Record Short Selling Hits Hong Kong's AI Giants: MiniMax Short Ratio Reaches 20%
The crowd sees a narrative of Chinese AI dominance. I see a ledger of leveraged liabilities.
On the Hong Kong exchange, the data is unambiguous. MiniMax, the once-celebrated AI unicorn, now carries a short interest of 20%. That is not a rounding error. That is a declaration of war by the smart money. Zhipu AI sits at roughly 6%, a level that would be alarming for a mature industrial stock but is merely the opening salvo for a pre-earnings tech name. This is not a dip. This is a structural repricing.
Let me be clear about what is happening. The market is not punishing these companies for a missed quarter. It is punishing them for a fundamental flaw in their business model. The era of the pure-play large language model (LLM) company, funded by narrative and valuation multiples, is ending. The market is now demanding proof of unit economics. And the market does not believe the proof is coming.
The catalyst was not a bad earnings report. It was a competitor's product launch. In July, Moonshot AI released Kimi K3. The market reaction was instantaneous and brutal. Zhipu AI's stock dropped 24%. MiniMax fell 18%. This is the market's version of a technical knockout. A single product release from a rival was enough to erase a quarter of a company's market capitalization. This tells you everything about the fragility of the perceived competitive moats in this sector.
A price drop of that magnitude in response to a competitor's launch signals that the market believes Kimi K3 represents a generational leap, not an incremental improvement. This is the market pricing in a paradigm shift in model architecture or training efficiency. Moonshot AI has not just gotten ahead. It has changed the game.
Now, let's look at the responses. Zhipu AI, through the lens of Jefferies' analysis, has positioned its GLM-5.3 model as offering "similar performance at 19% lower cost." This is the classic strategy of a follower. It is the admission of a company that knows it cannot win the race for intelligence and is pivoting to a race for efficiency. The logic is sound in a vacuum. In a market, it is a trap. A 19% cost advantage is not a structural moat; it is an engineering margin. It is achievable through quantization, speculative sampling, or optimized batching. It is a lead that can be erased in a single engineering sprint by a competitor with deeper pockets.
Hedgeye's assessment of MiniMax is even more damning. They describe the company as being neither the smartest nor the cheapest. In the business of AI models, this is the kiss of death. It is the "stuck in the middle" strategy that Michael Porter warned about in the 1980s. You have no pricing power because you are not the best, and you have no volume advantage because you are not the cheapest. You are squeezed from both sides.
The market mechanics reinforce this pessimism. Short sellers are not just increasing positions; they are doing so ahead of the interim earnings reports. MiniMax reports on August 26th. Zhipu AI reports on August 31st. This is a calculated bet. The shorts are betting that the financials will expose the weakness in the narrative. They are betting on a confirmation of the "stuck in the middle" thesis.
The price action tells a similar story. Both stocks are down over 50% from their peaks. Yet, Zhipu AI's stock is still 800% above its IPO price. This is the context that is often lost. We are not talking about distressed assets. We are talking about valuations that have been reset from the stratosphere to the merely elevated. The early investors have massive gains on paper. The question is whether they want to realize those gains. Given the lock-up expiry in July, with 25.68 million shares for Zhipu AI and 150 million for MiniMax, the supply overhang is enormous. The pressure to sell, even at prices 50% below the peak, is immense for investors sitting on 800% gains.
Southbound capital, the mainland money flowing through Stock Connect, has been buying the dip. They now hold approximately 12% of Zhipu AI and 8.1% of MiniMax. This is often cited as a sign of confidence. I view it with suspicion. This buying has not been able to prop up the price. The selling pressure is overwhelming the buying. This suggests that the Southbound flows are not institutional conviction but a form of value investing based on a thesis that the market does not share. They are catching a falling knife, not building a floor.
Now, let's address the contrarian angle. The narrative is overwhelmingly bearish, which is precisely when the risk of a short squeeze emerges. A 20% short interest is a powder keg. If the interim earnings report on August 26th is not a catastrophe, if there is even a hint of improving margins or better-than-feared revenue, the shorts will be forced to cover. The speed and violence of a short squeeze in a relatively thin market like this cannot be overstated. The risk/reward is asymmetric. The downside for the shorts is a quick 20% move up. The downside for the longs is a slow grind down. The smart play is not to join the crowd on either side but to recognize the timing catalyst.
The crowd sees art; I see a leveraged liability. The crowd sees a technology race; I see a capital allocation problem. The fundamental question is whether a pure-play LLM company can generate a sustainable profit in a market where the core technology is rapidly commoditizing. The data suggests the market is saying no. It is saying that the value accrues to the application layer or to the hyperscalers who can subsidize model development with other revenue streams.
Zhipu AI and MiniMax are caught in the middle. They lack the technical superiority of Moonshot AI and the cost structure of the mega-caps. They are being forced into a price war they cannot win and a quality war they are losing.
Based on my experience auditing trading systems and market structure, the next few weeks will define the trajectory of these names. The earnings reports are the catalyst. The short interest is the fuel. The question is whether the spark will ignite a rally or a rout.
Optionality is the shield against the black swan. In this case, the black swan is a positive earnings surprise that triggers a short squeeze. The white swan is a confirmation of the bearish thesis, leading to a further de-rating. The prudent move is to respect the data. The short interest is high. The valuation is in a correction phase. The market is telling you it does not believe in the business model. It is my job to listen.