The Ledger of Fear: How Record Short Interest on AI Unicorns Is Rewriting the Valuation Matrix

The ledger shows a curious divergence. While the narrative around artificial intelligence remains firmly bullish, the order books for two of China's most prominent AI pure-plays tell a different story. Over the past thirty days, short interest in MiniMax has surged to a record 20% of float. Zhihu AI, trading at a valuation still 800% above its IPO price, has seen its share price sliced in half from its peak. This is not a blip. This is a structural repricing of what the market believes a large language model company is actually worth. The narrative says these are foundational assets. The ledger says they are becoming commodities. I have spent nearly two decades tracing capital flows across digital assets, and the pattern here is unmistakable. When the gap between narrative and on-chain (or in this case, on-exchange) reality widens, the correction is rarely gentle. We are witnessing the crypto-native dynamics of a liquidity crunch and a narrative unwind playing out in the equities of the AI sector. The only difference is the ticker symbol. Context is necessary here. MiniMax and Zhihu AI are not fringe players. They are part of the so-called "AI Dragon" cohort, companies that emerged from the 2023-2024 funding boom with billion-dollar valuations and grand ambitions. They went public in Hong Kong to much fanfare, riding a wave of optimism about the democratization of artificial intelligence. The promise was simple: they would build the foundational models that power the next generation of applications, and in doing so, capture the value of the entire AI stack. The market bought that story. For a while. Then the data started to shift. In July, the release of Kimi K3, a rival model, was expected to be a catalyst. Instead, it became a sell signal. Zhihu AI's stock dropped 24% in the weeks following the release. MiniMax fell 18%. Why would a competitor's product launch decimate the stock price of two other companies? The answer lies in the incentive structure that the market has finally begun to dissect. This brings us to the core of the analysis. The prevailing assumption in the AI bull market was that technical leadership translated directly into pricing power and, ultimately, profitability. The on-chain evidence from the equity market suggests that assumption is flawed. Let's break down the evidence chain, piece by piece. First, there is the issue of technological parity. Jefferies, in a research note, pointed out that Zhihu AI's GLM-5.3 model performs at a level comparable to Kimi K3, but at a cost that is 19% lower per task. In a rational market, this would be a positive catalyst for Zhihu. It demonstrates engineering efficiency and cost leadership. Yet the stock fell. This tells us that the market is no longer pricing in incremental performance improvements. The marginal utility of a 1% accuracy gain is approaching zero. The market is looking for something else: a clear, defensible path to sustained profitability. Second, the cost advantage cuts both ways. While a 19% cost advantage gives Zhihu ammunition in a price war, it also signals that the price war is the battlefield. Hedgeye, a research firm known for its short calls, has explicitly stated that Zhihu is under pricing pressure that limits its ability to raise prices and expand margins. This is the core of the bear thesis. When you compete on cost, you are competing in a race to the bottom. The market is not rewarding the lower price; it is punishing the implied margin compression. It is the classic crypto dilemma of "yield or die" translated into the AI sector. You can chase market share by undercutting competitors, but if your yield vectors don't point toward profitability, the capital will eventually flee. Third, we have the supply-side shock. In July, the lock-up periods for both companies expired. Zhihu AI saw 25.68 million shares unlock. MiniMax saw 150 million shares unlock. Combined, this represented a potential sell pressure of approximately $11.5 billion. This is not a technicality; it is a fundamental shift in supply. Early investors, who got in at a fraction of the current price, have a strong incentive to realize their gains. The ledger shows that they have been doing exactly that. This overhang is not a one-day event. It is a persistent weight on the stock price, a constant reminder that insiders do not see the same value that the market initially did. This leads to a critical question: who is buying? The data shows that Southbound capital—funds flowing from mainland China into Hong Kong—has been a consistent buyer. Zhihu AI is now 12% owned by Southbound investors, and MiniMax is 8.1%. Yet, despite this steady inflow, the prices have fallen. This suggests that the buying pressure is being overwhelmed by the selling pressure. It is a tug-of-war, and the sellers are winning. This is a classic value trap signal. Money is flowing in based on a long-term thesis about AI's importance to China's technological sovereignty, but the short-term reality of earnings and cash burn is dragging the price down. The core insight, however, goes beyond these company-specific factors. The record short interest is a bet on the failure of the entire pure-play LLM business model. The market is effectively saying that a company whose only asset is a model—no matter how good—will not be able to generate sufficient returns to justify its valuation. This is a profound shift from the earlier days of the AI boom, where any company with a decent model could raise capital at ever-higher valuations. Let's map the yield vectors here. In traditional finance, a company creates value by generating a return on invested capital (ROIC) that exceeds its cost of capital. For most pure-play AI companies, this equation is inverted. The cost of capital is high, and the ROIC is deeply negative. The primary driver of value is no longer the model itself, but the network effects, the distribution channels, and the proprietary data that can be wrapped around the model. This is why giants like Alibaba and ByteDance are seen as safer bets. They have the ecosystem to subsidize model development and a distribution network to monetize it at scale. MiniMax and Zhihu AI lack this moat. They are caught in the middle. Not smart enough to command a premium, not cheap enough to win a race to the bottom. Based on my experience auditing the 2020 DeFi Summer, I see a direct correlation here. During that period, we saw hundreds of protocols launch with high yields and grand promises. The ones that survived were not the ones with the most sophisticated smart contracts. They were the ones with the most sustainable tokenomics—a clear plan for how value would accrue to the token holders. The same logic applies here. The market is asking for a clear path to value accrual. They want to see a transition from a cost center (R&D) to a profit center (products). The companies that fail to articulate this transition will be left behind. This brings us to the contrarian angle. The popular narrative is that this short interest is a clear negative signal, a death knell for these companies. I disagree with the absoluteness of this conclusion. The market is pricing in a high probability of failure, but it may be overcorrecting. There is a real possibility that the August earnings reports could deliver a positive surprise. Here is the blind spot the short sellers might be missing: the cost of compute is falling. The ledger shows that the cost per unit of intelligence is dropping at a rate faster than Moore's Law. If Zhihu AI can maintain its cost advantage while the absolute cost of compute plummets, its unit economics could improve dramatically. The 19% cost advantage is a point-in-time metric. The trend line is what matters. If both companies can harness the falling cost curve, they could achieve profitability faster than the market expects. Furthermore, the correlation between short interest and future returns is not linear. Record short interest is often a contrarian indicator. It means that most of the selling pressure has already been expressed. The marginal seller has already sold. The remaining shareholders are the true believers, the Southbound funds with a long-term mandate. If the earnings report is not a disaster, the short squeeze potential is enormous. A 20% short interest ratio is a powder keg. There is also the possibility of consolidation. At current valuations, these companies are becoming attractive takeover targets for larger tech firms or even private equity. A company trading at a fraction of its peak valuation, with a world-class model and a decent customer base, has inherent strategic value. An acquisition would instantly re-rate the stock and render the short thesis moot. This is a tail risk for the short sellers and a potential catalyst for the longs. However, we must also examine the counter-thesis to my contrarian view. The most significant risk is that the August earnings reveal a deepening of losses. The AI arms race requires constant, massive capital expenditure. If revenue growth is not accelerating fast enough to offset the R&D burn, the losses will widen. The market is not forgiving of widening losses in a high-interest-rate environment. The price could fall further, and the shorts would be vindicated. We also need to consider the macro environment. Global capital is becoming more risk-averse. Money is flowing to safety. Unprofitable, high-growth companies are being de-rated across the board, not just in the AI sector. This is a systemic headwind that no company can escape. The short sellers are simply riding a macro wave, and their timing has been impeccable. What is the market telling us about the next six months? The key signal will be the revenue quality. We need to look beyond the top-line number and examine the composition of that revenue. How much is coming from enterprise contracts with long-term commitments versus short-term API calls? How much is coming from government subsidies or strategic investors? The market will reward high-quality, recurring revenue. It will punish one-off deals and unsustainable promotional pricing. The second signal is the management guidance. What are they saying about the future? Are they guiding for continued heavy investment in frontier models, or are they signaling a shift toward more profitable, application-focused work? The latter would be a positive surprise. It would indicate that the management is listening to the market and adapting their strategy to the new reality. The former would be a confirmation of the bear thesis. Here is what I will be watching for in the next few weeks, from a purely on-chain perspective. First, the initial reaction to the earnings release. A pop and a fade is a bearish signal. A steady climb on volume is a bullish signal. Second, I will be watching the Hong Kong exchange data for block trades. If we see large blocks of shares changing hands at a discount, it means the overhang is being distributed. If the selling is done via open market sales, it is a more gradual, but persistent, pressure. Third, I will be monitoring the rhetoric from the short sellers. If Hedgeye and others double down with additional research reports, the pressure will continue. If they go quiet, it might mean they are covering their positions and taking profits. Do not mistake volatility for signal. The stock price will swing wildly in the coming weeks. The signal will be in the data: the quality of earnings, the behavior of insiders, and the response of the short sellers. The ledger does not lie, only the narrative does. And the narrative is currently dominated by fear. Let's take a step back and consider the broader implication for the blockchain and AI intersection. I have been analyzing the convergence of these two technologies since 2026, tracking autonomous agents interacting with DeFi protocols. One of the key findings from that research is that AI agents are far more rational than human traders. They do not get caught up in narratives. They respond to data. The current repricing of AI equities is, in a sense, a very rational, agent-like response to the underlying fundamentals. It is a correction of a narrative-driven excess, a return to fundamentals. The lesson for the crypto market is clear. The same fate awaits projects that rely purely on narrative without a sustainable business model. The era of the whitepaper is over. The era of the income statement has begun. This is the most important takeaway from the current situation. The market is no longer paying for potential. It is paying for proof. Proof of users, proof of revenue, proof of a path to profitability. In the long run, this is a healthy correction. It will separate the wheat from the chaff. It will force companies to focus on building real businesses rather than chasing benchmarks. It will drive capital toward the most efficient operators. It will, ultimately, lead to a stronger and more sustainable AI ecosystem. But in the short term, it is painful. The blood is in the streets, and the shorts are licking their wounds. Or are they? The question is whether this is the beginning of the end for these two companies, or the start of a new chapter. The answer will be written in the next earnings release. The market is a harsh editor. It does not care about good intentions. It only cares about results. And the results are coming. I will be mapping the yield vectors before the August peak. I will be tracing the capital flows through the Hong Kong exchange. I will be watching the behavior of the early investors. And I will be reading the hashes of the earnings reports, line by line, to find the signal in the noise. The truth is out there, and it is always found in the data. The narrative will follow. It always does.