Short-selling for Hong Kong-listed AI companies Zhipu and MiniMax plummeted 50% last week, signaling a shift in market sentiment. The retreat comes as investors re-evaluate the prospects of independent AI labs amid a broader sector reset.
TL;DR
- Short-selling for Zhipu and MiniMax halved in the week ending September 8, 2026.
- Investment banks are downgrading valuations for independent AI labs, favoring full-stack cloud platform giants.
- Zhipu's August ARR reached $1.6 billion, but growth was driven by new product launches and a low base.
What happened
Hong Kong-listed AI companies Zhipu (02513.HK) and MiniMax-W (00100.HK) experienced a significant decline in short-selling activity last week. According to Hong Kong Stock Exchange data, short-selling volumes for both companies fell by approximately 50% between September 4 and September 8, 2026.
This decline in short-selling activity coincides with a broader market shift away from independent AI labs towards full-stack cloud platform giants. International investment bank Jefferies recently cut Zhipu's cloud business price-to-sales multiple from 50x to 30x, reflecting this change in market sentiment.
Why it matters
The retreat of short sellers may signal a phased rebalancing after a round of full pricing-in, rather than a turn toward optimism about independent LLM companies' prospects.
For developers and startups, this shift highlights the growing importance of ecosystem integration and commercialization channels. Independent AI labs may face increased scrutiny and pressure to demonstrate sustainable growth and profitability.
Investors should take note of the market's transition from narrative-driven to fundamentals-verification. This shift underscores the need for careful evaluation of AI companies' computing-power supply, customer concentration, gross margin, and profitability.
Key facts
- Short-selling volumes for Zhipu fell from 1.3083 million shares to 672,900 shares between September 4 and September 8, 2026.
- Short-selling volumes for MiniMax fell from 1.8875 million shares to 857,100 shares in the same period.
- Jefferies cut Zhipu's cloud business price-to-sales multiple from 50x to 30x.
- Zhipu's August ARR reached $1.6 billion, with a year-end target of $2.4 billion.
- Zhipu's top two customers each contributed more than $250 million in ARR, accounting for at least 31% of total ARR.
- Zhipu's cloud gross margin for the first half of 2026 rose to 24.6% from 18.9% for full-year 2025.
- Jefferies expects Zhipu's second-half cloud gross margin to fall back to 20.2%.
- China's Ministry of Industry and Information Technology aims to reach 9,800 EFLOPS of intelligent computing capacity by 2030.
Context
The volatility in Hong Kong's LLM sector is not an isolated event. In China's A-share market, domestic GPU leader Moore Threads is likewise facing the impact of lock-up share expirations.
The market's shift towards full-stack cloud platform giants reflects the inherent self-sustaining advantages these companies possess in underlying computing power acquisition, deep ecosystem integration, commercialization channels, and balance-sheet health.
As the AI sector matures, the market is increasingly focused on fundamentals such as computing-power supply, customer concentration, gross margin, and profitability. This shift underscores the need for AI companies to demonstrate sustainable growth and profitability.