The Short-Seller's Consensus: Dissecting the Market's Verdict on MiniMax and Zhipu AI

Look at the short interest on MiniMax. Twenty percent. That is not a hedge; that is a verdict. In my years tracing the gas trails back to the root cause of market dislocations, I have rarely seen such a concentrated bet against a company's survival. The market is not just betting on a price drop; it is betting on a fundamental failure of the business model. The code does not lie, but the auditor must dig. Here, the data is screaming from the order books of the Hong Kong exchange. The narrative is familiar to anyone who has watched the AI arms race. Zhipu AI and MiniMax, two of China's "AI Dragon" quartet, went public to much fanfare. Now, they face a record short-selling wave, a massive share unlock, and a stock price that has been cut in half from its peak. The trigger for this latest slide was the release of Kimi K3 by Moonshot AI in July. The market's reaction was not to reward innovation, but to punish the perceived intensification of a war of attrition. Zhipu's stock fell 24%, MiniMax fell 18%. The market is shifting the consensus layer, one block at a time, and the new consensus is that being a pure-play large language model (LLM) company is a losing game. To understand this, we must shift our focus from the model weights to the market mechanics. The core issue is not the technology; it is the unit economics. Jefferies, a broker, noted that Zhipu's GLM-5.3 model performs on par with Kimi K3 but at a 19% lower cost per task. This is a technical achievement. It demonstrates engineering efficiency and cost control. Yet, the stock did not rally. Why? Because in the current market phase, a 19% cost advantage is not a moat; it is a survival tactic. It is the difference between bleeding out in six months versus nine. The market is pricing in a future where these models become commoditized, and the only differentiator is price, which leads to a race to the bottom. This brings us to the core of the bear thesis, articulated by firms like Hedgeye. They argue that Zhipu is under pressure from a price war, which caps its ability to raise prices and expand margins. The question is no longer "who has the smartest model?" but "who can make money at scale?" The answer, for the pure-play companies, is increasingly "no one." The market has shifted from a "story-driven" to a "data-driven" regime. The story was "AGI is coming, buy the picks and shovels." The data is "revenue growth is slowing, losses are mounting, and the path to profitability is a mirage." Let's dissect the numbers. The share unlock is a critical piece of the puzzle. In July, after the IPO lock-up period expired, Zhipu had 25.68 million shares and MiniMax had 150 million shares become eligible for sale. At the time, this represented a combined market value of approximately $11.5 billion. This is a supply shock. It means early investors, venture capitalists, and employees have a strong incentive to cash out. The market is not just absorbing new supply; it is absorbing the signal that insiders do not believe the current valuation is sustainable. This is a classic "smart money" exit, and the retail and Southbound capital are left to catch the falling knife. Southbound capital, which is money flowing from mainland China into Hong Kong-listed stocks, has been a buyer. Zhipu's Southbound holding is around 12%, and MiniMax's is around 8.1%. This is often cited as a sign of long-term confidence. But I see it differently. It is a potential "value trap." These investors are buying based on a narrative of national AI champions and long-term growth. They are ignoring the technical reality of the balance sheet. They are buying the dip, but the dip is a waterfall. The data suggests that the selling pressure from the unlock and the short sellers is far outweighing the buying pressure from the Southbound bulls. The competitive landscape is even more unforgiving. MiniMax and Zhipu are stuck in the "middle ground." They are not the smartest (that title is contested by DeepSeek and others), and they are not the cheapest (that title is also contested). They lack the ecosystem advantages of the tech giants. Alibaba has Qwen, ByteDance has Doubao, and Tencent has Hunyuan. These companies have massive distribution networks, vast data resources, and the financial firepower to subsidize their AI efforts indefinitely. They can afford to run their model divisions at a loss because they are part of a larger, profitable enterprise. MiniMax and Zhipu do not have that luxury. They are standalone public companies, and their R&D budgets are directly tied to their stock price. A falling stock price means less capital for the next model, which means falling behind in the arms race, which means a further falling stock price. It is a vicious cycle. Now, let's consider the contrarian angle. The market's pessimism might be overdone. The short interest is at a record high, which historically can be a contrarian indicator. If the upcoming interim earnings reports, due on August 26th for MiniMax and August 31st for Zhipu, show better-than-expected revenue growth or a clear path to profitability, we could see a massive short squeeze. The market is pricing in a disaster. If the companies deliver anything less than a disaster, the stock could rally violently. This is the "sell the rumor, buy the news" dynamic, but in reverse. The rumor is bankruptcy; the news might be merely "bad, but not fatal." Furthermore, the market is ignoring the potential for these companies to be acquisition targets. At their current valuations, they are attractive assets for larger tech players or private equity firms. A company with a strong model, a decent user base, and a beaten-down stock price is a prime candidate for consolidation. The AI industry is heading for a shakeout, and the strong will absorb the weak. Zhipu and MiniMax could be the acquired, not the acquirers. This is a real possibility that the short sellers are not pricing in. Another blind spot is the assumption that the price war is a zero-sum game. The cost of inference is falling, but the demand for AI services is growing exponentially. The "Kimi K3" effect might not be a sign of a saturated market, but rather a sign of a market that is about to explode. As the cost of models drops, the barrier to entry for AI applications falls. This could lead to a surge in demand for API calls, which would benefit all model providers, including MiniMax and Zhipu. The short sellers are focused on the supply side (too many models) and ignoring the demand side (too many use cases). The market is also ignoring the potential for these companies to pivot. They are not just model providers; they are also application companies. MiniMax has its own consumer app, Talkie, which has a significant user base. Zhipu has a suite of enterprise solutions. If they can successfully monetize these applications, they can reduce their dependence on the commoditized API business. The market is treating them as pure-play model companies, but they have the potential to be more. The code does not lie, but the auditor must dig. The market is looking at the current income statement and ignoring the potential of the future balance sheet. In the chaos of a crash, the data remains silent. But the data is clear: the market has lost faith in the pure-play LLM business model. The question is whether this faith can be restored. The upcoming earnings reports are the first test. If they show a path to profitability, the shorts will be squeezed. If they show a widening loss, the shorts will be vindicated. The market is a consensus mechanism, and right now, it is reaching a consensus that these companies are overvalued. The question is whether that consensus is correct. Based on my analysis of the market mechanics, the competitive landscape, and the financial pressures, I believe the market is right to be skeptical, but it may be too pessimistic in the short term. The next few weeks will be a fascinating test of the market's conviction. This is not a story about technology. It is a story about capital allocation. The market is shifting its capital from companies that promise the future to companies that deliver the present. MiniMax and Zhipu are caught in the middle of this shift. They are the canaries in the coal mine for the entire AI industry. If they cannot find a way to make money, it will send a chilling signal to every other AI startup in the world. The takeaway is not to short these stocks, but to understand the fundamental shift in market sentiment. The era of "growth at all costs" is over. The era of "profitability at all costs" has begun. The market is shifting the consensus layer, one block at a time, and the new block is a demand for earnings. The question is, can these companies mine it?