.75 Billion
The math is unforgiving. For every yuan of revenue generated by Zhipu, it burned through more than four yuan in losses. MiniMax followed a similar trajectory, spending heavily on the computational infrastructure required to train its proprietary multimodal architectures.
These losses are structural. Building foundational models requires massive clusters of advanced graphics processors, top-tier engineering talent, and immense electrical power. In mainland China, securing this hardware has become increasingly complex and expensive due to international trade restrictions. The operational costs are fixed, while revenues remain highly speculative and dependent on corporate clients who are themselves experimenting with tight budgets.
The Commodity Trap and Price Wars
The revenue outlook has grown grimmer due to a savage price war initiated by traditional technology giants. In the second quarter, companies like Alibaba, Tencent, and ByteDance slashed the prices of their large language model application programming interfaces by up to 90 percent.
This aggressive discounting forced the hands of pure-play startups. To remain competitive and keep developers inside their ecosystems, Zhipu and MiniMax had to match those cuts.
Price cuts hit these startups disproportionately. While Alibaba can subsidize its AI infrastructure through its profitable cloud and e-commerce divisions, standalone AI firms have no secondary engine to absorb the blow. Valuations for these enterprises are traditionally calculated as a multiple of sales. When price reductions depress future revenue growth projections, the entire valuation structure collapses.
The technological differentiation is also narrowing. MiniMax focused heavily on a native multimodal approach, developing language, voice, and video elements concurrently. Yet, institutional analysts have noted a lag in domestic benchmarks. Without a definitive, category-defining model that establishes clear technical superiority over cheap corporate alternatives, the product risks becoming a commodity.
Searching for Capital Anchors
The inclusion of both firms into the Hang Seng Tech Index in June provided a temporary buffer. Index tracking funds were legally obligated to purchase shares to match the benchmark composition, creating non-discretionary buying pressure. Morgan Stanley estimated that passive inflows provided a temporary floor, but passive capital cannot permanently offset a fundamental shift in sentiment among active institutional managers.
Faced with a cooling Hong Kong market, both companies are already looking toward domestic listings in mainland China. Zhipu has signaled intent to pursue an A-share listing to raise an additional 15 billion yuan, while MiniMax has explored listing on Shanghai’s STAR Market.
These moves suggest that the current capital pools are insufficient for the long-term survival of independent model developers. A secondary domestic listing can take a long time to clear regulatory review, and mainland public markets have become strict regarding unprofitable enterprises.
The next few weeks will determine if the current prices represent a stable baseline or the beginning of a deeper structural decline. Cornerstone investors who backed these listings are looking at a market environment completely altered from January. If large blocks of shares begin moving to the sell side, the resulting liquidity strain will redefine how the entire market values artificial intelligence.
The period of valuing AI companies based on scarcity is over. The sector has entered an execution phase where survival depends on unit economics rather than capital raises. Investors are discovering that the cost of intelligence is high, and the margins are thinner than anyone anticipated.
