Record short interest in MiniMax and Zhipu AI signals a market that has stopped believing the narrative.
MiniMax's short ratio just hit 20%. Let me say that again: one in five shares outstanding is now borrowed and sold by investors betting on a decline. In the modern era of Hong Kong-listed AI companies, that number is unprecedented. It's a number that tells a story the glossy IPO prospectuses never mentioned—and it's a story that's far from finished.
Here's the uncomfortable truth that 20% represents: it isn't just about MiniMax. It's about the entire "pure AI model company" species.
Zhipu AI's short interest hovers near 6%, which sounds modest until you realize the context: both companies are down over 50% from their peaks, yet Zhipu still trades at 800% above its IPO price. The shorts aren't just betting on a stock decline; they're betting on a narrative collapse. And as someone who has spent years decoding market narratives, I can tell you this is a familiar pattern.
The Context: When the Narrative Stops Compounding
The AI model industry in China has been built on a narrative of endless growth. "We're doubling revenue!" "Our model is smarter!" "We're China's answer to OpenAI!" These are the whispers that pushed valuations into the stratosphere. But in late 2025, the narrative has shifted to something far more concerning: "Where's the profit?"
The truth is, the market has moved from a phase of narrative-driven valuation to a phase of earnings-driven reality. The shorts are simply the first to act on the realization that the Emperor is wearing very expensive, very flashy clothes with no pockets.
My 2020 experience with DeFi's "Yield Trap" taught me a crucial lesson: when a sector's core economic loop is driven by capital inflows rather than organic revenue, the crash is rarely about the technology—it's about the narrative decay. When new investors stop believing in the growth story, the only metric that matters is the burn rate. And in the case of pure AI model companies, the burn rate is a furnace.
The Core Insight: The Unspoken "Stuck in the Middle" Problem
The most revealing data point in this entire story isn't the short ratio itself—it's the analysts' assessment of the companies' positioning. Hedgeye's commentary on MiniMax is brutal: "Neither the smartest nor the cheapest." This is a death sentence in a market where differentiation is the only lifeline.
Let's dissect this with a "narrative decay" framework. The "pure AI" story has three traditional pillars: technology, cost, and market positioning.
- Technology: The market has seen Kimi K3 emerge from Moonshot AI, a model that caused a 24% single-day drop in Zhipu and an 18% drop in MiniMax when it launched. That's not a minor scare—that's the market pricing in a generational gap. If K3 was a simple incremental improvement, the market wouldn't have reacted so violently. The market is telling you that a true "frontier" model exists in China, and it doesn't come from the shorted names.
- Cost: Zhipu's strategy is "performance similar to K3, but 19% lower cost per task." That sounds smart. But it's a follower's strategy. It admits you can't win on pure intelligence, so you're competing on economics. The problem is that cost advantages in AI are engineering, not science. Any optimization you discover can be copied by a better-funded competitor within months. If Moonshot AI's next move is a 25% cost cut, Zhipu's entire positioning gets evaporated overnight.
- Market: MiniMax's position is the worst. In a market where you need to be either the "smartest" or the "cheapest," being neither is a trap. This is the classic "stuck in the middle" business school dilemma, and it's fatal in a sector with high fixed costs and brutal price competition.
The market isn't just selling a stock; it's selling the narrative that "a pure model company without a moat is a commodity." And commodities are priced for their margins, not their potential.
The Core Problem: Unit Economics Can't Beat the Price War
The technical analysis is compelling, but the unit economics are where the bones are buried. The shorts are betting on the financial statements, and they have a rational basis.
The price war problem: The API token price has been dropping steadily. The race to the bottom in pricing is squeezing gross margins. If you're a "pure model" company, your revenue is directly tied to API calls or enterprise deals. You need to either increase volume at a faster rate than prices drop, or you need to find a way to differentiate your product to justify a premium. If you can't do either, you're just bleeding cash on a treadmill.
The gross margin mystery: We have no direct data from these companies on their gross margins. But given the aggressive pricing in the market and the cost of serving large models, the analyst community is increasingly concerned about the gap. When a company's technical value proposition is "we're cheaper," it's inherently margin-crushing.
The lock-up clock: July's lock-up expiry released a flood of shares—25.68 million shares for Zhipu and 150 million for MiniMax, totaling roughly $11.5 billion at the time. This creates a classic supply-overhang scenario. The shorts know that early investors have massive unrealized profits (Zhipu is still 800% above IPO price), so the incentive to sell is strong. When you have a sell-side supply surge and a buyer-side narrative crisis, the price can only fall.
The Southbound capital puzzle: This is the most fascinating detail. Mainland investors via Southbound Stock Connect have been buying. Zhipu's holding is around 12%, MiniMax around 8.1%. Yet the price still falls. This is a classic "catching a falling knife" scenario. It shows that the sell-side pressure is far more dominant than the buy-side support. And when a "value" buyer is the only one holding the bag, the floor is often lower than they think.
The Contrarian Angle: The Empty Sky Above the Shorts
Let me be the devil's advocate for a moment, because that's what a narrative hunter does. The short interest of 20% is a powder keg for a short squeeze.
The market is positioned for the half-year earnings report. MiniMax reports on August 26, Zhipu on August 31. If these reports show anything better than the consensus bottom, the shorts will have to cover their positions quickly. A 20% short interest can be a ticking time bomb. If the stock moves up by even 10%, the short squeeze forces rapid buying, which accelerates the price rise.
I've seen this pattern in my audit of the crypto markets. When a project is overleveraged in one direction, the market loves to create a "liquidity shock" to punish the consensus. The shorts are all on one side, and the market can move fast to make them pay for their arrogance.
But let's not overstate it. The short squeeze is a trading event, not an investment thesis. The structural problem remains: in a market where a stronger model exists, being the "cheap copy" is not a sustainable business model.
The Decay of the "Pure Play" Narrative
What we're seeing in Hong Kong is the "narrative decay" of the pure-play model company. The story used to be "AI is the future, so these companies are valuable." The new story is "AI is the future, but the future belongs to the platforms that can integrate it, not the companies that just rent out the models."
The industry is moving toward a "model-as-infrastructure" play. The question is no longer "who has the best model?" but "who can build the best business on top of it?"
This is a hard lesson I've learned from watching the DeFi sector collapse in 2020. The "yield farming" protocols offered high APYs, but the value wasn't being created by the protocol itself—it was just a flow of tokens from the treasury. When the token price fell, the APY fell, and the narrative collapsed. The same logic applies here: if the model isn't generating organic revenue at a margin that covers the cost of capital, the whole thing is just a subsidized growth play.
The shorts are not just betting on a stock decline; they're betting that the "pure model" company is a structurally flawed business model. They're betting that the cost of training and inference is too high to be offset by API revenue, and that the price war is only going to accelerate. And for a company like MiniMax, which lacks a clear edge, they might be right.
The Bottom Line: The Hunt for the Next Narrative
I hunt for the story the data refuses to tell. And the data here tells a story of a market that has lost faith in a certain kind of company. But the market is also a forward-looking mechanism. The shorts have their thesis, but the thesis is a snapshot, not a permanent state.
The key is to watch the "new narrative" that's forming. The one that says "a pure AI company is a commodity business, and the value lies in the applications built on top of it." If that narrative wins, the pure model companies will be forced to pivot, either into specialized verticals or into becoming a platform for others to build on.
The question I'm asking myself is: What is the endgame for the pure play model companies? Will they survive as independent entities, or will they be absorbed by the larger tech conglomerates? Will they find a way to monetize their models in a way that justifies their current (or even their reduced) valuations? Or will they become the cautionary tale of the 2025 AI bubble?
Decode the script before you bet on the actor. The shorts are decoding a story that looks like a tragedy. The bulls are hoping for a plot twist. The truth is likely somewhere in the middle: the company will survive, but the "pure model" label will become a relic of the past.
The narrative has decayed. The only question is: what's the next act?
