The most honest signal in a market is often the one it refuses to speak. Over the past weeks, the Hong Kong stock exchange has been humming with a quiet, persistent rhythm of doubt. Short interest in two of China's most prominent AI 'giants'—MiniMax and Zhipu AI—has climbed to historic levels. MiniMax's short ratio now sits at a remarkable 20%, a figure so elevated it stops being a trading metric and becomes a statement of collective belief. For Zhipu, the ratio hovers around 6%, a level that, while lower, still speaks of a consensus that is not about technology, but about the void between the promise and the product. This is not a story about numbers. It is a story about a covenant broken, or perhaps, a covenant that was never really made.
The market's silence here is deafening. Open source is not a license; it is a covenant. It is the unspoken agreement between a builder and a user, that the code will be honest, that the data will be truthful, that the value is not in the token but in the trust. When the market votes with such a heavy hand against these AI firms, it is not just voting against a balance sheet. It is voting against a fundamental question of integrity: are these companies weavers of conviction, or just weavers of tokens?
To understand this, we must first look at the genesis of the sell-off. It did not come from a regulatory crackdown, nor from a sudden loss of technical talent. It came from a single event: the release of Kimi K3. On the day of its release, the stock prices of both Zhipu AI and MiniMax dropped by approximately 24% and 18%, respectively. This market reaction was not a whisper; it was a gavel strike. It was the market's judgment that the Kimi K3 was not an incremental improvement but a generational leap, a leap that re-categorized the entire competitive landscape.
In my experience, auditing dozens of protocols and whitepapers over the years, I've learned that the market is often slow to price in true innovation, but it is lightning-fast to punish the absence of it. The Kimi K3 release was not just about a better model. It was about the opening of a gap in the technical fabric. The short sellers, who are often the most cynical of the market participants, are often the first to see this gap. They don't see a model; they see a cost structure. They see the future.
For Zhipu, the strategy in the face of this was a retreat to the familiar. Jefferies noted that their new model, GLM-5.3, positions itself as having "similar performance" to Kimi K3, but at a 19% lower cost per task. On the surface, this seems like a pragmatic, even brilliant, move. In a world of brutal price wars, being the cost leader is a position of power. But my years in the DeFi ecosystem, watching liquidity farming schemes, have taught me to be wary of the simple narrative of the subsidy. The 19% cost advantage is often not a result of a new architectural truth, but of engineering—inference optimization, quantization, batch processing, or speculative decoding. It is a synthetic, fabricated advantage, not a natural one. This kind of advantage is not a moat; it is a sandcastle that the tide of a competitor's next release will wash away. The market knows this. The market is not stupid. That is why the stock did not rebound on this news. It sees through the optimization, and it knows that the underlying asset—the model itself—is not the most intelligent.
For MiniMax, the situation is even more stark. Hedgeye, a well-known research firm, put it in the most brutal of terms: MiniMax is "neither the smartest nor the cheapest." In a market where the future is defined by either a superior intelligence or a superior cost, MiniMax has planted its flag in the middle of the void. This is a fatal position, a "stuck in the middle" scenario. In our technical world, we often talk about the power of the niche. Nurture the niche, and the forest will follow. But MiniMax is not a niche. It is a broad, generic, undifferentiated product. It is a quote from a non-poet. It has no edge, no values, and no reason to be chosen over the more intelligent or the more affordable.
As we move into the core of this analysis, we must ask: what does the short seller see that the retail investor does not? They see the unit economics. They see the numbers that the press release hides. We do not write code; we weave conviction. But the short sellers are looking at the ledger, and the ledger is silent. It does not speak of the innovation; it speaks of the cost of the compute, the cost of the talent, and the cost of the token.
Let's dive into the specific data. This is not a market brief for the faint of heart. This is a data-driven analysis of the core flaw.
The Cost of the Token (or the Lack Thereof)
The fundamental flaw lies in the business model of the "pure play" large language model (LLM) company. The narrative is simple: build a giant model, and the world will come to your API. But the reality is that these models are not a scarce resource. They are a commodity, and commodities suffer from price wars. The price of API calls has been falling for months, a positive for consumers, but a knife to the throat of these companies. If the model is not the best in the market, you have to compete on price. But if you have no cost advantage, you are just losing money with every query.
Here is the first-person technical signal: In my years of auditing open-source code, I've learned that the value of a protocol is not in the code itself, but in the active state of the community. The same applies here. The value of a model is not in the weight, but in the adoption. When you look at the short interest, you are looking at the community's belief. The short interest on MiniMax is a direct reflection of the market's belief that their community is not growing, that their cost is not falling, and that their technology is not leaping. The short sellers are not just betting on a price drop; they are betting on a failure to reach the point of the total available market.
The Institutional Shift
But there is a deeper, more dangerous shift. It is the shift in the valuation logic. The market is moving from a "story-telling" valuation (based on the potential of AI) to a "net present value" (based on the reality of the financials). This is the transition from the "market cap to the sky" to the "profit and loss" account. The shorts are based on the second, and the first is now being rejected. This is a market-wide shift, and it is the reason why the prices have been cut in half from their peaks.
For Zhipu AI, the stock price is still 800% above its initial public offering (IPO) price. This is a staggering number, but it also reveals the enormous weight of the valuation that the market is trying to digest. The early investors are sitting on massive, but the entrance of the new investors is at a price that is not a function of the current reality but a function of a future that is now in doubt. The short sellers see this. They see the 800% number, and they see the 50% drop from the peak, and they know the window for the current valuations is closing.
Then there is the overhang of the lock-up. In July, the lock-up period ended for both companies. Zhipu and MiniMax had 25.68 million shares and 150 million shares unlocked, respectively, totaling around $11.5 billion at the time. The pressure of this massive supply of shares, combined with the short interest, is a double-edged sword. It is a pressure cooker. The shorts are betting that the share supply will overwhelm the buy-side, and that the valuation will be repriced downward.
The Southbound capital: A missing trust?
The mainland China-based capital, through the Stock Connect, has been buying the dip. Zhipu's ownership by Southbound investors is around 12%, and MiniMax is around 8.1%. This is a classic "catching the falling knife" scenario. But the market is not listening to the story of the retail. The stock price has not recovered despite this buying. This suggests that the sellers, or the short sellers, are much more aggressive than the buyers. It is a story of the lack of trust. The market is not trusting the narrative of the "value" as much as it is trusting the narrative of the "short." The short sellers are not just a signal; they are a prophecy.
The Contrarian Angle: The Short Squeeze is a Silent Niche
Now, let me offer a contrarian view. As I have often said, the void between tokens holds the true value. There is a possibility that the market is too pessimistic. A 20% short interest is extreme. It is a high-risk position, and it is vulnerable to a "short squeeze." If the upcoming financial reports (MiniMax on August 26, Zhipu on August 31) show better-than-expected gross margins, or a clear path to profitability, the shorts will be forced to cover their positions, leading to a sudden, violent upward price spike. In the world of DeFi, we have seen this dynamic play out with the "rekt" and the "pump." The short interest is a bet on failure. But the company can decide to succeed. The company can decide to be more than a "pure" model, to become a "solution."
But this is a hope, not a strategy. The most important thing is that the market is waiting for a proof of profitability. The market is waiting for a sign that the "silence in the ledger" is not a silence of a void, but the silence of a quiet preparation.
The Takeaway: The Value of the Void
I look at this, and I see a fundamental shift in the AI ecosystem. It is a shift from the "learning" phase to the "commercialization" phase, and the phase is brutal. The short sellers are not the enemy; they are the mirror. They are reflecting the truth that the companies have not yet found a way to turn their technology into a sustainable and ethical economic engine. The value is not in the short; the value is in the void between the promise and the reality. The void between the token and the value.
For the long-term, I see a consolidation coming. The AI market in China is moving from "a superpower and many" to a "survival of the fittest." The companies that will survive are the ones that are not just the smartest or the cheapest, but the ones that have a narrative, a covenant. They must be the ones that are not just building a model, but building a story that the user can trust. The ones that can listen to what the repository refuses to say.
As I look at the numbers, I am reminded that the deep, silent, and honest ones are the most important. The short sellers are speaking. The market is speaking. The question is: will the AI companies listen? The void between tokens holds the true value. Will they fill the void with a product, or will they fill it with a vision? The short sellers believe they will fill it with the loss. The future will tell us if they are right. The future will tell us if the silence in the ledger is the sound of a death or the sound of a birth.