Let's look at the data first. The numbers are not ambiguous. MiniMax, listed in Hong Kong, now carries a short interest ratio of 20%. Zhipu AI sits at approximately 6%. These are not normal levels. A 20% short ratio is a declaration of war by the market. It signals a collective, data-driven consensus that these companies' current valuations are built on narrative, not on fundamentals.
Verify this: the market has already voted with its wallet. Zhipu AI and MiniMax have both fallen more than 50% from their peak prices. Yet, Zhipu AI's stock still trades 800% above its IPO price. The disconnect is staggering. This is not a dip; this is a repricing. The market is moving from the 'story stock' phase to the 'audit' phase.
Check the chain, not the hype. The on-chain and market data are telling a clear story about the viability of pure-play large language model companies in China.
Context: The Post-IPO Reality Check
The context here is the brutal transition from private-market exuberance to public-market scrutiny. For years, Chinese AI labs raised massive capital rounds at ever-increasing valuations. The narrative was simple: AI is the future, and these are the leaders. The IPO window opened in 2025, and the market finally got a chance to price these companies with real numbers. The result has been a reality check.
The short sellers are not irrational actors. They are taking a position based on a clear thesis: these companies are burning cash in a hyper-competitive market with no clear path to profitability. The trigger for the latest surge in short interest was the release of Kimi K3 by Moonshot AI in July. The market reaction was immediate and violent. Zhipu AI dropped 24%, and MiniMax fell 18% in the immediate aftermath. This was the market's way of saying that Kimi K3 was not an incremental update; it was a generational leap that changed the competitive landscape.
Jefferies' assessment of Zhipu AI's new GLM-5.3 model as offering 'similar performance at 19% lower cost' is a classic follower strategy. It is an admission, whether intentional or not, that Zhipu AI cannot compete on pure capability. The company is pivoting to a cost-efficiency narrative. Hedgeye's assessment of MiniMax is even more damning: it is 'neither the smartest nor the cheapest.' This is the 'stuck in the middle' trap. In a market where you cannot compete on product leadership or price, you have no moat.
Core: The Data Integrity Check and the Short Thesis
Let's dissect the core thesis of the short sellers. It is not just about valuation; it is about the fundamental unit economics of these businesses.
1. The Pricing Power Paradox
From my audit experience in 2017, when I reviewed early-stage tokenomics, I learned that a project's sustainability hinges on its ability to generate value. In the AI model market, the value is being commoditized. The market is in a price war. API costs are falling, and companies are forced to match prices to retain customers. Hedgeye notes that Zhipu AI is under 'price war pressure,' which limits its ability to raise prices and improve margins. This is a 'growth without profit' risk. Revenue might be increasing, but the cost of acquiring that revenue is high, and the ability to convert it into profit is structurally limited.
2. The 'Middle Trap' and the Lack of Differentiation
MiniMax's position is analytically the most dangerous. A company needs either a technological lead (to command a premium) or a cost advantage (to compete on price). MiniMax has neither. The data does not lie: if you are not the best and you are not the cheapest, your commercial space is squeezed from both ends. The short sellers are betting that this positioning is not a temporary phase but a structural weakness.
3. The Unlock Overhang
The data on share unlocks is critical. In July, the lock-up periods ended. Zhipu AI had 25.68 million shares unlock, and MiniMax had 150 million shares unlock. At then-current prices, this represented a combined supply of approximately $11.5 billion. This is a massive overhang. Early investors, who are sitting on enormous paper profits (Zhipu AI is still up 800% from IPO), have a strong incentive to sell. The data suggests that the selling pressure will continue to suppress the stock price.
4. The 'Southbound' Flow Illusion
There is a popular narrative that mainland Chinese investors are 'supporting' these stocks through the Stock Connect. The data shows that southbound funds hold about 12% of Zhipu AI and 8.1% of MiniMax. However, this buying has failed to lift the stock price. This is a critical data point. It proves that the sell-side pressure from short sellers and unlock holders is far exceeding the buy-side pressure. The 'dip-buying' narrative is not holding up against the fundamental selling pressure. This is not support; it is a potential 'catch the falling knife' scenario.
Contrarian: The Squeeze Risk and The Cost Narrative
Now, let me apply my structural skepticism. The data is bearish, but the trade is crowded. A 20% short ratio is not just a bearish signal; it is also a measure of potential volatility. The risk here is a short squeeze. If the upcoming interim earnings reports (MiniMax on August 26, Zhipu AI on August 31) contain any positive surprise, the short sellers will be forced to cover their positions. This forced buying can trigger a rapid, sharp price increase. The contrarian angle is that the market may be pricing in a complete collapse, leaving no room for error. The data suggests that the risk/reward for new short positions is now poor.
Furthermore, the '19% lower cost' claim by Zhipu AI needs scrutiny. Rigour over rumour. My experience with on-chain data and yield aggregation taught me to look at the methodology. Is this cost advantage derived from a fundamental architecture innovation, or is it from engineering optimizations like model quantization, speculative sampling, and batch processing? If it is the latter, it is a temporary advantage. It is an engineering problem, not a science problem. Moonshot AI can easily replicate these optimizations, eroding Zhipu AI's only stated advantage. The market is skeptical of this cost narrative, and the data supports that skepticism: the stock did not rally on the GLM-5.3 announcement.
The fundamental question is whether the 'pure-play' large model company is a viable business model. The short sellers are betting that it is not. They are betting that the technology will continue to commoditize, prices will continue to fall, and only companies with massive scale or a unique, defensible application layer will survive. The on-chain data from DeFi protocols in the 2020 cycle showed a similar pattern: yield follows logic, not luck. The same principle applies here. The logic of the AI market says that pure model providers are in a race to the bottom.
Takeaway: The Next Signal
The interim reports are the catalyst. The data to watch is not just the revenue growth but the gross margin and the loss per token. The market wants to see if the 'cost advantage' translates into a margin advantage. If Zhipu AI can show that its 19% cost advantage is leading to a 19% margin improvement, the short thesis is weakened. If the reports show that revenue is growing but losses are widening, the short sellers will be validated, and the stock will likely see another leg down.
The signal to track is the short ratio after the earnings announcement. If the ratio drops, it means the bears are covering and the risk is passing. If the ratio increases, it means the market sees the earnings as confirmation of their thesis. The data is clear. The narrative is dead. The era of the 'story stock' is over. We are now in the era of the audit. The question is, can these companies pass the test?